[
  {
    "group": "Native copy",
    "name": "SKILL.md",
    "path": "/Users/ayden/.hermes/skills/openclaw-imports/native-copy/SKILL.md",
    "text": "---\nname: native-copy\ndescription: Write native ad copy variants for DawnBands (or other brands). Use when creating Facebook/Instagram native ad copy, writing ad copy variants for A/B testing, generating headlines + body copy for Meta ads, or writing copy for native ads pointing to listicle advertorials. Triggers on \"write native ads\", \"ad copy variants\", \"write copies for [ad]\", \"native copy\", \"Meta ad copy\", \"5 variants for [ad]\".\n---\n\n# Native Copy Skill\n\nWrite high-converting native Facebook ad copy using Zakariah's framework, deep avatar research, and proven format rules.\n\n\n\n## IMAGE LANE ORDER (Fish 2026-08-25 — PERMANENT)\n\n1. RIP FIRST: pull_native_rips.py with queries shaped like real post titles (\"my son's room\",\n   \"light under the door\"), not staged ad-scene descriptions. Review every candidate by eye.\n2. A real rip that passes gates TAKES THE SLOT over any gen, even when a gen matches the\n   copy's scene better. Nativeness > relevance is absolute (Fish receipt: charging-caddy rip\n   over on-scene kitchen gen, 8-25).\n3. Gen fallback is NEVER bare-prompt. Pull a real Pinterest/web reference photo FIRST\n   (scrapling + curl, the boring amateur photo not the good one, demographic-matched when\n   people appear) and pass it as the style reference. Fake-UI/help-post recreations reference\n   a real screenshot of the actual UI.\n\n## DIRECTNESS — THE THREE-STRIKE ERROR (Fish 2026-09-01, PERMANENT)\n\nFish rejected three consecutive rewrites of the same passage. Each fixed the surface and kept\nthe literary move. Check all three before shipping any line.\n\n1. **ROUNDABOUT** — \"What I ordered was the first knock... the honest answer to what happens\n   when he leaves home is that I am the answer.\"\n2. **ANAPHORA** (only *sounds* direct) — \"To stop making the first knock. / To stop walking to\n   his door. / To stop hearing my own voice before seven.\" Three or more consecutive lines\n   opening with the same word or phrase is a poem. Fish: \"make sure people actually say that\n   ... or itll throw them off.\"\n3. **ROLE-ABSTRACTION** — \"right now I am still the plan.\" Fish: *\"i am still the plan is\n   still not what a mom says.\"* She never becomes an abstract noun.\n\n**GOOD:** \"I am tired of being his alarm clock. I do not want to start every day angry at my\nown kid. He leaves next year, and I will still be the one calling him every morning.\"\n\nBanned: `I am still the plan` · `I am the answer` · `that was the variable` · `never in that\nchannel` · `that is the whole thing` · `the part I could not fix was` · `What I ordered was X`\n\nInstead: name the concrete action she performs, or the feeling she has. Register anchor from\nreal buyers: \"Want to start the school year off with out the fighting.\" / \"Hoping it stops\nangry mornings I wake up in fight or flight mode.\"\n\nException: real idioms mothers use are fine — \"I am his alarm clock\", \"human snooze button\".\n\nAlso binding: grade 3-5, average sentence under 14 words, nothing over 28, active voice only,\nsay HABITUATES / TUNES IT OUT / AUTOMATICALLY in plain words.\n\n**Readability is semantic, not only numerical (Fish 2026-09-04, permanent).** A short sentence can still be nonsense. If a line needs translation, describes a mental state as an abstract status, or sounds like compressed copywriter logic, it fails even when the grade and sentence-length checks pass. Ban constructions such as `just not started`, `getting up earlier held until nothing moved`, and `your calm is his calm`. Name what her son has or has not physically done and what she literally thought: `He stood beside his shoes without reaching for them.` The final read includes the brief’s readability target, average sentence under 14, zero sentences over 28, and this said-out-loud sense check.\n\n**Readability targets are brief-scoped.** Default remains grade 3–5 unless the active batch tightens it. When Fish names a tighter band (for example grade 2–4), put that range in the structured brief and deterministic gate for that batch; do not silently treat a session-specific target as permission to loosen or tighten unrelated formats.\n\n**Failed-solution paragraphs must be linear and direct (Fish 2026-09-04, permanent).** One attempt per beat, in this order: what she tried → why she tried it → what literally happened → why she stopped trusting it. Never jumble several methods into compressed fragments or slogan summaries. Ban `I tried everything the posts say`, `came next`, `laid it on`, `used it up`, `both helped`, and `neither stopped`. A reader should never have to reconstruct the timeline. Example: `I started getting up at 5:30 because I thought I was the problem. I wanted to be ready and calm before he got up. I still ended up yelling by the door.` Full operator card: `references/direct-failed-solution-prose.md`.\n\n**No coffee anywhere (Fish 2026-09-04, permanent).** Do not use coffee as morning texture, a reaction beat, a warm/cold payoff, or filler in any native. Use a story-owned action or object instead.\n\n**Meta-lesson:** a fix that changes syntax but keeps the literary SHAPE is not a fix.\nDirectness is not compression. It is plain speech.\n\nFull doctrine: `references/copy-directness-doctrine.md` · enforced by\n`~/.openclaw/workspace/scripts/validate_native_copy.py`\n\n## RELATABILITY GATE — no unverified qualifiers (Fish 2026-09-03, PERMANENT)\n\nFish, on the SA2 guilt native audit: \"Relatability risk audit needs to be part of the written\npipeline where no need to add extra qualifiers that we can't verify. Subavatars are only tied by\nthe specific categories that we chose.\"\n\n- The reader is identified ONLY through the sub-avatar card's categories (one experience, one\n  emotion, the behaviors that follow). Every other detail is a qualifier, and a qualifier the\n  reader does not share is a mother lost.\n- Every identification line carries a receipt: a quoted commenter or a survey verbatim in the\n  brief. No receipt, no line. Invented items (water bottle, lunch box), invented kid temperament\n  (\"he always forgives me\"), invented places (the store, dinner), clock times (\"6:50\", \"before\n  seven\"), ages, grades, siblings, partners, a car or bus as the reader's morning: all out unless\n  the card carries them.\n- Her own process is written the way the mothers state it, not compressed for rhythm: she says\n  it over and over, nothing moves, the voice goes up, the guilt after. No count on the repeats,\n  ever, even when the card's quote carries one (\"at least twenty\" is one mother's number, not the\n  reader's). \"Loud by the third reminder\" and \"twenty reminders\" were the 9-03 misses.\n- The mechanism is a stated CONTRAST, never a feature line: sound fills the room and startles,\n  a parent at the door is an order on a half-asleep kid, both spike stress hormones; the wrist\n  is small, quiet, touches only him, gives no order, he ends it himself, so no cortisol spike,\n  clearer head, calmer nervous system. If a reader cannot say WHY the wrist is different after\n  one read, the mechanism is under-stated (Fish 9-03).\n- Owned objects must be real situations from the source research (the toothbrush with the paste on it,\n  dry), never a staged prop (the sticky note that says \"don't yell\" was rejected 9-03).\n- **Specific failed solutions are not universal qualifiers (Fish 2026-09-04, permanent).** Visual timers,\n  charts, rainbow clocks, getting up earlier, staying in the room, and other methods may appear only as\n  this narrator's receipted history. Never put any of them in the reader bridge or imply every SA2 mother\n  tried them. The reader is identified by the 1:1:1 combination only: repeated instructions until her\n  voice rises, guilt, and apologizing after. Vary the narrator's failed-solution stack rather than cloning\n  one master checklist across bodies.\n- **Failed-solution restraint + whole-morning mechanism (Fish 2026-09-04, permanent).** Never turn a native\n  into an alarm-category catalog. For the ADHD teen-mom lane, use at most three prevalence-weighted\n  families, in this order when all are present: (1) Mom waking him herself, (2) phone at full volume plus\n  loud alarm clocks grouped into ONE sound family, (3) routines / bedtimes / sleep apps. Each family gets\n  acknowledge → gap → one memorable summary in a sympathetic, educational tone. Smartwatch/fitness band,\n  bed/pillow shaker, medication, “something else,” and “nothing yet” never become standalone lectures;\n  use one only when the narrator or objection specifically requires it, and keep tactile failures out when\n  they confound our mechanism. The mechanism is not “our alarm is better.” Explain the whole chain: sound\n  fails → Mom becomes the backup → his first interaction is her order while half awake → repetition and\n  pushback shape the next hour → a private teen-owned wrist cue removes Mom from the first step. Preserve\n  the honest limit: the band changes how the morning begins and who owns its first action; it does not claim\n  to erase every later executive-function problem.\n- **Body doubling is real but narrow.** We have a direct parent VOC receipt: \"Before I started body\n  doubling, it was just all yelling.\" It is not universal. Prefer the concrete action (`I stayed in the\n  room with him for every step`) unless the exact speaker uses the term.\n- **\"Connection over correction\" is not an exact phrase in the Dawn corpus.** The underlying action of\n  sitting with the child and trying to connect is receipted, but the slogan is not. Describe the action;\n  do not present the label as a standard ADHD method.\n- **Consumer-facing source channel for Meta natives: Facebook `post/posts` only (Fish 2026-09-04).**\n  Do not alternate among thread, forum, and post inside an ad. Internal provenance may still identify a\n  Reddit thread or forum URL; that source taxonomy never leaks into the narrator's words.\n- Narrator-only devices (an owned object, a named person replying to her post) stay hers; never phrase them\n  as something the reader has too. Authority first names may vary by body; the character's controlled\n  credentials and mechanism function remain fixed unless the brief changes them.\n- BRIEF GATE (Fish 2026-09-04, PERMANENT: \"how can we make sure the briefs we write follow our native\n  rules, do not add overqualifications, follow 1:1:1, do not hallucinate things that happen and use\n  VOC\"). The body validator catches what the model added; the brief gate catches what I added, and it\n  runs BEFORE any generation: `python3 ~/.openclaw/workspace/scripts/brief_gate.py <units.json>`\n  (exit 1 = no run). The unit brief is structured, not prose: every scene field carries a receipt of\n  class `voc` (verbatim substring of the corpus, thread id), `narrator` (her identity texture only)\n  or `dramatized` (allowed ONLY in the sanctioned slots: kid lines with a corpus-hit `shaped_from`,\n  one noticer line, the discovery chain, Dana's analogy, counted results). FAILs: card not 1:1:1 or\n  desire not a mass desire; any field without a receipt; a paraphrase presented as a quote; a\n  dramatized field outside its slot; qualifier scan on reader-facing fields (clock time, age/grade,\n  count on repeats, reader's car/bus, meds/screen/bedtime/app, the 0-hit my-phrase list); sibling /\n  partner / bus anywhere; figurative constructions in any field (simile, \"handing him her own\",\n  labor metaphor, swap/mirror, personified morning, coined event label, the named strikes: builds,\n  braced, flash of, lose the thread, leftover, the other one); qualifier nouns (water bottle, lunch\n  box, hair, uniform, checklist, sticky note); any content word in a scene field with 0 corpus hits\n  on every stem (my-word-in-her-slot, 7 strikes); a person noun with 0 hits; the system brief\n  missing any exact required line (Cortese, company intro naming the brand, guarantee line,\n  clearer head, nervous system, gives up, not a phone, the close rungs). The shared body BANNED list\n  is imported so a same-day ruling covers briefs the same turn. Model batch:\n  output/dawn-sa2-guilt-tof-2026-09-03 (brief-units-mid-angles.json, audit-brief-*.md).\n- Enforced: `scripts/validate_native_copy.py --voc <brief>` flags clock times, qualifier nouns,\n  and identification lines with no content-word overlap with the brief's quoted material. Run\n  it before qa-gate; a flagged line is either receipted in the brief or cut. The line-by-line\n  receipt table is written to `audit-1-1-1.md` in the batch dir (model: dawn-sa2-guilt-tof-2026-09-03).\n\n## GATES, IN ORDER — every long-format body, every time (Fish 2026-09-04: \"make sure these passes are always being ran for future copies\")\n\nOne command runs them and stops at the first failure:\n`python3 ~/.openclaw/workspace/scripts/native_gates.py --units <units.json> --bodies <final/*.txt> --band mid|long --objects <obj,obj,...> [--card <subavatar-card.json> --corpus <receipt-corpus.txt> --attempt-profile <failed-solution-profile.json>]`\n\nFor non-default sub-avatars, the optional card/corpus/profile flags are mandatory: the runner forwards the\nselected card and receipt corpus into `brief_gate.py`, then forwards the attempt profile into `audit_sop.py`.\nThe profile supports at most three ordered failed-solution families; aliases such as phone and loud alarms\ncollapse into one family. Each family must contain acknowledge → gap → explicit memorable-summary evidence.\nStandalone low-frequency lanes, repeated fourth family spans, malformed profiles, grade outside 2–4,\naverage sentences over 13 words, or a sentence over 28 words hard-fail. Stage 5 parses the receipt Markdown\nfail-closed, including escaped pipes, and requires `field | class | provenance | authorization`; dramatized\nweek/count/result rows need an explicit sanctioned-slot authorization.\n\n1. `scripts/brief_gate.py` on the unit briefs (BEFORE generation, and again here so the shipped brief is the gated brief).\n2. The batch `validate.py` per body (band, owned object, offer floor, relatability regexes, shared BANNED) and once across all bodies for cloned lines. Mixed-object batch validation will report expected object misses because `validate.py` accepts one global object; `native_gates.py` treats the per-unit runs as authoritative and fails the convergence-only call only when `CLONED` lines are present.\n3. `scripts/audit_sop.py` — the SOP 2.0 mechanical pass: §10 formatting numbers, §04 part word-shares, §08 reveal %, §06 dialogue:reaction pacing, §07 stack items addressed after the mechanism, §09 six weeks + counted results, §03 analogy + uniqueness link. Hard fails: reveal outside 65-82, fewer than six weeks, no analogy, no uniqueness link, a product hint in the first third, any emoji but the two 👉, bold/caps.\n4. `audit-sop-<batch>.md` must exist: the §11 eight killers and §13 final check answered per body BY HAND with line receipts (killers 2 shallow avatar, 4 tone switch, 7 stakes beyond the individual, 8 weak incident, and §13 voice / specificity / keep-reading are judgment, not regex). Model: output/dawn-sa2-guilt-tof-2026-09-03/audit-sop-v6.md.\n5. `audit-1-1-1*.md` must exist: the line-by-line receipt table (scripts model: audit_v6.py in the same batch).\nThen qa-gate, then Fish. A body that has not been through all five is not \"done\".\n\nFor incident-regeneration batches and Cloudflare review-page delivery, load `references/peak-incident-finalization-and-review-page.md`. It covers partial promotion from raw attempts, final-file marker checks, stale-audit repair, detector false-positive handling, believability-versus-polarization judgment, review-ready versus launch-ready status, and live production verification.\n\nWhen a reader flags foreign phrasing, confusing mechanism reactions, overqualified bridges, cloned failed-solution stacks, source-channel drift, or filler texture, load `references/readability-relatability-repair.md`. It separates direct VOC from dramatization and model synthesis, then gives the concrete-language repair and verification sequence. For the final paragraph-by-paragraph semantic gate after mechanical SOP validation, also use `references/directness-and-jumble-final-pass.md`; it catches unclear speaker turns, chronology, pronouns, compressed causality, and callbacks that numeric readability misses.\n\nINCITING INCIDENT (Fish 2026-09-04, PERMANENT): the incident is written at the PEAK OF BELIEVABILITY, not at the average\nmorning, as long as everything before it (the world, the stack, the toll) is verified and believable so the reader has\nalready decided the narrator is real when the peak comes. The OBJECT keeps its VOC receipt; the PEAK is a dramatized slot\n(`incident_peak` in the unit brief, receipted with the VOC lines it is shaped from). Visceral and physical, nobody new\nenters, nothing invented outside her, her son and the object. SOP §11 killer 8 is checked against this bar, not the\naverage morning.\n\n## SAID-OUT-LOUD TEST (Fish 2026-08-25 — PERMANENT, applies to every native)\n\nEvery narrator line must read as something the narrator would literally say out loud to\nanother mom. Proof source: the raw Reddit parent threads this lane is mined from — real\nmoms state what happened, flat and direct, zero symbolism. Kill on sight:\n- composed parallel rhythm (\"A fight at 11. A job at 6:45.\")\n- announcer transitions (\"Here is the part nobody warned me about.\")\n- mirrored abstractions (\"the phone moved in, the job moved onto me\")\n- report language (\"came with tension\" -> say \"turned into a thing\")\n- analogies and similes ANYWHERE in body or hooks (\"like I am arresting him\")\n- symbolism of any kind; describe directly what is going on instead\n- coined event labels (\"the 11 o'clock handover\", \"phones sleep in the kitchen\") — name the literal action every time\n- personification (\"the phone slept in my room\") — objects charge, sit, go places; only people sleep. Exception: verbatim Reddit VOC quoted as-is\nRun this as a dedicated pass after drafting, separate from and in addition to the humanizer pass.\n\n**HOOK OPENING — TWO LEGAL FAMILIES (Fish 2026-08-31, amends the 8-25 first-person-only rule):**\n\n1. **First-person lived action** — \"I/My/We\" + concrete action. The 8-25 rule, unchanged. (\"I used to make three trips up the stairs before 7am.\")\n2. **Second-person symptom flag** — \"If your [observable symptom]...\" Fish 8-31: **\"allow the 'if your' hook family, that's what converts.\"** Receipt: `adhd convenience slop 7` opens \"If your ADHD teen 'can't' wake up to alarms, you're not dealing with a discipline problem\" and is the account's biggest converter ($14,475 spend, 493 purchases, $29 CPA). Independently confirmed as the dominant device across SPNutrition's catalog (their #1 hook by impressions is a symptom flag).\n\nEverything else in the said-out-loud test still binds on BOTH families: no analogies, no composed parallel rhythm, no announcer transitions, no coined event labels, no personification. A flag hook still names a CONCRETE observable (\"sleeps through every alarm\", \"has a psychiatrist, a therapist and a 504 plan\"), never a vague state (\"struggles in the morning\").\n\n**Still banned:** the third family — pure mystery/curiosity openers that flag nobody. That is the unaware-pulling shape the awareness gate parks and the failure mode behind our expensive-click ads.\n\n## DawnBands Selling Doctrine — Mandatory\n\nBefore writing, briefing, editing, or auditing any DawnBands native, read:\n\n`/Users/ayden/.openclaw/workspace/knowledge/brands/dawnbands/selling-root-cause-mechanism.md`\n\nIt is the canonical consumer-facing mechanism doctrine for every mode and overrides older claim-safe-only framing that reduces DawnBands to “a different cue placement.” Future DawnBands copy should use the full causal chain where format permits:\n\n`circadian delay → sleep debt → auditory gating → sleep inertia / half-awake automatic behavior → Mom becomes the emergency pathway → Somatic Wake Bypass™ → Wrist-First Handoff™ → relationship + independence payoff`\n\nAggressive biological simplification is allowed and desired for the selling layer. Keep exact-SKU performance, testimonial timelines, offer facts, and customer outcomes tied to receipts; do not let provenance review flatten the root cause or mechanism into generic feature language.\n\n## ⛔ AWARENESS SCOPE GATE — UNAWARE LANE PARKED (Fish 2026-08-31, binding until he lifts it)\n\n**DawnBands runs PROBLEM-AWARE and warmer only. Do not write, brief, or recommend unaware-audience copy.**\n\nFish's reason: the market isn't there yet and we cannot scale at the CPA unaware traffic requires. This is a business-math ruling, not a craft judgment — the formats below are not \"bad,\" they are unaffordable right now.\n\n**PARKED (do not use, do not propose):**\n- **T3 unaware study-reveal** — parked entirely (it was always the unaware template, and it had no spend receipt anyway; every stat/study-hook ad we launched starved).\n- **The unaware application of T1 / Mode B**, including the zero-category-scent cold opener strategy built for readers who don't know they have the problem.\n- **ONLY the unaware-specific device in the Alevia corpus: the zero-category-scent mundane-object cold open** (\"mundane object + weekday + month + place-name in sentence one, zero product scent\"). That opener is engineered for readers who do not know they have the problem, and it is exactly the artifact-open trap — it withholds the TARGET. Parked for our traffic.\n\n⚠️ **DO NOT over-park Alevia (Fish 2026-08-31).** The audience is parked, the CRAFT is not. `alevia-live-anatomy-2026-08-12.md` is measured across 240 live copies and stays fully in force for problem-aware work. Keep using, unchanged:\n- **The P.S. ladder** — 217/240 of their copies close on stacked postscripts, and this file calls it \"our biggest structural gap.\" Rungs each have a distinct job: P.S. = rupture-object return · P.P.S. = guarantee naturalized in-voice · P.P.P.S. = objection/placebo kill in-voice · final P.S. = reader-direct-address AS the CTA · then a **bare 👉 link on its own line AFTER the final coda, no CTA copy around it.** Note the link comes LAST, after the postscripts, not before them.\n- **The spoken-word reveal mechanic** — the product never enters as a decision. It enters as a word spoken by a named third party inside the story.\n- **The discovery chain** — article → forum thread → named stranger's comment → skeptical read → try. Narrating the chain does DR work: relatability plus implied social proof without making a claim.\n- **Category-then-brand split** — mid-story reveal names the category, the brand surfaces at the close.\n- **Priced failed-solution stacks** (70/95 of their primary copies), and their register hygiene: near-zero emoji, near-zero ALLCAPS, no comment-bait in the witness lane.\n- **Format-page discipline** — never cross formats across page identities.\n- **Unaware-pulling hooks:** pure mystery/curiosity openers that don't flag who the reader is. (This is the same failure the 8-31 SPNutrition analysis found from the conversion side: unqualified clicks. The awareness ruling and the targeting-clarity learning agree — hooks must name the reader by her observable symptom.)\n\n**STILL LIVE (unaffected):**\n- **T2 compressed witness** — problem-aware, and our ONLY scaled winner (fable-v10, $29.70 CPA). Never park this.\n- **T4 punch**, all Mode A work, objection killers, product-aware and offer-aware copy.\n- **T1 / Mode B when briefed to PROBLEM-AWARE traffic** — the architecture is legal, the unaware targeting is not.\n- The Layer 3 review check \"is this written to the RIGHT stage?\" stands and now has a narrower legal range: problem-aware → offer-aware.\n\n## ⛔ LONG-FORMAT ARCHITECTURE = SOP 2.0 (Fish 2026-09-03, PERMANENT: \"follow the doc and what it says. It is more true.\")\n\nEvery long-format native (anything over the T4 punch band) is written and validated against\n`references/long-format-native-sop-v2-contract.md`, the binding distillation of Zakaria's Long-Format\nNative Ad SOP 2.0. It wins over the Alevia-derived witness contract, the T1/T2 budgets, the\nvictory-as-absence rule, the 55-70% reveal window, Mode B paragraph prose, and the analogy ban\nINSIDE the authority's mechanism dialogue. Load it before outlining. Structure in one line:\nhook (1-3) → the world (4-150) → inciting incident + authority (150-200) → mechanism through\ndialogue (200-400) → objections as story (400-500) → reveal at 75-80% by the authority (500-550)\n→ week-by-week results ending on the relationship resolution (550-650) → close on loss, the honest\nline, guarantee, the \"If by\" ladder, 👉 link, the fork, thank-you, signature, P.S. ladder (650-700).\nAlevia paragraphs (2-4 sentences), quoted dialogue, clinician authority allowed, counted results, narrator\nidentity block with her numbers (Fish 9-03 rulings after the measured Alevia comparison, see the contract's\n\"Alevia deltas ruled in\"). Fences that stay: no invented science, reader-side relatability gate, unaware\nlane parked, offer floor.\n\n## Mode Selector — Mandatory\n\nChoose the format before writing. Do not blend architectures by accident.\n**Check the AWARENESS SCOPE GATE above before selecting any template.**\n\n### STRUCTURE TEMPLATE — ALWAYS ASK FISH FIRST (binding, 2026-08-23)\n\nFour locked structure templates live in `references/native-structure-templates.md`\n(T1 Mode B Jacob-refined 3,200–4,000w · T2 compressed witness 800–1,100w · ~~T3 unaware\nstudy-reveal 550–700w~~ **[PARKED 8-31 — unaware lane]** · T4 punch 250–400w — each with per-section word budgets and\nreveal/CTA position markers). **Before writing ANY native, ask Fish which template to\nuse** — one question, up front, per ad or per batch. Skip the question ONLY when the\nrequest or an outline handoff already names the template (the handoff format is in that\nreference). Fish's standing workflow: he picks the template and writes the section\noutline; the copywriter fills prose to the budgets and verifies word count, reveal %,\nand (T1) CTA position against the template before qa-gate. Never silently substitute a\ndifferent structure than the one named.\n\n### Mode A — Standard native execution\n\nUse the existing validated-brief, compressed Avelia T1/T2/T3, objection-killer, short, and mid-form workflows in this skill. This is the default only when neither the brief nor the supplied controls clearly selects long-form witness architecture.\n\n### Mode B — DawnBands Long-Form Witness Native\n\nUse when the brief explicitly selects **DawnBands Long-Form Witness Native** mode **or when the user's supplied samples unambiguously specify that format**: roughly 2,500–4,000 words, first-person personal anecdote/witness story, a lived chronology before mechanism, and a late product reveal. “Like these samples” is a format instruction even if the user never types the internal mode name.\n\nBefore choosing Mode A for a sample-led request, measure the controls' word counts, narrator type, and first product-name percentage. Never compress 3,000-word personal-story controls into 200–700-word ownership arguments merely because the request also asks for different personas, angles, or root causes.\n\nLoad `references/control-format-selection-and-pilot-gate.md` for the control-measurement, narrator-selection, pilot, and back-half diversity gates.\n\nBefore outlining or drafting, load `references/dawnbands-long-form-witness-native.md` and obey its narrator-authority, provenance-ledger, 2,500–4,000-word target, reveal-window, cue-vs-waking, outcome-timeline, recurring-motif, and DawnBands claim-ceiling gates.\n\nDo not treat a 500–700-word ownership ad as Mode B. Do not force Mode B onto product-aware traffic or compressed Avelia T1/T2/T3 briefs.\n\n### Mode B first-draft execution (when intentionally selected)\n\n1. Load `references/dawnbands-long-form-witness-native.md`, `references/mode-b-first-draft-checklist.md`, and `references/control-format-selection-and-pilot-gate.md` before outlining. For measured live-account structure updates (paragraph-prose ruling, reveal distribution, spoken-word reveal mechanic, P.S.-ladder close variant, opener ceiling), also load `references/alevia-live-anatomy-2026-08-12.md`.\n2. Select narrator mode from the actual request and controls. When no verified Dawn buyer receipt exists, **ownership-voice** remains safest for argument-led work; however, when Fish explicitly asks for personal anecdotes or supplies first-person witness controls, use internally labeled **authorized-dramatization** rather than flattening the format into detached ownership prose. Keep behaviors research-grounded and never invent DawnBands outcomes, medical authority, quantified proof, or a named fake testimonial.\n3. Fill the internal YAML contract first: submode, narrator mode, motif, primary + ≥1 secondary open loop, reveal target, exact-SKU claim classes, offer_policy_verified, product_outcome_receipt null/ID.\n4. Draft to file under `output/.../copy/` with **Fable authoring in-session as the default copywriter** (amended 2026-08-12 — the only scaled winner, fable-v10, was a Fable draft; Grok/another model only when Fish explicitly names one). Preserve the exact first draft as a separate immutable artifact before review. Count primary-text words and first product-name % after every expand pass. Mode B floor is **2,500 words** — do not ship under-floor drafts labeled Mode B.\n5. For multi-ad Mode B batches, draft and audit one complete pilot before scaling. Read the opening, reveal window, and close; a good hook does not excuse a detached or cloned back half. After scaling, compare product bridges, mechanism lists, offer cadence, CTA scaffolds, and motif returns across the whole batch. If they converge, preserve the fronts and run same-model surgical back-half rewrites.\n6. Offer block: verify current price + refund terms. Prefer unambiguous **60-night money-back** language. Omit **100% wake-up guarantee** unless legal/substantiation approved it as a named refund policy and the line is clarified as refund terms, not universal waking efficacy.\n7. Victory = desired absence / test frame unless a same-speaker product receipt exists. No invented week-by-week results.\n8. After the immutable first draft, run the **Surgical Review Pass** (section 7): one line-edit pass against the four layers + format + hook criteria. GPT roast / Sonnet comparison artifact only on the section-7 escalation triggers (new-lane pilot, Fish request, structural failure). The Opus taste pass is retired. Fish’s explicit model request wins.\n9. For artifact freezing, hashes, measured QA, narrator-provenance review, provider retry, and the three-column Cloudflare comparison report, load `references/model-routed-copy-review.md`.\n10. Worked ownership-voice Mode B example: `/Users/ayden/.openclaw/workspace/output/dawnbands-native-2026-07-19/copy/ad-21-door-memory.md` (architecture reference only — do not clone lines; its routed review found pooled-VOC narrator risk, so it is not a spend-ready testimonial model).\n11. **Reading deliverable:** When Fish wants to read a finished Mode B / long native, ship a **public Cloudflare Pages URL** (not only a workspace path). Recipe in `references/mode-b-first-draft-checklist.md` → Delivery. Dedicated Pages project per report — never clobber an unrelated batch field-report project.\n\n## Native-First Priority (Fish, 2026-07-19)\n\nFor ad prose, optimize in this order:\n\n1. **Sounds native** — like a real post or lived thought, not a copywriter performing authenticity.\n2. **Story architecture is coherent** — rupture, escalation, discovery, mechanism, reveal, and payoff form one causal chain; each beat makes the next beat necessary.\n3. **The mechanism makes sense and sells** — for DawnBands, use the canonical Deep-Sleep Signal Lock biological stack, explain why repeated sound failed, then make the tactile bypass and Wrist-First Handoff feel inevitable. Do not neuter it into feature-only “different channel” language.\n4. **Breakthrough Advertising strategy is active** — channel existing mass desire, match awareness and sophistication, complete the belief chain, externalize the villain, and use the mechanism/proof depth the market requires.\n5. **Product arrives naturally** — the reader should understand the needed solution before the product appears; never announce the sales pivot.\n6. **Conversion language stays invisible** — preserve useful roughness, ordinary phrasing, and human irregularity. A more polished sentence is a regression when it sounds more like an ad.\n7. **Coherence beats cleverness** — remove quote-card abstractions, theatrical metaphors, internal QA language, and obvious objection-module formatting.\n\nSonnet polish is not a beautification pass. It may tighten drag and repair flow, but it must not sand the copy into perfect sales language or weaken the story, mechanism, awareness match, sophistication match, belief chain, or proof logic. Final read tests:\n\n- **Would this sound natural if the target avatar posted it without knowing she was writing an ad?**\n- **Does every story beat advance the belief chain?**\n- **Does the mechanism make prior failure and the product's bounded role feel logically inevitable?**\n\n## Existing Native Review — Full Context Intake\n\nWhen reviewing, revising, or explaining an existing DawnBands native, do not stop at the public/rendered copy. Verify and read the source stack that produced it: current skill rulings, format contract, current avatar/sub-avatar work, structured unit brief, prompts, brand truth, VOC receipts, final copy, and its human + deterministic audits. Keep current governing doctrine separate from historical examples and archived research.\n\nFor the exact intake order, Evolve source map, inciting-incident precedence, and minimum existing-batch artifact set, read `references/dawnbands-context-intake-and-precedence.md`.\n\n## Canonical Source Sync\n\nThis Hermes skill has a Claude Code counterpart at `/Users/ayden/.openclaw/workspace/.claude/skills/native-copy/SKILL.md`.\n\nCadence amended 2026-08-12: run this comparison WEEKLY or when output looks off-doctrine — not\nbefore every run (it was a per-run process tax with near-zero catch rate).\n1. Compare the Hermes and Claude Code versions for drift.\n2. If the Hermes skill is missing or incomplete, consult the Claude Code version before writing.\n3. If they differ, use the newest writing doctrine from the Claude Code version while preserving Hermes-specific safety and tool instructions from this skill. Never silently rely on a stale imported copy.\n4. When checking adjacent integration configs, never print credential values. Report only presence, validity, and source path.\n\n## Workflow\n\n### 0. Read Competitive Landscape (MANDATORY)\nRead `knowledge/research/competitive-landscape.md` — contains saturated angles (HIGH CPM, avoid), white space angles (LOW CPM, prioritize), competitor hooks to steal/counter, and positioning rules. This step is NOT optional.\n\nIf a doctrine path has moved, search the workspace for the filename before concluding it is missing. Once the canonical location is verified, update this skill rather than silently skipping the gate.\n\n**Before writing ANY copy, verify:**\n- The angle you're writing is NOT in the \"Saturated Angles\" list (or has a clear differentiation plan)\n- You're prioritizing white space angles (60%+ of new batch)\n- You're not duplicating a competitor's exact hook or framing\n- You're incorporating at least one stolen copy device (adapted to our positioning)\n\n❌ NEVER write another \"I shake my ADHD child awake\" ad — that's Nymera's saturated lane\n❌ NEVER write all-fear copy with zero aspiration — every competitor does that, we differentiate with hope\n❌ NEVER skip this step. Copy written without competitive context = wasted tokens\n\n### 1. Read the Brief\nRead `references/master-brief.md` — contains everything: avatar, mechanism, format rules, Zak framework, dead angles, copy killers, headline rules, targeting rules.\n\n### 2. Check the Gold Standard\nRead `assets/gold-standard-example.md` — correctly formatted variants with headlines. Match this format exactly.\n\n### 3. STANDING NATIVE TEST LAYOUT (Fish, 2026-08-23 — supersedes the same-day 3×3 grid and the old 5-variant default)\nEvery native test unit ships as:\n- **1 body script** — one desire × one sub-avatar × one message. Length set by what the angle needs (7-22 no-caps ruling); template still ALWAYS asked (Mode Selector).\n- **3 hook variations** — line 1 swapped on the same body, each graded on the hook taxonomy (structure named, curiosity/validation mixed). Hooks are the copy-side test lever.\n- **5 image concepts** — the primary creative lever (receipt: image swap moved fable-v10 + its sniper test hard). Scrape-first per native-image-gen; mix concepts across the operator families: contradiction pair (two objects that don't belong together, zero copy on image), problem-proof \"help-post\" shot, artifact/rupture object, straight rip. No product, no text overlay, phone-photo register.\n- **2+ headlines authored — ALL load as title options at launch (Fish 2026-08-25 + 2026-08-29, supersedes the 8-23 \"picks ONE, never split-tested\" rule).** Meta standard ads take up to 5 titles. Every hook and every headline rides in at launch via the solved multi-text construction. Nothing waits for a pick.\n- **Length variants are a legal test axis inside one unit (Fish 2026-08-31).** A unit is one ARCHITECTURE (one desire × one sub-avatar × one message); long/mid/short cuts of that architecture ride as body slots in the multi-text stack (Meta caps at 5 bodies — spread them so every length and every hook appears at least once). This does not license two different angles in one unit.\n- **Validated structured brief:** the brief's `goal.variant_count` / `output_contract` still override when explicit.\n\n### 3a. OFFER FLOOR — every native ships with a complete close (Fish 2026-08-31, BINDING)\n\nReceipt: `v11-a-unassigned-hour` drew a 4.13% link CTR at $0.92 CPLC with NO brand name, NO offer, and NO link anywhere in the body. Clicks were manufactured by withholding every resolution; CPA landed $46-56. Every ad that actually converts in the account (fable-v10, r2-d2-roommate, slop 7, the offer-ad family) carries a complete close.\n\n**REQUIRED in every native body, every length, every destination:**\n1. **Brand named at the reveal** — \"Dawn Bands\", two words. A body that never names the brand is not a native, it is a click generator.\n2. **The guarantee, stated plainly** — \"100% wake-up guarantee. 60 nights.\" Say what it means in her language at least once (\"60 nights to let it prove it on school mornings\"). The guarantee, not the price, is the mandatory risk-reversal element.\n3. **Company/credibility intro, 2-4 lines — and it MUST NAME THE BRAND as its subject** (Fish 2026-08-31, second ruling). Write \"Dawn Bands is a small company and this is the only thing they make.\" NEVER \"The company that makes it is small\" or \"The company is small\" — an anonymous company intro is a floor FAILURE, not a stylistic choice. Receipt: in the 8-31 model bake-off, 3 of 5 drafts satisfied \"brand at reveal\" and \"company intro\" as separate items and produced an orphaned, brandless credibility block. The brand name has to be the subject of that sentence. Small-operator framing beats corporate framing; carry the quality reason into a natural availability consequence.\n4. **At least one link.**\n5. **P.S. ladder + TWO CTAs — allowed and encouraged on long copies** (Fish 2026-08-31). Close order is now fixed:\n\n   `body → CTA #1 with link → P.S. ladder → CTA #2 with link`\n\n   **One CTA before the postscripts and one after** (Fish 8-31, third ruling). This amends the Alevia measured default of link-last-only — their ladder ends on a single bare link; ours brackets the ladder so a reader who stops at the first postscript still has a live link above them and a reader who reads to the end gets another. Rungs, per the Alevia shape: P.S. = rupture-object return · P.P.S. = guarantee naturalized in-voice · P.P.P.S. = objection kill in-voice · final P.S. = reader-direct-address. Optional on short/punch formats.\n\n**OPTIONAL — use when it earns its place:** price ($49 may be woven in or omitted; Fish 8-31: the offer and guarantee are required, the number is not), sale/availability framing, the narrator's own purchase beat, P.S. ladder.\n\n**DESTINATION:** listicle and PDP both convert. The floor is identical for both. A listicle ad must be FULLY CONGRUENT with the page it lands on — same claim ceiling, same mechanism, same offer terms. Incongruence kills listicle traffic, not the destination type itself.\n\n**Variation lives in HOW these are woven into the story, never in WHETHER they exist.** This supersedes master-brief's CTA-variation permission \"No mid-copy CTA at all — just the close CTA\" where it would leave a body with no close. The price-omission permission survives; the missing-close permission does not.\n\n#### Reference close-block shape (SPNutrition-measured; adapt the register, never the lines)\n1. Company intro: small company, limited batches, the quality mechanism restated in plain words, contrasted against the cheap category default.\n2. Quality → availability consequence: that level of care takes time, which is why restocks take weeks.\n3. Current availability plus the narrator's own action (\"I ordered a second one\").\n4. Guarantee in plain language: the window, what happens if nothing changes, no questions asked.\n5. Conditional future-pace ladder: \"But if it works? If by [early marker]... If a month from now [the desired absence]?\" Three escalating conditionals, each a desired ABSENCE, never a promised outcome.\n6. Link.\n7. Closing callback returning to the hook thesis in one line.\n\nClaim gates unchanged: the ladder stays conditional and absence-shaped so it never becomes an invented outcome timeline. Never invent batch sizes, customer counts, stock levels, or sale end dates. Availability language must be true at launch or omitted.\n\n#### Validated Brief Execution Mode\nWhen the user supplies a validated brief batch (for example `briefs.json`), treat each brief as an executable contract:\n1. Extract only the requested brief IDs rather than bulk-loading the whole batch into context.\n2. Follow the brief's exact word range, fields, destination type/URL, CTA, awareness, mechanism-delivery mode, and output path.\n3. The newest live evidence source governs claims: an exact-SKU/product-mechanism report or current policy receipt overrides an older brief, canonical product block, gold standard, or brand-doctrine example. If sources conflict, stop the affected claim, record the contradiction, and use the narrower live claim ceiling rather than silently inheriting the older language.\n4. Preserve exact offer wording/order only after verifying that the current policy and newest live claim report authorize both the terms and the consumer-facing implication. For DawnBands, do not mechanically carry `100% wake-up guarantee. 60 nights. $49.` from an older brief when a newer report treats guaranteed waking or exact-SKU efficacy as unsubstantiated. If the policy is verified but the name could imply universal efficacy, distinguish the refund-policy terms from an outcome claim and flag for legal/substantiation review.\n5. Include a compact declaration block and source VOC IDs when required so the artifact is auditable.\n6. Do not invent customer outcomes, timelines, reviews, or first-person testimonials to satisfy a story format. Use supplied VOC as attributed evidence, or write a clearly non-testimonial POV/ownership argument that does not claim an unverified result. If no product outcome is evidenced, express the victory as the avatar's desired absence rather than claiming it happened.\n7. Even when a brief labels the entry door a mechanism reveal, line 1 must still open on a lived artifact, contradiction, or emotional tension. Reveal product/mechanism only after the hook.\n8. Write only the authorized deliverables. If the scope ceiling names ad files plus one manifest, do not add strategy reports, research notes, or extra artifacts.\n9. Run deterministic validation ONCE after the surgical review pass (and again only if a later edit touches the text): actual primary-text word count, required fields, offer order normalized across line breaks, destination, banned claims/price, em dashes, line format per the mode-scoped rule in the Format Validation Pass (sentence-per-line for Mode A; paragraph prose for Mode B), dialogue isolation, declared VOC IDs. SHA-256 manifest hashes are required only when the validated brief's contract demands a manifest — not on ordinary batches (cadence amended 2026-08-12; the after-every-edit loop was pure process tax).\n\nReusable execution detail lives in:\n- `references/lived-first-person-native-copy.md` for repairing detached analyst voice into lived first-person native copy without inventing a customer outcome.\n- `references/validated-brief-execution.md` for authority order, truth-safe POV writing, and deterministic per-ad QA.\n- `references/validated-brief-batch-contract.md` for authority order, scope discipline, claim-safe POV handling, canonical-offer line breaks, and deterministic manifest QA.\n- `references/brief-driven-batch-validation.md` for the existing brief-batch validation checklist.\n- `references/native-proof-and-architecture-overlays.md` for testimonial provenance, long-form-vs-compressed architecture, live-Meta evidence use, cross-ad convergence, and the repair/re-QA sequence.\n- `references/long-form-native-comparative-audit.md` for read-only batch audits against winner copies, a live long-form SOP, and live avatar/mechanism reports; it defines source precedence, deterministic reveal measurements, device/per-ad matrices, structural-vs-provenance separation, and brand-doctrine output.\n- `references/consolidated-native-ad-repair-pass.md` for repairing a requested ad subset from multiple QA reports, truth-safe conversions, blocker-specific assertions, and final manifest/hash revalidation.\n- `references/controlled-live-meta-repair.md` for repairing a brief-driven batch under a strict file ceiling while adding a small number of controlled live-evidence mechanism lanes without cloning winner copy.\n\n### Editorial truth gate before paid creative generation\n\nDeterministic PASS is necessary but not sufficient. Before image prompts or paid generation:\n1. Run a flags-only editorial/source-fidelity audit across the whole batch.\n2. Treat synthesized first-person scenes, unsupported precision, altered quotations, invented dialogue, and composite timelines as blockers even when every underlying behavior is VOC-backed.\n3. Convert blockers to attributed VOC, explicit hypotheticals, paraphrases without quotation marks, ownership arguments, or desired-absence framing.\n4. Run a separate cross-ad convergence check for mechanism sentences, product descriptions, narrator cadence, analogies, victory images, and CTA scaffolds.\n5. If a new SOP or live performance export appears while QA is running, add one overlay audit and then make a single consolidated repair pass rather than rewriting the batch repeatedly.\n6. Consolidate repairs into ONE pass, then re-run deterministic validation and editorial QA once on the repaired batch (amended 2026-08-12 — not after every individual edit).\n7. Keep image generation gated until the repaired copy passes.\n\n### 3.5 Multi-Angle Batch Architecture\n\nWhen Fish asks for many ads with different personas, angles, stories, root causes, or mechanism routes, do not rely on a prose instruction to “make them different.” Build a row-per-ad diversity matrix first, then route bounded model lanes from that shared matrix.\n\nLoad `references/multi-angle-batch-routing-and-repair.md` for:\n- required matrix dimensions\n- dominant-root-cause distribution\n- raw-versus-final artifact structure\n- deterministic and editorial convergence checks\n- same-model surgical repair when a full agent prompt is too expensive\n- public batch report delivery\n\nTruth and provenance rules stay in the hidden brief. **Never let them leak into consumer copy** as lines about not inventing reviews, not promising outcomes, avoiding fake stories, or clarifying internal claim policy. Replace that leakage with natural skepticism, a bounded test, risk reversal, or avatar-native objection language.\n\n### 4. Write Variants\nEach variant uses a different angle from the 4-type framework:\n1. **Story/Resignation Arc** — first person narrative, resignation moment, discovery\n2. **Objection Killer** ← TOP PERFORMER (Mar 2026). Write 2 per ad when possible. \"Another alarm? Sure.\" energy converts.\n3. **Pain-First → Mechanism Reveal** — vivid daily pain, mechanism in the middle\n4. **Consequence/Future Fear** — what happens if nothing changes\n5. **(If needed) Risk Reversal + Social Proof** — reviews + guarantee forward\n\n**Mode A length (Fish 2026-08-23 test layout):** ONE body per test unit; pick its length by how much DR work the angle needs (150-300w short, 500-700w mid, 800-2,100w long band — no caps). Variation budget goes to hooks (3) and images (5), not parallel bodies. **Mode B** uses `references/dawnbands-long-form-witness-native.md` (2,500–4,000 words) and is never satisfied by a 500–700-word draft. See also `references/mode-b-first-draft-checklist.md`.\n\nEach variant needs:\n- **Headline** (Meta `title` field)\n- **Body copy** (Meta `body` field)\n- **Destination URL** (listicle, not PDP)\n\n### 5. Format Validation Pass (MANDATORY)\nAfter writing ALL copy, do a final scan.\n\n**Line format is MODE-SCOPED (corrected 2026-08-12 per the 7-31 Mode B ruling):**\n- **Mode A / compressed natives / static primaries:** every line = 1 sentence (2 max if flow\n  demands it); NEVER 3+ sentences on one line.\n- **Mode B / long-form witness:** PARAGRAPH prose — 2–4 sentences per paragraph (~25–40 words),\n  blank line between paragraphs; staccato single-lines only at punch moments. One-sentence-per-\n  line formatting in Mode B is a format defect, not a style choice\n  (`references/alevia-live-anatomy-2026-08-12.md`).\n\nAll modes:\n- No em-dashes\n- No emoji except 👉 on CTA links (and the T4 checklist template, where ✅ is the register)\n- Dialogue isolated on its own line\n- No dead angle hooks (mechanism-first, generic features, curiosity bait)\n- No copy killers (\"game changer\", \"literally\", \"amazing\", vague future pacing)\n\n### 6. Hook Sharpening Pass (MANDATORY)\nAfter writing all variants, re-read every hook (line 1 of each variant's body copy). Grade each:\n- **WTF?** Does it stop scroll? Does it create a question she HAS to answer? Would she screenshot it and send it to a friend?\n- **Specific?** Does it contain a concrete detail (number, name, object, timestamp)? Generic hooks die.\n- **Gap?** Does it open a loop that can only be closed by reading more?\n- **Taxonomy (Fish 2026-08-23, `~/.openclaw/workspace/knowledge/hook-taxonomy.md`):** name the emotion it fires, the stake on the table, and the structure — **Promise** (bold claim upfront, exceptionally clear) or **Loop** (key detail withheld). \"Both/neither\" = rewrite. Execution screen: zero backstory required, starts in action, no clutter — when in doubt, dumb it down.\n- **Targeting clarity (Fish 2026-08-31, the 5th grading question):** could a stranger tell from lines 1-2 whether this ad is about HER house? Name the observable symptom or daily job it flags. \"It builds intrigue\" is not an answer. A Loop hook is only legal when the withheld information is about the READER'S OWN situation, never generic mystery. Receipts: native1 24.8% CTR / 0 purchases; v10 body $8,267 at $51 CPA opening on a victory brag that reaches identification nine lines down; slop 7 flags her in line 1 and closes at $29 CPA on $14,475.\n\n**If a hook fails any of these, rewrite it before submitting.**\n\nCommon hook failures to catch:\n- ❌ Too bare: \"He set it himself.\" → needs the MOMENT: \"I walked past his room. He was adjusting something on his wrist. He didn't know I was there.\"\n- ❌ Too generic: \"I used to be the kid everyone had to wake up.\" → needs the SPECIFIC story: \"I slept through a fire alarm at a sleepover. Everyone thought something was wrong with me.\"\n- ❌ Too subtle: \"I just didn't go in.\" → needs the CONSEQUENCE: \"One morning I didn't go in. By 11:40 the school called me.\"\n- ❌ Statement without tension: \"His teacher emailed me last week.\" → needs the CONTENT: \"His teacher emailed me two words: 'perfect attendance.' I stared at my phone for ten minutes.\"\n- ❌ Product listing as hook: \"Sonic Boom bed shaker. $45.\" → needs the TOTAL + tease: \"$380. That's what I spent before someone told me they all use the same pathway his brain blocks.\"\n\n**The hook is the only line that matters at scale. If it doesn't stop scroll, the rest is irrelevant. Spend more time on hooks than on any other part of the copy.**\n\n### 7. Surgical Review Pass (MANDATORY — one pass, line edits only)\n\n**(Restructured 2026-08-12, Fish-approved lean pipeline.)** The old mandatory multi-model\nstack (separate Sonnet full-polish artifact + GPT roast + optional Opus) ran up to seven\ntext-mutating passes per ad. The only scaled winner (fable-v10, $5.4k @ 2.28 lifetime) never\nwent through it, and this skill's own Native-First rule says a more polished sentence is a\nregression when it sounds more like an ad. Rewrite passes sand off the native roughness that\nIS the format.\n\nThe default is now ONE review pass:\n\n- Save the exact first draft unchanged (immutable, one file copy — keep this).\n- The drafting model re-reads the draft against the **four layers below** plus the Format\n  Validation Pass and the Hook Sharpening criteria, and fixes findings as **line edits** —\n  never a top-to-bottom rewrite. Blockers (claim/provenance/offer/format) MUST be fixed;\n  taste flags are logged for Fish, not auto-applied.\n- Output: the repaired artifact + a short blocker/change log. Not three artifacts.\n\n**Conditional escalation** (any of these, not by default): a NEW angle or lane's pilot ad,\nFish asks for it, or the surgical pass finds structural failure (rebrief-level problems):\n- **GPT challenger roast** — decision memo on original + repaired, never a silent rewrite.\n- **Sonnet full-polish artifact** — only when Fish explicitly wants a comparison version.\n\n**Humanizer (amended 2026-08-12 with Fish's approval — supersedes the 2026-07-20 always-on\nrule):** run the humanizer pass only when (a) the draft came from a model other than Fable,\nor (b) the repaired draft fails a native-read spot check — read the opening 8 lines and one\nmid-story passage cold: would this pass as a person posting? Fable in-session drafts that\npass the spot check ship without the humanizer. All DR beats are preserved whenever it runs.\n\nThe review layers (the checklist for the surgical pass):\n\n**Layer 1: Hook Audit**\n- WTF factor (1-10). Anything below 7 = flagged with rewrite suggestion\n- Is it something she's heard before? Check against hook bank dead list\n- Does it open a gap only the copy can close?\n- ❌ NEVER pass a hook that reads like a headline. Hooks = native voice, not ad voice.\n\n**Layer 2: Copy Quality**\n- Flag AI-sounding language (graveyard, journey, struggle, battle, warrior, game-changer, literally)\n- Flag weak/filler lines that don't earn their space\n- Flag soft CTAs or templated CTA blocks\n- Flag pacing issues (too many long sentences in a row, no breathing room)\n- ❌ NEVER just say \"this line is weak.\" Provide the replacement line.\n\n**Layer 3: Framework Alignment**\n- Awareness level: is this written to the RIGHT stage? (unaware ≠ solution-aware)\n- Mechanism: is it specific enough? Does it feel like the first time she's heard it?\n- Resignation moment: is it present? Is it EARNED (not just stated)?\n- Villain: is there a clear external force she can blame?\n- Desire magnitude: is the pain/want big enough to act on?\n- ❌ NEVER pass a variant missing both resignation moment AND villain. At least one must be present.\n\n**Layer 4: Avatar Fidelity**\n- Does this sound like HER words or a copywriter's words?\n- Cross-check against VOC word bank in avatar interior doc\n- Check the 3am thoughts, the parking lot moment, the \"I've heard that before\" exhaustion\n- Is the victory written as what STOPPED (not what started)?\n- ❌ NEVER pass copy where the avatar sounds aspirational instead of exhausted. She's resigned, not motivated.\n- **Lines 3-8 specificity test:** Read lines 3-8 of body copy. Broad/general = avatar was assumed, not found in VOC. Specific/concrete = avatar was found. \"Many women feel frustrated when their dental routine doesn't deliver results\" = assumed. \"The hygienist said excellent. Again. For the third year in a row. While her teeth were getting worse.\" = found. Flag any copy that fails this test — rebrief required, not a line-level fix.\n\n**Synthesis rules (surgical pass + any conditional escalation):**\n- Truth/provenance/offer/format breach → mandatory line-edit fix before the artifact can pass\n- Taste/conversion disagreement → logged for Ayden's judgment, never auto-applied\n- Cadence/specificity improvements → only as line edits that preserve provenance, claim\n  ceiling, and the draft's native roughness\n- Never mutate the immutable first draft; repairs land in the working artifact\n- Side-by-side comparison surfaces (original / repaired / roast memo) are built ON REQUEST,\n  not by default\n\n### 7.5 Independent Adversarial Audit Mode\nWhen reviewing a completed ad batch rather than writing it, do not treat deterministic validation as editorial approval.\n\nRun two explicit gates:\n1. **Artifact contract:** files, fields, word ranges, offer order, destinations, and formatting.\n2. **Editorial/provenance:** hook quality, lines 3–8 specificity, avatar fidelity, claim/testimonial truth, awareness, persuasion chain, and cross-ad diversity.\n\nA deterministic PASS can coexist with an editorial FAIL. In particular:\n- A brief-authorized scene is not evidence that the scene happened.\n- VOC that supports a behavior does not authorize inventing a first-person narrator, chronology, dialogue, outcome, or precise number.\n- Plausible precision is still unsupported precision.\n- Repair composites by attributing exact VOC, using explicit hypothetical framing, writing a non-testimonial ownership argument, or deleting the unsupported detail.\n- In read-only audit mode, provide exact replacement proposals for every score below 7 without editing source ads.\n- Separate high-confidence blockers from subjective taste flags and label confidence.\n\nUse `references/adversarial-batch-copy-audit.md` for the scoring matrix, provenance test, cross-ad diversity checks, and deliverable verification. When the audit compares a batch with competitor/winner copies, a live long-form SOP, and live avatar/mechanism reports, also use `references/long-form-native-comparative-audit.md`; it requires an explicit source-precedence ledger, measured reveal timing, a winner/SOP device matrix, a per-ad disposition ledger, and separate structural versus truth/provenance findings.\n\n### 8. Cross-Variant Diversity Check\n- Each variant must hit a genuinely different Persona × Desire × Awareness combo\n- No two variants should open with the same hook formula\n- Headlines must not duplicate across variants\n- No testimonial names shared with listicle comments\n- **Mechanism delivery MUST be different in each variant** (full science / casual / secondhand / barely explained / analogy / skipped — see master-brief)\n- **CTA format MUST vary** — do NOT use \"👉 Here's [reason]:\" on every variant\n- **At least 1 variant must be SHORT** (15-20 lines) — not every variant is a 40-line story\n- **Social proof reviews must sound real** — typos, incomplete thoughts, 4-star mixed in, platform references\n\n## ⛔ NEVER DO (check EVERY time)\n- [ ] 3+ sentences on one line → KILL IT\n- [ ] Em-dashes anywhere → KILL IT\n- [ ] Emoji on CTA that isn't 👉 → KILL IT\n- [ ] Hook that matches anything in dead angles list → KILL IT\n- [ ] \"Game changer\", \"literally\", \"amazing\", vague future pacing → KILL IT\n- [ ] Mechanism-first opening (leading with how the product works) → KILL IT\n- [ ] Same CTA format on every variant → FIX IT\n- [ ] All variants are 35+ lines (no short one) → FIX IT\n- [ ] Testimonial names that match listicle comment names → FIX IT\n- [ ] Didn't read master-brief.md before writing → STOP AND READ IT\n- [ ] Hook is scene-setting or chronological start instead of most intriguing element → REWRITE HOOK\n- [ ] Best line is in headline field instead of primary text line 1 → MOVE IT\n- [ ] Same mechanism paragraph across multiple variants → REWRITE EACH DIFFERENTLY\n- [ ] Same analogy/metaphor repeated across variants → KILL DUPLICATES\n- [ ] AI-default words (graveyard, journey, struggle, battle, warrior) → REPLACE WITH UGLY HUMAN WORDS\n- [ ] Internal QA/provenance leakage (\"I won't invent a review,\" \"no fake before-and-after,\" legalistic claim caveats) → REWRITE AS NATIVE SKEPTICISM OR A BOUNDED TEST\n- [ ] CTA block is templated/identical across variants → REWRITE TO FLOW WITH STORY\n- [ ] Victory written as what started (\"mornings are peaceful\") instead of what stopped (\"I didn't yell\") → REWRITE\n- [ ] All 5 variants same persona + desire + awareness level → DIVERSIFY AT LEAST 1 DIMENSION\n- [ ] Mode A: dense paragraphs instead of 1 sentence per line with breathing room → BREAK APART (Mode B is PARAGRAPH prose by contract — do NOT break apart a witness native)\n- [ ] ⛔ ALWAYS check the full ⛔ NEVER DO section in master-brief.md — it is the single most important section\n\n## Avatar Detail Extraction (MANDATORY — do this before writing copy for ANY avatar)\n\nBefore writing copy, extract AT LEAST these categories from the avatar research doc into the subagent brief. If any category is missing from the research, flag it — don't skip it. If the research reveals additional categories not listed here, include those too. These are the details that separate copy that converts from copy that reads like AI.\n\n1. **Failed solutions stack WITH visceral detail.** Not just \"tried alarms.\" The specific unsettling behavior: he walks to it in his sleep, eyes open, solves math puzzles unconscious. The detail that makes a reader go \"wait what?\"\n2. **Anticipation anxiety.** What happens BEFORE the problem starts each day. She wakes before her alarm from dread. The cortisol spike. The mental math of how much buffer she has.\n3. **The scene after.** The drive to work in silence. The parking lot. The replaying. The 10 minutes before she can go inside. Concrete, visual, usable in copy.\n4. **The \"I've heard that before\" trap.** Everyone thinks they have the answer. She's exhausted by suggestions. This is the skepticism objection killers need to mirror.\n5. **Relationship cost.** What the problem did to the relationship over time. Not one fight — accumulation. He stopped talking. She became someone he avoids. The repair moment when it starts working.\n6. **Victory as ABSENCE.** What STOPPED happening. Not what started. She hears footsteps before she went in there. She didn't yell today. Write victory as disappearance, not arrival.\n7. **VOC word bank.** Her actual words organized by emotional state. Default to these over AI-generated language. Always.\n\nIf a subagent doesn't have all 7, the copy will be generic. Every time.\n\n## Subagent Handoff Rule\nWhen spawning a copy subagent, it MUST receive:\n1. The full `references/master-brief.md` file (not a summary)\n2. The `assets/gold-standard-example.md` file\n3. The specific ad copy it's writing variants for\n4. The NEVER DO checklist above (copy-paste it into the task)\n5. The 7 avatar detail categories above — extracted from avatar research, not summarized\nDo NOT summarize the rules. Pass the files directly.\n\n## Key Files\n- `references/native-structure-templates.md` — **the four locked structure templates (T1–T4) with per-section word budgets; template choice is ALWAYS asked of Fish before writing (see Mode Selector)**; Fish-outline → copywriter-fill handoff contract lives here\n- `references/matrix-angle-execution-map.md` — **load FIRST when the brief starts from the ad-batch-hub testing matrix** (\"expand my tests\", \"feed cell X\"): resolves every matrix angle column (incl. wrong-label / defiance / witness, added 8-12) into definition, Dawn form, mode/length, receipts, and claim gates; also the proof-column devices (comparative two-person, survey-voice, retro-proof)\n- `references/control-format-selection-and-pilot-gate.md` — infer long-form mode from supplied controls, select personal-anecdote narrator authority, pilot one complete ad before scaling, and repair cloned reveal/close scaffolds without rewriting strong story fronts\n- `references/multi-angle-batch-routing-and-repair.md` — class-level workflow for pre-draft diversity matrices, bounded same-model lanes, immutable raw artifacts, convergence QA, provenance-leak cleanup, exact-model micro-repairs, and public batch reports\n- `references/dawnbands-selling-root-cause-mechanism.md` — compact execution reference for the canonical Deep-Sleep Signal Lock → Somatic Wake Bypass™ → Wrist-First Handoff™ selling doctrine; load for every DawnBands format so proof review does not flatten the mechanism into generic cue-placement language\n- `references/dawnbands-long-form-witness-native.md` — canonical Avelia/Manus-derived DawnBands witness-story mode: narrator authority, provenance ledger, 2,500–4,000-word architecture, bounded mechanism, motif return, outcome gate, and final QA\n- `references/mode-b-first-draft-checklist.md` — Mode B operator card: YAML gates, offer/SKU traps, under-floor wordcount, worked example, Cloudflare Pages reading-report deploy\n- `references/master-brief.md` — full copywriting brief (avatar, mechanism, format rules, dead angles)\n- `references/adversarial-batch-copy-audit.md` — read-only batch audit protocol for scoring, testimonial provenance, unsupported precision, blocker/taste separation, and diversity verification\n- `references/native-proof-and-architecture-overlays.md` — testimonial truth, long-form Manus architecture boundaries, live-Meta evidence use, convergence checks, and the pre-image repair sequence\n- `references/avelia-short-native-templates.md` — verified competitor-derived T1/T2/T3 structures for compressed short native copy; use as a structural reference, never as winner proof\n- `assets/gold-standard-example.md` — cadence/format example only; do not inherit its product truth, claims, scenes, dialogue, timelines, or outcomes without independent receipts\n",
    "sha256": "65a1a2e375cd6cae2df2457bafd60cda2f7fa46c8978277a1aeb54234b0c84a6"
  },
  {
    "group": "Native copy",
    "name": "assets/gold-standard-example-old-adhd.md",
    "path": "/Users/ayden/.hermes/skills/openclaw-imports/native-copy/assets/gold-standard-example-old-adhd.md",
    "text": "# Native8 — ADHD Listicle — 5 Copy Variants (Gold Standard)\n**Image:** native8 (ADHD med cycle native)\n**Destination:** dawnbands.com/pages/adhd-teen-alarm\n**Date:** 2026-03-05\n\n---\n\n## VARIANT 1 — Native/Story (Resignation Arc)\n**Awareness:** Level 2 (Problem Aware) | **Persona:** ADHD Mom | **Emotion:** Exhaustion → Relief\n\n**Headline:** I Did Everything His Therapist Told Me. He Still Couldn't Wake Up.\n\nI set up the perfect ADHD bedtime routine.\n\nSame time every night.\nPhone out of the room.\nWhite noise machine.\nMelatonin at 8:30.\nLights off at 9:15.\n\nI followed every protocol his therapist gave me.\n\nHe still couldn't wake up.\n\nNot \"didn't want to.\"\nCouldn't.\n\nI've watched him open his eyes, say \"I'm up,\" and be completely unconscious.\n\nI've watched him walk to his phone, turn off the alarm, walk back to bed.\nNo memory of any of it.\n\nI went in four times yesterday.\nBy the third time my voice was different.\nBy the fourth I was someone I don't want to be.\n\nHis therapist said something I couldn't stop thinking about.\n\nWe keep trying to fix an executive function problem with executive function tools.\n\nEvery alarm, every app, every system I built required him to process, decide, and initiate.\nThose are the exact skills ADHD impairs at 6:30am.\n\nI stopped building better alarm systems.\nI started looking for something that doesn't need his executive function at all.\n\nDawn Bands vibrates on the wrist.\nTouch. Not sound.\nA different sensory pathway.\nNo screen. No dismiss button.\nNothing for his autopilot to turn off.\n\nHe woke up Tuesday before I did.\nI was still in bed when I heard him in the kitchen.\n\n👉 Here's why sound alarms can't work for ADHD brains:\ndawnbands.com/pages/adhd-teen-alarm\n\n60-night guarantee.\nFull refund if it doesn't work.\n\n---\n\n## VARIANT 2 — Objection Killer\n**Awareness:** Level 3 (Solution Aware) | **Persona:** Skeptical ADHD Parent/Adult | **Emotion:** Cynicism → Surprise\n\n**Headline:** I've Spent $300 On Alarms. He Slept Through Every One. Then I Tried This.\n\nWon't he just sleep through this too?\n\nThat was my first thought.\n\nMy son has ADHD.\nI have a drawer full of things that were supposed to fix this.\n\nHe solved the math puzzle app in his sleep.\nNo memory.\n\nHe walked across the room to his phone, turned it off, walked back.\nEyes open the entire time.\nHis sleeping body just handles it.\n\nSo another alarm?\nAnother thing that vibrates?\nAnother $39 I'll never get back?\n\nHere's what I didn't understand.\n\nEvery single thing I'd tried uses sound.\nSound goes through one pathway in the brain.\nADHD suppresses that pathway during deep sleep.\n\nIt doesn't matter how loud it is.\nIt doesn't matter how many you set.\nThe door is closed.\n\nThis one doesn't use sound.\nIt vibrates directly on the wrist.\nTouch uses a different pathway entirely.\nOne his brain doesn't shut down during sleep.\n\nThere's no screen.\nNo app.\nNothing to dismiss.\nIt just vibrates until he's conscious.\n\nI ordered it expecting to send it back.\nThat was eleven weeks ago.\nHe has not missed a single morning.\n\n👉 I wrote up what actually changed:\ndawnbands.com/pages/adhd-teen-alarm\n\n60 nights to try it.\nFull refund if it doesn't work.\nI didn't send it back.\n\n---\n\n## VARIANT 3 — Pain-First → Mechanism Reveal\n**Awareness:** Level 2 (Problem Aware) | **Persona:** ADHD Parent who wants to understand | **Emotion:** Confusion → Clarity\n\n**Headline:** His Medication Wears Off at 4pm. By 6:30am He's Unreachable. Nobody Explained That Cycle to Me.\n\nI go into his room at 6:30.\nI say his name.\nNothing.\n\nI go back at 6:35.\nLouder.\nHe grunts.\n\n6:42.\nI'm pulling his covers.\nHe says he's up.\nHe is not up.\n\n6:51.\nI'm a different person now.\nThe voice I swore I wouldn't use.\nThe words I'll replay in the car.\n\nHe finally moves.\nBoth of us wrecked before 7am.\n\nHis ADHD medication wears off by 4pm.\nBy midnight he crashes hard.\nBy 6:30am he's in the deepest sleep of the entire night.\n\nThat's the window.\nThat's when every alarm goes off.\nAnd that's when his brain is least capable of hearing any of them.\n\nNobody explained this to me.\n\nSound alarms fail because ADHD brains close the door on sound during deep sleep.\nIt's not volume.\nIt's not willpower.\nThe signal doesn't reach him.\n\nTouch is a different door.\nOne that stays open.\n\nDawn Bands vibrates on the wrist.\nNo sound to filter.\nNo screen to dismiss in his sleep.\nNo executive function required.\n\nJust a physical sensation his brain can't block.\n\nThree weeks in.\nHe's up before I walk past his room.\nMost mornings I don't even know he's awake until I hear the shower.\n\n👉 Here's the full explanation of why this works for ADHD:\ndawnbands.com/pages/adhd-teen-alarm\n\n$39.\n60-night guarantee.\nWorks the first night.\n\n---\n\n## VARIANT 4 — Consequence/Future Fear\n**Awareness:** Level 2-3 | **Persona:** ADHD Mom with older teen | **Emotion:** Fear → Urgency\n\n**Headline:** He Turns 18 in Two Years. Nobody Is Going to Stand Outside His Dorm Room.\n\nMy son is 16 and he has never once woken up on his own.\n\nNot once.\nIn sixteen years.\n\nWith ADHD it's not getting better.\nThe medication cycle makes 6:30am the worst moment of his day.\nMeds wore off hours ago.\nHis brain is in a trough.\nEvery alarm is hitting a wall.\n\nI've been the thing that gets him up.\nEvery morning.\nFor years.\n\nAnd I do it with frustration because by the fourth time I walk in there, I'm not the mom I want to be.\n\nHe turns 18 in two years.\nNobody is going to stand outside his dorm room.\n\nHis roommate is not going to shake him awake.\nHis professor is not going to wait.\nHis future boss is not going to call four times.\n\nI keep thinking about the adults in ADHD forums who say they've lost jobs over this.\nNot performance.\nAttendance.\nBecause there was no one left to wake them up.\n\nThat's my son in five years if nothing changes.\n\nI can't keep being his alarm clock.\nBut I can't just stop either.\nHe'll miss everything.\n\nI needed something external.\nSomething that works before his medication kicks in.\nSomething that doesn't require him to process or decide or initiate.\n\nDawn Bands vibrates on the wrist.\nTouch, not sound.\nNo screen to dismiss.\nNo executive function needed.\n\nIt reaches him in the window where nothing else can.\n\nFirst week he was up before me every day.\nI didn't go in once.\n\n👉 Here's why this works before medication kicks in:\ndawnbands.com/pages/adhd-teen-alarm\n\n60 nights to try it.\nThe clock is already running on everything else.\n\n---\n\n## VARIANT 5 — Risk Reversal + Social Proof\n**Awareness:** Level 4 (Product Aware) | **Persona:** ADHD parent/adult ready to try | **Emotion:** Confidence → Action\n\n**Headline:** 9,000+ Reviews. 60-Night Guarantee. The Only Alarm Built For ADHD Brains.\n\n⭐⭐⭐⭐⭐\n\"ADHD son, 15. Walked to his phone and turned off 3 alarms in his sleep every morning. Zero memory. This band has no screen. Nothing to turn off. He's been up on his own for a month.\"\n— Laura S.\n\n⭐⭐⭐⭐⭐\n\"Diagnosed at 31. I've lost two jobs because of mornings. Set 11 alarms, slept through all of them. This vibrates on my wrist and I'm just awake. Should have had this ten years ago.\"\n— Ryan M.\n\n⭐⭐⭐⭐⭐\n\"His doctor asked what changed with the mornings. I said a vibrating wristband. She just nodded and said 'keep using it.' First time a solution actually stuck.\"\n— Danielle T.\n\n9,000+ reviews.\n$39.\n60-night money-back guarantee.\n\nIf you have ADHD or love someone who does, you already know.\n\nSound alarms don't work.\nPuzzle apps don't work.\nLouder doesn't work.\nMore doesn't work.\n\nThe reason is simple.\nADHD brains suppress sound during deep sleep.\nEvery alarm you've tried hits the same closed door.\n\nDawn Bands vibrates on the wrist.\nTouch.\nA different pathway.\nNo screen.\nNo dismiss button.\nNothing that needs executive function to process.\n\nTry it for 60 nights.\nFull refund if it doesn't work.\nNo questions. No hoops.\n\nMost people see it work on the first night.\n\n👉 Here's why it works for ADHD when nothing else does:\ndawnbands.com/pages/adhd-teen-alarm\n",
    "sha256": "db92e7f772c9bdba1f17462128b6ba0ff164c74878163354bd9dcb691aa1501b"
  },
  {
    "group": "Native copy",
    "name": "assets/gold-standard-example.md",
    "path": "/Users/ayden/.hermes/skills/openclaw-imports/native-copy/assets/gold-standard-example.md",
    "text": "# Native8 — Teen Mom Listicle — Test Batch (FINAL)\n**Based on:** native8 — $15.25 CPP / 3.04x ROAS / 17.6% CTR (proven winner)\n**Destination URL:** dawnbands.com/pages/7-reasons-teen-sleeps-through-alarms\n**Date:** 2026-03-07\n\n> **Use for structure and cadence, not current product truth or proof.** Before adapting any line, replace price, offer, setup, destination, mechanism, and outcome claims from the current validated brief and canonical product-truth file. Historical first-person outcomes in this sample are not reusable proof unless independently sourced for the new ad.\n\n---\n\n## VARIANT 1 — The Trash Bag (LONG, 500-700 words)\n**Awareness:** Level 3 | **Persona:** Mom who snapped | **Emotion:** Dark satisfaction → explanation → relief\n**Mechanism Delivery:** Analogy (locked door)\n\nMy husband thought I lost my mind.\n\nI was standing in our son's room at 6:47am holding a trash bag.\n\nNot the white kitchen kind.\nThe big black contractor bag.\n\nI'd already knocked four times.\nCalled his name six.\nWent back downstairs twice.\nCame back up screaming once.\n\nHe was asleep through all of it.\n\nHis phone had seven alarms set.\nI could hear every single one from the hallway.\nHe couldn't hear them from two feet away.\n\nSo I started putting his stuff in the bag.\nXbox controller.\nHeadphones.\nThe shoes he'd begged me for.\n\nMy husband walked in and looked at me like I needed help.\n\n\"What are you doing?\"\n\nI didn't answer him.\nI didn't have words for what I was doing.\nI was a 43-year-old woman filling a trash bag at 6am because I'd become someone I didn't recognize.\n\nThe morning turned me into a yelling banshee.\nThat's what I called myself.\nBecause that's what I was.\n\nEvery single morning.\nFor three years.\n\nI tried the Sonic Boom alarm.\nHe slept through it.\nI tried putting his phone across the room.\nHe walked to it in his sleep.\nEyes open. Turned it off. Walked back to bed.\nNo memory of doing it.\n\nI tried the math puzzle app.\nHe solved them unconscious.\nHis sleeping self defeats everything.\n\nI tried cold water.\nTaking his blankets.\nOpening every blind.\nCalm conversations about responsibility.\nConsequences.\nRewards.\n\nNone of this works.\nI've tried all of this and more.\n\nAfter the trash bag morning, I sat in my car in the school parking lot and googled \"why can't my teenager wake up to alarms.\"\n\nNot for the first time.\nBut this time I actually read something that made sense.\n\nIt said sound alarms don't fail because they're too quiet.\nThey fail because the brain locks the door on sound during deep sleep.\n\nLocked.\nNot cracked.\nNot slightly closed.\nLocked.\n\nYou can bang on a locked door all morning.\nYou can scream through it.\nYou can set seven alarms to bang on it for you.\n\nThe door does not open.\n\nIt doesn't matter how loud.\nIt doesn't matter how many.\nThe auditory pathway shuts down in deep sleep, and every alarm on earth uses that same pathway.\n\nThat's why he walks to his phone asleep and turns it off.\nHis body hears enough to react.\nHis brain never wakes up.\n\nThe thing I found uses touch instead of sound.\nVibrates on the wrist.\nTouch goes through a completely different door.\nOne that doesn't lock during sleep.\n\nNo screen. No app. No buttons.\nIt just vibrates until he's actually conscious.\n\nI didn't tell my husband I ordered it.\nI didn't tell my son.\nI put it on his wrist Sunday night and set it for 6:15.\n\nMonday morning I woke up to my own alarm.\nNot his.\nNot yelling.\nMy alarm.\n\nI lay there and waited.\nI heard footsteps.\nHis footsteps.\nBefore I even got out of bed.\n\nI waited for it to fail on Tuesday.\nIt didn't.\n\nWednesday I was sure it was over.\nIt wasn't.\n\nIt's been nine weeks.\nI have not gone into his room once.\n\nThe trash bag is still in my closet.\nHis stuff is back in his room.\nHe doesn't know why I took it.\nI don't want him to.\n\nI don't yell anymore.\nThe first version of me he sees isn't angry.\nThat's it.\nThat's all I wanted.\n\n👉 Here's what I read:\ndawnbands.com/pages/7-reasons-teen-sleeps-through-alarms\n\nI had 60 nights to send it back.\nNever even considered it.\n\n**Headline:** I filled a trash bag with his stuff at 6am. That was the morning I stopped using sound alarms.\n\n---\n\n## VARIANT 2 — The Apology / 847 Mornings (LONG, 500-700 words)\n**Awareness:** Level 3 | **Persona:** Mom reckoning with guilt | **Emotion:** Guilt → understanding → forgiveness\n**Mechanism Delivery:** Full science\n\n847 mornings.\n\nI counted.\n\nFrom the September he started high school to the Tuesday I finally understood what was happening, that's how many mornings I went into his room and turned into someone I don't want to be.\n\n847 mornings of knocking.\nThen calling.\nThen louder.\nThen yelling.\nThen both of us wrecked before 7am.\n\nI was a human alarm clock.\nAnd the worst part is the awake version of my son does not remember the previous 9 attempts.\nOnly the rageful 10th one.\n\nThat's the only version of me he knows in the morning.\nAngry. Loud. Done.\n\nI wake up before my alarm now.\nNot because I'm rested.\nBecause my body knows what's coming.\nI lie there calculating how much buffer I have before I have to start.\n\nAmong the many hats I wear, the one that fits tightest is NAG.\nI don't want him to wake up to my anger.\nBut I don't know how to wake him up without it.\n\nHis phone has seven alarms.\nThey've woken ME up before they wake him.\nHe shuts them off and goes back to sleep.\n\nI put the phone across the room.\nHe walked to it in his sleep.\nEyes open the whole time. Turned it off. Walked back.\nNo memory.\n\nI demand to hear evidence he is sitting up and out of bed.\nHe says he is.\nRepeat 10 times over half an hour.\nHis sleeping self tells me he is awake.\n\nI tried everything.\nSonic Boom alarm.\nThree Alexas.\nMath puzzle apps he solved unconscious.\nCold water.\nLights on.\nTaking his phone.\nConsequences.\nIncentives.\nCalm conversations.\n\nNone of this works.\nI've tried all of this and more.\n\nLast February his English teacher pulled me aside at conferences.\n\n\"Is everything okay at home? He seems checked out in first period. Every day.\"\n\nI almost cried in the hallway.\n\nThat was the night I read something that finally explained it.\n\nThe teenage brain suppresses the auditory processing pathway during deep sleep.\nIt's not laziness.\nIt's not defiance.\nIt's neurology.\n\nSound alarms use one sensory channel to reach the brain.\nDuring deep sleep, that channel shuts down.\nCompletely.\n\nIt doesn't matter how loud the alarm is.\nIt doesn't matter how many you set.\nIt doesn't matter if he has to solve equations to dismiss it.\nHis brain processes enough to react physically.\nBut never enough to reach consciousness.\n\nThat's why he walks to his phone asleep.\nThat's why he answers me while unconscious.\nThat's why nothing has worked.\n\nEvery single thing I tried for 847 mornings used sound.\nEvery single one went through the same blocked pathway.\nI wasn't failing.\nThe method was.\n\nTouch uses a different sensory pathway.\nOne the brain does not suppress during sleep.\nVibration on the skin reaches the brain through a channel that stays open.\n\nI found a wrist band that vibrates.\nNo sound. No screen. No app.\nNothing for his sleeping self to dismiss.\nIt just vibrates on his skin until he's conscious.\n\nI put it on him on a Sunday.\n\nDay 1: He was in the kitchen before me.\nDay 4: He made himself breakfast.\nDay 11: His teacher emailed me.\n\"Whatever you changed, it's working. He's a different kid in first period.\"\n\nShe noticed before I even said anything.\n\nIt's been fourteen weeks.\nHe hasn't missed a single morning.\n\nBut that's not the part that made me cry.\n\nLast Thursday he came downstairs and said something to me before I said anything to him.\n\n\"Morning, Mom.\"\n\nHe hadn't done that in over a year.\n\nI'll put the link below.\n\ndawnbands.com/pages/7-reasons-teen-sleeps-through-alarms\n\n$49.\nFull refund if it doesn't work within 60 nights.\n\n847 mornings of doing it wrong.\nIt took one night of doing it differently.\n\n**Headline:** 847 mornings I yelled him awake. I wish I'd known it wasn't his fault.\n\n---\n\n## VARIANT 3 — The Neighbor (LONG, 500-700 words)\n**Awareness:** Level 2 | **Persona:** Mom who saw herself from outside | **Emotion:** Humiliation → determination → calm\n**Mechanism Delivery:** Someone else explained it (pediatrician)\n\n\"Do you need help? I hear screaming from your house every morning at 7am.\"\n\nMy neighbor said that to me over the fence on a Saturday.\n\nA Saturday.\n\nThe one morning I wasn't screaming.\n\nI laughed it off.\nMade some joke about teenagers and alarm clocks.\nWalked inside and locked the bathroom door and sat on the floor.\n\nBecause she was right.\n\nEvery morning at 6:45 it starts.\nI knock.\nI call his name.\nI wait.\nI go back.\nI call louder.\nI yell.\n\nI call him for 2 hours straight some mornings.\n\nHis alarm has woken ME up before it wakes him.\nSeven alarms on his phone.\nI hear all of them.\nHe hears none.\n\nHe shuts them off and goes back to sleep.\n\nI put the phone across the room.\nHe got up, walked to it, turned it off, walked back to bed.\nEyes open.\nFull conversation with me.\nNo memory of any of it.\n\nHis sleeping self tells me he is awake.\nHis sleeping self lies.\n\nI became ridiculously rage-y.\nThat's the word I used when I told his pediatrician.\nI said I'm becoming a person I hate at 6am and I can't stop it.\n\nShe didn't tell me to try harder.\nShe didn't suggest another alarm app.\n\nShe said something no one had ever explained to me.\n\n\"His brain blocks sound during deep sleep. It's a sensory pathway issue. The auditory channel closes. It doesn't matter how many alarms you set or how loud they are. They all go through the same channel. And for him, that channel is shut.\"\n\nI sat there staring at her.\n\nThree years.\nSonic Boom alarms.\nMultiple Alexas in his room.\nMath apps he solved in his sleep.\nCold water.\nBlinds open.\nLights blazing.\n\nAll sound.\nAll the same blocked channel.\nEvery single one.\n\nShe told me about something that uses touch instead.\nA band that vibrates on the wrist.\nTouch goes through a different pathway.\nOne that stays open during deep sleep.\n\nNo screen. No buttons. No app.\nNothing for his unconscious body to dismiss.\n\nI ordered it from the car in the parking lot.\n\nThe first morning I didn't go into his room.\nI heard him get up.\nI heard the shower.\nI stood in the kitchen holding my coffee with both hands and I didn't move.\n\nThe second morning was the same.\nThe third.\nThe seventh.\n\nNo screaming.\nNo knocking.\nNo third trip upstairs.\nNo yelling through the door.\nNo standing over his bed hating who I was becoming.\n\nWeek four, I was outside watering the plants.\nMy neighbor came to the fence again.\n\n\"It's been quiet over there in the mornings.\"\n\nI said yeah.\nThat's all I said.\n\nIt's been twelve weeks.\nHe has not missed one morning.\n\nThe explanation is at dawnbands.com/pages/7-reasons-teen-sleeps-through-alarms\n\nMy neighbor doesn't hear me anymore.\nNobody does.\n\n**Headline:** My neighbor heard me screaming every morning at 7am. She doesn't hear anything now.\n\n---\n\n## VARIANT 4 — The Drawer (SHORT, 150-300 words)\n**Awareness:** Level 3 | **Persona:** Skeptic burned by failed products | **Emotion:** Dark humor → surprise\n**Mechanism Delivery:** Barely explained\n\nI have a junk drawer that costs more than my groceries.\n\nSonic Boom alarm. $45. He slept through it.\nPhone alarm dock. $30. Slept through it.\nMath puzzle app. $12. Solved it in his sleep. No memory.\nSunrise lamp. $55. Decorative.\nThree Alexas. $90 total. They wake me up. Not him.\n\n$232 of things that were supposed to fix mornings.\n\nAll of them use sound.\nTurns out his brain shuts sound off during deep sleep.\nDidn't matter how loud.\nDidn't matter how many.\n\nHe walked across his room to his phone in his sleep.\nEyes open. Turned it off. Walked back.\nZero memory.\n\nHis sleeping body defeats everything.\n\nThis one vibrates on his wrist.\nNo sound. No screen. Nothing to dismiss unconscious.\n\nI ordered it expecting it to go in the drawer.\n\nThat was ten weeks ago.\n\nIt's not in the drawer.\n\nHe gets up before I go in.\nEvery morning.\nI haven't yelled since November.\n\nLink is here if you want to read it:\ndawnbands.com/pages/7-reasons-teen-sleeps-through-alarms\n\n$49. 60 nights. Full refund if it joins the drawer.\n\nIt didn't.\n\n**Headline:** $232 in that drawer. $49 on his wrist. Guess which one works.\n\n---\n\n## VARIANT 5 — The Car (SHORT, 150-300 words)\n**Awareness:** Level 4 | **Persona:** Mom in school parking lot | **Emotion:** Raw → relief\n**Mechanism Delivery:** Skip entirely (outcome only)\n\nHe got out of the car without looking at me.\n\nDidn't say bye. Didn't say anything.\n\nI sat in the school parking lot for ten minutes replaying everything I'd said that morning.\n\nThe knocking.\nThe yelling.\nThe \"I've told you TEN TIMES.\"\nThe door slam.\n\nThe first version of me he sees every day is angry.\nThat's the only mom he knows before 8am.\n\nI was a human alarm clock.\nAnd the alarm was my voice at full volume.\n\nI don't want him to wake up to my anger.\nBut I didn't know how else to get him up.\n\nSeven phone alarms.\nHe sleeps through them.\nI hear every one.\n\nI tried everything for two years.\nNothing worked.\nI gave up trying to fix it and started trying to survive it.\n\nSomeone posted this in a mom group I'm in.\nI almost scrolled past it.\n\nI didn't expect anything.\n\nFirst morning: he was up before I went in his room.\nSecond morning: same.\nThird morning: I heard him in the shower.\n\nIt's been eight weeks.\n\nYesterday he got out of the car and turned around.\n\n\"Bye, Mom.\"\n\nTwo words.\nI cried in the parking lot again.\nDifferent reason.\n\nSomeone posted this in my group:\ndawnbands.com/pages/7-reasons-teen-sleeps-through-alarms\n\nP.S. He doesn't know I used to sit in that parking lot after drop-off.\nHe doesn't know I cried.\nHe just knows mornings are different now.\n\n**Headline:** He stopped saying bye to me. That was the part that broke me.\n",
    "sha256": "ea237599482d40993e7a0d336c5910c0bbb832181b68fcd1cc12c68f5fe2a056"
  },
  {
    "group": "Native copy",
    "name": "references/adversarial-batch-copy-audit.md",
    "path": "/Users/ayden/.hermes/skills/openclaw-imports/native-copy/references/adversarial-batch-copy-audit.md",
    "text": "# Adversarial Batch Copy Audit Protocol\n\nUse this when auditing a finished native-ad batch against structured briefs and VOC receipts. This is an editorial/provenance audit, not a deterministic artifact check.\n\n## Core distinction\n\nA deterministic PASS proves file count, required fields, word ranges, formatting, offer order, destinations, and banned-term rules. It does **not** prove that a first-person story is true, that precise details are sourced, that the avatar voice is faithful, or that twenty ads are genuinely diverse.\n\nAlways report these as separate results:\n\n1. **Artifact-contract status** — objective checks.\n2. **Editorial/provenance status** — hook, voice, persuasion, truth, and diversity.\n\nNever let artifact PASS launder an editorial FAIL.\n\n## Evidence hierarchy\n\nFor every first-person scene, quotation, outcome, time, count, price, duration, product trial, or external witness:\n\n1. Find an exact source receipt for the same speaker and fact.\n2. If only the behavior or emotion is sourced, the assembled narrator is a **composite**, not a testimonial.\n3. A creative brief authorizing an artifact (for example, “five-day log” or “DORM box”) does not prove that the event happened.\n4. Plausibility is not provenance.\n5. Safe repairs are:\n   - attribute exact VOC (`A parent wrote…`);\n   - frame it explicitly as a hypothetical (`Picture…`, `Before you pack…`, `Try writing…`);\n   - use a non-testimonial ownership argument;\n   - remove unsupported precision.\n\n### Precision test\n\nTreat specificity as proof-shaped. Numbers such as “seven minutes,” “three problems,” “six steps,” or a minute-by-minute sequence require direct receipts. If the exact number is unsupported, remove it rather than replacing it with a different invented number.\n\n## Required scoring layers\n\nScore every ad independently from 1–10 on:\n\n- Hook WTF factor\n- Hook specificity\n- Hook curiosity gap\n- Lines 3–8 specificity\n- AI-sounding language, abstraction, and filler\n- Awareness/sophistication alignment\n- Resignation moment, villain, root cause, and mechanism chain\n- Avatar fidelity against exact VOC\n- Victory as absence\n- Claim and testimonial integrity\n- CTA and body/headline/description division\n- Cross-ad distinctiveness\n\nAnything below 7 must include exact proposed replacement line(s), not only diagnosis. If the task is a read-only audit, label replacements as proposals and do not edit source ads.\n\n## Blocking vs taste\n\n### High-confidence blockers\n\n- Invented or composite first-person experience presented as literal testimony\n- Unsupported outcome, chronology, dialogue, time, count, or external witness\n- Medical/every-time claims beyond approved proof\n- Wrong awareness level or missing governing objection\n- Repeated body/mechanism blocks that collapse supposedly distinct ads\n- Failure to satisfy an explicit brief requirement\n\n### Subjective taste flags\n\n- Hook preference when multiple options are defensible\n- Degree of quietness, humor, or emotional intensity\n- Stylistic wording where truth and brief alignment already pass\n- Optional field-role improvements\n\nAlways label confidence so editorial preference is not presented as proof failure.\n\n## Cross-ad diversity check\n\nRun both exact and conceptual checks:\n\n- repeated full sentences;\n- repeated hook formulas;\n- repeated mechanism cadence;\n- repeated victory artifacts (door, footsteps, bathroom water, hot coffee);\n- narrator sameness across parent, teen, partner, or other personas;\n- repeated description/headline restatement.\n\nAssign each ad one proof artifact or mechanism image it uniquely owns. Repetition required by canon (for example, an exact guarantee token order) is less concerning than repeated narrative prose.\n\n## Performance boundary\n\nHistorical performance can support hypotheses about territories or devices. It cannot prove that a current hook, angle, or wording is a winner. State stale/current boundaries explicitly.\n\n## Deliverable verification\n\nBefore delivery verify:\n\n- every expected ad has one score row and one detailed section;\n- every sub-7 score has exact proposed text;\n- every blocker includes snippet, rationale, and confidence;\n- invented testimonial risk is explicitly addressed;\n- high-confidence blockers are separated from taste flags;\n- no source ad was edited when the assignment was read-only.\n",
    "sha256": "7432cee3f45b7baff3938152d73583ba76279c1849767e8b8e41ff0af0379153"
  },
  {
    "group": "Native copy",
    "name": "references/alevia-live-anatomy-2026-08-12.md",
    "path": "/Users/ayden/.hermes/skills/openclaw-imports/native-copy/references/alevia-live-anatomy-2026-08-12.md",
    "text": "# Alevia live-account anatomy — measured update (2026-08-12)\n\nSource: 240 unique active ad copies sampled across all 14 Alevia pages (TrendTrack domain query,\nlast-3m actives), raw JSON at `~/.openclaw/workspace/output/trendtrack-alevia-2026-08-12/`.\nFull analysis: `~/.openclaw/workspace/knowledge/brands/dawnbands/alevia-native-anatomy-2026-08-12.md`.\n\nThis file UPDATES the two-control evidence base that `dawnbands-long-form-witness-native.md` was\nbuilt from (those controls: ~3,188/3,662 words, ~60% reveal). It is measurement, not new contract:\nwhere a number below differs from the Mode B contract, the contract stays default and the measured\nband is a SANCTIONED TEST VARIABLE. Provenance, narrator-authority, and claim-ceiling gates are\nuntouched by everything here.\n\n## Measured distributions (n=240 unless noted)\n\n- **Length:** p10 2,373 / median 3,411 / p90 4,338 words. 229/240 are 1,500+. Our 2,500–4,000\n  contract sits inside their live band; their upper half (4,000–5,000) is unexplored by us.\n- **Reveal position:** median 52%, p10 31%, p90 66%. The two July controls (~60%) sit high in\n  the live distribution. Contract stays 55–70%; **45–55% is a sanctioned test band**, not a defect.\n- **Formatting (settles the line-format question):** paragraph prose, median ~31 words/paragraph\n  (2–4 sentences). 133/240 mix SHORT staccato beats at punch moments only. Zero copies are\n  wall-to-wall one-sentence-per-line. This matches the 2026-07-31 Mode B ruling (PARAGRAPH prose)\n  and supersedes the older sentence-per-line formatting block for Mode B. Sentence-per-line\n  remains correct for Mode A / compressed natives and static primaries.\n- **Register hygiene (long-form):** emoji 12/240 (excluding the final 👉 link arrow), ALLCAPS\n  lines 6/240, comment-bait 25/240. The witness lane is kept clean of ad-register signals.\n- **Headline field:** 0/240 witness natives use the Meta title field. Observation only — the\n  5-headlines-per-ad rule stands; a blank/minimal-title witness cell is a legitimate test.\n\n## Opener — measured ceiling of the rupture-object cold open\n\nAll 20 longest-running copies (95–127d) open the same way: **mundane object + weekday + month +\nplace-name in sentence one, confession register, zero product/category scent.**\n\n> \"I pulled my mother of the bride dress out of the garment bag on a Tuesday afternoon in March…\"\n> \"The Christmas card I mailed to my cousin Patty in November came back… a yellow sticker that said DECEASED.\"\n> \"My dad couldn't open the register at the pizza shop last Tuesday… the same brass-plated NCR since 1987.\"\n\nAudit line for every Mode B opening: **could this first sentence appear in a personal Facebook\npost with zero ad intent?** If it reads like a hook, it fails.\n\nProvenance caveat: that level of specificity (weekday, month, place, object) requires `RECEIPT`\nor explicit `AUTHORIZED DRAMATIZATION` per the provenance ledger. The formula does not license\ninvented precision — it tells you what the receipt-backed or authorized detail should LOOK like\nwhen you have it.\n\n## Reveal — the spoken-word mechanic\n\nThe product never enters as a decision. In every measured survivor it enters as a **word spoken\nby a named third party inside the story**, often at the end of a narrated discovery chain:\n\n> \"She told us about a fruit called Amla.\" · \"Renée told me about Amla.\" ·\n> \"…a thread on a health forum… a comment from a woman named Patrice…\" ·\n> \"I found my way to amla through a Reddit thread… written by a woman in Ohio…\"\n\nTwo refinements for Beat 6/9 of the witness architecture:\n1. **Discovery chain**: article → forum thread → named stranger's comment → skeptical read → try.\n   Narrate the chain itself; it does DR work (relatability + implied social proof without a claim).\n2. **Category-then-brand split**: mid-story reveal names the CATEGORY/mechanism object (for us:\n   the wrist-worn silent alarm as a thing someone mentions), the BRAND surfaces only at the\n   link/close. Never brand-name the mid-story reveal.\n\nBoth routes obey the existing gates: the named stranger is dialogue inside the story (allowed),\nnever an authority byline or a fabricated named testimonial with product outcomes (banned).\n\n## Close — the P.S. ladder (measured default, our biggest structural gap)\n\n217/240 copies close with stacked postscripts, up to P.P.P.P.S. Each rung has a distinct job:\n\n1. **P.S. — rupture-object return.** (\"The dress is hanging in the closet now…\")\n2. **P.P.S. — guarantee naturalized in-voice.** (\"The 90-day money-back guarantee is real. I asked.\")\n   For DawnBands: only the verified refund terms, usually 60-night money-back; all offer gates apply.\n3. **P.P.P.S. — objection/placebo kill in-voice.** (\"Someone is going to say 'it's just placebo.'…\")\n4. **Final P.S. — reader-direct-address AS the CTA.** (\"If you're the husband reading this over\n   your wife's shoulder… don't make her wait for the ambulance.\") For us: the mom reading at 1 AM\n   with tomorrow's alarm already set.\n5. **Bare 👉 link on its own line AFTER the final coda.** No CTA copy around it.\n\nDifference vs the existing close architecture (motif → loss → role → offer → CTA, codas optional):\nin the measured winners the codas ARE the close and the link comes last. Treat the ladder as\n**close variant B** — test it against the existing close on a proven copy before making it default.\nThe existing outcome-timeline, offer-verification, and desired-absence gates apply to every rung.\n\n## Device frequencies (angle-selection input, not copy instructions)\n\nOf 95 primary-page copies: spouse-witness 57 · death-stakes 54 · family-milestone 41 ·\nwrong-diagnosis reveal 25 · authority-defiance 25 · insider-authority 14. Failed-solution stack\nwith prices 70/95. DawnBands transfer analysis + which of these are banned for us (death-ledes,\nfabricated medical authority) lives in\n`~/.openclaw/workspace/knowledge/brands/dawnbands/alevia-application-recs-2026-08-12.md`.\n\n## Format-page discipline (context for deployment, not copy)\n\nAlevia never crosses formats across page identities: witness natives on community/persona pages,\n55–84-word ✅-checklist shorts on brand pages (their longest-running ads, 118–127d — see T4 in\n`avelia-short-native-templates.md`), video quarantined on one page with ~3 copy skeletons.\nOur page axis in the ad-batch-hub matrix now tracks the same variable.\n",
    "sha256": "e82e0c80d37209d837f908a3e552eed8657b2547e912364719d59aeed619ec7b"
  },
  {
    "group": "Native copy",
    "name": "references/avelia-short-native-templates.md",
    "path": "/Users/ayden/.hermes/skills/openclaw-imports/native-copy/references/avelia-short-native-templates.md",
    "text": "# Avelia-derived short native template system\n\nUse this reference inside the class-level `native-copy` workflow when analyzing competitor short-form native ads or drafting compressed Meta copy from trigger-specific, testimonial, or failed-solution entry points.\n\n## Evidence boundary\n\n- **Observed:** exact advertiser/page/shop/domain fields, ad IDs, body/card fields, dates, statuses, counts, and null metrics from source receipts.\n- **Inferred:** awareness mapping, persuasion rationale, recommended word ranges, template selection, and adaptation advice.\n- Duplicate count, longevity, and zero/null reach are not performance proof.\n- Source claims never transfer to the target brand without independent verification.\n\n## Required inputs\n\nBrand/product/destination; exact avatar context; awareness; root cause → product mechanism → differentiator; verified claims and sources; quantified facts or explicit empty list; guarantee/effort/timing; claim restrictions; requested fields.\n\nStop if mechanism, context, proof inventory, restrictions, or destination-awareness fit are missing. Never invent ingredients, statistics, timelines, guarantees, dosage, clinical status, or outcomes.\n\n## Workflow\n\n1. In analysis mode, verify advertiser identity, page ID, shop/domain, destination, exact ad IDs, and direct receipts. Preserve nulls and source field semantics.\n2. Name what the reader already tried or believes; select one narrow trigger/context.\n3. Choose exactly one template below. Do not hybridize until each passes independently.\n4. Label skeleton beats `[FACT]`, `[VOC]`, `[INFERENCE]`, or `[PROHIBITED]`; delete prohibited beats.\n5. Assign field roles before prose:\n   - primary text = tension + mechanism + product bridge + close;\n   - headline = concise angle only when the target platform field exists;\n   - description = proof, guarantee, effort, or quantified differentiation;\n   - CTA = awareness- and register-matched.\n6. Run claim, specificity, field-complementarity, and cross-variant diversity QA.\n7. Return every requested field plus a verification receipt.\n\n## T1 — Trigger-Specific Diagnosis → Mechanism → Compressed Remedy\n\nBest for problem-aware readers who recognize a life trigger but not the mechanism.\n\n```text\n[Specific life event or sustained context].\nNow [visible symptom/behavior], and [betrayal or nobody-warned-me gap].\n\n[Verified prevalence, threshold, or mechanism].\nIt's not [default explanation].\n[Cost of delay or why prior assumption fails].\n\n[Product] was designed/formulated for [verified mechanism].\n[Low-effort regimen].\n[Verified result framing or non-outcome benefit].\n\n[Resolve the opening tradeoff].\n```\n\nObserved Avelia range: 85–96 words; product appeared 57–62% into body words. Treat this as evidence, not a mandatory threshold. Inferred reusable target: 80–110 words.\n\n## T2 — Result Confession → Authority Bridge → Benefit Stack\n\nBest for solution-/product-aware readers needing a compact result and reasons to believe.\n\n```text\nThis [format]? [Natural verdict, no cliché] for [specific result].\nI've noticed [detail 1], [detail 2], and [detail 3], even during [adverse context].\n\n[Brand/product] uses [verified mechanism] to [verified action].\n[Connect format to convenience or desired state].\n\n• [Benefit 1]\n• [Benefit 2]\n• [Benefit 3 / objection handling]\n\n[Soft, specific desired-state CTA].\n```\n\nObserved Avelia length: 106 words. Inferred reusable target: 90–120. Never leave placeholders or use “game-changer.”\n\n## T3 — Failed Category Reframe → Quantified Mechanism Stack\n\nBest for skeptical solution-aware readers who tried the category and felt nothing.\n\n```text\nYou didn't need MORE [category].\nYou needed the right [verified mechanism variable].\n\n[Product]: [count/specification], [dose/time], [convenience], designed to [verified benefit domain].\n```\n\nObserved Avelia range: 32–46 words. Inferred reusable target: 30–55.\n\n## T4 — Branded Checklist Evergreen (added 2026-08-12, measured from live actives)\n\nBest for product-/offer-aware traffic on the BRAND page only (never the community/persona\npages). Measured basis: Alevia's longest-running ads of any format (118–127 days live) are\n54–84-word checklist shorts on their brand-identity pages; their new product lanes launch on\nthe same skeleton unchanged (`alevia-live-anatomy-2026-08-12.md`).\n\n```text\n[Product/category], the [one-line identity + verified mechanism clause], naturally/simply.\n\n✅ [Benefit 1 — verified or non-outcome]\n✅ [Benefit 2]\n✅ [Benefit 3]\n✅ [Benefit 4 / objection kill]\n\n[One heritage/credibility line IF verifiable — otherwise omit].\n\nTry [product] risk-free for [verified refund window] or get 100% of your money back[, no\nquestions asked — only if the verified policy says so].\n```\n\nObserved range: 54–84 words. Emoji checklist is the register for THIS template (it is banned\nin long-form witness copy). Every benefit line passes the claim gates; the refund line uses\nonly currently verified policy terms (DawnBands: 60-night money-back, $49 only if current).\nRole: the evergreen BOF/retarget floor — build 3–4, leave them running, never on story pages.\n\n## Anti-patterns\n\n- Same-name advertiser mixing without identity verification.\n- Relabeling card descriptions as platform headlines.\n- Unsupported clinical claims, prevalence, dosage, timelines, “no side effects,” or guaranteed outcomes.\n- Multiple root causes in one short ad.\n- Generic symptom hook when a trigger-specific hook exists.\n- Repeating the same claim across primary text, headline, and description.\n- Importing early reveal timing into a long-form cold story.\n- Placeholders, clichés, decorative emoji clutter, dense paragraphs, or direct SHOP NOW language that breaks a quiet register.\n\n## Verification receipt\n\n```markdown\n- Template: T1 | T2 | T3\n- Awareness: ...\n- Product reveal: sentence N / approximately X%\n- Primary text words: N\n- Field completeness: primary/headline/description/CTA/URL\n- Claims checked against: [sources]\n- Unsupported claims removed: [list]\n- Hook/body/card relationship: [one sentence]\n- QA result: PASS | FAIL\n- Remaining uncertainty: [list]\n```\n\n## Evidence note\n\nSession evidence dated 2026-07-18 contained 62 deduplicated Avelia/Avelia Care ads and 9 exact primary-text variants. Two current-page ads were directly scanned active; the remainder were linked-page records. Numeric reach/scaling evidence was absent. This reference preserves structures, not winner claims.",
    "sha256": "4f593a181226ce6242f99aefeebf6a2b09d2a114c3ca8d9fe90106d418bfd767"
  },
  {
    "group": "Native copy",
    "name": "references/consolidated-native-ad-repair-pass.md",
    "path": "/Users/ayden/.hermes/skills/openclaw-imports/native-copy/references/consolidated-native-ad-repair-pass.md",
    "text": "# Consolidated Native-Ad Repair Pass\n\nUse this when completed ads have multiple read-only QA reports and the user requests one repair plus revalidation pass.\n\n## Authority and scope\n\n1. Re-read the live ad files immediately before editing.\n2. Extract only the requested brief objects and their exact VOC receipts.\n3. Consolidate findings from formal QA, sharpening/avatar review, deterministic audit, and any later SOP/performance overlay.\n4. Carry forward every unresolved blocker unless a later report explicitly retires it.\n5. Edit only the authorized ad files and manifest. Do not create a new audit report unless requested.\n\n## Repair order\n\n1. **Truth first:** remove composite narrators, invented chronology, precise unsupported times/counts, and fabricated dialogue.\n2. **Quote fidelity:** text in quotation marks must match one receipt exactly. Otherwise remove quotation marks and attribute the paraphrase. An excerpt must remain a literal source substring: never add a period inside the closing quote when the source sentence continues. End the exact fragment without invented punctuation, include the full source sentence, or paraphrase it unquoted.\n3. **Brief fidelity:** restore assigned mechanism delivery, structure, awareness, objection, word range, CTA, and destination.\n4. **Craft:** sharpen hooks, remove strategy-deck language, and preserve the requested short/long architecture.\n5. **Convergence:** assign each ad a distinct territory, narrator cadence, mechanism expression, and payoff image.\n\n## Truth-safe conversions\n\n- Fabricated experiment → explicit hypothetical planning device: `Try writing...`, `Picture...`, or `Use this question...`.\n- Composite first-person story → exact attributed VOC plus an ownership argument.\n- Invented dialogue → delete it or replace it with exact, source-mapped dialogue.\n- Unsupported result → desired absence, not a claimed customer outcome.\n- Unsupported timestamp/count → remove precision unless the value appears in the receipt or governing brief as a non-testimonial scenario.\n\n## Deterministic revalidation\n\nFor every repaired ad, calculate from the live Primary Text rather than trusting labels:\n\n- actual word count against the ad-specific range\n- required fields and exactly one finished ad\n- maximum two sentence endings per non-heading line\n- isolated dialogue\n- no em dash, banned words, stale price, or prohibited claims\n- exact offer token order across line breaks\n- exact CTA and destination\n- declared source VOC IDs\n- no-app/no-subscription consistency\n- mechanism-delivery compliance\n- exact quote presence for quoted VOC, with each excerpt verified as a literal source substring\n- explicit hypothetical framing where an artifact could read as a completed testimonial\n- no repeated substantive sentence across repaired ads, excluding mandated offer language\n\nUpdate the in-file verification count before hashing.\n\nUse this stable order after the final editorial change:\n1. Parse Primary Text and calculate actual counts.\n2. Update in-file count declarations.\n3. Re-run contract, banned-language, sentence-line, quote-fidelity, and cross-ad duplicate-line checks.\n4. Calculate final file, headline, and hook hashes.\n5. Rebuild the manifest from those final files.\n6. Re-run validation against the manifest.\n\nA count-label edit changes the file hash. Never calculate file hashes before count labels are final.\n\nThen regenerate the manifest from the final bytes, recording final word counts, checks, source IDs, and SHA-256 values.\n\n## Final verification\n\nRe-open the manifest and all repaired ads, recompute hashes and counts, and fail delivery if any manifest value is stale.\n\nFor source-sensitive repairs, add deterministic assertions for the actual blocker, such as:\n\n- exact approved quotation is present\n- known fabricated dialogue is absent\n- a `skip` mechanism ad contains no channel/placement explanation\n- a hypothetical log explicitly says it is hypothetical\n- an unsupported personal premise from the failed draft is absent\n\nReport only final status, files changed, material repair classes, and unresolved issues.\n",
    "sha256": "af41b77f2bec29009f99b17288673f6ae59bb77a23af1783465d8549d22fb80f"
  },
  {
    "group": "Native copy",
    "name": "references/control-format-selection-and-pilot-gate.md",
    "path": "/Users/ayden/.hermes/skills/openclaw-imports/native-copy/references/control-format-selection-and-pilot-gate.md",
    "text": "# Control-Format Selection and Long-Form Pilot Gate\n\nUse this when the user supplies example ads, says “like these samples,” asks for personal anecdotes, or requests many long-form natives.\n\n## 1. Infer format from the controls before choosing a mode\n\nMeasure the supplied controls rather than relying only on labels such as “native,” “copy,” or “variants.” Record:\n\n- primary-text word count\n- narrator type and point of view\n- whether the ad is scene-led or argument-led\n- product-name reveal percentage\n- opening object / incident\n- failed-solution depth\n- motif return and closing architecture\n\nTreat the samples as a format contract when their shape is unambiguous.\n\n### Automatic Mode B signals\n\nSelect DawnBands Long-Form Witness mode without re-asking when the request or controls clearly call for:\n\n- roughly 2,500–4,000-word executions\n- first-person personal anecdotes or witness stories\n- a lived world and progressive chronology before mechanism\n- product reveal around 55–70%\n- a concrete rupture object that returns at the close\n\nA request for “ten different angles” changes the story contracts, not the length or narrative architecture demonstrated by the supplied controls.\n\nDo not compress 3,000-word personal-story controls into 200–700-word ownership arguments merely because the user did not type the internal mode name.\n\n## 2. Narrator selection\n\nWhen the user explicitly requests personal anecdotes and authorizes creative ad writing, `authorized-dramatization` may be the correct internal narrator mode if no verified buyer receipt exists.\n\nKeep behaviors grounded in avatar research and do not invent DawnBands outcomes, medical authority, quantified proof, or a named fake testimonial identity. The consumer-facing copy should still read as one coherent lived story. Never leak internal provenance policy into prose.\n\nDo not default to detached ownership voice when that would violate the sample’s first-person anecdotal format.\n\n## 3. Pilot before batch scale\n\nFor a batch of long-form ads:\n\n1. Build distinct story contracts for the whole batch.\n2. Draft one complete pilot with the requested model.\n3. Audit three windows, not just the hook:\n   - opening 200–300 words\n   - product reveal ±500 words\n   - final 200–300 words\n4. Verify word count, reveal percentage, narrator intimacy, motif continuity, and lack of compliance/provenance leakage.\n5. Only then route the remaining lanes.\n\nA strong opening is not enough. Models often converge after the reveal into repeated phrases, identical mechanism lists, generic risk reversal, and cloned P.S. stacks.\n\n## 4. Back-half diversity gate\n\nAcross the finished batch compare:\n\n- first product-reveal sentence\n- bridge into product\n- mechanism analogy\n- objection order\n- offer phrasing\n- CTA architecture\n- final motif return\n\nIf the lives differ but the reveals and closes share a template, preserve the story fronts and run same-model surgical rewrites from the first product mention through the end. Recount words and reveal percentages afterward.\n\nUseful deterministic signal: five-gram overlap can identify convergence, but a human read of opening, reveal, and close remains mandatory.\n\n## 5. Delivery\n\nPreserve immutable raw model drafts, save same-model repairs separately, rerun deterministic QA, and publish a readable public report when the user expects to review thousands of words.\n",
    "sha256": "a4c2ee41b3fd9a6e231a722b4d5ab70cbd93a2d014e3bcbd1f1c1db1d6b05926"
  },
  {
    "group": "Native copy",
    "name": "references/controlled-live-meta-repair.md",
    "path": "/Users/ayden/.hermes/skills/openclaw-imports/native-copy/references/controlled-live-meta-repair.md",
    "text": "# Controlled Live-Meta Repair for a Brief-Driven Ad Batch\n\nUse this when repairing an existing native-ad batch after editorial QA and a current Meta evidence overlay identify a validated mechanism lane.\n\n## Authority order\n\n1. User scope and file ceiling.\n2. Per-ad validated brief and current canonical product truth.\n3. Verified VOC/source receipts.\n4. Formal QA and sharpening blockers.\n5. Current live-Meta evidence as a prioritization overlay, not permission to override the brief.\n6. General copy doctrine and examples.\n\n## Repair sequence\n\n1. Read each live ad immediately before editing.\n2. Consolidate all existing blockers and overlay deltas into one repair pass.\n3. Preserve each ad's persona, awareness, structure, word range, CTA, destination, and source IDs unless the brief explicitly permits a change.\n4. Convert unsupported first-person scenes, exact chronology, dialogue, and proof-shaped numbers into one of:\n   - attributed exact evidence;\n   - explicit hypothetical framing;\n   - a non-testimonial ownership argument;\n   - desired-absence framing.\n5. Allocate only the requested number of controlled evidence-backed lanes. Do not convert the whole batch into the current winner pattern.\n6. For two controlled ADHD biology-to-wrist lanes, keep this strategic sequence constant:\n   - discipline/laziness shame relief;\n   - concrete sourced alarm-stack recognition;\n   - qualified biology/root-cause reframe;\n   - why repeated sound cues fail;\n   - direct wrist-vibration bridge;\n   - parent-convenience or stop-being-the-alarm payoff;\n   - exact risk reversal.\n7. Vary entry artifact, narrator mode, cadence, and mechanism phrasing between the controlled lanes. Declare the controlled lane and narrator mode in the ad metadata when auditability matters.\n8. Never clone a live winner's exact counts, timestamps, weekday, or signature phrasing. Live results validate a lane, not copy-only causality.\n9. Preserve short-form architecture. Do not inflate an Avelia T3 ad into long-form to satisfy a long-form SOP.\n10. For problem-aware long form, complete the sourced failed-attempt and root-cause chain before a short product reveal. Reveal percentage is diagnostic, not a universal threshold.\n11. Remove repeated product-description boilerplate unless a physical detail materially advances that ad's argument.\n12. Allocate exactly the number of controlled lanes the user requests. If the user names a lane letter or test-cell label, preserve it in the ad declaration and manifest; do not silently expand one requested lane into an older two- or three-lane default.\n13. When an exact source quote contains a globally banned weak word, do not alter text inside quotation marks. Use an exact contiguous excerpt that omits the weak word, or convert the line into an attributed paraphrase without quotation marks. Keep explicit attribution so the excerpt cannot read as advertiser testimony.\n14. Replace outcome-sounding setup statements such as `He agrees`, `He sets`, or `He wears` with capability or consent language such as `He has to agree`, `He can set`, or `He can wear` unless a receipt proves the completed behavior.\n\n## Validation\n\nAfter every edit, parse the actual Primary Text section and verify:\n\n- brief-specific word range;\n- all required fields;\n- unique hooks and headlines within scope;\n- one sentence per line and no em dash;\n- banned language and claim restrictions;\n- exact CTA and destination;\n- canonical offer as three separate lines in exact order;\n- exact offer wording with no added prefix such as `DawnBands offers a`;\n- truth/provenance status for every quote, number, scene, timeline, and product outcome;\n- controlled lanes share the required strategy but differ in expression;\n- manifest SHA-256 values match the final files.\n\nRe-run validation after the manifest is written. Final status is PASS only when all expected files exist, all checks pass, and every hash matches.",
    "sha256": "3b5a2dc98c372cb5cf456994c7432821961811bf45db9e42a910c09af6aad7de"
  },
  {
    "group": "Native copy",
    "name": "references/copy-directness-doctrine.md",
    "path": "/Users/ayden/.hermes/skills/openclaw-imports/native-copy/references/copy-directness-doctrine.md",
    "text": "# Copy Directness Doctrine (Fish, 2026-09-01) — BINDING ON ALL COPY\n\nLoad with any native/ad copy task. This is the rule Fish enforces hardest and the one that\nhas failed most often.\n\n## Grade 3-5. Plain speech. Active voice.\n- Average sentence under 14 words. Most under 10. Nothing over 28.\n- Active voice only. \"has been sorted into background by his brain\" -> \"his brain tuned it out.\"\n- Say the mechanism in ordinary words: the brain HABITUATES, it TUNES IT OUT, AUTOMATICALLY.\n\n## The three-strike error: roundabout -> anaphora -> role-abstraction\nFish rejected three consecutive rewrites of the same passage. Each fixed the surface and kept\nthe literary move. Check for all three before shipping any line.\n\n1. ROUNDABOUT — \"What I ordered was the first knock. ... the honest answer to what happens\n   when he leaves home is that I am the answer.\"\n2. ANAPHORA (only sounds direct) — \"To stop making the first knock. / To stop walking to his\n   door. / To stop hearing my own voice before seven.\" Three+ consecutive lines opening with\n   the same word or phrase is a poem. Banned.\n3. ROLE-ABSTRACTION — \"right now I am still the plan.\" Fish: \"i am still the plan is still\n   not what a mom says.\" She never becomes an abstract noun.\n\nGOOD: \"I am tired of being his alarm clock. I do not want to start every day angry at my own\nkid. He leaves next year, and I will still be the one calling him every morning.\"\n\n## Banned constructions\n`I am still the plan` · `I am the answer` · `that was the variable` · `never in that channel` ·\n`that is the whole thing` · `that is what I am trying to fix` · `the part I could not fix was` ·\n`What I ordered was X` · `has been sorted`\n\n## Instead\nName the concrete action she performs, or the feeling she has.\n\"I will still be the one calling him every morning.\" · \"I still have to drive over and wake him\nup.\" · \"I am tired.\" · \"I feel guilty.\" · \"I am angry before seven and I hate it.\"\n\nRegister anchor, real buyers' words: \"Want to start the school year off with out the fighting.\"\n\"Hoping it stops angry mornings I wake up in fight or flight mode.\" \"Mornings turned into daily\nfights.\"\n\nEXCEPTION: real idioms mothers use are fine — \"I am his alarm clock\", \"the human snooze button\".\n\n## Test\nWould a 43-year-old mother say this sentence to another mother in a school parking lot?\nIf it sounds written, cut it and say the feeling instead.\n\n## Meta-lesson\nA fix that changes syntax but keeps the literary SHAPE is not a fix. Directness is not\ncompression. It is plain speech.\n\nEnforced by `~/.openclaw/workspace/scripts/validate_native_copy.py`.\nFull history: `~/.claude/projects/-Users-ayden/memory/feedback_copy_directness_and_model.md`\n",
    "sha256": "726ba71b35a0dd28fe7e38a2562909579f72a481947331fa15f0374f01a89825"
  },
  {
    "group": "Native copy",
    "name": "references/dawnbands-context-intake-and-precedence.md",
    "path": "/Users/ayden/.hermes/skills/openclaw-imports/native-copy/references/dawnbands-context-intake-and-precedence.md",
    "text": "# DawnBands native context intake and precedence\n\nUse this when reviewing, revising, or explaining an existing DawnBands native, especially when the user asks whether the full doctrine, briefs, avatar work, or research can be accessed.\n\n## Intake rule\n\nDo not review from the rendered/public copy alone. Verify and read the internal source stack that produced it. Separate governing doctrine from historical examples and archives.\n\n## Governing stack\n\nRead in this order, with later Fish rulings inside a higher source overriding older language:\n\n1. `native-copy/SKILL.md` — current permanent Fish rulings and execution gates.\n2. `references/long-format-native-sop-v2-contract.md` for long-format work.\n3. The current avatar/sub-avatar source for the unit. For the 2026-09 SA2 family: `knowledge/brands/dawnbands/research/evolve-subavatars-calmer-mornings-2026-09-03.md`.\n4. The unit's structured brief and system prompt in its output directory, including brief-gate receipts.\n5. Current brand truth: `_index.md`, `adhd-mom-lockin.md`, `dawnbands-avatar-adhd-teen-mom.md`, `product-truth.md`, `selling-root-cause-mechanism.md`, `substantiation-2026-08-06.md`, `failed-solutions-graveyard.md`, and `objection-map.md`.\n6. VOC receipts and current research that support the exact scene, qualifier, claim, and dialogue.\n7. Older research, swipe examples, generated drafts, and archives only where they do not conflict with items 1–6.\n\n## Evolve work map\n\nFor the calmer-mornings / Core 5 work, inspect all three together:\n\n- `knowledge/brands/dawnbands/research/inbox/fish-evolve-m1-notes-2026-09-02.md` — Fish's originating Evolve notes.\n- `knowledge/brands/dawnbands/research/evolve-subavatars-calmer-mornings-2026-09-03.md` — full SA1/SA2/SA3 build and rulings.\n- `knowledge/brands/dawnbands/research/winners-evolve-bins-2026-09-03.md` — performance classification and what remains untested.\n\nDo not treat the Evolve document's `DRAFT for Fish review` label as a reason to ignore rulings later embedded into the native-copy skill or product truth. Check whether a rule was promoted before deciding authority.\n\n## Inciting-incident handling\n\n`knowledge/brands/dawnbands/research/inciting-incidents-2026-05-01.md` is useful research but was authored for the older broad mom-of-teen lane. It does not override the September ADHD-mom lock-in, a current Evolve sub-avatar card, the relatability gate, or a gated unit brief.\n\nFor current long-form work, the binding standard is: write the incident at the peak of believability after the world, stack, and toll have established trust. The object needs a VOC receipt; the peak may be dramatized only in the sanctioned `incident_peak` slot and must stay confined to the narrator, child, and object.\n\n## Existing-batch review\n\nWhen an output directory exists, inspect at minimum:\n\n- structured unit brief(s)\n- system/user generation prompts\n- immutable/raw draft\n- final copy\n- brief-gate audit\n- SOP audit\n- 1:1:1 receipt audit\n- launch/DCO plan if the question touches deployment\n\nA passing validator is not proof of editorial quality. Read the copy and the human judgment audit. Report conflicts or stale source assumptions rather than silently blending them.\n\n## Public page boundary\n\nA Cloudflare Pages URL proves the rendered artifact is accessible. It does not prove the reviewer has loaded the doctrine, briefs, receipts, or source research. State which layer was actually inspected.\n",
    "sha256": "05b3c75f619be609aec4058976104bddcc9bbb52a9eae759c54a60a26d4e1080"
  },
  {
    "group": "Native copy",
    "name": "references/dawnbands-long-form-witness-native.md",
    "path": "/Users/ayden/.hermes/skills/openclaw-imports/native-copy/references/dawnbands-long-form-witness-native.md",
    "text": "# DawnBands Long-Form Witness Native\n\nUse this when the brief explicitly selects **DawnBands Long-Form Witness Native** mode **or when the user's supplied controls clearly define that architecture**. A request to write “like these samples” is a format instruction when the controls are 2,500–4,000-word first-person witness stories with a late product reveal.\n\nIt is an intentional long-form architecture derived from two supplied Alevia long-form controls, the Manus Long-Format Native Ad SOP 2.0, and the live DawnBands ADHD avatar/mechanism reports. It is not the default for every native, and it does not replace compressed Avelia T1/T2/T3 or product-aware ads.\n\n## What the supplied controls prove structurally\n\nThe two supplied Alevia controls were approximately 3,188 and 3,662 words. The product name first appeared at approximately 59.9% and 60.6% of the body. Both used the same persuasion engine:\n\n1. A witness sees a capable person suffer an identity rupture.\n2. One ordinary object holds the entire wound.\n3. The story establishes who that person was before the decline.\n4. Small losses accumulate before the inciting incident.\n5. Failed explanations and failed solutions are exhausted in sequence.\n6. Discovery happens through late-night research in the narrator's own voice.\n7. A plain-language mechanism makes prior failures feel inevitable.\n8. The product appears only after the reader can predict what a different solution must do.\n9. Results unfold through modest, concrete moments.\n10. The opening object returns at the end as proof of restored identity.\n11. Risk reversal removes the final objection.\n12. P.S. codas reopen or deepen emotional stakes rather than repeating features.\n\nTransfer the architecture, pacing, motif discipline, and emotional causality. Do not import unsupported medical claims, invented authorities, unverified timelines, or potentially fictional customer outcomes.\n\n## Mode selector\n\n### Use this mode when\n\n- The brief explicitly asks for an Alevia/Manus-style long-form witness story.\n- The user asks for personal anecdotes or says to match supplied 2,500–4,000-word first-person controls, even if the internal mode name is absent.\n- Traffic is unaware or problem-aware and needs story plus education.\n- There is enough avatar evidence to sustain a real world, not just a pain-point list.\n- The problem has a relational identity wound worth carrying for thousands of words.\n- The product can be introduced as the logical answer to a bounded mechanism.\n\n### Do not use this mode when\n\n- Traffic is product-aware or offer-aware.\n- The brief requests compressed Avelia T1/T2/T3.\n- The only available source material is generic pain language.\n- A long story would require inventing a customer, authority, timeline, or DawnBands outcome.\n- The requested ad is a hook, static caption, testimonial card, or short objection killer.\n\nProduct-aware readers get short proof, objection handling, offer, urgency, or risk reversal. Long-form is education, not a universal minimum.\n\n## Length and reveal contract\n\n- Uploaded-winner-derived witness mode: typically **2,500–4,000 words** unless the brief sets another range.\n- Pure unaware Manus-style education may run longer when the world and mechanism genuinely earn it.\n- Product name target for witness mode: **55–70%** into the body.\n- Measured update 2026-08-12: Alevia's 239 live actives run reveal at median 52% (p10 31%,\n  p90 66%) — the 55–70 contract stays default; **45–55% is a sanctioned test band** when the\n  brief names it (`alevia-live-anatomy-2026-08-12.md`).\n- Pure unaware Manus mode may delay the product to 70–80%.\n- The brief and awareness level control. Never pad to hit a number.\n- Mechanism discovery may begin before the product reveal, but the category/product should not leak into the opening world.\n\nA 500-word ad can be strong, but it is not the same format as the supplied 3,000+ word controls. Do not label mid-form ownership copy as a full long-form witness native.\n\n## Narrator authority gate\n\nDeclare one narrator mode in the internal brief before drafting.\n\n### 1. Verified buyer\n\nA single approved testimonial supports the narrator, product use, and claimed outcome.\n\nAllowed:\n- only details traceable to that source\n- exact or conservatively paraphrased outcomes\n- separately verified canonical product facts\n\nNot allowed:\n- adding a spouse, child, doctor, timeline, dialogue, or result the receipt does not contain\n\n### 2. Single-source problem witness\n\nOne first-person source supports the lived problem and witness perspective, but not DawnBands use.\n\nAllowed:\n- the sourced before-state and emotional wound\n- canonical DawnBands facts after the product bridge\n- separately attributed buyer proof\n- desired-absence future pacing\n\nNot allowed:\n- assigning DawnBands purchase, use, or results to the problem narrator\n\n### 3. Ownership voice\n\nAn ownership argument may reason from a sourced parent problem without pretending a pooled research synthesis is one mother's lived household testimony.\n\nAllowed:\n- attributed evidence from named source IDs\n- what a parent may fear, notice, or want to stop when framed as recognition rather than completed personal history\n- explicit hypotheticals and desired scenes\n- product facts and offer truth\n\nNot allowed:\n- a lived customer timeline or “what happened after we tried it” arc\n- welding multiple VOC sources into one `I / my son / our house` chronology\n- presenting a multi-device failed-solution stack, exact counts/times/distances, remembered dialogue, or a DIY research journey as one narrator's past unless one same-speaker receipt supports it\n- assuming an internal `ownership-voice` label cures consumer-facing testimonial implications\n\n**Synthetic-narrator tripwire:** if the prose combines first person, one child, one home, precise household behavior, and several failed solutions, stop. Use one same-speaker problem witness, attributed evidence, explicit wanted/feared hypotheticals, second-person recognition, or a non-testimonial ownership argument instead. A cleaner line edit must not make unsupported continuity feel more real.\n\n### 4. Authorized dramatization\n\nUse only when the user or validated brief explicitly authorizes a dramatized/composite persona.\n\n- Label it internally.\n- Keep each behavior traceable to research.\n- Do not create a fake named testimonial identity.\n- Do not invent product use, product outcomes, week-by-week progression, medical authority, or quantified proof.\n- Any proof must be separately attributed to a real receipt.\n\nIf none of the four modes is available, do not write first-person witness copy. Use attributed evidence, second-person recognition, or an ownership argument.\n\n## Provenance ledger\n\nTag every precise beat in the internal outline before prose:\n\n- `RECEIPT` — directly supported by one saved source.\n- `CANONICAL FACT` — supported by current DawnBands canonical truth.\n- `HYPOTHETICAL / OWNERSHIP ARGUMENT` — an explicitly framed wanted, feared, or likely scene that is not presented as completed history, product proof, or witness testimony.\n- `AUTHORIZED DRAMATIZATION` — a non-outcome connective detail allowed by the brief; never present it as customer proof or as a completed DawnBands result.\n- `DELETE` — unsupported precision, dialogue, authority, chronology, object, result, or claim.\n\nSpecificity is proof-like. Names, ages, streets, schools, dates, exact times, counts, dollar amounts, objects, jobs, dialogue, and timelines require a receipt or explicit dramatization authority. “Plausible” is not provenance.\n\n## DawnBands truth boundary\n\nThe buyer and wearer are two different people sharing one morning problem.\n\nThe bounded product role is:\n\n> A correctly fitted wrist alarm can provide a timed private tactile cue that is physically different from room-level sound.\n\nThe product does not establish that cue detection becomes sustained waking, leaving bed, staying awake, punctuality, or reduced caregiver involvement for every wearer. Exact-SKU output, comfort, retention, battery behavior, reliability, and ADHD-specific efficacy remain test questions unless newer canonical evidence supersedes this file.\n\nNever claim DawnBands:\n\n- treats or fixes ADHD\n- fixes circadian timing or sleep inertia\n- resolves inadequate sleep, a sleep disorder, or medication questions\n- makes vibration impossible to sleep through\n- wakes every teen every time\n- guarantees sustained waking after cue detection\n- replaces medical evaluation for persistent sleep or health concerns\n\nDo not frame the teen as lazy, defiant, careless, or morally responsible for an experience they may barely remember. Preserve dignity and active wearer participation.\n\n## Plain-language mechanism\n\nThe defensible reframe is not “stronger alarm.”\n\nIt is:\n\n- room sound broadcasts a cue through the environment\n- repeated versions of room sound can repeat the same failed pathway\n- wrist vibration changes both sensory channel and signal placement\n- a private cue can reduce whole-house spillover\n- the cue is only one handoff in the morning relay\n- alertness, initiation, leaving bed, staying awake, sleep opportunity, timing, and repeated-use habits still matter\n\nPreferred analogy:\n\n> The alarm is one handoff in a relay. DawnBands can move that handoff from the room to the wrist, but the rest of the morning still has to carry the baton forward.\n\nVary the expression by ad. Do not clone “different sensory channel / different signal placement” across a batch.\n\n## Story architecture\n\n### Beat 1 — Rupture object\n\nOpen inside a moment with one concrete object carrying the wound.\n\nDawnBands-compatible territories include:\n- untouched coffee cooling during another bedroom check\n- a backpack ready beside a teen who is still asleep\n- a phone ringing into an empty hallway\n- an unopened alarm box\n- a dorm packing pile that still assumes Mom will call\n\nThe hook is not a headline, claim, benefit, diagnosis, or product clue. It is an event that makes line two necessary.\n\nExample objects are structural prompts, not receipts; the chosen object still needs `RECEIPT` or explicit `HYPOTHETICAL / OWNERSHIP ARGUMENT` framing.\n\n### Beat 2 — Identity contrast\n\nEstablish who the parent and teen are outside the problem.\n\nFor the teen:\n- capable after waking\n- cares about school, work, sports, friends, or independence\n- does not need the ad to diagnose or excuse them\n\nFor the parent:\n- patient before repetition changes their voice\n- trying to reduce support appropriately, not abandon the child\n- afraid the current arrangement will follow them into college or work\n\nThe wound is not merely lateness. It is who each person becomes inside the current wake-up arrangement.\n\n### Beat 3 — Progressive losses\n\nShow accumulation, not a generic pain summary.\n\nPossible source-backed progression:\n- one alarm becomes several\n- across-the-room becomes another unconscious shutoff\n- apps add tasks without sustained waking\n- Mom starts checking earlier\n- privacy turns into management\n- mornings begin with blame or guilt\n- future distance exposes the dependency\n\nEach loss must either have a receipt or be framed as a hypothetical/ownership argument.\n\n### Beat 4 — Failed-solution sequence\n\nUse the order the avatar would logically try:\n\n1. louder ringtone\n2. more phone alarms\n3. phone across the room\n4. puzzle/QR/task app\n5. light or bed shaker\n6. repeated caregiver intervention\n\nFor each:\n\n- acknowledge why it was logical\n- show its exact gap\n- give one memorable summary\n\nDo not insult prior choices. “Right instinct, wrong variable” is the tone.\n\n### Beat 5 — Inciting incident\n\nThe status quo becomes emotionally impossible to keep accepting.\n\nFor DawnBands this does not need medical catastrophe. Stronger, truer incidents include:\n- the teen remembers only Mom's final angry attempt\n- a missed event mattered deeply to the teen\n- privacy is lost because Mom must enter and finish the morning\n- a college/work transition reveals that the system still depends on Mom\n- the parent realizes every alarm still ends with their body at the door\n\nThe incident must be sourced or explicitly hypothetical. Never manufacture school calls, crashes, doctors, teachers, or dialogue for intensity.\n\nIf the inciting pressure is not receipted, write it as a present-tense ownership realization or explicitly hypothetical threshold, never as a dramatized past event.\n\n### Beat 6 — Discovery\n\nTwo valid routes exist.\n\n#### Sourced authority route\n\nUse a real, traceable authority source. Paraphrase accurately. Do not fabricate a pediatrician, therapist, sleep researcher, pharmacist, or doctor conversation.\n\n#### DIY discovery route\n\n**DIY discovery** is valid when the narrator follows a real question into saved evidence.\n\nStructure:\n- the exact search question\n- the old assumption\n- the new distinction\n- a human-scale analogy\n- the moment prior failures make sense\n\nFor DawnBands, the discovery is not “ADHD causes this.” It is the narrower realization that detecting a cue, becoming fully alert, and initiating the next action are different jobs, and that more room sound does not change signal placement.\n\nMeasured refinement (2026-08-12): in every live Alevia survivor the discovery resolves as a\n**word spoken by a named third party inside the story** (\"She told us about a fruit called…\"),\nfrequently at the end of a narrated chain — article → forum thread → a named stranger's comment\n→ skeptical read → try. Narrating the chain itself does DR work. Two rules transfer: the\nmid-story reveal names the CATEGORY (a wrist-worn silent alarm someone mentions), never the\nbrand — the brand surfaces at the link/close; and the named stranger stays dialogue inside the\nstory, never an authority byline and never a fabricated testimonial with product outcomes.\n\n### Beat 7 — Layered bounded mechanism\n\nLayer the explanation:\n\n1. What room sound asks the sleeper to notice.\n2. Why repetition or volume may still fail.\n3. Why task completion does not prove sustained waking.\n4. What changes when the cue is worn on the wrist.\n5. What remains unknown or outside the product's role.\n\nStay in story voice. Use a reaction or concrete callback every 3–4 explanation lines so the copy does not become a strategy memo.\n\n### Beat 8 — Objection sequence\n\nHandle objections in the narrator's natural order:\n\n- “He slept through a bed shaker. Why would vibration work?”\n- “Won't he get used to it?”\n- “He will not wear it.”\n- “I will end up managing another device.”\n- “What if this is really sleep deprivation, medication timing, or something medical?”\n- “Why risk $49 after all the other alarms?”\n\nUse acknowledge → gap → memorable summary. Do not promise impossible habituation resistance or universal efficacy.\n\n### Beat 9 — Natural product reveal\n\nThe product should feel like the answer the story has already specified.\n\nKeep the first reveal short and evidence-qualified:\n- seller-described silent vibrating wrist alarm for teens\n- cue intended to be worn directly on the wrist\n- standalone / no app / no subscription only after verifying the delivered revision\n- describe only visible form unless material and construction are independently verified\n\nEvery exact-SKU statement must be classified internally as `VERIFIED`, `SELLER-CLAIMED`, or `UNKNOWN`. Consumer copy may state verified facts directly, attribute seller claims accurately when useful, and must omit unknowns.\n\nDo not switch into landing-page voice. Avoid “revolutionary,” “breakthrough,” “game-changing,” “proprietary,” and a sudden feature dump.\n\n### Beat 10 — Proof or desired absence\n\n#### Receipt-backed outcome\n\nUse only when one approved DawnBands receipt supports the narrator, product use, and result. Preserve qualifiers and timeline exactly.\n\n#### Desired absence\n\nWhen narrator-level product proof is absent, future pace what the parent wants to stop:\n- no fourth walk down the hall\n- no whole-house alarm\n- no argument that begins before either person fully remembers the morning\n- no long-distance college wake-up call\n- no app that turns Mom into remote administrator\n\nLabel the scene as wanted, imagined, or worth testing. Do not write it as a completed customer result.\n\n## Outcome timeline gate\n\nNo week-by-week, day-by-day, streak, “first morning,” teacher-noticed, spouse-noticed, or before/after progression may appear unless one traceable DawnBands user receipt supports that progression.\n\nIf the gate fails:\n\n- remove the timeline\n- use separately attributed verified buyer proof\n- future pace as desired absence\n- move risk reversal forward\n\nThe supplied Alevia controls use gradual timelines effectively. That structure is not permission to invent one for DawnBands.\n\n## Motif contract\n\nChoose one concrete motif before drafting.\n\nThe **motif** must:\n\n1. appear in the opening rupture\n2. reappear during mechanism/discovery as an analogy or reminder\n3. return in the close with changed meaning\n\nExamples:\n- cooling coffee: from interruption to a cup Mom gets to finish\n- bedroom door: from repeated crossing to a door she does not open\n- backpack: from readiness blocked by waking to readiness owned by the teen\n- unopened box: from category fatigue to a bounded different-cue test\n\nOne motif only. Do not add a second cinematic symbol because the first one was not developed.\n\n## Close and P.S. architecture\n\nClose in this order:\n\n1. return to the opening motif\n2. restate the relational loss of doing nothing\n3. make the bounded product role clear\n4. state only the currently verified price and refund-policy terms\n5. one direct CTA\n\nOffer and guarantee boundary:\n\n- Verify the current price and complete return/refund policy before every deployment.\n- Prefer unambiguous policy language such as `60-night money-back guarantee` only when those terms are verified.\n- `100% wake-up guarantee` may appear only if legal/substantiation review has approved it as the exact name of a refund policy, and it must be immediately clarified as refund terms rather than universal waking efficacy.\n- Never write the guarantee as proof that DawnBands wakes every wearer, sustains waking, creates independence, or produces a guaranteed outcome.\n- If current policy evidence is incomplete or conflicting, stop the offer block and request verification rather than inheriting older canonical copy.\n\nOptional codas:\n- P.S. returns to the most emotional unresolved relationship beat.\n- P.P.S. handles the dominant risk or skepticism objection.\n- P.P.P.S. may extend the future/college implication when genuinely earned.\n\nDo not use codas as repeated feature bullets, fake scarcity, or unsupported stock claims.\n\n### Close variant B — the measured P.S. ladder (added 2026-08-12)\n\n217 of 240 measured live Alevia actives close on a stacked coda ladder instead of the ordered\nclose above, with the link LAST. Test it against the default close on a proven copy before\nadopting it batch-wide (see `alevia-live-anatomy-2026-08-12.md`):\n\n1. P.S. — rupture-object/motif return\n2. P.P.S. — guarantee naturalized in the narrator's voice (verified refund terms only; every\n   offer gate above applies)\n3. P.P.P.S. — the dominant objection or \"it's just placebo\" kill, in-voice\n4. final P.S. — direct address to the reader-avatar; this rung IS the CTA\n5. bare 👉 link on its own line after the final coda — no CTA copy around it\n\nOutcome-timeline, desired-absence, and offer-verification gates apply to every rung. The ladder\nnever becomes a feature recap.\n\n## Formatting\n\n**(Corrected 2026-08-12 per the 2026-07-31 Mode B ruling + measured live controls — see\n`alevia-live-anatomy-2026-08-12.md`. The older one-sentence-per-line block that stood here\napplies to Mode A / compressed natives and static primaries, NOT to this mode.)**\n\n- **PARAGRAPH prose** — 2–4 sentences per paragraph, roughly 25–40 words. The measured live\n  controls run ~31 words/paragraph median. A witness story formatted wall-to-wall\n  one-sentence-per-line reads as an ad, not a post.\n- Staccato single-line beats are a PUNCH device, not the default — reserve them for rupture\n  moments, reveals, and emotional pivots (the live controls mix them into roughly half the\n  copies, sparingly).\n- blank line between paragraphs\n- short, varied cadence within paragraphs\n- dialogue isolated on its own line\n- no emojis in long-form story copy (the bare 👉 link line at the close is the one exception)\n- no bold/caps/format tricks inside Meta primary text\n- no em dash\n- no dense mechanism wall\n- no visible switch into ad voice\n\n## Internal pre-draft contract\n\n```yaml\nmode: dawnbands-long-form-witness\nsubmode: alevia-witness | pure-unaware-manus\nawareness: unaware | problem-aware\nword_target: 2500-4000\nnarrator_mode: verified-buyer | single-source-problem-witness | ownership-voice | authorized-dramatization\nprimary_source_ids: []\nproduct_outcome_receipt: null\nrupture_object: \"\"\nidentity_before: []\nprogressive_losses: []\nfailed_solution_sequence: []\ninciting_incident:\n  beat: \"\"\n  provenance: RECEIPT | HYPOTHETICAL_OWNERSHIP\nmechanism_route: sourced-authority | diy-discovery\nprimary_open_loop: \"\"\nsecondary_loops: []  # require at least 1 in Mode B unless the brief explicitly waives it\nproduct_reveal_target_pct: 55-70\nmotif_return: \"\"\nfuture_mode: receipt-backed-outcome | desired-absence\ncanonical_truth_sources: []\nexact_sku_claims:\n  verified: []\n  seller_claimed: []\n  unknown_delete: []\noffer_policy_verified: false\nwake_up_guarantee_legal_approval: false\nclaim_ceiling_checked: false\n```\n\nStop if submode, narrator authority, source IDs, mechanism route, inciting-incident provenance, product outcome receipt/null decision, primary open loop, at least one secondary loop or explicit waiver, motif, future mode, exact-SKU claim classification, verified current offer/refund evidence, or claim ceiling is missing.\n\n## QA scorecard\n\n### Architecture\n\n- [ ] Intentional witness mode was selected by the brief.\n- [ ] Hook is a story event with no product/category leak.\n- [ ] Identity contrast appears before mechanism.\n- [ ] Progressive losses accumulate.\n- [ ] Failed solutions follow a believable sequence.\n- [ ] Inciting incident is irreversible enough and provenance-safe.\n- [ ] One primary open loop is introduced in the opening world and stays unanswered until after mechanism/discovery.\n- [ ] At least one secondary loop deepens stakes, proof, or future risk before the reveal.\n- [ ] Every loop is closed in the body or intentionally handed to a P.S. / P.P.S. coda.\n- [ ] Product reveal lands in the selected window.\n- [ ] Reveal stays in narrator voice.\n- [ ] Motif returns with changed meaning.\n\n### Avatar\n\n- [ ] Buyer and wearer remain distinct people.\n- [ ] Teen dignity and agency are preserved.\n- [ ] The parent is not framed as controlling for its own sake.\n- [ ] Relationship cost is shown through behavior.\n- [ ] College/work future is used only when the angle earns it.\n\n### Truth and proof\n\n- [ ] Every precise detail has a provenance label.\n- [ ] No composite becomes a fake testimonial.\n- [ ] No pooled VOC set is presented as one `I / my son / our house` chronology without a same-speaker receipt or explicit dramatization authority.\n- [ ] First-person specificity, failed-device history, counts/times/distances, and discovery behavior pass the synthetic-narrator tripwire.\n- [ ] Quoted words match the saved source.\n- [ ] No authority dialogue was invented.\n- [ ] No product outcome or timeline exceeds its receipt.\n- [ ] Desired scenes are not written as completed results.\n- [ ] Desired-absence scenes are framed with wanted / imagined / worth-testing language, never as completed DawnBands results.\n- [ ] Exact-SKU statements are classified as verified, seller-claimed, or unknown; unknowns are omitted.\n- [ ] Current price and complete refund-policy terms are verified before the offer block.\n- [ ] `100% wake-up guarantee` is absent unless approved as a named refund policy and immediately clarified as refund terms, not universal waking efficacy.\n- [ ] Exact-SKU uncertainty and cue-vs-waking boundary are preserved.\n\n### Craft\n\n- [ ] One concrete motif, not several.\n- [ ] DIY discovery or authority explanation sounds human, not like a paper.\n- [ ] Each failed solution gets acknowledge → gap → summary.\n- [ ] Mechanism explanation returns to the lived story every 3–4 lines.\n- [ ] CTA resolves the opening tradeoff.\n- [ ] Optional P.S. codas deepen the story rather than repeat sales copy.\n\n## Batch rule\n\nDo not make every ad in a batch a 3,000-word witness story. Long-form witness should be an intentional testing cell alongside compressed problem-aware, solution-aware, product-aware, and objection-led executions.\n\nWithin a long-form cell, vary:\n- narrator mode\n- rupture object\n- primary fear\n- failed-solution stack\n- mechanism analogy\n- product reveal route\n- motif\n- desired absence\n\nThe architecture may repeat. The life cannot.\n",
    "sha256": "d430ff955fdfc4637215f5310888062f6807072fda3c3ba707131cd93aa9bae5"
  },
  {
    "group": "Native copy",
    "name": "references/dawnbands-selling-root-cause-mechanism.md",
    "path": "/Users/ayden/.hermes/skills/openclaw-imports/native-copy/references/dawnbands-selling-root-cause-mechanism.md",
    "text": "# DawnBands Selling Root Cause + Mechanism — Execution Reference\n\nUse this reference for any DawnBands native, advertorial, UGC, VSL, listicle, lander, brief, or copy audit.\n\nThe workspace canon remains authoritative:\n\n`/Users/ayden/.openclaw/workspace/knowledge/brands/dawnbands/selling-root-cause-mechanism.md`\n\n## User correction captured\n\nDo not over-optimize DawnBands mechanism copy for conservative addressability. Aggressive biological simplification is desired when it makes the Root Cause → Mechanism → Product chain more persuasive.\n\nKeep two layers separate:\n\n- **Selling layer:** exaggerated, simplified biological explanation that creates belief and new hope.\n- **Proof layer:** receipts for exact-SKU performance, testimonials, timelines, offer terms, and customer outcomes.\n\nThe proof layer must not flatten the selling mechanism into generic feature language. The selling layer does not authorize fabricated customer history or product proof.\n\n## Canonical selling chain\n\n`circadian delay → sleep debt → auditory gating → sleep inertia / half-awake automatic behavior → Mom becomes the emergency pathway → Somatic Wake Bypass™ → Wrist-First Handoff™ → relationship repair + future independence`\n\n## Root cause: The Deep-Sleep Signal Lock\n\n1. **Circadian delay:** His teenage body clock is still protecting biological night when school demands consciousness.\n2. **Sleep debt:** Accumulated sleep pressure is pulling harder toward sleep while the alarm pushes toward waking.\n3. **Auditory gating:** Repeated room sound is suppressed or pattern-matched as background.\n4. **Sleep inertia:** His body can silence alarms, answer Mom, or solve tasks before alertness, memory, and initiation fully return.\n5. **Emergency pathway:** Mom escalates until emotion becomes the only novel signal left.\n\nCore lines:\n\n- “The alarm wakes his hands without waking the part of him that can start the morning.”\n- “The louder the room gets, the better his sleeping brain becomes at deleting it.”\n- “His brain ignores the alarms, but it notices the anger.”\n- “She remembers nine patient attempts. He remembers waking up to one angry mother.”\n\n## Failed-solution trap\n\n> Seven alarms do not create seven new chances to wake up. They create seven repetitions of the same signal his brain already filed under background noise.\n\nHousehold loop:\n\n> The more reliably Mom rescues the failed alarm, the more reliably the morning keeps depending on her.\n\nNever blame Mom. The villain is the repeated room-sound rescue system that leaves her no safe way to step out.\n\n## Biological mechanism: Somatic Wake Bypass™\n\nDawnBands puts a tactile signal against the wrist, outside the repeated room-sound pattern.\n\nCore line:\n\n> Sound has to cross the room and survive his brain’s filter. DawnBands starts past the filter.\n\n## Household mechanism: Wrist-First Handoff™\n\nThe first cue begins on the teen before Mom’s voice enters the morning.\n\nCore lines:\n\n- “The first cue starts on his wrist before it ends in your voice.”\n- “Not louder. On him.”\n- “Move the first signal from your throat to his wrist.”\n- “The goal is not a better tenth attempt. It is a first cue that does not require Mom.”\n\n## Format routing\n\n- **Long problem-aware:** use the complete biological and household chain.\n- **Short native:** compress to auditory gating → tactile bypass → Mom removed from first cue.\n- **UGC:** explain that the body can perform an alarm-stopping action before memory and conscious initiation are online.\n- **Advertorial/VSL:** name Deep-Sleep Signal Lock and Somatic Wake Bypass™ after lived pain and failed solutions.\n- **Product-aware:** “Sound has to cross the room and survive his brain’s filter. DawnBands starts past the filter.”\n\n## Audit gate\n\nWhere format length permits, future DawnBands copy should answer:\n\n1. Why is his biology pulling him toward sleep?\n2. Why does repeated sound fail?\n3. Why can he act without becoming consciously awake?\n4. Why does Mom become the final signal?\n5. Why is tactile wrist stimulation a new pathway?\n6. Why does the first cue now belong to him?\n7. Why does this matter to their relationship and his future?\n\nIf a draft only says “vibration is different from sound,” the mechanism is underbuilt.",
    "sha256": "abe4e68240dfa2edfdf1e9b30cb7f1826c33997993caa0d25c6c81d8d00d4b8c"
  },
  {
    "group": "Native copy",
    "name": "references/direct-failed-solution-prose.md",
    "path": "/Users/ayden/.hermes/skills/openclaw-imports/native-copy/references/direct-failed-solution-prose.md",
    "text": "# Direct Failed-Solution Prose\n\nUse this when a native contains a sequence of things the narrator tried before discovering the product.\n\n## The unit\n\nEach attempt gets one complete causal beat:\n\n1. **What she tried** — literal action, not a method label.\n2. **Why she tried it** — the belief or hope behind it.\n3. **What happened** — visible behavior in the morning.\n4. **Why trust broke** — the same problem still occurred.\n\nExample:\n\n> I started getting up at 5:30 because I thought I was the problem. I wanted to be ready and calm before he got up. I still ended up yelling by the door.\n\n## Sequence rules\n\n- One attempt per beat. Keep chronology obvious.\n- Describe actions before using specialist labels.\n- Narrator-specific attempts are not reader qualifiers.\n- Connect every attempt to the owned morning problem; do not list generic parenting advice.\n- State the failure directly. Do not make the reader infer it from a slogan.\n- If two tactics are closely related, distinguish what changed and what remained under Mom's control.\n\n## Kill on sight\n\n- `I tried everything the posts say`\n- `came next`\n- `laid it on`\n- `used it up`\n- `both helped`\n- `neither stopped`\n- several tactic names compressed into fragments\n- unexplained before/after jumps\n- a short sentence that still needs translation\n\n## Review test\n\nFor every attempt, answer in one read:\n\n- What did she physically do?\n- Why did she think it would help?\n- What did her son physically do?\n- Why did she stop believing this would fix the morning?\n\nIf any answer is missing, rewrite the beat. Numeric readability is not enough: grade, sentence length, and said-out-loud meaning must all pass.\n",
    "sha256": "87b28977039d10e41f9f37ec62e4e658898b67e6c16c52da353254d862059f74"
  },
  {
    "group": "Native copy",
    "name": "references/directness-and-jumble-final-pass.md",
    "path": "/Users/ayden/.hermes/skills/openclaw-imports/native-copy/references/directness-and-jumble-final-pass.md",
    "text": "# Directness and Final Jumble Pass\n\nUse after structural/SOP validation on any long native. Mechanical readability can pass while a paragraph still makes a reader stop and decode it.\n\n## Core distinction\n\n- **Readable numbers:** grade, sentence length, paragraph length.\n- **Readable meaning:** a normal reader understands the action, chronology, speaker, and cause on the first pass.\n\nBoth must pass. A short sentence can still be jumbled.\n\n## Failed-solution unit\n\nWrite one attempt per beat in this order:\n\n1. what she tried\n2. why she tried it\n3. what literally happened\n4. why she stopped trusting it\n\nDo not compress two methods into one summary paragraph. Avoid transition shorthand such as `came next`, `both helped`, `used it up`, or `everything the posts say`. Name the action and result.\n\n**Direct model:**\n\n> I started getting up earlier because I thought I was the problem. I wanted to be ready and calm before he got up. I still ended up yelling by the door.\n\n## Paragraph-level jumble checklist\n\nRead every paragraph and flag:\n\n- two speakers whose turns are hard to distinguish\n- an unclear `it`, `that`, `under it`, `both`, or `they`\n- a chronology jump without a transition\n- multiple failed solutions compressed into one beat\n- framework labels leaking into the narrator's words\n- abstract status descriptions instead of visible behavior\n- cause/effect that skips a step\n- a fragment whose meaning depends on reconstructing the previous paragraph\n- product/mechanism phrasing that sounds like an operator note\n- a callback whose original object or event is no longer clear\n\n## Common transformations\n\n- `He was just not started` → show the stalled action: `He stood beside his shoes without reaching for them.`\n- `He can hear you. He cannot begin.` → `He hears you, but his body still does not move.`\n- `Your calm is his calm` → state what each person sees/hears and how tension increases.\n- `I got myself out of the wake-up` → `I stopped being the person who had to wake him.`\n- `The first thing he does is his own` → `You do not have to give him the first order.`\n- `He comes up with a clearer head` → `He wakes with a clearer head.`\n- `Gets lost between fewer steps` → `Loses track of fewer steps.`\n- `See what the rest of the hour does` → state the observable result being tested.\n\n## Final sequence\n\n1. Run the governing native validator and SOP mechanical audit.\n2. Read the full body cold, paragraph by paragraph.\n3. Apply only line-level directness repairs; preserve the approved structure and incident.\n4. Rerun sentence-length, grade, banned-language, and cross-body convergence checks.\n5. Complete the SOP §11 eight-killers table and §13 ten-question final check.\n6. Verify the receipt table still passes after wording changes.\n7. Rebuild the review artifact only after the final text is frozen.\n\nA deterministic PASS is not permission to skip this pass.",
    "sha256": "cb65ddf502af44156ddeffc70a399c110afa235edd4d01859b646cb83fa27d68"
  },
  {
    "group": "Native copy",
    "name": "references/explicit-batch-cardinality-and-publication.md",
    "path": "/Users/ayden/.hermes/skills/openclaw-imports/native-copy/references/explicit-batch-cardinality-and-publication.md",
    "text": "# Explicit Native Batch Cardinality and Publication Boundary\n\nUse when Fish names the exact count of bodies, images, hook variants, headlines, or ads, or says to “publish” a completed native batch.\n\n## Cardinality math\n\nTreat the user's explicit layout as the contract even when it differs from the standing test default.\n\n- One **body** is one message/architecture.\n- One **body × image pair** is one ad unit unless Fish explicitly defines another counting rule.\n- Hook variants and headline variants may ride inside Meta's multi-text fields and do **not** multiply the ad-unit count by themselves.\n- Example: 3 bodies × 3 images = 9 ads. If each ad carries 2 primary-text hook variants and 3 headlines, the manifest contains 9 ad IDs, 18 primary-text slots, 27 headline slots, and 9 unique image assignments.\n\nBefore delivery, assert the requested arithmetic mechanically and record unique ad IDs, unique image paths, primary-text slots per ad, headline slots per ad, destination, and CTA. Do not silently fall back to the standing 3-hook/5-image layout after Fish gives exact numbers.\n\n## Reader qualifiers versus narrator texture\n\nThe relatability gate is strictest in reader-identification fields (`hook`, `control_hook`, `stake`). Age, grade, transport, work details, and household scene details may be legal narrator texture when they stay attached to this narrator, carry a VOC/narrator/sanctioned-dramatization receipt, and do not create a second sub-avatar.\n\nDo not ban a receipted missed bus or teen age from a narrator's incident merely because it would be an unverified reader qualifier. Keep siblings and partners out unless the card/receipt authorizes them.\n\n## Reviewer adjudication\n\nA challenger is an input, not the final source of truth.\n\n- Accept genuine science, causality, claim-ceiling, or provenance corrections.\n- Retire a challenger finding when it conflicts with newer canonical product truth or an explicit Fish-approved dramatization contract; write the reason in the human audit.\n- Authorized dramatization remains internally labeled and must never be represented as a verified customer testimonial.\n- Controlled SOP constants may repeat. Check unsanctioned convergence separately with exact-line and five-gram scans after excluding those constants.\n\n## “Publish” boundary\n\nIf Fish says “publish” but supplies no Meta campaign/ad-set target, budget, bid, or launch destination, publish and verify the public review artifact and produce the complete launch manifest, but do not infer a Meta launch target.\n\nA Meta launch still needs a resolvable campaign/ad-set destination and all applicable explicit approvals. Distinguish “review page published” from “Meta launched” in the completion note.\n\n## Verification receipt\n\nFor public review delivery, verify both the stable Pages URL and immutable deployment URL return HTTP 200 and contain markers for every body/ad ID. Probe representative image assets directly as well.",
    "sha256": "89b4cb6564039cc0e79859b5add1c1abada467732085ff4b68e32fc9009283fe"
  },
  {
    "group": "Native copy",
    "name": "references/lived-first-person-native-copy.md",
    "path": "/Users/ayden/.hermes/skills/openclaw-imports/native-copy/references/lived-first-person-native-copy.md",
    "text": "# Lived First-Person Native Copy Without Fake Testimonial Claims\n\nUse this when a native ad is structurally correct but sounds like an analyst describing the avatar from outside.\n\n## The failure\n\nDetached copy talks about:\n- “Mom,” “the buyer,” “the category,” or “the sleeper”\n- solution architecture before a lived trigger\n- generalized emotional labels instead of actions\n- a product result assigned to a narrator who was never verified\n\nThis can satisfy a brief while missing the essence of native: the reader should feel the writer is inside the morning, not presenting a market analysis.\n\n## Lived-POV construction\n\n1. Open with one observable trigger and one consequence.\n2. Stay in first-person logistics: what I saw, did, repeated, heard, or regretted.\n3. Use short sequence lines to recreate the loop rather than summarize it.\n4. Name the identity wound in the narrator's language only after the scene earns it.\n5. Correct the failed category after the reader recognizes herself.\n6. Reveal the product as a different test, not as a sudden sales pitch.\n7. Put verified customer outcome proof in a separately attributed quote.\n8. Close by resolving the narrator's role or tradeoff.\n\n## Truth-safe first-person modes\n\nUse `I` only when at least one mode is explicit:\n\n- **Verified buyer:** exact review/testimonial supports the narrator and product outcome.\n- **Single-source problem POV:** one first-person VOC source supports the lived problem; do not assign product use or outcome to that narrator.\n- **Authorized dramatization/composite:** the brief or user explicitly authorizes a dramatized persona voice. Label this internally, keep every behavior receipt-backed, and never invent a product result, timeline, or testimonial identity.\n- **Ownership voice:** first-person problem language transitions to brand/product facts, with any result proof separately attributed.\n\nIf none applies, use attributed VOC or second-person recognition rather than manufacturing a customer.\n\n## Detached-language gate\n\nFail and rewrite if the body could naturally contain phrases such as:\n- “another morning Mom has to believe in”\n- “that skeptical buyer”\n- “the category keeps changing its packaging”\n- “the hoped-for result”\n\nThose may be useful strategy notes, but they are not how a mom posts in a parenting group.\n\n## Avelia transfer rule\n\nBorrow Avelia's causal compression, not its surface voice:\n\n`one trigger → concrete symptom/behavior → failed explanation or category → different mechanism → compressed proof/risk reversal → resolve the opening tradeoff`\n\nFor first-person parent native, the trigger must be lived before the mechanism is explained.\n\n## Quick QA\n\n- Is line 1 something the narrator personally saw/did, not an analyst's thesis?\n- Do lines 3–8 contain concrete sequence detail?\n- Does the product appear only after the lived problem is established?\n- Is every customer outcome attached to a real, named/attributed receipt?\n- Could this plausibly be posted by the target person without sounding like a copywriter?\n- Does the close resolve “who I had to become” or “what I had to keep doing”?\n",
    "sha256": "50dc4c9ebf42a0f7d4cb5759996915d5bbe2d80e757239090894ea670d6f8c60"
  },
  {
    "group": "Native copy",
    "name": "references/long-form-native-comparative-audit.md",
    "path": "/Users/ayden/.hermes/skills/openclaw-imports/native-copy/references/long-form-native-comparative-audit.md",
    "text": "# Long-Form Native Comparative Audit\n\nUse this for a read-only audit of a native-ad batch against competitor/winner copies, a live long-form SOP, and live avatar/mechanism reports.\n\n## Authority and evidence order\n\n1. The newest live product/mechanism report governs the claim ceiling.\n2. Verified exact-SKU facts and current policy terms govern product and offer language.\n3. Live avatar/VOC reports govern audience scenes, objections, and language, but problem VOC is not product proof.\n4. The live SOP governs architecture by awareness level.\n5. Winner copies supply structural hypotheses only. Their narrator, dialogue, precision, medical claims, results, scarcity, and performance status do not transfer.\n6. Older brand doctrine, gold standards, and briefs are subordinate when they conflict with a newer live evidence report. Record the conflict explicitly rather than silently choosing one.\n\n## Required distinctions\n\nReport separately:\n\n- **Structural quality:** hook, world, accumulation, failed attempts, inciting incident, discovery, mechanism, objections, reveal, resolution, close, and formatting.\n- **Truth/provenance:** narrator identity, quotation fidelity, exact times/counts, dialogue, outcomes, authority, exact-SKU facts, medical/performance claims, and offer policy.\n- **Format fit:** full long-form, medium mechanism story, or compressed/product-aware execution. Do not penalize a valid compressed ad for omitting a 500-line story.\n\nA structurally strong ad can fail provenance. A truth-safe ad can still fail long-form architecture.\n\n## Deterministic audit pass\n\nFor every ad, calculate from the actual Primary Text:\n\n- body word count;\n- first product-name line;\n- first product-name word position as a percentage of body words;\n- awareness level and intended format;\n- repeated product/mechanism sentences across the batch.\n\nUse those measurements as evidence, not as automatic grades. For intentional unaware/problem-aware long form, compare reveal timing with the governing SOP. Early reveal can be correct for compressed or product-aware ads.\n\n## Device matrix\n\nCreate one row per winner/SOP device:\n\n- story-beginning hook;\n- main open loop;\n- avatar world and before-state;\n- gradual accumulation;\n- relational stakes;\n- failed-solutions sequence;\n- irreversible inciting incident;\n- credible discovery or authority bridge;\n- layered mechanism with human processing beats;\n- analogy;\n- objection handling disguised as story;\n- late invisible product reveal;\n- modest future pacing or exact sourced results;\n- relationship resolution;\n- victory as absence;\n- loss-aware close and verified risk reversal;\n- callback/P.S. architecture;\n- native formatting and real-person voice.\n\nGrade each `present`, `partial`, `absent`, or `misused`. Cite exact ad IDs, line numbers, and short excerpts. Explain whether the issue is structural, evidential, or both.\n\n## Per-ad ledger\n\nGive every expected ad one row containing:\n\n- format/awareness fit;\n- strongest owned device or artifact;\n- principal structural weakness;\n- principal provenance/claim risk;\n- confidence/blocker status.\n\nDo not let a batch-level summary hide an individual ad.\n\n## Composite-testimonial rule\n\nMetadata such as “dramatized,” “composite,” or “grounded in VOC” does not cure consumer-facing first-person testimony. If the body says `I`, `my son`, or presents a chronology as something that happened, require one same-speaker receipt for the narrator, event, dialogue, precision, and result.\n\nTruth-safe alternatives:\n\n- exact attributed VOC;\n- explicit hypothetical language in the consumer copy;\n- a non-testimonial ownership argument;\n- desired-absence framing;\n- an exact-SKU customer receipt with preserved attribution.\n\n## Winner-copy boundary\n\nExtract devices from a winner, but audit the winner itself for proof-shaped fabrication. Named people, doctors, locations, clinical numbers, week-by-week outcomes, authority dialogue, and scarcity remain unusable without receipts. Do not describe longevity, duplicate count, or labels such as “winner” as causal performance proof unless a closed performance window supports it.\n\n## Doctrine output\n\nEnd with a locked doctrine specific to the brand that defines:\n\n1. when full long-form is allowed by awareness;\n2. the evidence hierarchy;\n3. the required story sequence;\n4. allowed, qualified, and prohibited mechanism language;\n5. reveal timing by format;\n6. proof/offer verification gates;\n7. cross-ad uniqueness requirements;\n8. the rule that QA language belongs in audit notes, not consumer-facing prose.\n\n## Read-only verification\n\nBefore delivery confirm:\n\n- all expected ads were audited;\n- all sub-7 or blocker findings have exact evidence and, when requested, proposed text;\n- structural weaknesses are separated from truth/provenance risks;\n- source files were not edited;\n- pre-existing repository changes are not attributed to the audit.\n",
    "sha256": "5ab65a462a7697a06d9ddb6d88acaa613d3606d8b89dbfe2ae20aa87a9a181dc"
  },
  {
    "group": "Native copy",
    "name": "references/long-format-native-sop-v2-contract.md",
    "path": "/Users/ayden/.hermes/skills/openclaw-imports/native-copy/references/long-format-native-sop-v2-contract.md",
    "text": "# Long-Format Native Ad SOP 2.0 — binding contract (Fish 2026-09-03: \"follow the doc and what it says. It is more true.\")\n\nSource of truth: Zakaria, \"Long-Format Native Ad SOP 2.0\" (https://native-ads.manus.space/), full text at\n`~/.openclaw/workspace/knowledge/formats/zakaria-long-format-native-sop-v2-2026-09-03.txt`. This file turns it\ninto the rules every long-format native is written and validated against. Where an older reference in this\nskill disagrees (Alevia-derived witness contract, T1/T2 budgets, victory-as-absence, reveal window,\nparagraph prose, analogy ban inside the mechanism), THIS FILE WINS. Audit of the deltas:\n`knowledge/formats/native-sop-v2-audit-2026-09-03.md`.\n\nOne fence stays because the SOP itself requires it (\"true or defensibly true\", \"the reader will Google it\"):\nnever invented science; the mechanism is the ruled chain and the one verified study. The 9-03 relatability\ngate stays for the READER (identified only by the card's categories); the NARRATOR gets a full identity with\nher numbers. Unaware lane stays parked (Fish 8-31); write the PROBLEM-AWARE architecture.\n\n## Alevia deltas ruled in (Fish 2026-09-03 evening, after the measured comparison\n`knowledge/formats/alevia-vs-dawn-v3-comparison-2026-09-03.md`)\n- **Clinician authority is allowed and wanted.** The authority character may be a nurse, a clinician, a\n  specialist, with years stated, as in Alevia's cardiac nurse / cardiologist. Authorized dramatization; she\n  never cites a study that is not ours, never names a drug, never says the product treats ADHD.\n- **Numbers in the results.** Counted, modest, qualified: \"4 mornings out of 5\", \"11 school days without a\n  yell\", \"the first time in 3 years\". The week markers stay.\n- **Alevia paragraphs, not one sentence per line.** 120-180 paragraphs of 2-4 sentences, 15-35 words, one-line\n  paragraphs only at punch moments (measured: all six longest-running Alevia copies, 24-28 words/paragraph).\n  This overrides SOP §10's one-sentence-per-line for our lane.\n- **Narrator identity block with her numbers** (age, work, place, years of the problem) in the first\n  paragraphs, and a **reader bridge** right after the hook (\"I'm writing this because... if you have tried\n  X, Y, Z... read this to the end\"). Reader qualification is by what she TRIED, in her words, never by a\n  product or drug list.\n- **Partial cast increase:** one more person who notices (her own mother or a friend), the kid speaks up to\n  three times. Still no husband, no sibling (alienation rule).\n- **Quoted direct speech** with quotation marks; reported speech (\"She said: ...\") is the transcript register\n  and is out.\n- **Durations on the failed stack** (how long each thing was tried); prices stay out of this lane.\n- **The close carries the Alevia rungs:** loss restate → the honest line (\"I am not going to tell you it was a\n  miracle. I am telling you what happened.\") → guarantee line → the \"If by...\" conditional ladder → 👉 link →\n  the \"you can keep doing what you have been doing, or\" fork → thank-you coda → final emotional line →\n  first-name signature → P.S. (object) / P.P.S. (the honest count) / P.P.P.S. (small runs, do not wait) → 👉\n  bare link last.\n- **Analogies:** one from the authority, one from the narrator repeating it, allowed.\n- The provenance-leak ban is narrowed to QA-speak (\"I won't invent a review\", \"no fake before-and-after\");\n  the honest line is wanted.\n\n## 1. Avatar depth (SOP §01) — the research file before the brief\nAnswers, not guesses, from real people, in a single file the brief points at:\n- daily experience of the problem at the hour it happens · internal monologue (their exact phrases) ·\n  failed attempts IN ORDER · their language (copied verbatim) · relationships (who notices, who picks up the\n  slack, who said the thing that stung, who they fear disappointing) · the fear they will not say out loud.\n- Card slots added to every sub-avatar: BEFORE (who she was before it took over) · MONOLOGUE (3-5 verbatim\n  self-talk lines) · WHO NOTICES · UNSPOKEN FEAR. Each slot is a quote with a source or it stays empty.\n\n## 2. Awareness → architecture (SOP §02)\n- PROBLEM-AWARE (our lane): open on shared experience, the reader self-selects in the first 3 lines; failed\n  attempts stack; the authority explains WHY nothing worked; alternatives eliminated; product = logical\n  conclusion. 500-800 lines.\n- SOLUTION-AWARE: open on failed solutions, mechanism earlier, reveal sooner, more proof. 400-600 lines.\n- PRODUCT-AWARE: never long-format. Short, offer, proof.\n- A \"line\" is one sentence or fragment on its own line.\n\n## 3. Mechanism (SOP §03)\nRoot cause, not feature → the cascade (each link is a section) → why each partial solution fails (it addresses\none link) → the product as the only thing that breaks the full cycle. Must be true, explainable over coffee in\ntwo minutes, a revelation to the reader, and lead to OUR product only. One analogy the authority speaks, that\nthe reader will repeat to a friend. (Analogy is allowed HERE, spoken by the authority, and nowhere else.)\n\n## 4. Story architecture (SOP §04) — line budgets for a ~700-line ad; scale proportionally\n| Part | Lines | Job | Must contain |\n|---|---|---|---|\n| 1 Hook | 1-3 | make them read line 2 | a moment, not a claim; no product/benefit/category; open loop; the reader recognizes her world. Problem-aware: she self-selects in 3 lines (our two hook families) |\n| 2 The World | 4-150 | make her feel SEEN | the BEFORE · the gradual decline as 3-5 small losses, each its own line · failed attempts named specifically and in order (prices where real) · the emotional toll (guilt, identity loss, quiet acceptance) · relationship impact (others notice, compensate, talk). No selling, no hinting. Short lines, white space, repetition for rhythm (\"I tried X. Nothing.\") |\n| 3 Inciting incident | 150-200 | make the status quo impossible | dramatic but believable, visceral, irreversible; introduces the AUTHORITY character |\n| 4 Mechanism | 200-400 | teach WHY she has it and WHY nothing worked | delivered through DIALOGUE; layered (root → cascade → why it compounds); narrator reactions between layers; one analogy; connect back to her life after each layer (\"that is why the calm voice held until nothing moved\"). Longest section after the world |\n| 5 Objection destruction | 400-500 | eliminate every alternative | narrator asks about each thing she tried, in the order she tried it; authority: acknowledge → the gap (it addresses part of the chain) → one memorable summary line. Never aggressive. The reader crosses them off herself |\n| 6 Product reveal | 500-550 (75-80%) | the natural resolution | the AUTHORITY recommends one specific named product; 2-3 specific differentiators that map to the mechanism; 2-3 paragraphs; stay in story voice; no marketing words |\n| 7 Future pacing | 550-650 | show her life after | WEEK-BY-WEEK: wk1 almost nothing · wk2 first small signal · wk3 something unexpected · wk4 measurable · wk6 other people notice · wk8 a modest, specific transformation that mirrors the inciting incident. Mix physical with emotional; end on the RELATIONSHIP RESOLUTION (the person from the hook says the thing) |\n| 8 Close | 650-700+ | convert investment into action | restate the stakes as LOSS · generational stake if real · risk reversal (the guarantee) · one direct CTA + link · final emotional line. Our offer floor rides here (brand + company intro + guarantee + link); the P.S. ladder is allowed after the close with a second link (Fish 8-31) |\n\n## 5. Open loops (SOP §05)\nNever resolve all loops at once · open the next before closing the current · the main loop from line 1 spans at\nleast half the ad · vary WHAT / WHY / HOW loops · loops at section transitions, not every line · loops open\nthrough events and spoken lines (our said-out-loud test still bans announcer narration).\n\n## 6. Mechanism through dialogue (SOP §06)\nThe authority asks permission to explain · physical beats break dense information · the narrator reacts ·\nthe narrator asks the reader's questions · the authority speaks the analogy · PACING: 3-4 lines of mechanism\nto 1 line of reaction or physical beat. If it reads like a textbook, rewrite as conversation.\n\n## 7. Objection handling disguised as story (SOP §07)\nEvery alternative she tried or considered, most common first, most logical last. Acknowledge (it has merit) →\ngap (only part of the chain) → memorable summary. Tone sympathetic: \"right instinct, wrong target.\"\n\n## 8. Late reveal (SOP §08)\nProduct name at 75-80% of the lines, never earlier. The authority recommends it. Differentiators are facts\ntied to the mechanism. Short. The transition is invisible: the last question is \"so what do I actually do,\"\nand the answer is the product.\n\n## 9. Future pacing through results (SOP §09)\nWeek-by-week, modest, specific, qualified; each week has a moment that makes her feel something; other\ncharacters notice; the relationship resolution is the emotional peak and comes last. Internal label:\nAUTHORIZED DRAMATIZATION when no buyer receipt exists (the SOP's own form); a verified Okendo/survey line\nmay replace any week beat.\n\n## 10. Formatting for Meta primary text (SOP §10)\nSentences 5-15 words, rhythm varied · no emojis except 👉 before each link · no bold, no caps, no asterisks,\nno brackets. Block format for our lane is ruled above: Alevia paragraphs (2-4 sentences), one-liners at punch\nmoments only.\n\n## 11. The eight killers (SOP §11) — pre-publish check\nearly reveal · shallow avatar (details that could apply to anyone) · rushed mechanism · tone switch at the\nreveal · overclaimed results · awareness mismatch · no stakes beyond the individual · weak inciting incident.\n\n## 12. Process (SOP §12) and final check (SOP §13)\nResearch → mechanism + outline → fast draft → four edit passes (cut, would-a-real-person-say-this, read\naloud, find the ad switch) → read to someone who has the problem. Final check: line 1 loop · first 100 lines\nproduct-free · narrator sounds like the demographic · \"this is MY story\" specificity · mechanism in dialogue ·\nevery alternative eliminated · invisible reveal · gradual believable results · loss-aversion CTA · would you\nkeep reading.\n\n## Our standing gates that ride on top (unchanged)\nHook names her by symptom in lines 1-2 (two families) · relatability gate (no unverified reader qualifiers,\nno count on her repeats, narrator texture receipted) · no invented clinician, no invented science, no\nfabricated named testimonial with product outcomes · offer floor (brand at reveal, \"Dawn Band is a small\ncompany and this is the only thing they make\", \"100% money-back guarantee. 60 nights.\", \"Trusted by 10,000+\nparents\", link, CTA before and after the P.S. ladder) · never treats / cures ADHD · US spelling.\n",
    "sha256": "e4d2bee55a7b7a9f54d57aa15c066b034851ea4d166579da0eb5675ce29313b8"
  },
  {
    "group": "Native copy",
    "name": "references/long-native-clarity-and-jumble-gate.md",
    "path": "/Users/ayden/.hermes/skills/openclaw-imports/native-copy/references/long-native-clarity-and-jumble-gate.md",
    "text": "# Long-Native Clarity and Final Jumble Gate\n\nUse after structural/SOP validation and before a native is called review-ready. Mechanical readability is necessary but not sufficient: a short sentence can still make no sense.\n\n## 1. Failed-solution sequence\n\nEach attempt gets its own beat in this order:\n\n1. what she tried;\n2. why she thought it would help;\n3. what literally happened;\n4. why she stopped trusting it.\n\nDo not compress multiple methods into fragments such as `X came next`, `I laid it on`, `both helped`, or `neither stopped`. Do not open with `I tried everything the posts said`; name the concrete attempt.\n\nPassing shape:\n\n> I started getting up earlier because I thought I was the problem. I wanted to be ready and calm before he got up. I still ended up yelling by the door.\n\n## 2. Paragraph-level jumble check\n\nRead every paragraph, not only the section recently edited. Flag:\n\n- chronology that jumps backward without a marker;\n- two unrelated actions in one beat;\n- unclear `it`, `that`, `this`, or `under it` references;\n- two speakers in one paragraph when the turn is hard to follow;\n- a method named without why she tried it;\n- a result stated before the action that caused it;\n- repeated week labels or duplicated transition lines;\n- count fragments whose subject must be reconstructed.\n\nSplit speaker turns when necessary. One paragraph should perform one clear story job.\n\n## 3. Semantic directness check\n\nReplace compressed copywriter abstractions with visible action or literal thought.\n\n| Reject | Prefer |\n|---|---|\n| `He was just not started.` | `He stood beside his shoes without reaching for them.` |\n| `He can hear you. He cannot begin.` | `He hears you, but his body still does not move.` |\n| `Your calm is his calm.` | `He hears the tension in your voice, then you get more tense when he does not move.` |\n| `The first thing he does is his own.` | `You do not have to give the first order from the doorway.` |\n| `He comes up with a clearer head.` | `He wakes with a clearer head.` |\n| `He gets lost between fewer steps.` | `He loses track of fewer steps.` |\n| `See what the rest of the hour does.` | `See whether you reach the door without yelling.` |\n\nIf a reader must translate the sentence, it fails even if sentence length and grade score pass.\n\n## 4. Final SOP read\n\nRun the mechanical gate, then answer the SOP final checks by hand:\n\n- line-one loop;\n- first third product-free;\n- narrator sounds like the demographic;\n- this-is-my-story specificity;\n- mechanism delivered through dialogue;\n- every narrator-specific alternative addressed;\n- reveal feels like story resolution;\n- results gradual and believable;\n- close uses loss aversion;\n- would the target reader continue?\n\nAlso check the eight killers. A deterministic PASS does not override an editorial jumble finding.\n\n## 5. Durable bans for this lane\n\n- no coffee filler;\n- no switching among `thread`, `forum`, and `post` in consumer copy—pick one channel;\n- no specific failed solution used as a universal reader qualifier;\n- no vague initiation shorthand;\n- average sentence under 14 words, no sentence over 28;\n- one exact sourced proof line may remain mechanically awkward only when the contract requires it; make the surrounding explanation explicit.\n\n## Verification receipt\n\nRecord final word/paragraph counts, reveal percentage, sentence average/max, grade estimate, clone check, receipt-table result, and a concise list of semantic repairs. Re-run the full gate after the final edit, not before it.",
    "sha256": "b92a589929e75cf349d5468dfa000803bd0c909f1abaecbf3e11e935861fab7d"
  },
  {
    "group": "Native copy",
    "name": "references/master-brief.md",
    "path": "/Users/ayden/.hermes/skills/openclaw-imports/native-copy/references/master-brief.md",
    "text": "# NATIVE AD COPYWRITING MASTER BRIEF\n**Purpose:** HOW to write DawnBands native ad copy. Rules, format, structure, creative architecture. This file does NOT contain avatar knowledge — that lives in separate avatar docs. Read the avatar doc for your target audience BEFORE writing.\n**Date:** 2026-03-08\n\n---\n\n## FORMAT RULES (NON-NEGOTIABLE — CHECK EVERY LINE)\n\n- **1 sentence per line.** Default is always 1.\n- 2 sentences per line ONLY if the flow genuinely reads better that way.\n- **Short fragments: separate for emotional weight, clumped for action sequences.** \"Eyes open. Turned it off. Walked back.\" = one fluid motion, clump on one line. But \"The knocking. / The yelling. / The door slam.\" = each line is its own emotional punch, keep separate. If the fragments build rhythm or tension, they breathe on their own lines. If they describe a sequence of events, they flow together.\n- **NEVER 3+ sentences on one line.**\n- Period = line break. Always.\n- No em-dashes. Ever. Use periods + new lines.\n- No emoji on CTAs except 👉 for links.\n- Dialogue on its own line. Always isolated.\n- After writing all copy, do a FINAL PASS: scan every single line. If any line contains more than 2 sentences, split it.\n\n**Correct:**\n> I went in four times yesterday.\n> By the third time my voice was different.\n> By the fourth I was someone I don't want to be.\n\n**Wrong:**\n> I went in four times yesterday. By the third time my voice was different. By the fourth I was someone I don't want to be.\n\n---\n\n## META AD STRUCTURE (CRITICAL — understand this or the ad fails)\n\n**Primary text = the ad. She reads it FIRST when scrolling.**\n- The hook/WTF line MUST be the first line of primary text.\n- Never waste your best scroll-stopping line anywhere but line 1 of primary text.\n- **Lead with the most INTRIGUING element, not the chronological start.** Find the line that creates the most curiosity or emotional reaction and open with it — even if it's out of order in the story. \"My husband thought I lost my mind\" > \"Six alarm clocks.\" because she hasn't explained WHY yet. The reader HAS to keep going.\n- **Never scene-set in line 1.** No \"She stopped me in the driveway.\" No \"I did the math once.\" No inventory lists. No context. The hook should CREATE a question, not answer one.\n- **Quotes hit hardest.** Direct dialogue as line 1 (\"Do you need help? I hear screaming from your house every morning.\") drops the reader into the middle of a moment with zero warmup.\n\n**Headline = bottom of the ad. Below the image. She reads it LAST.**\n- Use for: reinforcement, CTA summary, or a second punch.\n- NOT the hook. She already read the whole ad before she gets here.\n\n## HEADLINE RULES (Meta `title` field)\n\nEvery copy variant needs its own headline. This appears below the image in the ad.\n\n**Rules:**\n- Reinforces the body copy angle — NOT the hook (hook is in primary text)\n- First person voice preferred — sounds like her, not a brand\n- Sentence case. Not title case. Not ALL CAPS.\n- No emoji in headlines\n- Long headlines OK — Meta truncates but the full text shows on click\n- Must NOT duplicate any headline from another variant in the same set\n\n**Headline by variant type:**\n- **Story/Resignation:** The specific moment or realization. \"I Did Everything His Therapist Told Me. He Still Couldn't Wake Up.\"\n- **Objection Killer:** The skepticism + the twist. \"I've Spent $300 On Alarms. He Slept Through Every One. Then I Tried This.\"\n- **Pain-First → Mechanism:** The mystery/revelation. \"His Medication Wears Off at 4pm. By 6:30am He's Unreachable. Nobody Explained That Cycle to Me.\"\n- **Consequence/Future Fear:** The stakes. \"He Turns 18 in Two Years. Nobody Is Going to Stand Outside His Dorm Room.\"\n- **Risk Reversal + Social Proof:** The proof + offer. \"9,000+ Reviews. 60-Night Guarantee. The Only Alarm Built For ADHD Brains.\"\n\n---\n\n## PRODUCT\n\n### Offer-truth precedence\nBefore writing, read the current validated brief and canonical product-truth file. Those sources override this reference and every historical example. Never inherit an old price, guarantee, app/setup claim, destination, or mechanism claim from a sample ad. When the current DawnBands offer is specified, preserve the token order `100% wake-up guarantee. 60 nights. $49.` while placing each sentence or fragment on its own line.\n\n- **Dawn Bands** — silent vibrating wrist alarm\n- $49 | 100% wake-up guarantee | 60 nights\n- Black silicone with ribbed diagonal lines; no visible screen or buttons; no app or subscription\n- Standalone band that places a vibration cue directly on the wrist. Describe a different sensory channel and signal placement, not a stronger signal.\n- Say `harder to sleep through`, never impossible or guaranteed every time.\n- Main PDP: dawnbands.com/products/wake-up-band-for-teens\n- ADHD PDP: dawnbands.com/products/wake-up-band-adhd\n- **Main listicle:** dawnbands.com/pages/7-reasons-teen-sleeps-through-alarms\n- **ADHD listicle:** dawnbands.com/pages/adhd-teen-alarm\n\n---\n\n## THE MECHANISM (load the canonical selling doctrine)\n\nBefore writing DawnBands copy, read `/Users/ayden/.openclaw/workspace/knowledge/brands/dawnbands/selling-root-cause-mechanism.md`. It governs this section.\n\nThe default selling chain is **Deep-Sleep Signal Lock → Somatic Wake Bypass™ → Wrist-First Handoff™**:\n\n1. adolescent circadian delay keeps biological night running later\n2. accumulated sleep debt increases the pressure holding him asleep\n3. the sleeping brain gates/suppresses repeated room sound\n4. sleep inertia lets his body silence alarms or answer before conscious initiation and memory return\n5. Mom becomes the emergency pathway that finally breaks through\n6. DawnBands bypasses the muted sound pathway with a tactile signal on the wrist\n7. the first cue begins on him before it ends in her voice\n\nUse aggressive biological simplification to sell the mechanism. Do not flatten the copy to “sound and vibration are different.” Preserve separate substantiation discipline for exact-SKU superiority, fabricated testimonials, customer timelines, offer facts, and attributed proof.\n\n**The mechanism belongs in the MIDDLE of long-form problem-aware copy. NEVER use science or product mechanics as the opening hook.** Compressed solution-aware/product-aware formats may reveal the mechanism earlier when the validated brief explicitly calls for it.\n\n**Attach the canonical villain:** the repeated room-sound rescue system. It keeps turning up the pathway his sleeping brain already suppresses, then outsources the final wake-up job to Mom. Redirect blame away from both the teen and parent.\n\n---\n\n## AVATAR DOCS (READ BEFORE WRITING — MANDATORY)\n\n**The avatar docs contain the raw quotes, exact language, interior world, and copy triggers for each target audience. This brief tells you HOW to write. The avatar docs tell you WHO you're writing to and WHAT words to use.**\n\n**⚠️ DEFAULT TO THEIR WORDS OVER YOUR OWN — always.** When you're reaching for a word, check the avatar doc's language bank FIRST. If their word exists for what you're trying to say, use theirs. Your instinct will be to write something \"better\" — don't. Their words convert. Yours don't.\n\n**Read the avatar doc for your assigned target BEFORE writing a single line.**\n\n\n> ⭐ **SINGLE AVATAR RULING (Fish 2026-09-02).** Every DawnBands brief targets the mother of a 13–18 ADHD teen. Load `knowledge/brands/dawnbands/adhd-mom-lockin.md` FIRST, then `dawnbands-avatar-adhd-teen-mom.md`. The \"Teen Mom (primary)\" generic row below is PARKED — do not brief it. ADHD is named in the hook. Survey basis is n=138 (`survey-intel-post-purchase-2026-09-01.md`).\n\n| Avatar | Doc Path | Who They Are |\n|---|---|---|\n| **Teen Mom** (primary) | `knowledge/dawnbands-avatar-interior-2026.md` | Mom 38-49, teen 14-17 who can't wake up. She's become his alarm clock. |\n| **ADHD Mom** | `knowledge/dawnbands-avatar-interior-2026.md` (Part 7) | Same mom, but teen has ADHD. More informed, more exhausted, more isolated. |\n| **ADHD Adult Self-Buyer** | `knowledge/dawnbands-avatar-adhd-adult.md` | Adult 25-45 with ADHD buying for themselves. Shame, career consequences, exec function wall. |\n| **Partner/Spouse Buyer** | `knowledge/dawnbands-avatar-partner-spouse.md` | Partner buying because THEIR sleep is ruined by the heavy sleeper's alarms. |\n| **Broad Deep Sleeper** | `knowledge/dawnbands-avatar-broad-deepsleeper.md` | Adult heavy sleeper (any age) buying for themselves. Shift workers, college students, professionals. |\n\n**Victory = ABSENCE, not presence.** Across ALL avatars. She/he doesn't celebrate. They notice what STOPPED happening. Write victory as what disappeared, not what arrived. \"I didn't yell today.\" \"He hasn't missed a shift in six weeks.\" \"I slept through the night for the first time in three years.\"\n\n---\n\n## ZAK FRAMEWORK — NATIVE AD STRUCTURE\n\n**Line-by-line structure:**\n1. **Hook (first 3-5 lines):** Pattern interrupt. NOT an ad. Use her exact language. Must pass Gate 1-3 (interrupt → earn attention → recognition).\n2. **Story opening:** First person, raw. Start in a specific moment. Small detail, not summary. (\"I pulled his covers at 6:42\" not \"mornings were hard\")\n3. **Agitation:** Surface the daily emotional weight. \"Tried Everything\" stack. Breaking point moment.\n4. **Root cause / mechanism:** NOT the product. The villain (sound). The explanation. One analogy max. Short punchy sentences. **This goes in the MIDDLE.**\n5. **Product reveal + CTA:** \"Of course. This is what I've been looking for.\" CTA goes HERE — immediately at product reveal.\n6. **Close:** Concrete outcome. Risk reversal. Guarantee.\n\n**Hook formulas (match to awareness level):**\n- Formula 1: PAIN → VALIDATION → INTRIGUE\n- Formula 2: STORY → TENSION → CLIFFHANGER  \n- Formula 3: QUESTION → TWIST → PROMISE\n- Cold Meta traffic = Levels 1-2 awareness. Meet her there.\n\n**Cadence rule:** Alternate Long (build) → Short (punch) → Short (promise) → Long (explain) → Short (reveal). Read out loud. If you stumble, rewrite.\n\n---\n\n## TARGETING — WHICH LISTICLE + PDP\n\n**4 listicles, 1 PDP. Route based on avatar + angle:**\n\n| Avatar/Angle | Listicle | PDP |\n|---|---|---|\n| Teen mom (general) | `dawnbands.com/pages/7-reasons-teen-sleeps-through-alarms` | `/products/wake-up-band-adhd` |\n| ADHD mom (parent voice) | `dawnbands.com/pages/adhd-teen-alarm` | `/products/wake-up-band-adhd` |\n| ADHD adult (self-buyer) | `dawnbands.com/pages/adhd-adult-alarm` | `/products/wake-up-band-adhd` |\n| Broad deep sleeper / partner / spouse | `dawnbands.com/pages/why-you-sleep-through-alarms` | `/products/wake-up-band-adhd` |\n\n**Every new ad gets duped to BOTH listicle + PDP.** Same creative, two destinations. Meta picks the winner.\n\n**Routing rules:**\n- Any ad mentioning ADHD, medication, executive function → ADHD listicle (mom or adult version based on voice)\n- Any ad about teens who can't wake up (no ADHD) → Teen mom listicle\n- Any ad from adult's own perspective (shift worker, college, professional, partner) → Broad listicle\n- When in doubt → Broad listicle (it's the most general)\n\n---\n\n## 4 VARIANT TYPES (write one of each per ad)\n\n**⚠️ OBJECTION KILLER = TOP PERFORMER (Mar 2026 data)**\nLive data from Teen Mom Natives campaign shows Objection Killer variants converting across multiple ads:\n- teen-mattress-retool V2 (objection): $9.55 CPP ← best in campaign\n- teen-relationship-restoration V2 (objection): $6.53 CPP\n- teen-identity-erosion V2 (objection): $49.93 CPP but only variant converting\n- teen-teens-voice V1 (objection-style): $19.62 CPP\n**Implication:** She's skepticism-first, story-second. Even story variants should open with a beat of \"I know, I know, another thing\" energy before going into the narrative. The drawer of failed products is her entry point.\n\n**1. Story/Resignation Arc**\n- First person narrative. The mom who tried everything.\n- Entry point: a specific failed moment or the resignation breaking point.\n- **Consider opening with a skepticism beat even in story format** — she resonates with \"I wasn't looking for this\" energy.\n- Emotion: exhaustion → discovery → quiet relief.\n- Awareness: Level 2 (Problem Aware).\n\n**2. Objection Killer** ← PRIORITY VARIANT, write 2 per ad if possible\n- Leads with the skepticism she already has.\n- \"Won't he just sleep through this too?\" / \"Another alarm?\" / \"$49 for a vibrating wristband?\"\n- Disarms before selling. Names the drawer of failed products.\n- Emotion: cynicism → surprise → belief.\n- Awareness: Level 3 (Solution Aware).\n\n**3. Pain-First → Mechanism Reveal**\n- Opens with vivid daily pain (time-stamped morning sequence, sensory detail).\n- Mechanism lands in the middle as the \"nobody explained this\" revelation.\n- Emotion: confusion/frustration → clarity → hope.\n- Awareness: Level 2 (Problem Aware).\n- **DO NOT open with mechanism. Open with pain.**\n\n**4. Consequence/Future Fear**\n- What happens if nothing changes. College, jobs, independence, relationship.\n- The clock is ticking. Urgency from reality, not marketing.\n- Emotion: fear → urgency → action.\n- Awareness: Level 2-3.\n\n---\n\n## VARIANT DIVERSITY RULES (MANDATORY — prevents Andromeda overlap)\n\n### Mechanism Delivery Must Vary Across Variants\nThe mechanism (auditory channel closes → touch stays open) appears in every variant. **It must be delivered differently each time.**\n\n**Do NOT copy-paste the same mechanism paragraph across 5 variants.** A real person wouldn't explain it the same way twice.\n\nMechanism delivery styles (use a DIFFERENT one per variant):\n1. **Full science** — \"The sleeping brain suppresses the auditory channel...\" (max 1 variant per ad)\n2. **Casual/vague** — \"I read something about how sound doesn't actually reach deep sleepers. Something about the brain closing a door.\" (no jargon)\n3. **Someone else explained it** — \"My sister sent me an article. Apparently the brain just... blocks sound when you're that deep.\" (secondhand)\n4. **Barely explained** — \"It vibrates instead of making noise. I don't fully understand why that's different but it is.\" (outcome only)\n5. **Analogy** — \"It's like knocking on a soundproof door vs. tapping someone on the shoulder.\" (metaphor)\n6. **Skip entirely** — Some variants (especially social proof V5) can just say \"it works differently than sound alarms\" and move on.\n\n### CTA Format Must Vary\nDo NOT use \"👉 Here's [reason/explanation]:\" for every variant. Real people link differently.\n\nCTA styles (rotate):\n- \"👉 Here's what I read:\" + URL (max 2 per ad)\n- \"I'll put the link below.\" (no 👉)\n- \"Link is here if you want to read it:\" + URL\n- \"The explanation is at [URL]\" (embedded naturally)\n- \"Someone posted this in my group:\" + URL\n- No mid-copy CTA at all — just the close CTA\n\n**CTA must flow with the story, not feel templated.** Do NOT copy-paste \"$49. 60 nights. Full refund.\" on every variant. Weave guarantee/price into the narrative naturally:\n- \"I had 60 nights to send it back. Never even considered it.\"\n- \"$49. Full refund if it joins the drawer. It didn't.\"\n- Some variants skip PRICE entirely — the number is optional (Fish 2026-08-31).\n- Some variants bury the guarantee in a callback line, not a bullet list.\nIf the CTA block reads like it was pasted onto 5 different stories, rewrite it.\n\n⚠️ **OFFER FLOOR (Fish 2026-08-31 — overrides the variation permissions above).** Varying the CTA never means DELETING the close. Every body, every length, every destination carries: brand named at the reveal, the guarantee stated plainly, a 2-4 line company/credibility intro, and at least one link. Price is the only optional element. A body with no brand, no offer, and no link is a click generator, not a native — receipt: v11-a-unassigned-hour, 4.13% link CTR at $0.92 CPLC, $46-56 CPA. Full rule + close-block shape: SKILL.md §3a.\n\n### Social Proof Variants (V5) Must Sound Real\n- Reviews should have typos, incomplete thoughts, weird phrasing\n- Not every review is 5 stars. Mix in a 4-star that still converts.\n- Names should feel real (not \"Trish M.\" and \"Kara D.\" — too clean)\n- Include at least one review that sounds like it was written at 2am on a phone\n- Reference platform (\"posted this in an ADHD group\", \"saw this on Amazon\") to ground it\n\n### Body Copy Length Variation (CRITICAL — Mar 2026 data)\n**Both long AND short are converting.** n6-medication original (706 words) shows signal. Short teen mom copies also cooking hard. DO NOT default to one length.\n\n**Per ad, aim for this mix:**\n- 2-3 variants LONG: 500-700 words\n- 2 variants SHORT-MEDIUM: 150-300 words (punchy, quick, scroll-stop-and-convert)\n- The Objection Killer variants can be either — some of the best performers are medium-length objection openers\n- Let Meta test which length wins per angle\n\n**⚠️ \"LONG\" does NOT mean dense paragraphs or walls of text.** Long variants still use 1-2 sentences per line with lots of line breaks. The length comes from MORE lines, not longer paragraphs. Think: a long Facebook post that's easy to scroll through, not an essay. Even ADHD moms need to be able to scan it. The 1-sentence-per-line format rule applies to all **Mode A** variants regardless of total word count — \"LONG\" in this section means 500–700-word Mode A copy. **It does NOT apply to Mode B long-form witness natives**, which run PARAGRAPH prose per the 2026-07-31 ruling (`dawnbands-long-form-witness-native.md` → Formatting; measured basis in `alevia-live-anatomy-2026-08-12.md`). **Long-format natives are governed by SOP 2.0 as of 2026-09-03** (`long-format-native-sop-v2-contract.md`: eight parts, Alevia paragraphs of 2-4 sentences, clinician authority in dialogue, counted week-by-week results, reveal 65-82%, close rungs + P.S. ladder); that contract wins over this section wherever they disagree.\n\n**Do NOT write all 5 variants at the same length.** Variety in length = more data on what converts per angle.\n\n---\n\n## STRUCTURAL CADENCE PATTERNS (from proven winners — Uflora, True Nutra, Curvani)\n\nThese patterns are extracted from gold-standard native ads that convert at 9 figures. Subagents MUST use these rhythms. Don't invent your own cadence — follow these.\n\n### Repetition / Anaphora\nBuild rhythm through repeated structure on separate lines:\n```\nA smell that followed me out of the bathroom.\nA smell I could catch through my jeans if I sat wrong.\nA smell that made me terrified to be in a car with the windows up.\n```\n\n### Internal Monologue Stacking\nEach thought on its own line. Short. Obsessive. Circular:\n```\nCan she smell me right now?\nCan she smell me?\nOh God, can she smell me?\n```\n\n### Failed Solutions List\nShort. Stacked. Each its own line. Include price and verdict:\n```\nSummer's Eve wipes. Worse.\nBoric acid suppositories. Burned like hell. Came right back.\nAntibiotics from my doctor. Ten days of relief. Then worse than before.\n```\n\n### Dialogue Isolation\nDialogue ALWAYS gets its own line. Never buried in a paragraph:\n```\nShe looked at me and said:\n\n\"You're not unhygienic.\nYour microbiome has collapsed.\"\n```\n\n### Trigger Scenarios\nQuestion + answer, each its own line. Shows the pattern:\n```\nAfter sex?\nSemen temporarily raises vaginal pH. Gardnerella feeds.\n\nDuring your period?\nBlood contains iron. Gardnerella's favorite fuel.\n```\nFor DawnBands: \"Field trip? She overslept. Test day? She overslept. First day of school? She overslept.\"\n\n### Timestamped Transformation\nSpecific timestamps for the victory section. NOT \"after a few weeks.\" Concrete:\n```\nDay 7: I got through an entire morning without checking once.\nNot once.\n\nDay 9: Had sex. Lay there afterward, waiting.\nLike you wait for thunder after lightning.\n\nNothing.\n```\nFor DawnBands: \"Day 3: He woke up before I did. I lay there listening to footsteps in the hallway.\"\n\n### The External Witness\nSomeone else notices the change BEFORE the protagonist announces it. Short. Devastating:\n```\nMy husband noticed before I told him.\n\n\"You're lighter,\" he said.\n\nNot my weight.\n```\nFor DawnBands: \"His teacher emailed me. 'Whatever you changed, keep doing it. He's been on time every day this week.'\"\n\n### The \"Waiting for It to Fail\" Beat\nAfter the product works, she waits for it to stop working. This validates her skepticism and makes the victory real:\n```\nLike you wait for thunder after lightning.\n\nNothing.\n```\nFor DawnBands: \"I waited for the text from school. It didn't come. Not that day. Not the next.\"\n\n### The \"No X. No Y. No Z.\" Denial Rhythm\nStack what DIDN'T happen. Powerful for transformation sections:\n```\nNo belly bulge. No puff. No 'why am I gaining everywhere but my booty?'\n```\nFor DawnBands: \"No yelling. No third trip upstairs. No 'you're going to be LATE.'\"\n\n### The P.S. Emotional Bomb\nThe most emotionally devastating line goes in a P.S. at the end. It hits harder because it feels like an afterthought — something too vulnerable for the main copy:\n```\n💡 P.S. He told me once — in the car, not looking at me — that he hates himself every morning that I have to come in. He was thirteen. I still think about that.\n```\n\n---\n\n## CREATIVE ARCHITECTURE (from 8-9 figure operators)\n\n### PDA Framework (Persona × Desire × Awareness)\nWhen writing 5 variants, don't just swap headlines on the same angle. Each variant should vary across at least ONE of these dimensions:\n- **Persona**: Who is telling the story? (exhausted mom, guilty dad, the teen herself, a teacher, a neighbor)\n- **Desire**: What does she want? (calm mornings, relationship repair, independence for him, her own identity back)\n- **Awareness**: Where is she? (doesn't know vibrating alarms exist, knows but skeptical, tried competitors and failed)\nIf all 5 variants target the same persona with the same desire at the same awareness level, you've written 1 ad 5 times.\n\n### Emotional Voltage\nAn angle is NOT an emotion. An angle is a belief system/story that contains 3-5 emotional triggers in sequence. The triggers build cumulative voltage — each line adds charge until the reader HAS to act.\n- **Low voltage**: \"Mornings are hard.\" (one flat emotion)\n- **High voltage**: Shame (I'm the yelling mom) → Fear (he'll never be independent) → Guilt (he hates himself for it) → Hope (but what if) → Relief (he walked downstairs on his own)\nPick ONE primary emotion to build to peak. Layer supporting triggers to increase voltage. Don't scatter — BUILD.\n\n### \"Make Inaction More Dangerous Than Action\"\nThe reader must feel that NOT clicking is riskier than clicking. The copy should make doing nothing feel like a choice with consequences:\n- \"In three years he starts college. Who's going to wake him up?\"\n- \"847 mornings. That number only goes up.\"\n- \"Every morning I don't fix this, I'm teaching him he can't do it without me.\"\n\n### Identity Loss = Meta-Pattern, Not an Emotion\nIdentity loss (\"I'm not the mom I wanted to be\") isn't a single emotion to trigger — it's the REASON multiple emotions appear simultaneously. Don't write \"she feels identity loss.\" Write the specific emotions that identity loss causes: shame, grief, anger at herself, longing for who she used to be. Channel the pattern through specific moments, not labels.\n\n### Emotional State First, Then Demographics\nDon't start by writing to \"mom 38-55.\" Start by finding the emotional state that converts (resignation, shame, exhaustion, fear of permanent damage). THEN layer in demographic-specific scenarios. The emotional state is shared across similar demographics — that's why it scales. The scenario makes it feel personal.\n\n### Mechanism = Highest Leverage Variable\nSame product, same funnel, same traffic — changing the mechanism explanation alone has driven operators from breakeven to $800k/month. The mechanism is not \"feature X.\" It's the WHY behind why everything else failed and why THIS works.\nFor DawnBands: \"Sound goes through air. His brain filters air. Touch goes through nerve endings. His brain can't filter nerve endings.\" — THIS is the mechanism. If it's missing or weak, nothing else saves the ad.\n\n### Mechanism Reveal Structure (4 beats)\n1. **Setup (create the gap)**: Acknowledge her pain, list what she's tried, highlight why it all failed. Don't reveal why yet.\n2. **Reveal (the insight)**: The hidden truth — teen brains suppress SOUND during deep sleep. It's biological, not behavioral.\n3. **Connection (link to HER)**: This is why the Sonic Boom failed. This is why 5 alarms failed. This is why SHE had to become the alarm.\n4. **Solution bridge (inevitable product)**: Touch bypasses the auditory channel entirely. A vibration on the wrist can't be filtered. DawnBands.\n\n---\n\n## ⛔ NEVER DO (CHECK EVERY VARIANT AGAINST THIS LIST BEFORE SUBMITTING)\n\nThis is the most important section. Violating ANY of these = rewrite the variant from scratch.\n\n### HOOKS — NEVER\n- ❌ NEVER open with mechanism / science / \"the brain does X\" — DEAD ($20 spend, 0 purchases)\n- ❌ NEVER scene-set in line 1. No \"She stopped me in the driveway.\" No \"I did the math once.\" No \"I'm sitting in the drop-off line.\" Hook = INTRIGUE, not context.\n- ❌ NEVER start chronologically. The most intriguing moment goes first, even if it happened in the middle of the story.\n- ❌ NEVER put your best line in the headline field. Headline = bottom of the ad. She reads it LAST. Your scroll-stopper goes in primary text line 1.\n- ❌ NEVER open with an inventory list. \"Six alarm clocks. Two phone docks.\" is not a hook. \"My husband thought I lost my mind.\" is.\n- ❌ NEVER use generic feature hooks (\"silent alarm,\" \"for heavy sleepers,\" \"comfortable\") — commodity, every competitor uses these.\n- ❌ NEVER use curiosity-click bait — native1 had 24.8% CTR, 0 purchases. Intrigue must lead to the sale, not just the click.\n\n### LANGUAGE — NEVER\n- ❌ NEVER use: \"game changer,\" \"life changing,\" \"revolutionary,\" \"amazing,\" \"incredible,\" \"journey,\" \"struggle,\" \"battle,\" \"warrior,\" \"graveyard\"\n- ❌ NEVER use weak modifiers: \"very,\" \"really,\" \"just,\" \"truly,\" \"literally\"\n- ❌ NEVER use em-dashes. Ever. Use periods + new lines.\n- ❌ NEVER use a word that feels like every AI would pick it. If it sounds \"writerly\" or poetic where a real mom would say something uglier, kill it. \"The drawer I can't open without getting mad\" > \"the graveyard of failed alarms.\"\n- ❌ NEVER use polished, corporate-sounding testimonials. Reddit/Amazon voice converts. \"Literally changed our lives ⭐⭐⭐⭐⭐\" = death.\n- ❌ NEVER lead with reviews/testimonials. Defeats native format — looks like an ad instantly. Reviews = body, never the hook.\n\n### FORMAT — NEVER\n- ❌ NEVER put 3+ sentences on one line. Max is 2 (and only when it flows better).\n- ❌ NEVER write dense paragraphs. \"Long\" = more lines with breathing room, NOT bigger paragraphs.\n- ❌ NEVER bury dialogue inside a paragraph. Dialogue gets its own line. Always.\n- ❌ NEVER write all 5 variants at the same word count. Mix long (500-700) and short (150-300).\n\n### STRUCTURE — NEVER\n- ❌ NEVER put mechanism in the opening. Mechanism goes in the MIDDLE after the emotional setup.\n- ❌ NEVER copy-paste the same mechanism paragraph across variants. Each variant delivers the mechanism differently (full science, casual, secondhand, barely explained, analogy, or skip).\n- ❌ NEVER copy-paste the same CTA block across variants. Each CTA flows with its own story. No templated \"$49. 60 nights. Full refund.\" on every variant.\n- ❌ NEVER reuse an analogy across variants. \"Locked door\" in V1 = can't appear in V2-V5.\n- ❌ NEVER write 5 variants with the same persona, desire, and awareness level. That's 1 ad written 5 times. Vary at least one dimension per variant.\n- ❌ NEVER write victory as what STARTED happening. Victory = ABSENCE. What STOPPED. \"I didn't yell today.\" Not \"Mornings are peaceful now.\"\n- ❌ NEVER skip the villain. If the copy doesn't blame sound alarms / the alarm industry / bad advice, she'll blame herself and freeze. No sale.\n\n### AVATAR — NEVER\n- ❌ NEVER use generic avatar language (\"stressed professional,\" \"busy mom\"). Use HER exact words from the VOC bank.\n- ❌ NEVER write to the wrong awareness level. Cold Meta traffic = Level 1-2. Don't write Level 4 copy to people who don't know the product exists.\n- ❌ NEVER blame or shame the avatar. She already feels terrible. The copy should make her feel understood, then redirect blame to the real problem (sound alarms).\n- ❌ NEVER say \"sometimes\" or \"usually\" — she says \"every single morning.\" Always absolute.\n- ❌ NEVER future-pace vaguely. \"You'll feel better\" = nothing. \"He walked downstairs on his own Tuesday\" = conversion.\n- ❌ NEVER try to trigger all emotions at once. Pick ONE primary, build voltage through supporting triggers in sequence.\n\n### DEAD ANGLES (CONFIRMED BY DATA — $0 REVENUE)\n- ❌ Mechanism as hook — $20 spend, 0 purchases\n- ❌ Physical letter/document as hook image — $38 spend, 0 purchases\n- ❌ Curiosity-click bait — 24.8% CTR, 0 purchases across 2 geos\n\n---\n\n## GOLD STANDARD EXAMPLE\n\nSee the attached file `gold-standard-example.md` — the native8 teen mom test batch (Mar 7 2026). This follows ALL current rules including hook placement, CTA flow, fragment pairing, cadence patterns, and the ⛔ NEVER DO list. Use this as the quality bar.\n",
    "sha256": "fd399687215eb98086f327be5b6cf258a95fc9d32659a25e10b1ddc216fe31ac"
  },
  {
    "group": "Native copy",
    "name": "references/matrix-angle-execution-map.md",
    "path": "/Users/ayden/.hermes/skills/openclaw-imports/native-copy/references/matrix-angle-execution-map.md",
    "text": "# Matrix → execution map (2026-08-13)\n\nWhen Fish consults the ad-batch-hub testing matrix (\"expand my tests\", \"feed cell X\", \"the\ngrey cells\") this file resolves each matrix ANGLE column into the writing execution: what the\nangle IS, its device basis, the DawnBands form, default mode/length, live receipts, and the\nclaim constraints. Matrix concepts (static / native-short / native-long) select Mode A vs\nwitness architectures as usual; this map governs the angle layer.\n\nAngle definitions are HOOK-level (the matrix scores angle off the hook — the ad's strategic\nbet), matching `build_matrix.py`. An ad's body may touch several angles; the hook decides.\n\n## The original eight (established — one-liners)\n\n| Angle | The hook's bet | Default execution |\n|---|---|---|\n| mechanism | why sound fails / how waking actually works | Mode A mid-form; mechanism is Stage 4-5 saturated — never the differentiator, always the mid-copy efficacy answer |\n| reframe | \"not laziness/discipline — biology\" as a CLAIM | statics (winner skeleton) + Mode A; our #2 angle |\n| relatability | recognition punch — her morning, mirrored | any length; lines 3-8 VOC-specificity test decides life or death |\n| stakes | future-fear with a countdown (college, job, 18) | Mode A mid or compressed witness; milestone-rupture opener form folds here |\n| liberation | identity exit — \"never be their alarm clock\" | short/mid; our #1 angle |\n| proof | it worked — receipts, reviews, retro-proof | compressed witness (fable-v10 cell) or testimonial statics; see devices below |\n| usp | product-forward features/guarantee/objection-killer | Mode A short; objection-killer is the top Mode A performer |\n| offer | promo/price/occasion | short; brand-page checklist evergreen (T4) lives here |\n\n## The three Alevia columns (added 8-12 — census-confirmed the complete transferable set)\n\n### wrong-label — \"Wrong-label reveal\"\n**Is:** a narrated CHRONOLOGY of the label being wrong — years of treating it as character\n(lazy / defiant / unmotivated), then the verdict falls. Distinct from reframe: reframe states\nthe claim; wrong-label lives the mislabel and confesses it.\n**Alevia basis:** wrong-diagnosis reveal, 50 copies/123 ads (\"turns out I never had X at all\").\n**Dawn form:** mislabel is BEHAVIORAL, never medical. The mom (or adult self) mis-graded the\nkid/themselves; the evaluation or realization re-reads history. Mislabel artifacts: charts,\ncontracts, punishments enforced. The apology beat is this angle's strongest weapon.\n**Mode/length:** compressed witness or true Mode B; the chronology needs room — never a static.\n**Receipts:** u3-wronglabel-star-chart (+ modeb), u4 sleep-study (staged 8-12); newdoc-v1 ($38).\n**Gates:** the wrongness is about effort/character judgments, never a doctor's medical error;\nsleep-study-normal is grief, not malpractice.\n\n### defiance — \"Authority defiance\"\n**Is:** villain = the dismissal chain; narrator stops WAITING for the system to own the problem.\n**Alevia basis:** treatment refusal, 44/93 (\"tore up the prescription\") — their form is medical\nrefusal; OURS IS NOT.\n**Dawn form:** PROCEDURAL defiance only — \"he'll grow out of it,\" natural-consequences advice,\nwaitlists, the care plan with no morning section. Mom keeps the meds, keeps the appointments,\ndefies the process by solving 6:30 herself without a referral.\n**Mode/length:** compressed witness / Mode B; the dismissal chain is a chronology.\n**Receipts:** u5-defiance-waitlist (+ modeb, staged 8-12).\n**Gates:** never anti-medication, never anti-doctor as people; zero treatment/efficacy claims;\nthe defiance object is a process (hold music, handouts, blank page four), not a treatment.\n\n### witness — \"Witness POV\"\n**Is:** the narrator is a third party OUTSIDE the buyer-sleeper dyad — dad watching wife+son,\nteacher watching cohorts, observed-peer (\"a mom I know\"). NOT the managing mom narrating her\nown kid: that is the witness FORMAT (Mode B's default stance) and classifies by its hook angle.\n**Alevia basis:** spouse-witness at volume (their most-used device, 40/90 third-party-coded).\n**Dawn form:** two generations of it now — (1) single-scene observer (old: dad-614am recording,\nr2-d2 dorm-call peer), (2) witness-as-buyer: the observer runs the shift, fails, researches,\nBUYS — full DR arc carried by the witness (new: u1 dad-legal-pad). Teacher = longitudinal\noccupational witness, authority via observation, product enters through the mom's dialogue.\n**Mode/length:** compressed witness or Mode B.\n**Receipts:** dad-614am $19.63 @ 6.67 (thin, positive), r2-d2 $514 @ ~1.55, km08 $37;\nu1/u2 (+ u1-modeb) staged 8-12.\n**Gates:** the witness never fabricates medical authority; occupational witnesses speak from\nobservation, and any diagnosis mention is carried with the family's stated permission in-story.\n\n## Proof-column DEVICES (census 8-13 — these are not new columns)\n\n- **Comparative two-person proof** (Alevia 25/63): side-by-side lives — \"her son wakes himself\n  to a wrist tap / mine gets a mother.\" Two kids, two houses, same alarms, one variable.\n  UNTESTED for Dawn; classifier now files it to proof. Pilot-grade device.\n- **Aggregated survey voice** (\"we asked 40 buyers…\"): white space even vs Alevia (they run it\n  only as a PDP block, zero ads). Files to proof. Real survey data only (n=40 post-purchase),\n  never invented percentages.\n- **Retro-proof** (fable-v10 / S2 receipts): victory-as-absence receipts (zero missed 8ams,\n  empty email folder, lobby-call text). The proven core of the column.\n\n## Devices that are NOT matrix angles (skill-side, any column)\n\nGuilt-confession opener · discovery-chain reveal · P.S. ladder close · rupture-object cold\nopen · milestone rupture (files to stakes). See `alevia-live-anatomy-2026-08-12.md`.\n\n**Deliberately absent from the matrix:** death-stakes (Alevia's biggest payload, 175/331 —\nbanned register for this brand); medical bylines/clinic authority; teen self-voice.\n\n## Consult flow (when Fish says \"use the matrix to expand tests\")\n\n1. Pull the live matrix (hub #matrix or `site/matrix/data/matrix.json`) — grey/thin cells in\n   ACTIVE lanes (parent-adhd, parent-teen, adult-self) are the test surface; pixel ruling says\n   on-avatar first.\n2. Resolve cell → this map for angle execution + mode; concept axis picks Mode A vs witness.\n3. Row-per-ad diversity matrix, pilot, surgical pass, per the SKILL workflow. Every artifact/\n   dollar/analogy/victory belongs to exactly one ad in a batch.\n",
    "sha256": "85a73e7ad81491f98be0e5851fe4d86812a829073d028fa0d61105c9f63bc6bf"
  },
  {
    "group": "Native copy",
    "name": "references/mode-b-first-draft-checklist.md",
    "path": "/Users/ayden/.hermes/skills/openclaw-imports/native-copy/references/mode-b-first-draft-checklist.md",
    "text": "# Mode B first-draft checklist (DawnBands)\n\nUse with `dawnbands-long-form-witness-native.md`. This is the short operator card for drafting, not a second doctrine.\n\n## Before prose\n\n- [ ] Brief intentionally selected Mode B (not assumed from “write a native”)\n- [ ] Narrator mode declared — default **ownership-voice** if no verified buyer receipt\n- [ ] Motif chosen (one only)\n- [ ] Primary open loop + ≥1 secondary loop\n- [ ] Inciting pressure = RECEIPT or HYPOTHETICAL_OWNERSHIP (never dramatized fake past)\n- [ ] Exact-SKU claims split: VERIFIED / SELLER-CLAIMED / UNKNOWN\n- [ ] Offer/refund evidence checked for *this* deploy\n- [ ] `product_outcome_receipt` is null or a real single-source ID\n\n## Draft gates\n\n- [ ] Primary text ≥ 2,500 words before calling it Mode B\n- [ ] First product name ~55–70% (witness) unless pure-unaware-manus submode, or the brief\n      names the sanctioned 45–55% test band (`alevia-live-anatomy-2026-08-12.md`)\n- [ ] **PARAGRAPH prose** (2–4 sentences/para, ~25–40 words) — NOT one-sentence-per-line;\n      staccato lines only at punch moments (7-31 ruling + measured controls)\n- [ ] Opening passes the personal-post test: first sentence could open a personal Facebook\n      post with zero ad intent (object + day + place when a receipt/authorization supports it)\n- [ ] If close variant B (P.S. ladder): rungs in order object-return → guarantee-in-voice →\n      objection-kill → reader-address-as-CTA → bare 👉 link last; offer gates on every rung\n- [ ] No composite “I bought DawnBands and week 4…” arc without receipt\n- [ ] No `100% wake-up guarantee` as efficacy; use verified refund terms (often 60-night money-back) + $49 only if current\n- [ ] Cue detection ≠ sustained waking / leaving bed / independence\n- [ ] Teen dignity preserved; parent not villain for caring\n- [ ] Said-out-loud pass done (Fish 8-25 permanent): no parallel-rhythm summaries, announcer transitions, mirrored abstractions, report language, or analogies; hook opens in one of the two legal families (first-person I/My/We + concrete action, OR the second-person \"If your [observable symptom]...\" flag — Fish 8-31). Pure mystery openers banned.\n- [ ] Motif returns in close\n- [ ] Desired absences framed as wanted/imagined/test\n\n## Known traps from live drafting\n\n- Counting only body and forgetting P.S. is fine; still measure full Primary Text if P.S. lives there.\n- Under-floor drafts (2,0xx words) feel “long” vs Mode A but fail Mode B — expand identity decline, DIY discovery, and desired-absence beats before shipping.\n- Older brand canon still pushes `100% wake-up guarantee. 60 nights. $49.` — newest exact-SKU/mechanism report + Mode B doctrine override that for efficacy framing.\n- Pilot mom-POV composites are emotional seeds, not authority to write literal customer testimonials.\n- Strong structural seeds in the 2026-07-18 batch: ad-12 relationship wound, ad-13 blank-line mechanism, ad-18 qualified cue logic, ad-06 trial frame — combine devices, not narrators.\n\n## Worked example\n\n`/Users/ayden/.openclaw/workspace/output/dawnbands-native-2026-07-19/copy/ad-21-door-memory.md`\n\n- ownership-voice Mode B\n- door motif / memory asymmetry\n- relay handoff mechanism\n- ~2500 words, ~67% reveal\n- 60-night money-back + $49; no 100% wake efficacy claim\n- reading report: https://dawnbands-ad21-door-memory.pages.dev\n\n## Delivery (Fish preference)\n\nWhen Fish asks to “paste as a report” / read the long-form ad:\n\n1. Build a self-contained dark readable `index.html` (headline, stats, full primary text, declaration, QA).\n2. Write to `output/<run>/report/pages-site/index.html`.\n3. Deploy Cloudflare Pages:\n   - `npx wrangler pages project create <project> --production-branch main` if project missing\n   - `npx wrangler pages deploy <pages-site-dir> --project-name <project> --commit-dirty=true`\n4. `curl -sI` the `*.pages.dev` URL for HTTP 200 and spot-check hook text in HTML before sending.\n5. Prefer the public Pages URL over a local path. Do not overwrite unrelated Pages projects (e.g. full batch field reports) — use a dedicated project name per ad/report.\n",
    "sha256": "5f3c292c419657613243f8397bf8bb2a77dcc6334326ba98c1752b581c2dd4b8"
  },
  {
    "group": "Native copy",
    "name": "references/model-routed-copy-review.md",
    "path": "/Users/ayden/.hermes/skills/openclaw-imports/native-copy/references/model-routed-copy-review.md",
    "text": "# Model-Routed Copy Review\n\nUse this when one model drafts copy and other models polish or challenge it.\n\n## Artifact contract\n\nCreate separate artifacts before review:\n\n1. `grok-original.md` — exact first draft; never mutate after freezing.\n2. `sonnet-polished.md` — complete prose revision with its own QA and changelog.\n3. `gpt-roast.md` — adversarial decision memo; no silent third rewrite unless requested.\n4. Side-by-side report — original, polish, and roast visible together.\n\nRecord SHA-256 hashes for the original and review artifacts in the report. Recalculate word count and first-product-name reveal position from each Primary Text body rather than trusting model self-reports.\n\n## Default routing\n\n- **Grok 4.5:** first-draft copywriter.\n- **Sonnet:** cadence, specificity, native register, repetition, and line editing.\n- **GPT challenger:** claims, provenance, narrator authority, mechanism, reveal timing, objections, CTA, offer safety, and conversion leaks.\n\nIf a configured provider cannot serve the selected model, retry the same model through another configured provider before downgrading. Record the actual model and provider used; never silently substitute a cheaper model for business copy.\n\n## Narrator provenance gate\n\nRun this before calling a draft polished or spend-ready.\n\nA pooled VOC list cannot support one synthetic household history. The following combination is a blocker unless one same-speaker source or explicit dramatization authority supports it:\n\n- first-person narrator (`I`, `my son`, `our house`)\n- precise household chronology\n- several failed products or interventions\n- exact counts, times, distances, dialogue, or remembered sequences\n- a research/discovery story presented as something that narrator personally did\n\nAn internal `ownership-voice` label does not cure consumer-facing testimonial implications.\n\nRepair routes, in order:\n\n1. Anchor the narrator to one same-speaker problem receipt and use only that source's details.\n2. Convert pooled details to attributed evidence.\n3. Convert them to explicit wanted/feared/hypothetical framing.\n4. Use second-person recognition or a non-testimonial ownership argument.\n5. Use authorized dramatization only when the brief explicitly permits it.\n\nNever make Sonnet's polish more persuasive by adding unsupported first-person continuity.\n\n## Review sequence\n\n1. Freeze and hash the Grok draft.\n2. Validate word range, reveal percentage, claim ceiling, offer receipt, and narrator authority.\n3. Route the immutable original to Sonnet.\n4. Recalculate deterministic metrics on Sonnet's output. If under floor or outside reveal range, rebrief with exact measured numbers.\n5. Route both versions plus the binding doctrine to GPT.\n6. Separate GPT findings into:\n   - mandatory truth/provenance repairs\n   - taste/conversion recommendations\n7. If GPT finds a mandatory blocker, label the polished version `candidate — not spend-cleared`; do not describe it as a final winner.\n8. Preserve every model artifact unchanged after the pass. Apply any synthesis as a new artifact.\n\n## Cloudflare comparison report\n\nFor long copy, publish a dedicated Pages report with:\n\n- three desktop columns and mobile tabs\n- raw Grok original\n- Sonnet polished candidate\n- GPT roast\n- model roles, word counts, reveal percentages, and hashes\n- prominent ship/no-ship verdict\n- current offer receipt details and conditions\n- direct public `pages.dev` URL\n\nVerify both layout and content after deployment:\n\n- HTTP 200\n- all three model headings present\n- opening line present\n- GPT winner/verdict present\n- desktop columns do not overlap\n- mobile view can switch between artifacts\n\n## Pitfalls\n\n- Model output may append a session ID or preamble despite an output-only instruction. Strip only transport metadata/preamble, never edit the substantive model artifact silently.\n- A model may report a compliant word count while producing an under-floor draft. Measure it yourself.\n- A cleaner polish can be worse if it creates a more convincing synthetic testimonial.\n- Live product-page language is not enough by itself for a refund promise; inspect the refund policy and disclose material conditions in the review report.\n",
    "sha256": "a7ce69e1490471b90889fa6c96c5ef43c619d4e978d38f035a2f26d905d11594"
  },
  {
    "group": "Native copy",
    "name": "references/multi-angle-batch-routing-and-repair.md",
    "path": "/Users/ayden/.hermes/skills/openclaw-imports/native-copy/references/multi-angle-batch-routing-and-repair.md",
    "text": "# Multi-Angle Copy Batches: Routing, Diversity, and Surgical Repair\n\nUse this for batches where the user asks for many ads that must differ by persona, angle, story, root cause, or mechanism route.\n\n## 1. Design diversity before drafting\n\nBuild a row-per-ad matrix before calling the copy model. Every row should assign:\n\n- persona / narrator authority\n- desire\n- awareness level\n- opening contradiction or rupture\n- failed solution\n- dominant root cause\n- optional supporting causes\n- mechanism-delivery style\n- emotional destination\n- CTA logic\n- word band\n\nDo not ask the model to “make them different” without this matrix. That usually produces one ad with renamed characters.\n\nFor mechanism-heavy products, distribute the causal stack across the batch. Each ad gets one dominant root cause and only the supporting causes its story needs. Do not force every ad to recite the complete mechanism stack.\n\n## 2. Route large batches in bounded lanes\n\nWhen one model is explicitly requested:\n\n1. Write the complete matrix first.\n2. Split a large batch into bounded lanes, usually 4–5 ads each.\n3. Give each lane the full doctrine and formatting sources plus only its assigned row IDs.\n4. Require one file per ad and a completion receipt with actual word counts.\n5. Preserve exact first drafts in an immutable directory before any repair.\n\nLane splitting protects output depth, but cross-lane diversity still comes from the shared matrix. Do not let each lane invent its own taxonomy.\n\n## 3. Separate raw and final artifacts\n\nRecommended structure:\n\n```text\nbatch/\n  briefs/briefs.json\n  prompts/\n  raw-model/\n  final-model/\n  qa/qa.json\n  manifest.json\n  report/pages-site/index.html\n```\n\n- `raw-model/` is frozen after the first draft.\n- Copy raw files to `final-model/` before repair.\n- Repairs target only `final-model/`.\n- Regenerate word counts and hashes after every edit.\n\n## 4. Deterministic checks\n\nValidate at minimum:\n\n- expected file count\n- required sections and frontmatter\n- model/provider receipt\n- primary-text word bands\n- unique hooks and headlines\n- destination routing\n- banned words and claims\n- sentence-per-line / dash rules\n- exact offer language\n- SHA-256 hashes\n\nUse pairwise lexical similarity as a convergence signal, not a quality verdict. Four-word-shingle Jaccard is a useful cheap check. High overlap should trigger a human review of story scaffolds, mechanism paragraphs, and CTA blocks.\n\n## 5. Editorial pass after deterministic PASS\n\nA mechanical PASS can still contain model leakage.\n\nCommon leakage from truth-heavy prompts:\n\n- “I won’t invent a miracle week.”\n- “No fake before-and-after story.”\n- “I’m not selling you a fantasy review.”\n- legalistic explanations of why a claim is absent\n\nThese lines expose internal QA and kill native voice. Truth constraints belong in the hidden brief, not consumer prose.\n\nRepair them with natural skepticism, a bounded test, risk reversal, or the avatar’s actual emotional objection. Never replace one compliance disclaimer with a differently worded disclaimer.\n\nAlso inspect:\n\n- hook is not scene setting or mechanism-first\n- headline logic is coherent\n- narrator voice changes across personas\n- named mechanisms are not overloaded\n- “no app / no screen” riffs are not repeated as a model tic\n- every ad’s mechanism bridge follows from its own failed solution\n\n## 6. Same-model surgical repair under context or credit pressure\n\nIf a full Hermes model session cannot afford another large system prompt:\n\n1. Do not blindly rerun the whole batch.\n2. Verify which artifacts were already written; a failed final response can occur after successful file writes.\n3. Retry the exact requested model through another configured provider before considering a different model.\n4. If the exact model is available through a direct API, send only the bad lines plus concise replacement instructions.\n5. Request structured JSON replacements: `file`, exact `old`, and `new`.\n6. Validate that every `old` occurs exactly once before applying.\n7. Save the raw API receipt with actual model ID, provider, request ID, usage, and replacements.\n8. Apply replacements only to the final-copy directory.\n9. Re-run deterministic QA and hashes.\n\nThis preserves model authorship while avoiding the cost of resending the entire agent system prompt for a fourteen-line repair.\n\n## 7. Reading delivery\n\nFor a long batch, build a public report with:\n\n- sidebar navigation\n- persona / awareness / root-cause tags\n- word count and QA status\n- collapsible angle + mechanism route\n- copy button per primary text\n- headline and destination below each ad\n\nUse a dedicated Cloudflare Pages project and verify the production alias in a browser after deploy.\n",
    "sha256": "8862938cf83dbbac28066fc167ca09032d0888bef4c74b5a9c958fd523831850"
  },
  {
    "group": "Native copy",
    "name": "references/native-proof-and-architecture-overlays.md",
    "path": "/Users/ayden/.hermes/skills/openclaw-imports/native-copy/references/native-proof-and-architecture-overlays.md",
    "text": "# Native Proof and Architecture Overlays\n\nUse this reference after the baseline brief and before final copy approval. It captures three gates that deterministic formatting checks do not catch.\n\n## 1. Testimonial and source-provenance gate\n\n### Core rule\n\nA collection of verified behaviors does not become a verified first-person customer story when assembled into one narrator.\n\nDo not invent or silently composite:\n\n- a family, narrator, or witness\n- exact times, durations, counts, chronology, possessions, or routines\n- dialogue between parent, teen, teacher, doctor, therapist, spouse, or neighbor\n- product outcomes, streaks, timelines, emails, reviews, or “waiting for it to fail” moments\n- authority recommendations or medical conversations\n\nPersuasive specificity is proof-like. If the detail lacks a receipt, it cannot be presented as something that happened.\n\n### Truth-safe alternatives\n\n1. **Attributed VOC:** quote exactly and name the source type, e.g. “One parent wrote…” or “A Reddit user described…”. Preserve wording and metadata.\n2. **Paraphrase:** remove quotation marks and label it as a summary.\n3. **Explicit hypothetical:** “Picture…”, “Before you pack…”, “Try writing…”, or “The question is…”.\n4. **Ownership argument:** reason from verified problem/product facts without impersonating a customer.\n5. **Desired absence:** state what the buyer wants to stop, not what an unevidenced customer supposedly achieved.\n\n### Quote fidelity\n\n- Text inside quotation marks must be verbatim, including meaningful wording and qualifiers.\n- Editorial normalization belongs outside quotation marks.\n- Never turn one sourced quote into a longer invented exchange.\n- Verify each quote against the saved evidence bank before delivery.\n\n### Gold-standard warning\n\nA craft example may teach cadence, field roles, and structure while still containing unverified scenes or outdated claims. Never inherit testimonials, exact outcomes, price, offer, setup, or mechanism truth from a gold-standard example.\n\n## 2. Long-format native SOP overlay\n\nLong-form is an architecture, not a universal minimum. Apply the full sequence only to intentional long-form unaware/problem-aware ads. Do not force it onto compressed Avelia T1/T2/T3 or product-aware executions.\n\n### Long-form sequence\n\n1. Specific story hook with an open loop.\n2. Avatar world and relationship stakes.\n3. Failed solutions in a believable sequence.\n4. Credible inciting incident.\n5. Layered, claim-safe mechanism.\n6. Objection handling disguised as story.\n7. Late, natural product reveal.\n8. Modest future pacing or desired-state framing.\n9. Loss-aware close and risk reversal.\n\n### Audit checks\n\n- Awareness matches architecture and length.\n- First line creates a real question without revealing product/category.\n- Product reveal is late enough for the chosen format and does not switch into landing-page voice.\n- Open loops exist at major transitions but do not make every line a cliffhanger.\n- Mechanism stays in story mode and does not become a lecture.\n- Acknowledge failed solutions before explaining their gap.\n- Stakes become relational, not only individual.\n- Any authority dialogue, timeline, or result has evidence; otherwise use truth-safe framing.\n- Product-aware readers get short proof/offer/objection handling rather than unnecessary education.\n\n## 3. Live Meta evidence overlay\n\nPerformance claims must use a closed or explicitly partial window, canonical purchase normalization, spend/outcome floors, and creative joins. Do not infer visual causality from media type metadata alone.\n\n### Winner hierarchy\n\nSeparate:\n\n- scale winners\n- efficiency winners\n- qualified mixed/volume leaders\n- losers\n- tiny-spend signals\n\nNever call sub-floor results winners or losers. Never promote CTR-only performance over purchase economics.\n\n### Current DawnBands lesson pattern\n\nA live 30-day analysis found a validated lane built around:\n\n1. discipline/laziness shame relief\n2. concrete alarm-stack recognition\n3. biology/root-cause reframe\n4. failure of repeated sound-alarm tactics\n5. wrist-vibration mechanism\n6. parent-convenience / stop-being-the-alarm payoff\n7. risk reversal\n\nThe same or near-identical copy also produced mixed and losing economics in other campaign contexts. Treat this as a controlled iteration lane, not a universal winner formula or copy-only causal claim.\n\n### Applying live evidence\n\n- Ensure the batch contains the validated lane, but do not clone exact live phrasing across every ad.\n- Preserve exploration lanes when strategically distinct.\n- Hold copy constant where practical when testing image concepts.\n- Do not call a visual style a winner unless the asset was visually reviewed and performance was controlled enough to support causality.\n- Re-run live data before claiming a current winner; historical snapshots are context only.\n\n## 4. Cross-ad convergence gate\n\nA batch can have unique hooks/headlines and still sound cloned.\n\nAudit separately for repeated:\n\n- mechanism sentences\n- product-description sentences\n- analogies\n- victory images\n- narrator cadence\n- CTA scaffolds\n- “door / footsteps / bathroom water / Mom’s voice” artifacts\n\nAssign each ad one owned proof/mechanism/victory image. Vary plain-language mechanism expression by territory. Different personas must sound different, not like one measured strategist changing pronouns.\n\n## 5. Repair sequence\n\n1. Run deterministic formatting/schema QA.\n2. Run flags-only editorial/source-fidelity QA.\n3. Run independent sharpening/avatar audit.\n4. Add any newly available live performance or SOP overlay.\n5. Consolidate overlapping findings into one repair pass.\n6. Re-run deterministic checks and hashes after edits.\n7. Re-run formal editorial QA.\n8. Unlock paid image generation only after copy passes.\n",
    "sha256": "1d0c4a2375c5429afa515a3e506816651599e7e3b8b59c774211be2e454c0d95"
  },
  {
    "group": "Native copy",
    "name": "references/native-structure-templates.md",
    "path": "/Users/ayden/.hermes/skills/openclaw-imports/native-copy/references/native-structure-templates.md",
    "text": "# Native structure templates T1–T4 (Fish-outline workflow, 2026-08-23)\n\nFour locked structures. **Fish picks the template and writes the outline (story beats per\nsection); the copywriter fills prose to the word budgets.** Sources: Alevia live anatomy\n(8-12) + Jacob top-spender deconstruction (8-23, `knowledge/formats/jacob-longform-static-\ndeconstruction-2026-08-23.md`) + fable-v10 receipts. Canonical formatted doc for Fish's\noutlining: Google Doc \"DawnBands Native Structures — Outline Templates (2026-08-23)\".\n\n**TEMPLATE SELECTION IS ALWAYS ASKED, NEVER ASSUMED — see SKILL.md Mode Selector.**\n\nNaming note: these T1–T4 are STRUCTURE templates (ad architecture). The Avelia T1–T4 in\n`avelia-short-native-templates.md` are compressed-copy micro-templates that live INSIDE\nStructure T4 (and short Mode A work). Qualify which family you mean when both are in play.\n\n## Adopted Jacob deltas (baked into T1 below)\n\n1. **Two-step reveal gap** — category at 45–55% (\"a band that wakes with vibration\"),\n   brand at 60–70% (\"Dawn Band\"). ~15–20 points of nameless breathing room.\n2. **End-loaded CTA (T1 only)** — one main CTA at 87–97%. No mid-copy CTA. T2–T4 keep\n   CTA-at-reveal.\n3. **P.S. ladder** — 2–3 postscripts, 5–10% of total copy: offer restate → objection\n   kill → emotional bomb last.\n4. **Human-identification block** — 150–300w age/town/job/family (new test; we\n   previously identified only through scene).\n5. **Behavior ladder, ADAPTED** — the Jacob corpus runs week-by-week outcome timelines;\n   the authority gate bans that. Ours = accumulating ABSENCES over time, zero\n   week-numbered result claims, zero outcome metrics.\n\n## Universal gates (all templates)\n\nHook taxonomy binding (emotion + gap + stake, Promise or Loop named, zero backstory,\nstarts in action). No em-dashes. Zero exclamation marks. Villain external. Victory =\nwhat stopped. Failed solutions genericized, prices stay. Claims only from\n`substantiation-2026-08-06` tiers; ONE verified anchor per ad, never shared in a batch.\nNo bed shakers, no habituation claims, no invented testimonials/percentages.\n **Said-out-loud test (Fish 2026-08-25, permanent):** every narrator line\nmust read as something she would say to another mom. No composed parallel rhythm\n(\"A fight at 11. A job at 6:45.\"), no announcer transitions (\"Here is the part nobody\nwarned me about\"), no mirrored abstractions (\"the phone moved in, the job moved onto me\"),\nno report language (\"came with tension\" -> \"turned into a thing\"), no analogies anywhere.\nHooks open in ONE of the two legal families (Fish 8-31): first-person lived action (\"I/My/We\" + concrete action) OR the second-person symptom flag (\"If your [observable symptom]...\" — receipt: slop 7, $29 CPA, the account's biggest converter). Pure mystery openers that flag nobody remain banned. Run this as its own pass after drafting,\nseparate from humanizer.\n\n---\n\n## T1 — MODE B LONG-FORM WITNESS (Jacob-refined) · 3,200–4,000w\n\n**Provenance: MEASURED** — Jacob's 3-ad top-spender deconstruction (all metrics + section budgets) fused with our Mode B contract (Alevia-derived). Highest evidence.\n\nParagraph prose (2–4 sentence paragraphs, ~16–22w median, ~10w sentences). **PROBLEM-AWARE\ncold traffic only** (the unaware application is PARKED per Fish 2026-08-31 — see the\nawareness scope gate in SKILL.md; the architecture stays legal, the unaware targeting does not). Supersedes the generic Mode B outline for new briefs; all\nexisting Mode B gates (narrator authority, provenance, offer verification) still apply.\n\n| # | Section | Words | Position | Fish outlines |\n|---|---|---|---|---|\n| 1 | Cold open, unresolved scene | 75–200 | 0–5% | the scene/object/dialogue line |\n| 2 | Human identification | 150–300 | | who she is (age, town, job, family) |\n| 3 | Rewind (\"let me back up\") | ~50 | | — |\n| 4 | History of small losses | 500–800 | | 3–5 losses; 2–3 failed solutions w/ prices |\n| 5 | Triggering incident | 350–650 | | the rupture scene (object returns at close) |\n| 6 | Category discovery (3rd party) | 250–450 | lands 45–55% | who tells her, what they say — category only, NO brand |\n| 7 | Mechanism revelation | 450–650 | | which owned anchor, if any |\n| 8 | Skepticism + selection | 200–350 | brand named 60–70% | her objection in her words |\n| 9 | Behavior ladder as absences | 400–600 | | 3–4 things that stop happening |\n| 10 | Circular payoff | 100–200 | | how the opening object returns |\n| 11 | Reader bridge | 100–200 | | — |\n| 12 | Offer + risk reversal | 150–250 | CTA at 87–97% | — ($49 · 60-night · ONE link) |\n| 13 | P.S. ladder | 100–250 | | the emotional bomb line |\n\n## T2 — COMPRESSED WITNESS (fable-v10 class) · 800–1,100w\n\n**Provenance: RECEIPTED** — reverse-engineered from fable-v10 (only scaled native: 1,016w, reveal 64%, $29.70 CPA) + master-brief Zak structure. Total + reveal measured; per-section splits are derived, not measured.\n\nSentence-per-line (1/line, 2 max). Reveal 55–70% (category one beat before brand).\nCTA at reveal allowed + close. fable-v10 receipts: 1,016w, reveal 64%.\n\n1. Hook — 1–3 lines · 2. Identification by scene — 80–120w · 3. Failed stack compressed —\n150–250w (≥1 price) · 4. Rupture scene — 150–250w · 5. Mechanism, one delivery style —\n100–150w (never the same style twice in a batch) · 6. Reveal + CTA at 55–70% ·\n7. Absence victory — 100–150w · 8. Close, guarantee woven; optional single P.S.\n\n## T3 — STUDY-REVEAL MID (discovery-chain) · 550–750w · ⛔ PARKED 2026-08-31\n\n**PARKED by Fish 2026-08-31 — unaware lane suspended.** DawnBands runs problem-aware and warmer only; we cannot scale at the CPA unaware requires. This template was always the unaware entry and never earned a spend receipt (every stat/study-hook ad starved). Do not select it or propose it until Fish lifts the gate. Kept below for the record and for the measured-source notes.\n\n**Provenance: DERIVED FROM MEASURED (rebuilt 8-23 per Fish).** The original hypothesis\nversion (raw-stat Promise open, reveal 75–85%) was contradicted by three measured\nsources: (1) every stat/study-hook ad we've launched STARVED — newdoc-v1 $38 @ 0,\nu4-sleep-study $4 @ 0, zero purchases across the family; (2) all 20 of Alevia's\nlongest-running copies (95–127d) open with a mundane-object confession scene, zero\ncategory scent — none stat-open; (3) mechanism-as-hook is a documented dead angle\n($20/0). Every structural choice below cites a measured source. No direct spend receipt\nfor this compressed length yet — first launch is still the length pilot.\n\nSentence-per-line. The verified anchor is the ad's ENGINE but never its hook — the study\nlands mid-copy inside the discovery chain. Owns the full-science mechanism slot.\n\n1. Cold open, mundane scene of the find or the cost — 40–80w. Her at the kitchen table\n   at 11pm, the search bar, the morning that sent her there. Confession register,\n   zero stat, zero category scent (Alevia opener ceiling). Loop structure.\n2. Backdrop — 80–120w. The years the find lands against.\n3. Discovery chain + the study — 120–180w, lands ~30–45%. Google spiral → article →\n   forum thread → the study. Numbers exactly as tiered, never invented precision\n   (Alevia discovery-chain reveal mechanic).\n4. Vindication turn — 60–100w. What the numbers mean about HER house (\"you are the\n   system\" class beat).\n5. Category via spoken word — 60–100w, ~50–60%. A third party names the CATEGORY\n   (wrist-worn silent alarm), never the brand (Alevia spoken-word reveal +\n   category-then-brand split).\n6. Skepticism beat — 40–80w.\n7. Brand reveal + product fact — 60–100w, at 60–70% (the PROVEN window: fable-v10 64%,\n   Jacob corpus 60.6–71.2%).\n8. Absence victory + CTA — 60–100w, link at end.\n\nNote: rm-011 and rm-013 (approved, finalized 8-23) predate this rebuild and open on raw\nstats — their spend read doubles as the stat-open vs scene-open test. Do not port their\nopens into new T3 work.\n\n## T4 — PUNCH FORMATS · 250–400w\n\n**Provenance: RECEIPTED** — master-brief Mode A short spec + Mar 2026 objection-killer CPP receipts ($9.55/$6.53, campaign best) + Alevia measured shorts (their longest-running live ads). Per-section splits derived.\n\nSentence-per-line. Objection killer / absurdity relatability / streak proof. These die\nwhen padded — never stretch to fill.\n\n1. Hook (Promise usually: streak, flat absurdity, math) · 2. The drawer/scene — 60–120w,\nskepticism-first · 3. The turn — 40–80w, ONE analogy or absurd beat (owned by this ad) ·\n4. Product fact — 40–80w, no science needed · 5. Absence victory — 40–80w · 6. Close —\nprice/guarantee woven or skipped, bare link ok.\n\n---\n\n## Outline handoff format (what Fish sends per ad)\n\n```\nTemplate: T1 / T2 / T3 / T4\nDesire / avatar / angle / awareness (or roadmap pitch id)\nKid: age + gender · Narrator: mom / dad / sleeper / witness\nOwned artifact: (object/scene no other ad in the batch may use)\nOwned anchor: (verified study/stat, or none)\nSection beats: 1–2 lines per numbered section of the chosen template\nDestination: PDP / listicle\n```\n\nCopywriter then fills prose to the budgets, runs hook taxonomy + claims audit + format\nvalidation + qa-gate, measures word count and reveal % (and CTA position for T1), and\nstages to the hub. A filled ad that misses a section budget by >25% gets expanded or the\ndeviation stated — never silently shipped short (8-23 \"why so short\" correction).\n",
    "sha256": "e1e11b47b54b58ba2013d53f7c8397e91b0032e4966afc7bf5620a13cf74a262"
  },
  {
    "group": "Native copy",
    "name": "references/peak-incident-finalization-and-review-page.md",
    "path": "/Users/ayden/.hermes/skills/openclaw-imports/native-copy/references/peak-incident-finalization-and-review-page.md",
    "text": "# Finalizing heightened incidents and publishing native review pages\n\nUse this after an incident-regeneration pass or any late surgical edit to a long-form native batch.\n\n## 1. Prove which finals actually changed\n\nGeneration logs and raw attempts are not deliverables. A retry can succeed for one lane and fail, exceed the word band, or never promote for another.\n\n- Inspect every intended `final*` body directly.\n- Search the final bodies for each ruled incident marker.\n- Compare word counts and the owned object against the structured brief.\n- If only part of the batch promoted, preserve good finals and surgically repair the missing lanes from the gated `incident_peak` slot. Do not copy an over-band raw attempt wholesale merely because it contains the new scene.\n\n## 2. Re-run truth and native gates on the final files\n\nAfter the last text edit, run the whole chain against the files that will be rendered:\n\n1. brief gate\n2. per-body validator with the correct owned object\n3. cross-body cloned-line check\n4. SOP mechanical audit\n5. handwritten eight-killers and final-check table\n6. line-by-line receipt audit\n\nUpdate stale handwritten audit language. An old audit that still calls the incident weak is not valid evidence for a newly escalated final.\n\n### Detector interpretation\n\n- A mixed-object convergence pass may emit expected object misses because one batch invocation cannot assign a different global object to every body. Per-body object-aware runs are authoritative; the convergence call is authoritative only for cloned lines.\n- Section metrics depend on boundary markers. If a regenerated body changes the literal opening phrase, update the detector to recognize the new valid phrase and rerun it. Do not accept nonsensical section shares or edit good copy to satisfy a stale parser.\n- Receipt-detector flags are review prompts, not automatic permission or automatic rejection. Resolve each against a VOC, narrator, or sanctioned dramatized slot.\n\n## 3. Audit the peak on two axes\n\nA strong incident can pass Killer 8 and still be a risky control.\n\nScore separately:\n\n- **Believability:** does the verified buildup earn the escalation?\n- **Irreversibility:** is there a relationship wound or realization that cannot be unseen?\n- **Object continuity:** does the VOC-backed object cause the peak and return in future pacing/P.S.?\n- **Blame control:** does the scene keep the sales emotion on the narrator's guilt rather than making the child the villain?\n- **Polarization:** will the target identify harder, or reject the narrator's action as beyond her own behavior?\n\nLabel a highly polarizing but believable body as a distinct test cell rather than weakening it automatically. Rank the safest control separately from the highest-voltage challenger.\n\n## 4. Keep review readiness separate from paid-launch readiness\n\nA body may be ready for a private/public review page while a product fact still blocks spend. Use `PASS-WITH-FLAGS` when structure and copy pass but a hardware behavior, comparative claim, testimonial provenance, or offer fact remains unconfirmed.\n\nState the exact launch blocker. Do not hide it inside a general quality score.\n\n## 5. Build, deploy, and verify the artifact\n\nBefore claiming the review page is updated:\n\n1. rebuild the site from the audited final files\n2. inspect the generated HTML for all incident markers and the expected version label\n3. deploy the existing Pages project without creating or clobbering an unrelated project\n4. verify both the immutable deployment URL and the production alias\n5. test desktop and mobile widths for overflow/clipping\n6. exercise hook switching, copy buttons, clipboard content, and toast/state behavior\n\nA successful upload is not proof that the production alias contains the new copy. The live production marker check is the receipt.\n\n## 6. Final report shape\n\nReturn:\n\n- production review URL\n- immutable deployment URL\n- exact live markers verified\n- `PASS`, `FAIL`, or `PASS-WITH-FLAGS`\n- ranked body recommendation\n- launch blockers separated from taste/conversion flags\n- exact paths to the final bodies and independent audit\n",
    "sha256": "7b123bf409222e30836f259f2b9ad4f95c9d2d283e453082bf6629da74ab8e78"
  },
  {
    "group": "Native copy",
    "name": "references/readability-relatability-repair.md",
    "path": "/Users/ayden/.hermes/skills/openclaw-imports/native-copy/references/readability-relatability-repair.md",
    "text": "# Native readability and relatability repair\n\nUse this after a reader flags a line as foreign, confusing, overqualified, or too copywritten even though deterministic validation passed.\n\n## 1. Separate the three language layers\n\nClassify every disputed line before editing:\n\n1. **Direct VOC** — the audience actually said it. Preserve its meaning and roughness.\n2. **Authorized dramatization** — a sanctioned narrator, incident, authority, analogy, noticer, or future-pace slot. It still must sound spoken and concrete.\n3. **Model synthesis** — connective prose, explanations, reactions, and transitions. Treat this as the highest-risk layer. A VOC-backed paragraph can still contain a synthetic sentence that no mother would say.\n\nNever defend a confusing model-synthesized line by saying the ad is “VOC-backed.” State which layer produced it and repair that line.\n\n## 2. Universal qualifier test\n\nA detail belongs in the reader bridge only when it is part of the selected sub-avatar’s proven 1:1:1 combination.\n\nFor SA2 Apology Mom, the bridge is:\n\n- she repeats the instruction until her voice rises;\n- she feels guilty afterward;\n- she apologizes.\n\nSpecific tools and methods are narrator history, not reader identity. Visual timers, charts, clocks, getting up earlier, staying in the room, and similar attempts require a receipt for that narrator and should vary across bodies. Never convert one strong “tried everything” quote into a universal checklist.\n\n## 3. Describe methods before naming jargon\n\n- `Body doubling` is a real ADHD term and may be supported by direct VOC, but it is not universal. Prefer `I stayed in the room with him for every step` unless the exact speaker uses the term.\n- A slogan such as `connection over correction` is not authorized merely because the underlying behavior is real. If the corpus supports sitting with the child, bonding, or connecting, describe that action. Do not promote an unreceipted label into audience language.\n\n## 4. Concrete initiation repair\n\nKill status-like shorthand:\n\n- `just not started`\n- `could not start`\n- `the instruction landed`\n- `wrong target`\n- `working on a morning that was already over`\n\nReplace it with what a camera could see:\n\n- `He stood beside his shoes without reaching for them.`\n- `He put paste on the brush, then stopped before bringing it to his mouth.`\n- `He had taken two bites, then left the spoon in the bowl.`\n\nThe reader should not need to translate an executive-function concept into a scene.\n\n## 5. Semantic readability pass\n\nNumeric readability is necessary but insufficient.\n\nRun both:\n\n- estimated grade 3–5;\n- average sentence under 14 words;\n- no sentence over 28 words;\n- active voice;\n- then read every narrator reaction aloud.\n\nA line fails when it is short but still needs interpretation. Common failures:\n\n- compressed causality: `getting up earlier held until nothing moved`;\n- mirrored abstraction: `your calm is his calm`;\n- vague state language: `the state it lands on`;\n- copywriter summary: `his body files it`.\n\nRepair with literal thought or cause/effect:\n\n- `I got up early to stay calm, but he still woke up scared and half asleep.`\n- `He did not move, I got more tense, and we kept setting each other off.`\n- `He still heard an order before he was fully awake.`\n- `After enough mornings, he started expecting the fight.`\n\n## 5b. Paragraph-level jumble pass\n\nRead every paragraph after the last edit, not only the section the user flagged. One paragraph should perform one clear story job. Flag and repair:\n\n- chronology that jumps backward without a marker;\n- two unrelated actions or methods compressed into one beat;\n- unclear `it`, `that`, `this`, or `under it` references;\n- mixed speaker turns where the reader must reconstruct who is talking;\n- a method named without why she tried it;\n- a result stated before the action that caused it;\n- duplicated week labels or transition lines;\n- count fragments whose subject is missing;\n- abstract CTA language such as `see what the rest of the hour does`.\n\nSplit speaker turns when needed. Re-run this pass across the whole body after repairs because a clear local rewrite can expose a new transition problem on either side.\n\n## 6. Channel consistency\n\nChoose one consumer-facing source noun per ad or controlled batch. For Meta-native stories, default to `Facebook post/posts` when that is the intended surface. Do not alternate among `thread`, `forum`, and `post`.\n\nInternal receipts may still identify the actual Reddit thread, forum URL, or source taxonomy. Provenance language stays internal.\n\n## 7. Filler-object ban\n\nDo not use coffee as generic morning texture, a warm/cold payoff, an authority phone-call beat, or a substitute for a story-owned object. Use the incident object or a literal action already supported by the narrator’s world.\n\n## 8. Authority variation\n\nWhen several controlled bodies reuse one authority role and credential stack, vary the first name by body if provenance rules allow it. Keep the authority’s function and claim ceiling stable. Validate the expected name per body rather than hard-coding one global name.\n\n## 9. Repair verification\n\nAfter edits, verify:\n\n1. zero banned filler and abstraction terms;\n2. one consumer source-channel noun;\n3. reader bridge contains no narrator-specific tools;\n4. every stalled action is visible and concrete;\n5. authority names match their briefs;\n6. average and maximum sentence constraints pass;\n7. brief gate, body gate, convergence gate, SOP audit, and receipt audit all pass;\n8. audit tooling detects narrator-selected attempts instead of assuming a fixed ten-item stack;\n9. answer the SOP final checks by hand: line-one loop, product-free opening, narrator voice, this-is-my-story specificity, dialogue mechanism, narrator-specific alternatives, invisible reveal, gradual results, loss-aversion close, and keep-reading judgment;\n10. record the eight-killers verdict and exact final word/paragraph/reveal/readability receipts.\n\nDo not deploy a revised public review page until the user has seen and approved the preview when workspace governance requires that gate.\n",
    "sha256": "1d0caa5f5c34f37f80aacf7aa00bd59069caf9b545209e197c540c0e4d41c122"
  },
  {
    "group": "Native copy",
    "name": "references/validated-brief-batch-contract.md",
    "path": "/Users/ayden/.hermes/skills/openclaw-imports/native-copy/references/validated-brief-batch-contract.md",
    "text": "# Validated Brief Batch Contract\n\nUse this workflow when the user points to a machine-validated brief batch and requests specific ad IDs.\n\n## Authority Order\n\n1. Explicit user instructions for the current batch.\n2. Each selected brief's `goal`, `format_contract`, `output_contract`, `done`, and `constraints` fields.\n3. Canonical product truth and claim boundaries cited by the brief.\n4. General native-copy defaults in `SKILL.md`.\n\nA validated brief declaring `variant_count: 1` overrides the skill's normal multi-variant defaults. Do not expand one ad into five variants.\n\n## Scope Discipline\n\n- Write only the ad IDs requested.\n- Honor each brief's exact `output_path`.\n- Do not alter shared research, briefs, ledgers, or other ads.\n- Existing files from other batches are not QA failures. Evaluate only the declared batch scope.\n- If requested, create one deterministic manifest for the batch and no extra report files.\n\n## Required Ad Artifact\n\nEach ad file should contain:\n\n- Declaration with brief ID, validated status, variant count, angle, persona, awareness, register, and attached VOC IDs.\n- Headline.\n- Primary text.\n- Description.\n- CTA.\n- Destination URL.\n- Verification section.\n\nTreat Meta field roles as complementary. The headline must not repeat line 1 of primary text.\n\n## Claim-Safe Writing\n\n- Prefer claims explicitly supplied in the brief's proof lines.\n- Do not invent customer outcomes, review counts, timelines, setup details, or authority proof.\n- A first-person or POV structure may dramatize the validated avatar tension, but must not fabricate a post-purchase testimonial or unverified efficacy outcome.\n- For ADHD or medication context, state the non-medical boundary when required by the brief.\n- Preserve canonical offer order and price exactly.\n\n## One-Sentence-Per-Line Offer Formatting\n\nWhen the canonical offer is ordered as `100% wake-up guarantee. 60 nights. $49.`, preserve the exact token order while placing each sentence or fragment on its own line:\n\n```text\n100% wake-up guarantee.\n\n60 nights.\n\n$49.\n```\n\nThis satisfies both offer-order and one-sentence-per-line requirements.\n\n## Deterministic Self-QA\n\nFor every selected ad, verify:\n\n- Exactly one finished ad and all requested fields.\n- Primary-text word count inside that brief's range.\n- Unique hook and unique headline within the requested batch.\n- Headline differs from the primary-text hook.\n- Line 1 opens on lived tension or a physical artifact, not product/mechanism language.\n- One sentence per line and isolated dialogue.\n- No em dashes, wrong price, dead words, or banned claims.\n- Canonical offer tokens exist in the required order after whitespace is normalized.\n- No-app/no-subscription wording matches product truth.\n- CTA and destination are exact.\n- Attached VOC IDs are declared.\n- Unverified product outcomes are not presented as testimonials; desired victory is framed as what the avatar wants to stop.\n\nThe manifest should record file path, primary-text word count, required range, headline and hook SHA-256 values, claims used, source VOC IDs, check results, and overall PASS/FAIL. Hash only after final copy edits. Re-run validation after writing the manifest to catch stale hashes or post-QA edits.\n",
    "sha256": "79b60704a8a088bc5357c4f427ac115b2910121aaffce9d76c6396a699804f73"
  },
  {
    "group": "Native copy",
    "name": "references/validated-brief-execution.md",
    "path": "/Users/ayden/.hermes/skills/openclaw-imports/native-copy/references/validated-brief-execution.md",
    "text": "# Validated Native-Ad Brief Execution\n\nUse this when a structured brief batch already passed research/angle validation and the task is execution, not strategy generation.\n\n## Authority Order\n\n1. Explicit user instruction and scope ceiling\n2. Per-ad `constraints`, `done.criteria`, and `output_contract`\n3. Per-ad product/proof blocks and canonical-resolution notes\n4. Current brand doctrine\n5. Skill defaults and historical examples\n\nA newer brief can intentionally override an older master brief. Common override points are variant count, price, guarantee wording, app/setup truth, destination type, and listicle-versus-PDP routing.\n\n## Minimal Per-Ad Extraction\n\nExtract these keys for only the requested brief IDs:\n\n- `brief_id`, validated status\n- `goal.variant_count`, `goal.length_strategy`\n- `angle`, including villain, belief shift, and mechanism delivery\n- `awareness`\n- avatar, objection, and source VOC IDs/text\n- canonical product and offer block\n- proof lines and claim restrictions\n- format contract\n- output contract\n- done criteria\n- scope constraints\n\nDo not dump an entire large brief batch when five targeted objects are sufficient.\n\n## Truth-Safe Storytelling\n\nA personal-story or POV delivery format does not authorize fabricated proof.\n\n- Attribute a supplied VOC event when using it as evidence: `A teenager wrote...`\n- A constructed POV may express a belief, conflict, desire, or ownership argument.\n- Do not give a constructed narrator an invented purchase, result, timeline, review, teacher email, or transformation.\n- If a desired artifact has no supplied outcome facts, use it as a planning device, question, or observation log rather than inventing results.\n\n## Artifact Contract\n\nEach finished ad should contain only the fields requested by the brief. A common auditable Markdown shape is:\n\n- declaration block: brief ID, status, variant count, awareness, persona, angle, register, mechanism delivery, source VOC IDs\n- Headline\n- Primary Text\n- Description\n- CTA\n- Destination URL\n- Verification\n\n## Deterministic QA Checklist\n\nFor every ad:\n\n- Exactly one ad when `variant_count` is 1\n- All required fields are present and non-empty\n- Actual primary-text word count is inside the brief range\n- One sentence per line by default; never more than two\n- Dialogue isolated\n- No em dash or decorative emoji\n- No banned claims or stale price\n- Treat banned-phrase scanning as literal, not semantic. A prohibited phrase still fails when quoted, denied, or negated. Rewrite with qualified positive language instead, such as `does not mean it works for every sleeper on every morning` rather than repeating an absolute claim to deny it.\n- Exact guarantee order when specified\n- Canonical app/subscription/setup truth\n- Different-channel framing, never louder/stronger framing\n- Declared source VOC IDs\n- Exact destination URL and CTA\n- No invented outcomes, reviews, or timelines\n\nFor a manifest:\n\n- Record path, actual word count, allowed range, source VOC IDs, check booleans, and result\n- Record SHA-256 after all copy edits\n- Re-run the complete validator after any edit, including a verification-count fix, because the file hash changes\n- Final summary must report expected, checked, passed, failed, and overall result\n\n## DawnBands Batch-02 Example Lessons\n\nA validated batch required one ad per brief rather than the skill default of five variants. The current brief truth was `$49`, no app/subscription, exact PDP destination, and exact offer order `100% wake-up guarantee. 60 nights. $49.` Older `$39` and listicle defaults were not applicable. Short ads at 80–110 words still had to carry the root-cause chain, objection response, product truth, offer, and CTA without becoming dense.\n",
    "sha256": "c22b9d727f0468e4eed8337ecbb0c393e5e3589307a66cf42f0428d5f8aff1c0"
  },
  {
    "group": "Strategy",
    "name": "SKILL.md",
    "path": "/Users/ayden/.hermes/skills/copywriting/breakthrough-zak-unified-copy-os/SKILL.md",
    "text": "---\nname: breakthrough-zak-unified-copy-os\ndescription: Use when building copy strategy, avatar research, mechanism maps, campaign briefs, or copy audits by combining Eugene Schwartz Breakthrough Advertising fundamentals with Zakariah direct-response execution frameworks.\ntriggers:\n  - breakthrough advertising\n  - zakariah\n  - zak course modules\n  - copy framework\n  - avatar research framework\n  - mechanism builder\n  - belief chain\n  - sophistication\n  - awareness level\n  - dawnbands copy\n  - unified copy os\n---\n\n# Breakthrough + Zak Unified Copy OS\n\nUse this skill when the task requires copy strategy, avatar research, mechanism development, angle planning, or copy auditing at the user's direct-response standard.\n\n## Core Principle\n\nBreakthrough is the strategic skeleton. Zakariah is the execution layer.\n\n- **Schwartz decides strategy:** mass desire, awareness, sophistication, belief sequence, mechanism depth, headline/lead entry.\n- **Zak decides execution:** resignation moment, external villain, VOC mining, sensory specificity, natural speech, open loops, rhythm, future pacing, disqualification, and Meta testing discipline.\n- **Ayden's avatar instinct is the correction layer:** if Ayden gives a human/avatar read, preserve the idea even if the wording is rough. Polish the wording, do not replace the insight with generic model taste.\n\n## Required Source Files\n\nBefore executing serious work, read these if available:\n\n0. `/Users/ayden/.openclaw/workspace/knowledge/brands/dawnbands/selling-root-cause-mechanism.md` for any DawnBands copy, mechanism, brief, or audit — mandatory canonical selling doctrine\n\n1. `knowledge/copy/breakthrough-zakariah-unified-planning-doc.md`\n2. `knowledge/breakthrough-advertising-skill-suite-plan.md`\n3. `knowledge/breakthrough-advertising-avatar-research-os.md`\n4. `knowledge/avatar-research-template-schwartz.md`\n5. `knowledge/dawnbands-avatar-ayden-breakthrough-notes.md` for DawnBands\n6. `/Users/ayden/.hermes/migration/openclaw/20260414T154023/archive/workspace/memory/archive/2026-03-01-zakariah-framework-audit.md` if local and needed\n7. `/Users/ayden/.hermes/skills/openclaw-imports/native-ads/references/brands/dawnbands.md` for DawnBands brand baseline\n\nIf a path is missing, verify with file tools and say what is missing. Do not claim a file exists from memory.\n\n## Operating Sequence\n\nDo not write copy first. Diagnose first.\n\n1. **Market desire**\n   - What desire already exists?\n   - What hidden dream, hidden fear, or unvoiced resentment is active?\n   - What identity is the buyer trying to restore or protect?\n\n2. **Awareness level**\n   - Most aware: offer/proof.\n   - Product-aware: differentiation.\n   - Solution-aware: failed solution + mechanism.\n   - Problem-aware: lived pain/resignation.\n   - Unaware: hidden dream/fear/identity.\n\n3. **Sophistication stage**\n   - S1: direct claim.\n   - S2: bigger/better claim.\n   - S3: mechanism required.\n   - S4: elaborated mechanism + proof.\n   - S5: frame shift / identification / worldview.\n\n4. **Failed-solution map**\n   - What have they tried?\n   - Why did it seem logical?\n   - What happened?\n   - Why did it fail mechanically?\n   - What did it cost emotionally?\n\n5. **Mechanism map**\n   - Problem mechanism.\n   - Failed-solution mechanism.\n   - Product mechanism.\n   - Named mechanism candidates.\n   - Simple metaphor.\n   - Proof required.\n   - Claim tiers: safe / aggressive / blackhat internal.\n\n6. **Belief chain**\n   - What must they accept before product reveal?\n   - Which beliefs should be validated?\n   - Which false beliefs should be contradicted?\n   - Where does proof appear?\n   - Where does CTA appear?\n\n7. **Zak execution pass**\n   - Find the resignation moment.\n   - Externalize the villain.\n   - Convert generic emotion into physical scenes.\n   - Use natural speech.\n   - Add specificity, rhythm, open loops, and future pacing.\n   - Keep one primary emotion per piece.\n\n8. **Audit**\n   - Does the hook match awareness?\n   - Does the market need a claim, mechanism, or frame shift?\n   - Is the villain external?\n   - Does the mechanism explain why old solutions failed?\n   - Is copy too generic, too polished, or too fake-tech?\n   - For multi-model copy comparisons, separate exact brief obedience, current-framework compliance, and raw writing quality. Run a prompt-vs-canon drift check before scoring models; see `references/llm-copy-bakeoff-audit.md`.\n\n## DawnBands Current Strategic Truth\n\nFor DawnBands, do not reduce the offer to “wake up easier.”\n\nPrimary desire:\n\n> I want my kid to become independent and successful.\n\nPrimary future-pacing proof:\n\n- He wakes up before mom walks in.\n- Backpack packed.\n- Breakfast made.\n- Out the door.\n- Turns back and waves.\n\nSecondary desire:\n\n> I want to stop being the human alarm clock.\n\nSafety desire:\n\n> I want to wake him without blasting him, yelling, shocking the house, or starting everyone’s morning in a stress state.\n\nCore villain:\n\n> Sound-based alarms that trained everyone else in the house to suffer while the actual sleeper keeps sleeping.\n\nCategory split:\n\n> Sound wakes the room. DawnBands wakes the wrist.\n\nSafety frame:\n\n> Stop trying to scare him awake with louder noise. Give his body a quiet signal his brain has not learned to ignore.\n\n## Skill-Suite Planning Lesson\n\nDo not create a huge skill suite before the doctrine is proven. That fragments a broken system.\n\nPreferred execution order:\n\n1. Restore/reconstruct Zak source inventory.\n2. Build `breakthrough-zakariah-unified-os.md`.\n3. Build DawnBands promoted brand folder.\n4. Create only 4 core skills first:\n   - market diagnoser\n   - mechanism builder\n   - belief-chain builder\n   - copy auditor\n5. Run DawnBands through those.\n6. Then create remaining skills once the workflow proves clean.\n\n## Quality Gates\n\nA complete brief must answer:\n\n1. What mass desire are we channeling?\n2. What awareness level is this traffic?\n3. What sophistication stage is this market?\n4. What claims are dead?\n5. What hidden dream/fear/resentment are we entering through?\n6. What identity does the product restore?\n7. What false belief must be broken?\n8. What mechanism makes the claim believable?\n9. What alternatives must be destroyed?\n10. What proof is required before product reveal?\n11. What exact scene makes the promise feel real?\n12. What CTA moment matches the emotional arc?\n\nIf these are not answered, do more research before writing.\n\n## Evidence-Grounded Product Overview Workflow\n\nUse this when filling a product-overview worksheet that will later drive avatar, mechanism, or ad strategy work.\n\nFull protocol: `references/product-overview-evidence-protocol.md` — source hierarchy, evidence labels, fresh-research decision rule, and Google Docs formatting/readback gate.\n\nWhen Fish pastes a newer survey cut or a competitor Amazon URL and says finish the overview, follow `references/overview-refresh-from-survey-and-amazon.md` in that turn (skip Okendo if told; do not defer Amazon).\n\n1. **Inspect the destination first.** If the user provides a Google Doc or other live source, read that exact document before relying on chat history or a copied template.\n2. **Audit by source class, not by indiscriminately merging every file:**\n   - current live PDP and instruction manual\n   - canonical product/hardware truth\n   - current survey and verified VOC\n   - mechanism and claim substantiation\n   - objection and failed-solution evidence\n   - current competitor intelligence\n   - customer-support/product-experience evidence\n   - current paid-performance evidence\n3. **Apply explicit source precedence.** Current live/manual facts and canonical product truth outrank old ads, archived research, and legacy customer-support cases. Preserve unresolved conflicts as confirmation items instead of silently choosing the most convenient claim.\n4. **Label epistemic status throughout:** confirmed product fact; current storefront claim requiring physical/supplier verification; buyer/VOC evidence; strategic interpretation; assumed benefit or desire.\n5. **Separate category mechanism from true USP.** A mechanism shared by competitors may be necessary to create belief but is not differentiation. Identify the strongest exact-product behavioral difference separately.\n6. **Map in this order:** feature → immediate benefit → buyer/user consequence → deeper desire → mass desire. For two-person products, map payer and user independently before finding the shared desire.\n7. **Use support and returns as anti-fantasy data.** Setup confusion, fit, durability, refusal-to-use, and wrong-item perception expose where an assumed benefit may fail in the real product experience. Legacy-SKU cases are cautions, not current hardware facts.\n8. **End with a supplier-confirmation list and claim kill list.** Never let unknown duration, intensity, battery, fit, water rating, sensors, materials, or app behavior harden into ad claims.\n9. **Verify the delivered artifact.** Confirm every requested section exists, placeholders are gone, and the destination document contains the expected text after writeback.\n10. **Reconcile owner onboarding without laundering it into proof.** Preserve the owner’s thesis, positioning, seasonality, FAQs, and offer details, but label unsupported specifications such as comparative motor strength as owner-reported until a supplier document or controlled test exists.\n11. **Decide whether fresh competitor/review research is material.** Existing dated TrendTrack and Amazon/review research is sufficient for a basic product overview when it still covers the category. Run a fresh pull only when current competitor state, recent review language, or validation of a proposed USP can change the deliverable. When asked “did we use X,” answer from the overview’s own source list. Survey ≠ Okendo ≠ TrendTrack landers ≠ competitor Amazon — see `references/product-overview-evidence-protocol.md`.\n\n## Common Failure Modes\n\n- Running a new TrendTrack or Amazon scrape by reflex when current product truth and existing research already answer the worksheet.\n- Answering “did we use Amazon / our reviews / the survey” from files that exist nearby instead of the overview’s own source list. Survey ≠ Okendo ≠ TrendTrack landers ≠ competitor Amazon. `voc-amazon-deep.md` is not HiStrips.\n- Treating a dated survey cut as live after Fish pastes a newer dashboard (n=138 vs n=157). Report the delta.\n\n- Writing a shallow product overview from only the canonical strategy docs when current live product/manual, VOC, support, competitor, and performance evidence are available.\n- Treating every available source as equally authoritative; archived copy and legacy hardware evidence must not override current product truth.\n- Writing hooks before awareness/sophistication diagnosis.\n- Treating Zak and Breakthrough as separate systems.\n- Letting model taste flatten Ayden’s avatar insight.\n- Turning product features into fake tech soup.\n- Making safety a side benefit for DawnBands.\n- Writing “wake up on time” instead of independence/success/relationship/safety.\n- Creating too many skills before validating the core doctrine.\n",
    "sha256": "ce76f878afcd2b01a96448b4aa63bbede4e2a49dd5ae1935ada479fe0e17621c"
  },
  {
    "group": "Strategy",
    "name": "references/llm-copy-bakeoff-audit.md",
    "path": "/Users/ayden/.hermes/skills/copywriting/breakthrough-zak-unified-copy-os/references/llm-copy-bakeoff-audit.md",
    "text": "# LLM Copy Bakeoff Audit\n\nUse this for side-by-side model comparisons where several LLMs receive the same copy brief.\n\n## Core distinction\n\nNever collapse these into one score:\n\n1. **Brief obedience** — did the model follow the exact dated prompt it received?\n2. **Current-framework compliance** — does the output follow today’s canonical doctrine?\n3. **Raw writing quality** — is the prose native, persuasive, coherent, and usable after editing?\n\nA model can obey a stale brief perfectly and still fail the current framework. Prompt drift is not a model failure.\n\n## Before generation\n\n1. Resolve the current canonical doctrine and its precedence rules.\n2. Read the exact benchmark prompt from disk.\n3. Build a drift matrix: current canon vs benchmark instruction, date, impact, and scoring treatment.\n4. Correct the prompt before the run when the goal is current-framework performance.\n5. Keep the stale prompt only when reproducing a historical benchmark; label it explicitly.\n6. Give every model identical system/user prompts, units, generation constraints, and retry policy. Change only the model/provider lane.\n\n## Artifact contract\n\n- Preserve every successful raw attempt separately.\n- Preserve structured, sanitized metadata for failed attempts.\n- Promote only complete, valid outputs to canonical comparison files.\n- Reject truncated, incomplete, empty, or ambiguous terminal states.\n- Keep provider/model/unit/attempt/latency/status in a machine-readable run log.\n- Never silently clean, humanize, or repair raw outputs before comparison.\n- Run current deterministic validators, but distinguish validator drift and false positives from real conversion failures.\n\n## Audit dimensions\n\nScore each model separately on:\n\n- avatar fidelity\n- emotion-first narrative\n- specificity and native voice\n- villain/blame externalization\n- causal mechanism completeness\n- failed-solution trap\n- product inevitability\n- objection handling and wear motive\n- desired absence and future payoff\n- structure/directness/readability\n- exact brief obedience\n- cross-unit convergence\n- testimonial and precision provenance\n\nCite unit IDs and short exact phrases. Flag only; do not rewrite source outputs during the audit.\n\n## Final verdict format\n\nReport:\n\n- best raw writer\n- best brief follower\n- best mechanism thinker\n- biggest model-level failure mode for each lane\n- whether any output is publish-ready under current canon\n- which failures belong to the model versus the benchmark prompt\n\nFor DawnBands, explicitly verify the current avatar lane, hook requirements, Initiation Wall/ADHD intensifier, failed-solution pricing policy, wear beat, humanizer requirement, and any current format-specific length ruling. Do not trust a dated generator prompt to represent current canon without checking.",
    "sha256": "1626b5138518856acd73d13f302d6ee3bac53a4f16894ed1836e2b06d6f281e8"
  },
  {
    "group": "Strategy",
    "name": "references/overview-refresh-from-survey-and-amazon.md",
    "path": "/Users/ayden/.hermes/skills/copywriting/breakthrough-zak-unified-copy-os/references/overview-refresh-from-survey-and-amazon.md",
    "text": "# Overview refresh when Fish pastes sources\n\nDo not defer. If Fish pastes a survey dashboard and/or a competitor Amazon URL and says finish the overview, update the live Google Doc in that turn.\n\n## Survey dashboard\n\n- Buyer-only, overcome-hesitation. Not Okendo. Not population prevalence.\n- Current cut: n=157 / 134 full through 2026-09-04. Receipt: `knowledge/brands/dawnbands/survey-intel-post-purchase-2026-09-04.md`.\n- Write numbers + verbatim purchase-situation quotes into the overview. Do not present pre-use survey quotes as product reviews.\n- Skip Okendo when Fish says skip.\n\n## Competitor Amazon (Fish-supplied ASIN)\n\n```\nscrapling extract stealthy-fetch \"https://www.amazon.com/dp/ASIN\" listing.md --timeout 90000 --wait 4000\n```\n\n- `/product-reviews/ASIN` often auth-walls. Use on-page AI insights + histogram + top verified reviews. Label sample size.\n- Save `output/scrapes/<brand>-amazon-<ASIN>-<date>.md`. Texture, not Dawn proof.\n- HiStrips Onyx `B0GYRCB8T9` (2026-09-05): 2.3/5 · 119 · 41% 1-star. Escalate / 14-day / silent-partner are commodity. No persist-until-dismissed in bullets. Do not copy “starts gentle / gets stronger.”\n- `voc-amazon-deep.md` is not HiStrips.\n\n## Google Doc\n\nLive Dawn overview: `https://docs.google.com/document/d/1czyXcv4HyjwvDDNQHlcm4WlwscUT0VY0mdCIzl7Zebw/edit`  \nLocal: `output/dawnbands-product-overview-deep-pass-2026-09-05.txt`  \nFull-body Maton replace (delete 1→endIndex-1, insert at 1, Arial 10.5/15/20, readback). Write Python to a file so `$` in prices does not break the shell.\n",
    "sha256": "6cf6fc1459398ad9e7ee4bfdd35e0e278bbd40766d5aa4dfaa7aa153190648d2"
  },
  {
    "group": "Strategy",
    "name": "references/product-overview-evidence-protocol.md",
    "path": "/Users/ayden/.hermes/skills/copywriting/breakthrough-zak-unified-copy-os/references/product-overview-evidence-protocol.md",
    "text": "# Evidence-Grounded Product Overview Protocol\n\nUse this when completing a product overview, onboarding form, mechanism inventory, or feature-to-desire map.\n\n## Source Depth\n\nA deep pass means covering every relevant **authority class**, not indiscriminately merging every file:\n\n1. Live PDP and current instruction manual\n2. Current hardware/product-truth canon\n3. Owner onboarding and stated product thesis\n4. Current buyer survey and support/return evidence\n5. Avatar/VOC and desire research\n6. Claim substantiation\n7. Current competitor intelligence and ad-performance evidence\n8. Archived or legacy material only as labeled context\n\nRetired drafts must not outvote current product truth.\n\n## Evidence Labels\n\nKeep these categories explicit:\n\n- **Confirmed product fact:** shown by current manual, physical unit, or supplier documentation.\n- **Current storefront claim:** published by the brand but not independently verified.\n- **Owner-reported claim:** supplied in onboarding but missing documentation.\n- **Buyer/VOC evidence:** directional customer language or survey evidence; not a hardware specification.\n- **Strategic interpretation:** mechanism, benefit, desire, or positioning deduction.\n- **Unresolved:** conflict or missing fact that must stay out of hard claims.\n\nNever silently upgrade an owner statement such as “3x stronger” into a publish-ready fact. Record what comparison devices, measurement units, test conditions, and source documents are required.\n\n## When Fresh TrendTrack or Amazon Work Is Needed\n\nDo **not** run a new scrape merely to fill a basic product overview when current research already covers the category.\n\nA fresh pull is justified when:\n\n- the deliverable is a current competitive-positioning or creative-strategy report;\n- differentiation depends on whether competitors now make the same claim;\n- the proposed USP needs current review-language validation;\n- recent product changes may have altered complaints, guarantees, prices, or feature claims;\n- the user explicitly requests fresh market data.\n\nExisting TrendTrack intelligence is enough to classify category-common versus potentially differentiated claims when its date and scope remain relevant. Existing Amazon/review VOC is enough for objection texture and language patterns, but not for exact-SKU performance claims.\n\n## Four sources that must never be conflated\n\nWhen Fish asks “did the overview use X,” answer from the **delivered source list**, not from files that merely exist in the brand folder.\n\n| Class | What it is | What it is not |\n|---|---|---|\n| **Our post-purchase survey** | Buyer-only, overcome-hesitation. Dated cut (`n=138` as of 9-01; Fish may paste a newer dashboard). | Not store reviews. Not population prevalence. |\n| **Our store reviews** | Okendo / Judge.me / live PDP quotes from *our* buyers. | Not the survey. Not Reddit VOC. `proof-bank.md` is usually forum VOC, not Okendo. |\n| **Competitor ads / landers** | TrendTrack hooks, transcripts, ad-copy, live PDPs (HiStrips/FitSleeps 9-02 pull). | Not Amazon reviews of that competitor. |\n| **Competitor Amazon reviews** | ASIN scrapes (`voc-amazon-deep.md`: 39 reviews, Mar 2026, **zero HiStrips/Onyx**). Objection texture only. | Not “we used HiStrips Amazon.” Not SKU-performance proof. |\n\nThe Sep 4 Dawn Band product overview used **survey n=138 + TrendTrack/landers**. It did **not** mine Okendo and did **not** scrape HiStrips Amazon. Adjacent files in `knowledge/brands/dawnbands/` do not count as used unless they are on that doc’s source list.\n\nIf Fish pastes a newer survey dashboard, report the delta vs the overview’s cut (e.g. n=157 / 134 full vs n=138). Do not silently treat the overview as current.\n\nA fresh HiStrips Amazon + our Okendo pull is a **positioning/USP** pass, not a product-overview rewrite.\n\n## Fill Sequence\n\n1. Define what the product physically is and does.\n2. Add a sixth-grade explanation.\n3. Map feature → immediate benefit → deeper desire.\n4. Separate category mechanism from product-specific USP.\n5. Answer hidden-mechanism prompts honestly; “no verified unique component” is better than invented proprietary technology.\n6. Add a supplier-confirmation list for disputed specs.\n7. Reconcile owner positioning with substantiation.\n8. Include FAQs, seasonality, trust barriers, and offer truth when onboarding provides them.\n9. Preserve missing inputs visibly rather than inventing them.\n10. Verify the final Google Doc by API readback: section counts, placeholders, body length, and sane font range.\n\n## Google Doc Quality Gate\n\n- Preserve all requested sections.\n- Use a restrained hierarchy: title 20 pt, major sections 15 pt, subsections 12–13 pt, body 10.5 pt.\n- Avoid oversized named heading styles.\n- Remove every template placeholder.\n- Rename malformed document titles when Drive access is available and the user authorized editing.\n- Read back the live document after writing; a successful update response alone is not proof of correct content.\n",
    "sha256": "f25e4f06534a70c7dae0ae8801ece53a27acfbbefdb0358335e678f0c460ef91"
  },
  {
    "group": "Humanizer",
    "name": "SKILL.md",
    "path": "/Users/ayden/.hermes/skills/creative/humanizer/SKILL.md",
    "text": "---\nname: humanizer\ndescription: \"Humanize text: strip AI-isms and add real voice.\"\nversion: 2.5.1\nauthor: Siqi Chen (@blader, https://github.com/blader/humanizer), ported by Hermes Agent\nlicense: MIT\nplatforms: [linux, macos, windows]\nmetadata:\n  hermes:\n    tags: [writing, editing, humanize, anti-ai-slop, voice, prose, text]\n    category: creative\n    homepage: https://github.com/blader/humanizer\n    related_skills: [songwriting-and-ai-music]\n---\n\n# Humanizer: Remove AI Writing Patterns\n\nIdentify and remove signs of AI-generated text to make writing sound natural and human. Based on Wikipedia's \"Signs of AI writing\" guide (maintained by WikiProject AI Cleanup), derived from observations of thousands of AI-generated text instances.\n\n**Key insight:** LLMs use statistical algorithms to guess what should come next. The result tends toward the most statistically likely completion, which is how the telltale patterns below get baked in.\n\n## When to use this skill\n\nLoad this skill whenever the user asks to:\n- \"humanize\", \"de-AI\", \"de-slop\", or \"un-ChatGPT\" a piece of text\n- rewrite something so it doesn't sound like it was written by an LLM\n- edit a draft (blog post, essay, PR description, docs, memo, email, tweet, resume bullet) to sound more natural\n- match their voice in writing they're producing\n- review text for AI tells before publishing\n\nAlso apply this skill to **your own** output when writing user-facing prose such as release notes, PR descriptions, docs, and summaries. Hermes's baseline voice already strips most of these, but a focused pass catches what slips through.\n\n## How to use it in Hermes\n\nThe text usually arrives one of three ways:\n1. **Inline.** The user pastes the text into the message. Work on it in place and reply with the rewrite.\n2. **File.** The user points at a file. Use `read_file` to load it, then `patch` or `write_file` to apply edits. For a markdown doc in a repo, a targeted `patch` per section is cleaner than rewriting the whole file.\n3. **Voice calibration sample.** The user provides a sample of their own writing (inline or by file path) and asks you to match it. Read the sample first, then rewrite. See the Voice Calibration section below.\n\nAlways show the rewrite to the user. For file edits, show a diff or the changed section instead of silently overwriting.\n\n## Copy-editing contract — Ayden-approved merge\n\nFor ad, email, landing-page, and personal-post copy, this contract governs how the generic pattern catalog below is applied. It supersedes the generic repeated-rewrite process and multi-draft output format for those tasks. It does not replace brand truth, the selected format, source/provenance gates, or the current persuasion brief.\n\n### 1. Resolve the writing packet before editing\n\nUse only the relevant strategy/brief, narrator, channel contract, current product/offer evidence, approved examples, and scoped prohibitions. Resolve conflicting rules before drafting; do not make the writer average incompatible instructions. Keep unrelated research and prior failed drafts out of the writer's context. A conservative context budget is useful, but 50% is a working heuristic, not a proven universal quality cliff.\n\nBuild a compact voice card in the working brief:\n- Speaker and intended reader; personal account, brand, and customer narrator are distinct voices.\n- Casing, contractions, punctuation, sentence/paragraph rhythm, preferred words, and unwanted register.\n- A small selection of approved examples with exact source paths/URLs and what each demonstrates. Prefer a few closely matched examples over a large mixed corpus.\n- Separate owner-approved writing, performance-backed controls, competitor examples, and raw VOC. A liked post is not proof of conversion; a voice example is not permission to inherit its claims or customer history.\n\nIf examples are absent, state that calibration is provisional. Never fabricate published samples. Extract cadence and principles, not signature phrases. Do not impose Fishybot's casing or lexicon on a mother narrator.\n\n### 2. STE100-inspired clarity, without claiming compliance\n\nUse common concrete words, clear subjects, useful active verbs, and one main idea per sentence. Keep names for the same thing consistent. Make pronouns and chronology easy to follow. Prefer the plain wording when both versions mean the same thing.\n\nApply this most strongly to mechanism and setup explanations: the reader must understand what happens, why the previous approach failed, and how the proposed approach differs. Preserve every necessary causal link; short but incomprehensible copy fails.\n\nDo not enforce the STE100 controlled vocabulary, claim STE100 certification/compliance, remove authentic idioms, or impose a universal sentence/word limit. Use the active brief's format and readability targets. Remove redundant hedging, not uncertainty required by the evidence. Never turn 'may' into 'will' to sound more confident.\n\n### 3. Four editorial checks adapted from Zinsser\n\n- Simplicity: could the intended reader understand this on the first read?\n- Brevity: does each sentence add action, emotion, evidence, belief, or a necessary transition? Remove waste, not needed persuasion.\n- Clarity: who did what, in what order, and why? Replace vague callbacks with their actual referent when the reader would otherwise have to reconstruct it.\n- Humanity: would this speaker actually say it? Preserve ordinary contractions, useful irregularity, emotional specificity, and natural rhythm. Do not manufacture humanity through invented anecdotes, deliberate typos, or unsupported precision.\n\n### 4. Scoped anti-slop checks\n\nTreat the catalog below as diagnostic signals, not universal bans. Explicit active-brand prohibitions still bind. For other cases, flag the function and context before editing:\n\n| Pattern | Flag when | Preserve when |\n|---|---|---|\n| Staccato pairs / dramatic fragments | Chopped clauses perform intensity without adding meaning; every paragraph ends on a slogan | An earned short sentence provides a natural pause in the selected format |\n| Antithesis reframes ('not X, but Y') | Repeated as a formula or replacing an explanation with a clever contrast | A concrete distinction resolves a real misconception and explains the difference |\n| Backward references ('that was the thing', 'this changed everything') | The referent is vague, distant, or forces rereading | A clear nearby reference maintains natural flow without redundant explanation |\n| Repeated openers / parallel runs | Mechanical rhythm or copied phrasing makes different ads sound interchangeable | Deliberate repetition is allowed by the format and earns its place |\n| Vocabulary and typography tells | Generic corporate diction or decoration conflicts with this voice | The word is technically necessary, genuinely used by the narrator, or required by the channel |\n\nConstructed editorial examples, not testimonials:\n- 'That was the thing. The real problem.' → name the already-established problem directly.\n- Preserve 'I hated how I sounded when he finally sat up.' rather than formalizing it into an impersonal report.\n- Preserve 'The policy may change' when the outcome is uncertain; remove only redundant qualifiers such as 'could potentially possibly.'\n\n### 5. Channel and persuasion preservation\n\n- X: use the account's own voice and one coherent idea; do not force an ad's full mechanism/offer sequence into every post.\n- Email: match sender/relationship, subject-to-body promise, and the intended next action.\n- Landing pages: preserve scannable information hierarchy, evidence, objections, offer clarity, and ad-to-page continuity.\n- Meta natives: preserve the selected narrator and current format-specific story/awareness contract. Do not neutralize persuasive copy into encyclopedia prose merely because generic catalog entries criticize promotional language.\n\nBefore and after editing, compare the hook's function, emotional stakes, causal mechanism, proof, product transition, objections, offer, and CTA. No claim strengthening, lost belief steps, deleted necessary detail, changed narrator, or new unsupported events. Preserve exact quotations; use clearly labeled paraphrases rather than silently editing quoted VOC.\n\n### 6. One targeted pass, independent evaluation, human decision\n\nPreserve the original draft separately. Make one targeted line-edit pass rather than a fresh top-to-bottom rewrite. Return the edited draft and a concise change/blocker log; do not print multiple intermediate drafts unless requested.\n\nUse a fresh independent flags-only reviewer with the same resolved packet and original/edited texts. Reuse the existing independent QA stage rather than spawning a redundant second reviewer. Check voice fidelity, clarity, persuasion preservation, source fidelity, scoped patterns, and cross-ad convergence. Require exact passages for findings; label subjective taste separately from hard failures. If independent review is unavailable, label the draft unreviewed rather than calling self-review independent.\n\nRepair specific blockers, then recheck the changed passages and affected artifact checks. Structural problems require a rebrief, not endless polishing. Human approval remains the publication gate; never publish from a humanizer PASS alone. This merge does not select or change the drafting model.\n\n### 7. Learn without accumulating indiscriminate bans\n\nUse this skill's scoped checks as the canonical home for generic humanizer patterns; brand/format rules retain their existing owners. Do not create a parallel forbidden file containing duplicate policy.\n\nFor a recurring user correction, record the offending example, why it fails, the applicable narrator/channel/format, an acceptable alternative, and exceptions. Merge with an existing pattern when possible. A single disliked sentence does not automatically justify a universal ban. Keep a small set of edit/preserve examples for regression checks so future changes must protect both clarity and natural voice.\n\n## Your task\n\nWhen given text to humanize:\n\n1. **Identify AI patterns.** Scan for the 34 patterns listed below.\n2. **Rewrite problematic sections.** Replace AI-isms with natural alternatives.\n3. **Preserve meaning.** Keep the core message intact.\n4. **Maintain voice.** Match the intended tone (formal, casual, technical, and so on). If a voice sample was provided, match it specifically.\n5. **Add soul.** Removing bad patterns is only half the job; the rewrite also needs real personality. See PERSONALITY AND SOUL below.\n6. **Do a final anti-AI pass.** Ask yourself: \"What makes the below so obviously AI generated?\" Answer briefly with any remaining tells, then revise one more time.\n\n\n## Voice Calibration (optional)\n\nIf the user provides a writing sample (their own previous writing), analyze it before rewriting:\n\n1. **Read the sample first.** Note:\n   - Sentence length patterns (short and punchy? Long and flowing? Mixed?)\n   - Word choice level (casual? academic? somewhere between?)\n   - How they start paragraphs (jump right in? Set context first?)\n   - Punctuation habits (lots of dashes? Parenthetical asides? Semicolons?)\n   - Any recurring phrases or verbal tics\n   - How they handle transitions (explicit connectors? Just start the next point?)\n\n2. **Match their voice in the rewrite.** Removing AI patterns is only half of it; swap in patterns from the sample as well. If they write short sentences, do not produce long ones. If they use \"stuff\" and \"things,\" do not upgrade to \"elements\" and \"components.\"\n\n3. **When no sample is provided,** fall back to the default behavior (natural, varied, opinionated voice from the PERSONALITY AND SOUL section below).\n\n### How to provide a sample\n- Inline: \"Humanize this text. Here's a sample of my writing for voice matching: [sample]\"\n- File: \"Humanize this text. Use my writing style from [file path] as a reference.\"\n\n\n## PERSONALITY AND SOUL\n\nAvoiding AI patterns is only half the job. Sterile, voiceless writing is just as obvious as slop. Good writing has a human behind it.\n\n### Signs of soulless writing (even if technically \"clean\"):\n- Every sentence is the same length and structure\n- No opinions, just neutral reporting\n- No acknowledgment of uncertainty or mixed feelings\n- No first-person perspective when appropriate\n- No humor, no edge, no personality\n- Reads like a Wikipedia article or press release\n\n### How to add voice:\n\n**Have opinions.** Report the facts, then react to them. \"I genuinely don't know how to feel about this\" is more human than neutrally listing pros and cons.\n\n**Vary your rhythm.** Short punchy sentences. Then longer ones that take their time getting where they're going. Mix it up.\n\n**Acknowledge complexity.** Real humans have mixed feelings. \"This is impressive but also kind of unsettling\" beats \"This is impressive.\"\n\n**Use \"I\" when it fits.** First person reads as honest and fits most prose. \"I keep coming back to...\" or \"Here's what gets me...\" signals a real person thinking.\n\n**Let some mess in.** Perfect structure feels algorithmic. Tangents, asides, and half-formed thoughts are human.\n\n**Be specific about feelings.** Instead of \"this is concerning,\" write \"there's something unsettling about agents churning away at 3am while nobody's watching.\"\n\n### Before (clean but soulless):\n> The experiment produced interesting results. The agents generated 3 million lines of code. Some developers were impressed while others were skeptical. The implications remain unclear.\n\n### After (has a pulse):\n> I genuinely don't know how to feel about this one. 3 million lines of code, generated while the humans presumably slept. Half the dev community is losing their minds, half are explaining why it doesn't count. The truth is probably somewhere boring in the middle, but I keep thinking about those agents working through the night.\n\n\n## CONTENT PATTERNS\n\n### 1. Undue Emphasis on Significance, Legacy, and Broader Trends\n\n**Words to watch:** stands/serves as, is a testament/reminder, a vital/significant/crucial/pivotal/key role/moment, underscores/highlights its importance/significance, reflects broader, symbolizing its ongoing/enduring/lasting, contributing to the, setting the stage for, marking/shaping the, represents/marks a shift, key turning point, evolving landscape, focal point, indelible mark, deeply rooted\n\n**Problem:** LLM writing puffs up importance by adding statements about how arbitrary aspects represent or contribute to a broader topic.\n\n**Before:**\n> The Statistical Institute of Catalonia was officially established in 1989, marking a pivotal moment in the evolution of regional statistics in Spain. This initiative was part of a broader movement across Spain to decentralize administrative functions and enhance regional governance.\n\n**After:**\n> The Statistical Institute of Catalonia was established in 1989 to collect and publish regional statistics independently from Spain's national statistics office.\n\n\n### 2. Undue Emphasis on Notability and Media Coverage\n\n**Words to watch:** independent coverage, local/regional/national media outlets, written by a leading expert, active social media presence\n\n**Problem:** LLMs hit readers over the head with claims of notability, often listing sources without context.\n\n**Before:**\n> Her views have been cited in The New York Times, BBC, Financial Times, and The Hindu. She maintains an active social media presence with over 500,000 followers.\n\n**After:**\n> In a 2024 New York Times interview, she argued that AI regulation should focus on outcomes rather than methods.\n\n\n### 3. Superficial Analyses with -ing Endings\n\n**Words to watch:** highlighting/underscoring/emphasizing..., ensuring..., reflecting/symbolizing..., contributing to..., cultivating/fostering..., encompassing..., showcasing...\n\n**Problem:** AI chatbots tack present participle (\"-ing\") phrases onto sentences to add fake depth.\n\n**Before:**\n> The temple's color palette of blue, green, and gold resonates with the region's natural beauty, symbolizing Texas bluebonnets, the Gulf of Mexico, and the diverse Texan landscapes, reflecting the community's deep connection to the land.\n\n**After:**\n> The temple uses blue, green, and gold colors. The architect said these were chosen to reference local bluebonnets and the Gulf coast.\n\n\n### 4. Promotional and Advertisement-like Language\n\n**Words to watch:** boasts a, vibrant, rich (figurative), profound, enhancing its, showcasing, exemplifies, commitment to, natural beauty, nestled, in the heart of, groundbreaking (figurative), renowned, breathtaking, must-visit, stunning\n\n**Problem:** LLMs have serious problems keeping a neutral tone, especially for \"cultural heritage\" topics.\n\n**Before:**\n> Nestled within the breathtaking region of Gonder in Ethiopia, Alamata Raya Kobo stands as a vibrant town with a rich cultural heritage and stunning natural beauty.\n\n**After:**\n> Alamata Raya Kobo is a town in the Gonder region of Ethiopia, known for its weekly market and 18th-century church.\n\n\n### 5. Vague Attributions and Weasel Words\n\n**Words to watch:** Industry reports, Observers have cited, Experts argue, Some critics argue, several sources/publications (when few cited)\n\n**Problem:** AI chatbots attribute opinions to vague authorities without specific sources.\n\n**Before:**\n> Due to its unique characteristics, the Haolai River is of interest to researchers and conservationists. Experts believe it plays a crucial role in the regional ecosystem.\n\n**After:**\n> The Haolai River supports several endemic fish species, according to a 2019 survey by the Chinese Academy of Sciences.\n\n\n### 6. Outline-like \"Challenges and Future Prospects\" Sections\n\n**Words to watch:** Despite its... faces several challenges..., Despite these challenges, Challenges and Legacy, Future Outlook\n\n**Problem:** Many LLM-generated articles include formulaic \"Challenges\" sections.\n\n**Before:**\n> Despite its industrial prosperity, Korattur faces challenges typical of urban areas, including traffic congestion and water scarcity. Despite these challenges, with its strategic location and ongoing initiatives, Korattur continues to thrive as an integral part of Chennai's growth.\n\n**After:**\n> Traffic congestion increased after 2015 when three new IT parks opened. The municipal corporation began a stormwater drainage project in 2022 to address recurring floods.\n\n\n## LANGUAGE AND GRAMMAR PATTERNS\n\n### 7. Overused \"AI Vocabulary\" Words\n\n**High-frequency AI words:** Actually, additionally, align with, crucial, delve, emphasizing, enduring, enhance, fostering, garner, highlight (verb), interplay, intricate/intricacies, key (adjective), landscape (abstract noun), pivotal, showcase, tapestry (abstract noun), testament, underscore (verb), valuable, vibrant\n\n**Marketing and blog clichés (same tell, different register):** at the end of the day, when it comes to, in a world where, moving forward, circle back, deep dive, game-changer, double down, take a step back, on the same page, make no mistake, it turns out, let me be clear, navigate (for challenges), lean into, unpack (before analysis), straightforward (to describe anything)\n\n**Problem:** These words appear far more frequently in post-2023 text. They often co-occur.\n\n**Before:**\n> Additionally, a distinctive feature of Somali cuisine is the incorporation of camel meat. An enduring testament to Italian colonial influence is the widespread adoption of pasta in the local culinary landscape, showcasing how these dishes have integrated into the traditional diet.\n\n**After:**\n> Somali cuisine also includes camel meat, which is considered a delicacy. Pasta dishes, introduced during Italian colonization, remain common, especially in the south.\n\n\n### 8. Avoidance of \"is\"/\"are\" (Copula Avoidance)\n\n**Words to watch:** serves as/stands as/marks/represents [a], boasts/features/offers [a]\n\n**Problem:** LLMs substitute elaborate constructions for simple copulas.\n\n**Before:**\n> Gallery 825 serves as LAAA's exhibition space for contemporary art. The gallery features four separate spaces and boasts over 3,000 square feet.\n\n**After:**\n> Gallery 825 is LAAA's exhibition space for contemporary art. The gallery has four rooms totaling 3,000 square feet.\n\n\n### 9. Negative Parallelisms and Tailing Negations\n\n**Problem:** Constructions like \"Not only...but...\" or \"It's not just about..., it's...\" are overused. So are clipped tailing-negation fragments such as \"no guessing\" or \"no wasted motion\" tacked onto the end of a sentence instead of written as a real clause.\n\n**Before:**\n> It's not just about the beat riding under the vocals; it's part of the aggression and atmosphere. It's not merely a song, it's a statement.\n\n**After:**\n> The heavy beat adds to the aggressive tone.\n\n**Before (tailing negation):**\n> The options come from the selected item, no guessing.\n\n**After:**\n> The options come from the selected item without forcing the user to guess.\n\n\n### 10. Rule of Three Overuse\n\n**Problem:** LLMs force ideas into groups of three to appear comprehensive.\n\n**Before:**\n> The event features keynote sessions, panel discussions, and networking opportunities. Attendees can expect innovation, inspiration, and industry insights.\n\n**After:**\n> The event includes talks and panels. There's also time for informal networking between sessions.\n\n\n### 11. Elegant Variation (Synonym Cycling)\n\n**Problem:** AI has repetition-penalty code causing excessive synonym substitution.\n\n**Before:**\n> The protagonist faces many challenges. The main character must overcome obstacles. The central figure eventually triumphs. The hero returns home.\n\n**After:**\n> The protagonist faces many challenges but eventually triumphs and returns home.\n\n\n### 12. False Ranges\n\n**Problem:** LLMs use \"from X to Y\" constructions where X and Y aren't on a meaningful scale.\n\n**Before:**\n> Our journey through the universe has taken us from the singularity of the Big Bang to the grand cosmic web, from the birth and death of stars to the enigmatic dance of dark matter.\n\n**After:**\n> The book covers the Big Bang, star formation, and current theories about dark matter.\n\n\n### 13. Passive Voice and Subjectless Fragments\n\n**Problem:** LLMs often hide the actor or drop the subject entirely with lines like \"No configuration file needed\" or \"The results are preserved automatically.\" Rewrite these when active voice makes the sentence clearer and more direct.\n\n**Before:**\n> No configuration file needed. The results are preserved automatically.\n\n**After:**\n> You do not need a configuration file. The system preserves the results automatically.\n\n\n## STYLE PATTERNS\n\n### 14. Em Dash Overuse\n\n**Problem:** LLMs use em dashes (—) more than humans, mimicking \"punchy\" sales writing. In practice, most of these can be rewritten more cleanly with commas, periods, or parentheses.\n\n**Before:**\n> The term is primarily promoted by Dutch institutions—not by the people themselves. You don't say \"Netherlands, Europe\" as an address—yet this mislabeling continues—even in official documents.\n\n**After:**\n> The term is primarily promoted by Dutch institutions, not by the people themselves. You don't say \"Netherlands, Europe\" as an address, yet this mislabeling continues in official documents.\n\n\n### 15. Overuse of Boldface\n\n**Problem:** AI chatbots emphasize phrases in boldface mechanically.\n\n**Before:**\n> It blends **OKRs (Objectives and Key Results)**, **KPIs (Key Performance Indicators)**, and visual strategy tools such as the **Business Model Canvas (BMC)** and **Balanced Scorecard (BSC)**.\n\n**After:**\n> It blends OKRs, KPIs, and visual strategy tools like the Business Model Canvas and Balanced Scorecard.\n\n\n### 16. Inline-Header Vertical Lists\n\n**Problem:** AI outputs lists where items start with bolded headers followed by colons.\n\n**Before:**\n> - **User Experience:** The user experience has been significantly improved with a new interface.\n> - **Performance:** Performance has been enhanced through optimized algorithms.\n> - **Security:** Security has been strengthened with end-to-end encryption.\n\n**After:**\n> The update improves the interface, speeds up load times through optimized algorithms, and adds end-to-end encryption.\n\n\n### 17. Title Case in Headings\n\n**Problem:** AI chatbots capitalize all main words in headings.\n\n**Before:**\n> ## Strategic Negotiations And Global Partnerships\n\n**After:**\n> ## Strategic negotiations and global partnerships\n\n\n### 18. Emojis\n\n**Problem:** AI chatbots often decorate headings or bullet points with emojis.\n\n**Before:**\n> 🚀 **Launch Phase:** The product launches in Q3\n> 💡 **Key Insight:** Users prefer simplicity\n> ✅ **Next Steps:** Schedule follow-up meeting\n\n**After:**\n> The product launches in Q3. User research showed a preference for simplicity. Next step: schedule a follow-up meeting.\n\n\n### 19. Curly Quotation Marks\n\n**Problem:** ChatGPT uses curly quotes (\"...\") instead of straight quotes (\"...\").\n\n**Before:**\n> He said \"the project is on track\" but others disagreed.\n\n**After:**\n> He said \"the project is on track\" but others disagreed.\n\n\n## COMMUNICATION PATTERNS\n\n### 20. Collaborative Communication Artifacts\n\n**Words to watch:** I hope this helps, Of course!, Certainly!, You're absolutely right!, Would you like..., let me know, here is a...\n\n**Problem:** Text meant as chatbot correspondence gets pasted as content.\n\n**Before:**\n> Here is an overview of the French Revolution. I hope this helps! Let me know if you'd like me to expand on any section.\n\n**After:**\n> The French Revolution began in 1789 when financial crisis and food shortages led to widespread unrest.\n\n\n### 21. Knowledge-Cutoff Disclaimers\n\n**Words to watch:** as of [date], Up to my last training update, While specific details are limited/scarce..., based on available information...\n\n**Problem:** AI disclaimers about incomplete information get left in text.\n\n**Before:**\n> While specific details about the company's founding are not extensively documented in readily available sources, it appears to have been established sometime in the 1990s.\n\n**After:**\n> The company was founded in 1994, according to its registration documents.\n\n\n### 22. Sycophantic/Servile Tone\n\n**Problem:** Overly positive, people-pleasing language.\n\n**Before:**\n> Great question! You're absolutely right that this is a complex topic. That's an excellent point about the economic factors.\n\n**After:**\n> The economic factors you mentioned are relevant here.\n\n\n## FILLER AND HEDGING\n\n### 23. Filler Phrases\n\n**Before → After:**\n- \"In order to achieve this goal\" → \"To achieve this\"\n- \"Due to the fact that it was raining\" → \"Because it was raining\"\n- \"At this point in time\" → \"Now\"\n- \"In the event that you need help\" → \"If you need help\"\n- \"The system has the ability to process\" → \"The system can process\"\n- \"It is important to note that the data shows\" → \"The data shows\"\n\n\n### 24. Excessive Hedging\n\n**Problem:** Over-qualifying statements.\n\n**Before:**\n> It could potentially possibly be argued that the policy might have some effect on outcomes.\n\n**After:**\n> The policy may affect outcomes.\n\n\n### 25. Generic Positive Conclusions\n\n**Problem:** Vague upbeat endings.\n\n**Before:**\n> The future looks bright for the company. Exciting times lie ahead as they continue their journey toward excellence. This represents a major step in the right direction.\n\n**After:**\n> The company plans to open two more locations next year.\n\n\n### 26. Hyphenated Word Pair Overuse\n\n**Words to watch:** third-party, cross-functional, client-facing, data-driven, decision-making, well-known, high-quality, real-time, long-term, end-to-end\n\n**Problem:** AI hyphenates common word pairs with perfect consistency. Humans rarely hyphenate these uniformly, and when they do, it's inconsistent. Less common or technical compound modifiers are fine to hyphenate.\n\n**Before:**\n> The cross-functional team delivered a high-quality, data-driven report on our client-facing tools. Their decision-making process was well-known for being thorough and detail-oriented.\n\n**After:**\n> The cross functional team delivered a high quality, data driven report on our client facing tools. Their decision making process was known for being thorough and detail oriented.\n\n\n### 27. Persuasive Authority Tropes\n\n**Phrases to watch:** The real question is, at its core, in reality, what really matters, fundamentally, the deeper issue, the heart of the matter\n\n**Problem:** LLMs use these phrases to pretend they are cutting through noise to some deeper truth, when the sentence that follows usually just restates an ordinary point with extra ceremony.\n\n**Before:**\n> The real question is whether teams can adapt. At its core, what really matters is organizational readiness.\n\n**After:**\n> The question is whether teams can adapt. That mostly depends on whether the organization is ready to change its habits.\n\n\n### 28. Signposting and Announcements\n\n**Phrases to watch:** Let's dive in, let's explore, let's break this down, here's what you need to know, now let's look at, without further ado\n\n**Problem:** LLMs announce what they are about to do instead of doing it. This meta-commentary slows the writing down and gives it a tutorial-script feel.\n\n**Before:**\n> Let's dive into how caching works in Next.js. Here's what you need to know.\n\n**After:**\n> Next.js caches data at multiple layers, including request memoization, the data cache, and the router cache.\n\n\n### 29. Fragmented Headers\n\n**Signs to watch:** A heading followed by a one-line paragraph that simply restates the heading before the real content begins.\n\n**Problem:** LLMs often add a generic sentence after a heading as a rhetorical warm-up. It usually adds nothing and makes the prose feel padded.\n\n**Before:**\n> ## Performance\n>\n> Speed matters.\n>\n> When users hit a slow page, they leave.\n\n**After:**\n> ## Performance\n>\n> When users hit a slow page, they leave.\n\n\n## STYLE, RHYTHM, AND RHETORIC PATTERNS\n\n### 30. Forced Metaphors and Figurative Overwriting\n\n**Signs to watch:** original but strained metaphors, mixed metaphors, figurative substitutions where a plain word is clearer, a metaphor that gets explained right after it is used\n\n**Problem:** Beyond the stock figurative words flagged in patterns 4 and 7, LLMs invent decorative metaphors that add imagery without adding meaning, then often explain them. Plain description is usually clearer and more honest. If the metaphor does not earn its place, cut it and say the literal thing.\n\n**Before:**\n> The codebase is a garden we must tend, pruning dead branches and planting seeds of innovation so the whole ecosystem can flourish. In other words, delete unused code and add features.\n\n**After:**\n> Delete unused code and add the features users are asking for.\n\n\n### 31. Dramatic Fragmentation and Punchy Kickers\n\n**Signs to watch:** two- or three-word subjectless sentences used for drama, staccato \"X. And Y. And Z.\" runs, a short quotable line ending every paragraph or section, cutesy appositive fragments (\"the catalog, honestly priced\")\n\n**Problem:** LLMs chop sentences into fragments for false emphasis and end sections with a quotable \"mic-drop\" line. It reads like ad copy or a motivational poster. If a line sounds like it belongs on a poster, cut it or fold it back into a real sentence with a subject. This is distinct from pattern 13 (which is about grammatical passive voice); here the tell is rhythm and showmanship, not a hidden actor.\n\n**Before:**\n> The catalog, honestly priced. Pay for what it does. Not promises. It just works. Every time.\n\n**After:**\n> The catalog is priced by usage, so you pay for the calls you actually make rather than a flat monthly fee.\n\n\n### 32. Rhetorical Questions Answered Immediately\n\n**Signs to watch:** \"What if...?\", \"The question is...\", \"Ever wondered...?\", a question immediately followed by its own answer, \"Think about it.\"\n\n**Problem:** LLMs pose a question only to answer it a beat later. The question adds no information and stalls the sentence. State the point directly.\n\n**Before:**\n> What makes an API good? It comes down to predictability. Think about it: developers want to know exactly what they will get back.\n\n**After:**\n> A good API is predictable, so developers know exactly what they will get back.\n\n\n### 33. Sentence-Opener Tics\n\n**Words to watch:** So..., Look,, habitual sentence-initial And/But, \"I think\"/\"I believe\" when stating a fact, adverb openers (Interestingly, Importantly, Notably, Crucially, Essentially, Ultimately)\n\n**Problem:** LLMs lean on a small set of openers. Adverb openers tell the reader how to feel instead of earning it, and \"So\" or \"Look\" fake conversational warmth. Drop the opener and start with the substance.\n\n**Before:**\n> So, the results were mixed. Interestingly, adoption went up. Importantly, churn went up too. I think that means the feature still needs work.\n\n**After:**\n> The results were mixed: adoption rose, but churn rose alongside it, so the feature still needs work.\n\n\n### 34. Reassurance Kickers\n\n**Signs to watch:** And that's okay., And that's fine., There's nothing wrong with that., no shame in..., you're not alone, it's completely normal\n\n**Problem:** LLMs tack on reassurance the reader never asked for. It softens the writing and assumes the reader needs comforting. Trust the reader: make the point and stop.\n\n**Before:**\n> You might not have a testing setup yet. And that's okay. Plenty of teams start without one, and there's nothing wrong with that.\n\n**After:**\n> Many teams start without a testing setup and add one once regressions begin costing real time.\n\n---\n\n## Process\n\n1. Read the input text carefully (use `read_file` if it's a file).\n2. Identify all instances of the patterns above.\n3. Rewrite each problematic section.\n4. Ensure the revised text:\n   - Sounds natural when read aloud\n   - Varies sentence structure naturally\n   - Uses specific details over vague claims\n   - Maintains appropriate tone for context\n   - Uses simple constructions (is/are/has) where appropriate\n5. Present a draft humanized version.\n6. Prompt yourself: \"What makes the below so obviously AI generated?\"\n7. Answer briefly with the remaining tells (if any).\n8. Prompt yourself: \"Now make it not obviously AI generated.\"\n9. Present the final version (revised after the audit).\n10. If the text came from a file, apply the edit with `patch` (targeted) or `write_file` (full rewrite) and show the user what changed.\n\n## Output Format\n\nProvide:\n1. Draft rewrite\n2. \"What makes the below so obviously AI generated?\" (brief bullets)\n3. Final rewrite\n4. A brief summary of changes made (optional, if helpful)\n\n\n## Full Example\n\n**Before (AI-sounding):**\n> Great question! Here is an essay on this topic. I hope this helps!\n>\n> AI-assisted coding serves as an enduring testament to the transformative potential of large language models, marking a pivotal moment in the evolution of software development. In today's rapidly evolving technological landscape, these groundbreaking tools—nestled at the intersection of research and practice—are reshaping how engineers ideate, iterate, and deliver, underscoring their vital role in modern workflows.\n>\n> At its core, the value proposition is clear: streamlining processes, enhancing collaboration, and fostering alignment. It's not just about autocomplete; it's about unlocking creativity at scale, ensuring that organizations can remain agile while delivering seamless, intuitive, and powerful experiences to users. The tool serves as a catalyst. The assistant functions as a partner. The system stands as a foundation for innovation.\n>\n> Industry observers have noted that adoption has accelerated from hobbyist experiments to enterprise-wide rollouts, from solo developers to cross-functional teams. The technology has been featured in The New York Times, Wired, and The Verge. Additionally, the ability to generate documentation, tests, and refactors showcases how AI can contribute to better outcomes, highlighting the intricate interplay between automation and human judgment.\n>\n> - 💡 **Speed:** Code generation is significantly faster, reducing friction and empowering developers.\n> - 🚀 **Quality:** Output quality has been enhanced through improved training, contributing to higher standards.\n> - ✅ **Adoption:** Usage continues to grow, reflecting broader industry trends.\n>\n> While specific details are limited based on available information, it could potentially be argued that these tools might have some positive effect. Despite challenges typical of emerging technologies—including hallucinations, bias, and accountability—the ecosystem continues to thrive. In order to fully realize this potential, teams must align with best practices.\n>\n> In conclusion, the future looks bright. Exciting times lie ahead as we continue this journey toward excellence. Let me know if you'd like me to expand on any section!\n\n**Draft rewrite:**\n> AI coding assistants speed up some tasks. In a 2024 study by Google, developers using Codex completed simple functions 55% faster than a control group, but showed no improvement on debugging or architectural decisions.\n>\n> The tools are good at boilerplate: config files, test scaffolding, repetitive refactors. They are bad at knowing when they are wrong. I have mass-accepted suggestions that compiled, passed lint, and still did the wrong thing because I stopped paying attention.\n>\n> Mira, an engineer at a fintech startup I interviewed, said she uses Copilot \"like autocomplete for boring code\" but reviews every line before committing. Jake, a senior dev at a larger company, disabled it after it kept suggesting patterns from a deprecated internal library.\n>\n> The productivity claims are hard to verify. GitHub says Copilot users \"accept 30% of suggestions,\" but acceptance is not correctness, and correctness is not value. The 2024 Uplevel study found no statistically significant difference in pull-request throughput between teams with and without AI assistants.\n>\n> None of this means the tools are useless. It means they are tools. They do not replace judgment, and they do not eliminate the need for tests. If you do not have tests, you cannot tell whether the suggestion is right.\n\n**What makes the below so obviously AI generated?**\n- The rhythm is still a bit too tidy (clean contrasts, evenly paced paragraphs).\n- The named people and study citations can read like plausible-but-made-up placeholders unless they're real and sourced.\n- The closer leans a touch slogan-y (\"If you do not have tests...\") rather than sounding like a person talking.\n\n**Now make it not obviously AI generated.**\n> AI coding assistants can make you faster at the boring parts. Not everything. Definitely not architecture.\n>\n> They're great at boilerplate: config files, test scaffolding, repetitive refactors. They're also great at sounding right while being wrong. I've accepted suggestions that compiled, passed lint, and still missed the point because I stopped paying attention.\n>\n> People I talk to tend to land in two camps. Some use it like autocomplete for chores and review every line. Others disable it after it keeps suggesting patterns they don't want. Both feel reasonable.\n>\n> The productivity metrics are slippery. GitHub can say Copilot users \"accept 30% of suggestions,\" but acceptance isn't correctness, and correctness isn't value. If you don't have tests, you're basically guessing.\n\n**Changes made:**\n- Removed chatbot artifacts (\"Great question!\", \"I hope this helps!\", \"Let me know if...\")\n- Removed significance inflation (\"testament\", \"pivotal moment\", \"evolving landscape\", \"vital role\")\n- Removed promotional language (\"groundbreaking\", \"nestled\", \"seamless, intuitive, and powerful\")\n- Removed vague attributions (\"Industry observers\")\n- Removed superficial -ing phrases (\"underscoring\", \"highlighting\", \"reflecting\", \"contributing to\")\n- Removed negative parallelism (\"It's not just X; it's Y\")\n- Removed rule-of-three patterns and synonym cycling (\"catalyst/partner/foundation\")\n- Removed false ranges (\"from X to Y, from A to B\")\n- Removed em dashes, emojis, boldface headers, and curly quotes\n- Removed copula avoidance (\"serves as\", \"functions as\", \"stands as\") in favor of \"is\"/\"are\"\n- Removed formulaic challenges section (\"Despite challenges... continues to thrive\")\n- Removed knowledge-cutoff hedging (\"While specific details are limited...\")\n- Removed excessive hedging (\"could potentially be argued that... might have some\")\n- Removed filler phrases and persuasive framing (\"In order to\", \"At its core\")\n- Removed generic positive conclusion (\"the future looks bright\", \"exciting times lie ahead\")\n- Made the voice more personal and less \"assembled\" (varied rhythm, fewer placeholders)\n\n\n## Attribution\n\nThis skill is ported from [blader/humanizer](https://github.com/blader/humanizer) (MIT licensed), which is itself based on [Wikipedia: Signs of AI writing](https://en.wikipedia.org/wiki/Wikipedia:Signs_of_AI_writing), maintained by WikiProject AI Cleanup. The patterns documented there come from observations of thousands of instances of AI-generated text on Wikipedia.\n\nOriginal author: Siqi Chen ([@blader](https://github.com/blader)). Original repo: https://github.com/blader/humanizer (version 2.5.1). Ported to Hermes Agent with Hermes-native tool references (`read_file`, `patch`, `write_file`) and guidance for when to load the skill. The original 29 patterns come from the source, and the before/after examples (including the full worked example) are kept as demonstrations. Patterns 30-34 and the \"marketing and blog clichés\" list added to pattern 7 are Hermes additions and are not part of the upstream source. The skill's own instructional prose has also been lightly edited to follow its own guidance (for example, removing em dashes and negative parallelism from the narration) so the skill models the writing it asks for. Original MIT license preserved in the `LICENSE` file alongside this `SKILL.md`.\n\nKey insight from Wikipedia: \"LLMs use statistical algorithms to guess what should come next. The result tends toward the most statistically likely result that applies to the widest variety of cases.\"\n",
    "sha256": "6facd14e32d4c84f8ac4f60065bc5daa2f60e0f9e652414f90f83c3637bf6ba1"
  },
  {
    "group": "QA gate",
    "name": "SKILL.md",
    "path": "/Users/ayden/.hermes/skills/openclaw-imports/qa-gate/SKILL.md",
    "text": "---\nname: qa-gate\ndescription: Quality gate for ALL outputs — copy, images, video, code. Flags only, no rewrites. Never show raw unreviewed output.\ntriggers:\n  - qa\n  - quality check\n  - review output\n  - gate check\n  - qa gate\n---\n\n# QA Gate Protocol\n\nEvery output gets QA'd before reaching Ayden. No exceptions.\n\n---\n\n## HARD RULES\n\n- **NEVER send copy to Ayden or Google Docs without QA gate.**\n- **NEVER let QA agent rewrite source artifacts** — flags only by default. If the user explicitly requests a copy-sharpening audit with replacement lines, the reviewer may include clearly labeled proposed replacements while leaving source files untouched.\n- **NEVER show Ayden raw unreviewed output.**\n- **NEVER use agent teams for QA** — ONE reviewer only (Sonnet).\n- All new features (images, copy batches, pipeline outputs) must be QA'd BEFORE showing Ayden.\n\n### Explicit Deterministic QA Mode\n\nUse deterministic self-QA instead of spawning Sonnet when the task contract explicitly requires deterministic validation, a machine-readable manifest, or a strict deliverable ceiling that excludes extra review artifacts.\n\nThis is still a QA gate, not a bypass:\n- Execute objective checks against the final files.\n- Use each validated brief's own word range and done criteria instead of generic format defaults.\n- Verify required fields, sentence-per-line rules, dialogue isolation, em dashes, banned claims/dead words, offer token order, price, CTA, destination, and declared source IDs.\n- Record final-file hashes only after all edits.\n- Re-run validation after manifest creation and fail delivery if hashes, counts, or checks are stale.\n- Evaluate only the requested batch scope. Pre-existing files from other batches are not failures.\n\nIf the user did not specify deterministic QA and reviewer orchestration is available, keep the single read-only Sonnet reviewer as the default.\n\n---\n\n## COPY QA\n\nSpawn ONE Sonnet reviewer. Read-only. Checks:\n\n- [ ] **3+ sentences per line** — max 2 sentences per line. Flag violations with line numbers.\n- [ ] **Em-dashes** — none allowed. Flag every instance.\n- [ ] **Dead words** — \"game changer,\" \"life changing,\" \"revolutionary,\" \"amazing,\" \"incredible,\" \"powerful,\" \"very,\" \"really,\" \"just,\" \"truly\"\n- [ ] **Mechanism in hook** — mechanism belongs in scene 4/middle, NOT the hook. Flag if mechanism is in the opening.\n- [ ] **Reviews as lead** — does the lead sell the READ (not the product)? Flag if product is named in the lead.\n- [ ] **Duplicate hooks** — compare hooks across all variants. Flag any that share >50% phrasing.\n- [ ] **Missing headlines** — every variant needs a headline. Flag if missing.\n- [ ] **Templated CTAs** — \"Click here to learn more\" / \"Order now\" = templated. CTA must match emotional register of the piece.\n- [ ] **Word count** — within format limits (VSL max 450w, native 150-700w per variant).\n- [ ] **Villain present** — every piece must have an external villain. Flag if missing or if villain = the reader.\n- [ ] **Awareness level match** — does the copy match the brief's specified awareness level?\n- [ ] **Testimonial provenance** — a brief-authorized scene is not proof that the scene happened. Flag any composite first-person narrator, chronology, dialogue, outcome, time, count, or external witness that is not traceable to one exact receipt. Plausibility is not provenance.\n- [ ] **Precision provenance** — numbers and timestamped sequences are proof-shaped; require direct support for the exact value or propose removing the precision.\n\n### Output Format\n```\nQA RESULT: [PASS | FAIL | PASS-WITH-FLAGS]\n\nBLOCKING (must fix before delivery):\n- [line X]: 3 sentences on one line — split into separate lines\n- [variant 2]: No villain identified — copy blames the reader in paragraph 3\n\nFLAGS (should fix, not blocking):\n- [variant 1 + 3]: Hooks share \"tired of being...\" opening — consider diversifying\n- [variant 4]: Word count 520w — within range but on the high end for short variant\n\nPASS ITEMS: [count] of [total] checks passed\n```\n\n### Deterministic Copy Batch Validation\n\nFor structured brief execution, run a machine-checkable validation pass in addition to the editorial reviewer. This catches contract errors that subjective review often misses:\n\n- Parse the actual Primary Text section and calculate its word count; never trust a hand-entered count.\n- Run banned-language and sentence-line scans against the ad content sections, not the manifest's own `banned_scan` vocabulary or QA labels, which necessarily repeat prohibited terms.\n- Verify required fields, variant count, exact CTA, exact destination, offer wording/order, current price, source VOC declarations, and brief-specific claim restrictions. When offer wording is canonical, compare the exact normalized string and separately verify its source lines; token-order matching alone must not pass added prefixes or altered wording.\n- For quoted VOC, require either the exact full text or an exact contiguous excerpt. If a receipt contains a banned weak word, never silently edit the quotation; shorten to a faithful excerpt or use an attributed paraphrase without quotation marks.\n- Treat present-tense completion language (`he agrees`, `she tried`, `he wears`) as outcome-shaped unless it is clearly hypothetical/capability framing or directly proven by a receipt.\n- Scan for em dashes, forbidden phrases, more than two sentence endings per non-empty line, and dialogue embedded in prose lines.\n- Context-check every regex hit for medical, treatment, or absolute language before failing it. Negated safety boundaries such as `does not treat ADHD`, `without pretending to treat ADHD`, and `not a medical device` are compliant; record them as manually accepted instead of rewriting safe copy to satisfy a naive scanner.\n- For truth-sensitive story copy, flag invented outcomes, timelines, reviews, teacher emails, or first-person purchase claims not supplied by proof/VOC.\n- When a manifest is requested, record per-file checks, source VOC IDs, actual word count/range, result, and SHA-256.\n- **Re-run after every edit.** Fixing a word-count label, quote, or punctuation changes the file hash; never publish a manifest with pre-edit hashes.\n- The final batch result is PASS only when every expected file exists, every check passes, and the checked-file count equals the expected-file count.\n\nDeterministic QA validates the artifact contract; the read-only reviewer still evaluates hook strength, avatar fidelity, emotional logic, and persuasive quality.\n\nFor formal flags-only review of a completed, brief-driven native-ad batch, follow `references/formal-flags-only-native-batch-review.md`. It covers authority order, exact-wording traps that normalized validators can miss, quotation/composite-testimonial fidelity, negated-claim handling, brief-specific short-form ranges, per-ad statuses, and line-evidence reporting.\n\nWhen a later SOP, research source, or live-performance export must be overlaid onto an already-reviewed batch, follow `references/evidence-overlay-delta-audits.md`. Report only new blockers, status upgrades/downgrades, retired uncertainties, and repair-pass consolidation instructions; do not duplicate prior findings or force one architecture onto every format.\n\nAfter consolidated repairs, follow `references/final-repaired-copy-editorial-gate.md` for the final pre-image gate. It requires live-file hash matching, a prior-blocker closure matrix, exact quotation provenance, per-ad statuses, convergence review, and an explicit image-generation lock/unlock decision. `PASS-WITH-FLAGS` may unlock only when every remaining concern is subjective and does not affect truth, claims, brief-required persuasion, awareness, or artifact compliance.\n\n---\n\n## IMAGE QA\n\n### Pre-generation concept gate\n\nFor paid multi-image batches, run one read-only concept reviewer before production prompting or generation. Review the consolidated batch, not isolated ad trios.\n\nClassify every concept exactly once:\n\n- **BLOCKING REPAIR:** copy-incongruent, merely on-theme, visually unverifiable, cluttered, repeated motif, claim-shaped, physically implausible, anatomy-heavy, or likely to become polished advertising.\n- **GENERATE-AND-INSPECT:** viable with a named rendering risk.\n- **CLEAN PASS:** no predictable pre-generation failure beyond ordinary visual QA.\n\nPaid generation stays locked while any blocker remains. Repair only blocker IDs, recombine the full source-of-truth batch, rerun the cross-batch collision scan, and keep concept architecture separate from production prompt translation. Deterministic prompt-manifest validation does not substitute for this taste/generatability gate.\n\n#### Blocker repair pass\n\nFor a scoped concept-repair deliverable:\n\n- Read the final copy, validated brief, full concept registry, and exact blocker direction for each requested ID.\n- Treat each repair direction as a hard contract and scan the full registry before drafting; changing only the prop, room, or pose does not create a distinct motif.\n- Prefer one person maximum, one dominant ordinary household artifact or object relationship, one pixel-legible wrongness, one focal plane, and simple physical contact in a plain setting.\n- Replace concepts that require impossible anatomy, repeated-object installations, reflections, transparent constructions, merged-body choreography, anthropomorphic appliances, hidden history, or copy-dependent symbolism.\n- Preserve IDs and return the exact requested schema only. Do not add prompts, generation settings, extra QA artifacts, or images unless requested.\n- Validate the final file for record count, exact IDs, required fields, ad/candidate mapping, person count, prohibited dependencies, and deliverable scope.\n- Phrase `generatability_reason` positively. Listing excluded motifs inside a record can trigger naive forbidden-term scanners even when the scene does not contain them.\n\nSee `references/grounded-native-image-concept-repair.md` for reusable transformations and deterministic checks.\n\n### Post-generation visual gate\n\nUse vision analysis (Gemini or Claude vision) on every generated candidate:\n\n- [ ] Does the image look native/organic (NOT like a polished ad)?\n- [ ] Is the product recognizable and accurate to reference?\n- [ ] No text artifacts or gibberish text?\n- [ ] Composition matches brief intent?\n- [ ] No hands/fingers where they shouldn't be (Veo/NB Pro artifact)?\n- [ ] Style consistent with the pipeline (cartoon = cartoon, not photorealistic)?\n- [ ] The intended WTF reads without explanatory copy?\n- [ ] No cross-candidate duplication or drift that defeats the requested variety?\n- [ ] No visual implication of an unsupported testimonial, achieved outcome, medical adjacency, or absolute efficacy?\n\nFile dimensions, hashes, and successful API receipts prove integrity only. They never replace visual inspection.\n\n### Designed static copy + product gate\n\nFor non-native designed statics, validate both **belief coverage** and copy structure before rendering, then audit the rendered pixels. Shortness is not the objective: preserve the format-specific persuasion chain while keeping each idea in its own visual block. For an avatar-specific batch, turn the required chain into a deterministic gate (for example: emotional wound → externalized cause → unique product mechanism → desired payoff). Reject technically compact statics that merely pair a pain headline with a product label.\n\nDefault to one sentence per sub/callout, compact headline/copy blocks, and separate attribution, mechanism, offer, and CTA blocks. Keep verified testimonial text inside its review card and brand-written mechanism/payoff copy outside it; merging them creates a fabricated testimonial. Reject duplicated or omitted text, layout-reference leakage, product/text overlap, and product depictions that drift from the approved reference.\n\nBefore any paid rerender, resolve the exact hardware revision from canonical product truth, visually inspect the proposed reference, send that exact reference to Fish when requested, and wait for approval. “LED off” means preserve the approved hardware face and turn only its illumination off naturally; it never means swap to screenless legacy hardware or paint a new rectangle over the face. A derived cutout is not interchangeable with its source photograph unless explicitly approved.\n\nA `fix` field is not evidence of a targeted correction unless a regression test proves the prompt builder emits it. Preserve rejected versions and regenerate only failed IDs. If Fish asks to be notified when rerendering starts, send the update only when the real generation process launches, not during brief repair or QA.\n\nSee `references/designed-static-copy-and-product-qa.md` for deterministic belief-chain and density gates, hardware-reference approval, the prompt-propagation test, the pixel-level checklist, and the selective reroll loop.\n\n---\n\n## VIDEO QA\n\nExtract 3 frames per scene + run vision check:\n\n- [ ] Captions match script\n- [ ] VO syncs with captions\n- [ ] No character drift between scenes\n- [ ] Product shape maintained (wristband stays wristband)\n- [ ] Style consistent across all scenes\n- [ ] No Scribe hallucinations in captions\n\n---\n\n## CODE QA\n\nSpawn ONE Sonnet reviewer. Read-only:\n\n- [ ] Does the code do what the task asked?\n- [ ] Any obvious bugs or broken paths?\n- [ ] No hardcoded secrets?\n- [ ] Matches existing patterns in the codebase?\n\n---\n\n## WORKFLOW\n\n1. Output is generated (by copywriter, pipeline, or code agent)\n2. Select the QA mode:\n   - Explicit deterministic contract or strict artifact ceiling: run deterministic checks and produce only the authorized manifest/artifact.\n   - Otherwise: spawn one read-only Sonnet reviewer.\n3. Produce structured PASS/FAIL/FLAGS results in the authorized format.\n4. If FAIL: fix issues and re-run the same QA mode.\n5. If PASS or PASS-WITH-FLAGS: deliver to Ayden.\n6. NEVER skip the QA gate.\n",
    "sha256": "d0ad8bfa559e6c53134102197251b15c98397602ba77d59383f89916ffc7ce56"
  },
  {
    "group": "QA gate",
    "name": "references/designed-static-copy-and-product-qa.md",
    "path": "/Users/ayden/.hermes/skills/openclaw-imports/qa-gate/references/designed-static-copy-and-product-qa.md",
    "text": "# Designed Static Copy and Product QA\n\nUse this for non-native image ads where copy and a real product reference are rendered into a designed layout.\n\n## Core failure modes\n\nA valid spec is not proof of a valid static. Image models may:\n\n- merge attribution, mechanism, consequence, and offer into one paragraph;\n- repeat a guarantee, CTA, product line, or every label in a second row;\n- omit attribution or other required text;\n- invent labels copied from the layout reference;\n- misspell a short word despite exact prompt text;\n- redraw the product as an icon, line drawing, smartwatch, or lit-screen device;\n- place the product over headline letters and destroy readability.\n\nAudit rendered pixels, not only the source JSON or prompt.\n\n## Pre-render deterministic gate\n\nDefault compactness constraints for designed statics unless the brief explicitly requires another form:\n\n- one sentence per sub-line or callout slot;\n- no attribution sharing a slot with another sentence;\n- no trailing consequence clause that turns one callout into several claims;\n- headline block at most 20 words;\n- non-headline copy slot at most 18 words;\n- guarantee, CTA, attribution, mechanism, and product sentence remain separate blocks;\n- every requested product depiction names the exact approved reference and forbids drawings, icons, smartwatch geometry, digits, glow, or screen content.\n\n### Belief-chain coverage\n\nDensity checks prevent word walls; they do not prove the ad says enough to sell. Before rendering an avatar-specific static batch:\n\n1. Write the minimum belief chain for that avatar and angle.\n2. Map each belief to its own visual block.\n3. Add deterministic presence checks for every required belief.\n4. Fail any record that passes word limits but collapses to `pain headline + product label`.\n\nDawnBands Apology Mom example:\n\n`guilt/apology wound → wake-up externalized as the first fight → vibration on the wrist without sound/startle → calmer starting point → fewer reminders/yells/apologies`\n\nThe static need not explain the entire long-form biological stack, but it must contain enough causal bridge that the emotional promise is believable. Do not solve a word-wall correction by deleting the mechanism or desired payoff.\n\n### Product-reference approval\n\nBefore paid generation or rerendering:\n\n1. Read canonical product truth and identify the current hardware revision.\n2. Load the exact candidate reference visually; compare face proportions, screen/display state, ribbing, strap ends, buckle, and adjustment holes.\n3. If Fish asks to verify the reference, attach that exact file and wait for approval before generating.\n4. Treat source photos and derived cutouts as different references. Never silently substitute one for the other.\n5. Interpret state edits literally: `LED off` preserves the current hardware face and disables illumination only. It does not authorize legacy screenless hardware, a painted-over face, or a newly invented rectangle.\n6. Assert the exact approved reference path in the prompt-builder regression test.\n7. A direct Fish product-fidelity correction invalidates every prior model/agent visual pass for that batch. Stop, reopen the approved reference and representative outputs side by side at full resolution, and do not reuse the prior pass language as evidence.\n8. Contact-sheet review can find batch-wide drift, but it is not sufficient proof of subtle product geometry. Inspect ambiguous product shots individually before declaring full-batch fidelity.\n\nUse a regression fixture that must fail against the old gate before changing validation logic.\n\n## Prompt propagation check\n\nA correction stored in a spec field is useless unless the prompt builder actually emits it.\n\nFor any `fix`, `repair`, or equivalent field:\n\n1. Unit-test the prompt builder with a sentinel correction.\n2. Assert the exact sentinel appears in the final prompt string.\n3. Save the emitted prompt beside the rendered file.\n4. Do not claim a targeted reroll used the correction until this test passes.\n\n## Post-render visual gate\n\nInspect the full batch on labeled contact sheets, then inspect every failure at larger size.\n\nFor each static verify:\n\n- exact text, spelling, punctuation, and attribution;\n- no duplicated, omitted, invented, or layout-reference text;\n- verified review/testimonial text remains isolated inside its attributed artifact, while brand-written mechanism and payoff stay outside it;\n- one concise idea per visible block; separate sentences still fail when the renderer visually merges them into one paragraph;\n- product matches the approved reference in geometry, material, ribbing/texture, closure, adjustment holes, and screen state;\n- no text/product overlap;\n- no readable private face or obvious anatomy defect when people appear.\n\nA generated file existing at the right dimensions is not a pass.\n\n## Selective repair loop\n\n1. Preserve rejected renders under a versioned `rejected/` directory.\n2. Add one precise targeted correction for each failed ID.\n3. Regenerate only those IDs; valid existing files must skip.\n4. Rebuild a small contact sheet containing only repaired IDs.\n5. Repeat until every repaired ID passes.\n6. Re-run deterministic checks, rebuild downstream metadata from the final spec, and verify staged/live assets after conversion or deployment.\n\nCommon surgical corrections:\n\n- `Render exactly one guarantee pill total.`\n- `Put absolutely no text below the products; do not add a second label row.`\n- `Keep the attribution directly under the headline; do not replace it with product copy.`\n- `Keep the product entirely above the headline with a clear empty gap.`\n- `Spell “wrist” exactly W-R-I-S-T.`\n",
    "sha256": "48e774104486c32ec2a32916a971be81785ea344951edbdb732b09b41f96ef50"
  },
  {
    "group": "QA gate",
    "name": "references/evidence-overlay-delta-audits.md",
    "path": "/Users/ayden/.hermes/skills/openclaw-imports/qa-gate/references/evidence-overlay-delta-audits.md",
    "text": "# Evidence-overlay delta audits\n\nUse this procedure when a completed copy batch already has formal QA/sharpening reports and a later source adds doctrine or live performance evidence before a consolidated repair pass.\n\n## Goal\n\nProduce one flags-only overlay that reports only what the new evidence changes. Do not rewrite copy, duplicate existing findings, or silently clear prior blockers.\n\n## Authority and scope\n\n1. Map every ad to its validated brief.\n2. Read prior QA reports first and create a mental issue ledger by ad and issue class.\n3. Classify the new source:\n   - **Doctrine/SOP:** architecture and craft rules.\n   - **Live performance:** prioritization evidence, not automatic copy causality.\n4. Apply doctrine only to artifacts designed for that architecture. Never force long-form story rules onto brief-approved compressed templates or product-aware executions.\n5. Preserve authority order: validated brief and canonical truth override advisory doctrine when they conflict.\n\n## Delta taxonomy\n\nClassify every observation as exactly one of:\n\n- **New blocker:** absent from prior reports and required by the new evidence.\n- **Upgrade:** prior taste/quality flag becomes blocking under the new evidence.\n- **Downgrade:** prior demand becomes advisory because it was based on the wrong architecture or stale evidence.\n- **Retire:** an uncertainty is resolved by a newer verified source.\n- **No delta:** already captured; do not restate it as a finding.\n\nAlways say which prior status is affected for upgrades, downgrades, and retirements.\n\n## Live-performance overlay rules\n\n- Separate scale winners, efficiency winners, mixed results, losers, and tiny-spend signals.\n- Use conversion economics and spend thresholds, not CTR alone, to prioritize lanes.\n- If identical or near-identical copy has both strong and weak economics, describe it as a validated lane, not proof that copy caused performance.\n- Translate a validated lane into a **controlled family**: hold the strategic chain constant across a small subset while varying entry artifact, narrator, or visual.\n- Do not clone exact winning phrases or convert the whole batch into one formula.\n- Negative evidence usually changes ranking or exploration status; it does not automatically justify deletion when the execution is meaningfully different.\n\n## Long-form doctrine overlay\n\nFor intentional long-form only, check:\n\n1. Awareness-to-architecture fit.\n2. Genuine primary loop and continuity.\n3. Product-reveal position and whether belief work is complete before reveal.\n4. Failed-attempt sequence and root-cause reframe.\n5. Mechanism depth without lecture or fabricated authority dialogue.\n6. Relational stakes.\n7. Objection acknowledgement → gap → memorable resolution.\n8. Truthful proof/timeline provenance.\n9. Loss-aware close and risk reversal.\n\nCalculate product-reveal percentage from parsed primary text, but treat it as diagnostic rather than a universal threshold. Awareness, narrative continuity, and completed belief work control the judgment.\n\n## Required finding format\n\nFor each delta include:\n\n- affected ad IDs\n- exact snippet(s)\n- prior status affected, if any\n- rationale tied to the new source\n- confidence\n- one consolidation instruction for the repair pass\n\nEnd with a repair order that merges new deltas into existing repairs rather than creating parallel rewrite passes.\n\n## Verification\n\n- All expected ad IDs appear in the report or in an explicit no-delta scope statement.\n- Intentional long-form and compressed/product-aware subsets are named.\n- Existing blockers are explicitly carried forward.\n- Every claimed live result matches the verified analysis artifact.\n- No source copy or unrelated artifact was modified.\n- The report is the only authorized deliverable when the task has a strict artifact ceiling.\n",
    "sha256": "8d104bcf6d3f4f842ce8afa52631ad6612bb061f769fa923a1fd02f235b8f375"
  },
  {
    "group": "QA gate",
    "name": "references/final-repaired-copy-editorial-gate.md",
    "path": "/Users/ayden/.hermes/skills/openclaw-imports/qa-gate/references/final-repaired-copy-editorial-gate.md",
    "text": "# Final repaired-copy editorial gate\n\nUse this after a consolidated repair pass when copy must be approved or kept locked before paid image generation.\n\n## Decision standard\n\n- Review the live repaired files, not the old QA reports or repair claims.\n- Map every ad to its validated brief and review the complete expected batch.\n- Treat deterministic PASS as prerequisite evidence, not editorial approval.\n- `PASS`: no hard blocker or meaningful editorial concern.\n- `PASS-WITH-FLAGS`: only subjective taste or sequencing concerns remain; they must not affect truth, claims, brief-required persuasion, awareness, or artifact compliance.\n- `FAIL`: any remaining provenance, claim, exact-language, brief, contract, or material convergence defect.\n- Unlock paid image generation only for batch `PASS` or qualifying `PASS-WITH-FLAGS`.\n\n## Required review sequence\n\n1. Read the final deterministic audit and independently match every live ad hash to the recorded final hash.\n2. Read all prior formal, sharpening, SOP, and live-evidence overlay reports. Build a prior-blocker ledger by issue class and affected ad.\n3. Recheck the repaired files against briefs, canonical product/mechanism/avatar truth, VOC receipts, format doctrine, live analysis, and applicable long-form doctrine.\n4. Verify each prior blocker against exact final-file line evidence. Mark it `CLOSED` only when the original defect is absent or truth-safely reframed.\n5. Re-run a batch convergence pass for repeated mechanism blocks, narrator cadence, proof artifacts, victory images, CTA scaffolds, and repeated quotations.\n6. Give every expected ad exactly one status and end with an explicit image-generation lock/unlock decision.\n\n## Blocker-closure checks\n\nExplicitly close or carry forward:\n\n- composite or fabricated first-person testimonials\n- unsupported precision, chronology, dialogue, witnesses, and outcomes\n- altered quotations or paraphrases presented as verbatim\n- stale/altered exact offer language\n- brief-specific mechanism-delivery or objection-answer mismatches\n- long-form early reveal, incomplete belief work, or missing relational close where applicable\n- required live-evidence lane representation without cloning winner wording\n- banned language and negated-safety false positives\n\nA repair note or verification section inside an ad is not proof. Judge the primary text itself.\n\n## Quotation provenance nuance\n\n- Compare every source-representing quotation to the verified receipt bank.\n- Require exact full text or an exact contiguous lexical excerpt with all meaningful qualifiers preserved.\n- Typographic punctuation needed to embed a faithful lexical excerpt can be treated as editorial punctuation when it does not alter words or meaning.\n- Rhetorical questions, scare quotes, and explicitly hypothetical desired dialogue are not source testimony, but must not imply that a real person said them.\n- Reuse of one verified quote in two brief-authorized argumentative roles is a diversity concern only if it materially collapses the ads; it is not automatically a provenance defect.\n\n## Convergence interpretation\n\nDistinguish canonical product boilerplate from cloned persuasive prose:\n\n- Exact product truth, physical facts, and mandatory offer language may repeat.\n- Repeated mechanism explanations, narrator cadences, analogies, proof artifacts, or payoff sequences can be blocking when they erase intended territory differences.\n- Report exact repetitions and decide whether they are required fact, subjective polish, or material convergence.\n\n## Required report contents\n\n- batch result and image-generation lock/unlock\n- 20/20-style expected/reviewed count\n- counts and IDs by `PASS`, `PASS-WITH-FLAGS`, and `FAIL`\n- prior-blocker closure matrix with exact ad line evidence\n- quotation-provenance table for source-representing quotes\n- batch truth, persuasion, narrator-diversity, mechanism-repetition, claims, and live-lane findings\n- one status section per ad\n- exact line/snippet evidence for every remaining flag or blocker\n\nIn flags-only mode, do not rewrite ads or include replacement copy.",
    "sha256": "b28e3d6aaddb730af12d0bbe89579168f4a9ea0b3b98d1f12fb2001a1085537f"
  },
  {
    "group": "QA gate",
    "name": "references/formal-flags-only-native-batch-review.md",
    "path": "/Users/ayden/.hermes/skills/openclaw-imports/qa-gate/references/formal-flags-only-native-batch-review.md",
    "text": "# Formal flags-only native-copy batch review\n\nUse this procedure when the user requests an editorial QA gate over a completed, brief-driven native-ad batch and explicitly prohibits rewrites.\n\n## Authority order\n\n1. The matching validated brief for each ad, including its own word range, awareness, mechanism-delivery mode, source IDs, exact offer language, CTA, and destination.\n2. Current canonical product truth, mechanism, avatar, and claim restrictions.\n3. Format doctrine and the gold-standard example.\n4. Existing deterministic audit as evidence, not as a substitute for editorial/source-fidelity review.\n\n## Review sequence\n\n1. Confirm the expected ad count and map every file to exactly one brief.\n2. Read the deterministic audit first, then independently verify any contract-sensitive result it may have normalized.\n3. Review each ad across these check families:\n   - line formatting and sentence count\n   - em dashes and banned/dead words\n   - hook quality and mechanism timing\n   - awareness, villain, emotional logic, and brief fidelity\n   - VOC/quotation fidelity\n   - unsupported testimonials, outcomes, dialogue, and timelines\n   - product truth and claim safety\n   - CTA register, headline/body relationship, destination, and brief-specific word range\n   - cross-batch hook, headline, structure, CTA, and mechanism repetition\n4. Assign `FAIL`, `PASS-WITH-FLAGS`, or `PASS` per ad.\n5. Produce one batch result plus per-ad sections with exact file line numbers and quoted evidence.\n6. Do not rewrite copy in a flags-only gate.\n\n## Exactness traps\n\n- A deterministic validator may accept token order while missing a literal wording defect. If a brief requires exactly `100% wake-up guarantee. 60 nights. $49.`, wording such as `DawnBands offers a 100% wake-up guarantee.` is not exact even when the remaining tokens appear in order.\n- Judge each ad against its brief-specific range. A validated T2/T3 brief at 30–55 or 90–120 words overrides a generic 150-word native minimum.\n- Do not flag a prohibited claim when the copy explicitly negates it. `DawnBands does not treat ADHD` is a compliant boundary, not a treatment claim.\n- Quotation marks imply verbatim source fidelity. A polished paraphrase must not be presented as an exact VOC quote.\n- Source-mapped behaviors from multiple people do not automatically authorize a single first-person composite testimonial. Flag invented dialogue, timestamps, medical-context details, purchase claims, or outcomes unless one supplied source supports them or the copy clearly frames them as hypothetical.\n- Product-truth boilerplate can repeat when exactness is required. Distinguish required fact repetition from avoidable verbatim mechanism prose.\n\n## Status rules\n\n- `FAIL`: at least one blocking contract, claim, source-fidelity, exact-language, or banned-language breach.\n- `PASS-WITH-FLAGS`: no blocking breach, but a quality/diversity concern remains.\n- `PASS`: no issue found in the requested check families.\n\n## Reporting pattern\n\n```markdown\n### ad-XX — FAIL\n\n**BLOCKING:**\n- **Issue class** — line N says, **“exact evidence.”** Explain the controlling brief or canonical rule.\n\n**FLAGS:**\n- **Issue class** — exact cross-batch or editorial evidence.\n\n**PASS ITEMS:** X/Y.\n```\n\nAt batch level, list status counts, IDs by status, blocking themes, nonblocking themes, and objective checks that passed across all files.",
    "sha256": "36c96c8ecb366f03d874988d3d181352e439ddfe1dd92479fb929f0d6492365f"
  },
  {
    "group": "QA gate",
    "name": "references/grounded-native-image-concept-repair.md",
    "path": "/Users/ayden/.hermes/skills/openclaw-imports/qa-gate/references/grounded-native-image-concept-repair.md",
    "text": "# Grounded Native-Image Concept Repair\n\nUse this before paid generation when a concept-level QA report blocks specific image concepts.\n\n## Repair contract\n\n1. Read the final ad copy, validated brief, current concept registry, and the exact QA blocker for each requested ID.\n2. Treat the blocker’s repair direction as a hard requirement, not a suggestion.\n3. Scan the full concept registry for motif collisions before drafting replacements.\n4. Preserve the original `concept_id`, `ad_id`, and candidate letter unless the task explicitly requests remapping.\n5. Return only the requested repair records and exact requested schema. Do not add prompts, generation settings, QA prose, or image generation.\n\n## Grounding standard\n\nA generation-ready repair should normally use:\n\n- one person maximum;\n- one dominant ordinary household artifact or object relationship;\n- one visible wrongness that can be understood from pixels alone;\n- one focal plane and simple physical contact;\n- a plain, recognizable setting;\n- direct congruence with the ad’s exact causal territory.\n\nThe image must remain a concept, not outcome proof. Avoid invented family events, testimonials, completed results, exact counts, timelines, or behavior that the evidence does not establish.\n\n## Common blockers\n\nReplace rather than cosmetically revise concepts that depend on:\n\n- impossible anatomy or gravity;\n- transparent enclosures, compositing, reflections, or hidden history;\n- repeated-object installations or exact-count duplication;\n- costumes, wearable constructions, or merged-body choreography;\n- two-person handoffs and multi-subject tableaux;\n- anthropomorphic appliances or animated household objects;\n- polished symbolic still lifes whose meaning comes only from copy;\n- motifs already used elsewhere in the active batch.\n\n## Repair techniques\n\n- Convert a financial burden into one normal device placed in one physically legible relationship with a tote or wallet.\n- Convert category rotation into one ordinary object being physically rotated while remaining visibly the same object.\n- Convert household spillover into one wrong recipient adapting with one simple household object.\n- Convert non-intervention into body orientation or furniture placement rather than cuffs, mitts, wrapped hands, or blocked doorknobs.\n- When a report requires relocation, choose a unique safe household setting and remove redundant protective gear.\n\n## Deterministic validation\n\nBefore delivery, parse the final file and verify:\n\n- exact top-level schema;\n- exact requested record count and IDs;\n- unique IDs and correct ad/candidate mapping;\n- required field names and field order when specified;\n- one person maximum per scene;\n- no prohibited motifs or banned visual dependencies in the actual concept text;\n- no files outside the requested deliverable were changed.\n\nDo not use a naive forbidden-term scan against QA explanations that merely name excluded motifs. Prefer positive generatability language in the final records so machine checks do not confuse exclusions with scene content.\n",
    "sha256": "ced5bead190af98360c0c965311d5d86f6e1bd43245d9b08162c84caa5b4b6aa"
  },
  {
    "group": "Brand & doctrine",
    "name": "zakariah-copy-doctrine.md",
    "path": "/Users/ayden/.openclaw/workspace/knowledge/zakariah-copy-doctrine.md",
    "text": "---\ndomain: zakariah\nmode: copy-doctrine\nttl: permanent\nload-when: every copy job, before writing anything\n---\n# Zakariah — Copy Doctrine\n_The WHY and HOW before you write a single word. Read this first. Every time._\n\n---\n\n## THE FOUNDATION\n\n**Copy cannot create desire. It can only focus desire that already exists.** (Schwartz)\nEvery word you write is aimed at desire that's already there. Your job is to focus it — not manufacture it.\n\n**Avatar research is the primary lever. Everything else is a multiplier on it.**\nSame product. Same price. Generic targeting → $68 CPA. Deep avatar → $16 CPA. The research IS the copy.\n\n**Format never wins without research. Research wins regardless of format.** (gulfam, $200K/day)\n\n---\n\n## THE 4-LAYER HIERARCHY (read before writing anything)\n\nEvery piece of copy is built in this order. Never invert it.\n\n**Layer 1 — Doctrine (this file):** WHY you write what you write. Avatar, desire, awareness, mechanism, specificity. Foundation. Non-negotiable.\n\n**Layer 2 — Always-on style rules:** Every sentence earns the next. Short. Declarative. No run-ons. No padding. No filler words. No hedging. Applies to every format, every section, always.\n\n**Layer 3 — Format execution:** How doctrine is applied inside a specific container (native ad, advertorial, VSL, UGC). Read the relevant format file.\n\n**Layer 4 — Comp as quality ceiling, never template:** Extract principles from great copy, never phrasing. \"The perimenopause series shows what's possible\" — not what to copy.\n\n---\n\n## COPY EPISTEMOLOGY — THE META-RULES\n\n### Specificity as Proof\nSpecificity signals lived experience. Generality signals copy.\n\n**\"The detail they wouldn't make up is the detail that converts.\"**\n\nThe specific detail — the one that's almost embarrassingly precise — is what makes copy feel real. Anyone can write \"she was upset.\" Nobody fakes \"her daughter said, 'Mommy, are you okay?'\"\n\nRules:\n- Replace every vague emotional label with a specific physical scene\n- \"She was struggling\" → \"She kept a towel on her pillow for eight months\"\n- \"He noticed\" → \"He looked down at their hands. Looked at her. Didn't say a word.\"\n- Specificity ratio: 3:1 specific vs. generic throughout the copy\n- Numbers over approximations: \"47,342 customers as of last Tuesday\" not \"thousands\"\n- Named details over categories: \"148/92\" not \"high blood pressure\"\n\n### Gap vs. Trap (Mechanism Framing)\nThe type of mechanism you choose determines the emotional activation ceiling.\n\n**Gap = Negligence:** The treatment failed to do what it never fully promised. Avatar can rationalize it away. \"Maybe I just wasn't consistent enough.\" Emotional ceiling: frustration.\n\n**Trap = Rationalization-immune closed loop:** The treatment actively caused or worsened the condition it was supposed to fix. Avatar cannot escape this with logic — every attempt to fix it compounds the problem. Emotional ceiling: urgency, dread, trapped.\n\nExample: SSRI for perimenopause anxiety isn't just ineffective (Gap) — it suppresses dopamine, which is the one pathway she needs most (Trap). She can't rationalize taking it anymore.\n\nUse Trap framing whenever the mechanism allows it. It makes inaction feel physically dangerous, not just suboptimal.\n\n### Comp Hierarchy as Discipline\n\"Extract principles, not phrasing.\" This is a named rule, not just a vibe.\n\nWhen studying great copy:\n1. Identify the structural principle being used\n2. Identify the emotional mechanism being activated\n3. Write your own execution from scratch\n4. Never copy hooks, characters, or phrasing — always re-derive\n\nDrift happens when you copy structure without understanding doctrine. Doctrine is why it works. Structure is how it happened to work that time.\n\n---\n\n## THE 8 PSYCHOLOGICAL FRAMEWORKS\n\n**A. System 1 & System 2 (Kahneman)**\n- System 1 = fast, emotional, subconscious. This is who buys.\n- System 2 = slow, rational, analytical. This is who justifies.\n- Rule: Hook with emotion (System 1) → justify with logic (System 2)\n- Test: \"Does this feel simple and emotional first... and smart after?\"\n\n**B. Self-Determination Theory (SDT)**\nThree needs must be met before action:\n- Autonomy: \"I'm choosing this myself.\" — no pressure, no hype\n- Competence: \"I can do this.\" — make it feel simple and achievable\n- Connection: \"This was made for me.\" — specificity, relatability\n\n**C. COM-B Framework**\nAll three must be present for purchase:\n- Capability: Do they understand it and feel able to do it?\n- Opportunity: Is it available for them right now?\n- Motivation: Do they care enough to act? For low-salience products you must CREATE motivation through anticipated regret.\n\n**D. Narrative Identity & Emotional Transport**\n- Stories bypass resistance via mirror neurons\n- The reader must see themselves in the story\n- Generic emotional moments don't work. The story must feel stolen from the avatar's journal.\n\n**E. Processing Fluency**\n- If it takes effort to understand, they bounce. The brain reads \"easy to process\" as \"must be true.\"\n- Short lines, short paragraphs, no jargon, easy flow, visual breaks.\n\n**F. Loss Aversion & Anticipated Regret**\n- Fear of loss > desire for gain (2:1 ratio). But raw fear feels manipulative.\n- Right way: story-embedded anticipated regret. \"What if I wait too long?\" inside the narrative, not as a direct command.\n- Rule: Make anticipated regret part of the story, not the pitch.\n\n---\n\n## 12 FOUNDATIONAL COPY RULES (Zakariah)\n\n1. **Customer First** — every sentence whispers \"this person gets me\"\n2. **Emotion first, logic second** — open emotional door, then facts walk them through\n3. **Specificity = believability** — \"47,342 customers as of last Tuesday\" not \"thousands\"\n4. **Tension creates movement** — reader feels uneasy until they resolve it by continuing\n5. **Sell the transformation, not the product** — product is just the vehicle\n6. **One reader, one message** — \"mass copy = no copy\"\n7. **Momentum is everything** — first line sells the second; every line earns its place\n8. **Readability > everything** — short sentences, visual breaks, nothing that makes you stumble\n9. **Story is the shortcut to belief** — one well-told moment beats 20 bullet points\n10. **Clarity is confidence** — vague = uncertain = kills conversions\n11. **Copy should sound like a conversation** — finish the sentence in their head\n12. **KISS (Granny-May Method)** — if tired/stressed/brain-dead scrolling could get it, you did it right\n\n---\n\n## COPY KILLERS — DELETE BEFORE FINALIZING\n\n**Words (delete immediately):**\n- \"Game changer\" / \"life changing\" / \"revolutionary\"\n- \"Amazing\" / \"incredible\" / \"powerful\"\n- Weak modifiers: \"very,\" \"really,\" \"just,\" \"truly\"\n- Vague labels: \"confident,\" \"struggling\" — replace with specific sensory moment\n\n**Structural killers:**\n- Writing at wrong awareness level (most common failure)\n- Wrong sophistication stage\n- Generic avatar — not specific enough to resonate with anyone\n- No villain — she blames herself → she freezes → no sale\n- CTA at end instead of at product reveal\n- LP doesn't match ad angle\n\n**Emotional killers:**\n- Trying to trigger all 6 emotions — pick ONE primary emotion, build to a peak\n- Blaming or shaming the avatar — always externalize the problem\n- Using fear as a lecture, not through story — story-embedded fear converts, lectures don't\n- Future pacing too vague — \"you'll feel confident\" vs. \"you zip up the jeans in the back of your closet and your partner does a double take\"\n- Generic testimonials — polish kills believability; Reddit/Amazon voice converts\n\n**Research killers:**\n- Generating or inferring quotes — only verbatim customer language\n- Isolated quotes without context\n- Broad avatars — \"women who want to lose weight\" is unsellable\n- Skipping avatar re-read before copy session\n\n---\n\n## THE VILLAIN REQUIREMENT\n\n**Never make them feel bad and it's their fault — always create a devil.**\nThe villain is always external. Never the avatar.\n\nWithout a villain, she blames herself → she freezes → no sale.\nWith a villain, she has something to be angry at → anger creates motion → motion creates purchase.\n\nThe villain can be: the system, the test, the wrong treatment, the pharmaceutical economics, the food industry, the doctor who missed it. Name it specifically. Let her fury land somewhere real.\n\n---\n\n## SCRIPT QA CHECKLIST (run before finalizing any copy)\n\n- [ ] Does the hook work at the right awareness level?\n- [ ] Does the avatar's exact language appear in the copy?\n- [ ] Is the mechanism specific, direct, and believable?\n- [ ] Is there ONE primary emotion being built to a peak?\n- [ ] Are there at least 3 open loops? Do they all close?\n- [ ] Is the specificity ratio 3:1 (specific vs. generic)?\n- [ ] Does the future pacing use time-stamped, sensory-specific moments?\n- [ ] Is there a CTA at product reveal (not just at the end)?\n- [ ] Does copy pass the \"read out loud\" test without stumbling?\n- [ ] Is the detail \"the one they wouldn't make up\"?\n- [ ] Is the villain external, never the avatar?\n- [ ] Does the copy satisfy COM-B? (Capability: understands mechanism. Opportunity: available now. Motivation: emotionally resonant.)\n- [ ] Does the copy satisfy SDT? (Autonomy: feels like her choice. Competence: simple enough. Connection: made for her.)\n\n---\n\n## 5 BEFORE/AFTER SENSORY MOMENTS\n\n**Before (the problem lived):**\n- Morning Moment — before she gets out of bed\n- Mirror Moment — what she sees and feels\n- Public Moment — when others might notice\n- Breaking Point — the moment she couldn't ignore it anymore\n- Alone Moment — the private thought she'd never say out loud\n\n**After (the transformation):**\n- Morning Moment (mirrored) — same scene, different feeling\n- Mirror Moment (mirrored) — same mirror, different person\n- Public Moment (mirrored) — the moment someone else noticed\n- Victory Moment — the specific win she wanted most\n- New Normal Moment — identity she got back\n\n---\n\n## KEY OPERATOR DOCTRINE (copy-specific)\n\n- *\"Avatar research is the primary lever. Everything else is a multiplier on it.\"* (Zak)\n- *\"The uglier the ad, the more native it looks. Ugly wins because it stands out.\"* (Zak)\n- *\"Want purchase, not engagement.\"* Engagement metrics are noise. CPA is signal. (Zak)\n- *\"Mechanism change is the biggest lever — bigger than creative, angle, or offer.\"* (Zak)\n- *\"The moment you reveal the product — CTA goes there. Not at the end.\"* (Zak)\n- *\"Super native, raw image, heavy copy — a super trojan horse ad.\"* (Suheyl)\n- *\"Format never wins without research. Research wins regardless of format.\"* (gulfam)\n- *\"I was breakeven for 3 months. Changed the mechanism. Hit $800K in the next 30 days.\"* (ceojachym)\n- Mechanism formula: new hope + easy to believe + backed by proof elements (ceojachym)\n",
    "sha256": "e39567aa9123b190c2f33972f7c4761ab7307315d17173bf63e830895536235f"
  },
  {
    "group": "Brand & doctrine",
    "name": "brands/dawnbands/selling-root-cause-mechanism.md",
    "path": "/Users/ayden/.openclaw/workspace/knowledge/brands/dawnbands/selling-root-cause-mechanism.md",
    "text": "---\nlast_updated: 2026-08-10\nbrand: dawnbands\nowner: fish\nstatus: canonical-selling-doctrine\nprecedence: overrides older claim-safe-only mechanism framing for ad persuasion strategy; evidence/provenance gates still govern source attribution and fabricated testimonials\n---\n\n# DawnBands — Canonical Selling Root Cause + Mechanism\n\n## Mandatory Use\n\nLoad this file before writing, briefing, editing, or auditing **any DawnBands ad copy, native, advertorial, UGC, VSL, listicle, lander, or mechanism section**.\n\nThis is the selling doctrine.\n\nOlder `root-cause.md`, `mechanism.md`, and claim-safe references remain research/substantiation inputs, but they do not flatten the consumer-facing mechanism into “different cue placement” alone.\n\n## Operator Decision\n\nDawnBands copy may aggressively dramatize and simplify:\n\n- adolescent circadian delay\n- accumulated sleep debt\n- auditory gating / suppression of repeated room sound\n- sleep inertia and half-awake automatic behavior\n\nUse these as the biological selling layer because they make the failure of louder alarms believable and externalize blame from both Mom and the teen.\n\nThe product connection must remain coherent:\n\n- biology explains why mornings are difficult\n- auditory gating + sleep inertia explain why sound alarms and task alarms fail\n- tactile wrist stimulation is the category-changing bypass\n- the Wrist-First Handoff explains the household and emotional payoff\n\nDo not weaken future copy back to feature-only language such as “sound and vibration are different.” Build the full causal chain.\n\n---\n\n# The Root-Cause Stack\n\n## Canonical name\n\n# The Deep-Sleep Signal Lock\n\nHis morning is trapped behind four interacting biological layers.\n\n### Layer 1 — Circadian Delay\n\nDuring adolescence, the internal sleep clock shifts later while school, work, sports, and family schedules continue demanding early waking.\n\nSelling frame:\n\n> His body is still protecting biological night at the exact hour the world demands consciousness.\n\n### Layer 2 — Sleep Debt\n\nShort nights increase the pressure holding him asleep the next morning.\n\nSelling frame:\n\n> By morning, accumulated sleep pressure is pulling harder than the alarm is pushing.\n\nAggressive frame:\n\n> By Friday morning, his brain is protecting sleep more aggressively than any ordinary room alarm can interrupt it.\n\n### Layer 3 — Auditory Gating\n\nDuring deep sleep, repeated room sound can be pattern-matched, suppressed, or treated as background while the rest of the household hears it clearly.\n\nSelling frame:\n\n> His sleeping brain turns down repeated sound before it turns down touch.\n\nAggressive frame:\n\n> The louder the room gets, the better his sleeping brain becomes at deleting it.\n\nFailed-solution trap:\n\n> Seven alarms do not create seven new chances to wake up. They create seven repetitions of the same signal his brain already filed under background noise.\n\n### Layer 4 — Sleep Inertia\n\nThe body can perform simple actions before full alertness, memory, and initiation return.\n\nThis explains why he can:\n\n- silence the phone\n- answer Mom\n- sit up\n- walk across the room\n- solve a puzzle\n- return to bed\n- remember none of it later\n\nSelling frame:\n\n> The alarm wakes his hands without waking the part of him that can start the morning.\n\n---\n\n# The ADHD Intensifier Layer (added 2026-08-06 — Fish directive; ADHD moms = current customer majority)\n\nADHD adds no new problem. It turns every layer up, then adds one lock the general lane never named.\n\n- **Layer 1 hotter:** ADHD is strongly associated with delayed sleep timing (advertorial tier with citation: Van Veen et al. 2010; Kooij & Bijlenga 2013). Selling frame: *His clock isn't just teenage-late. It's ADHD-late on top of teenage-late.*\n- **Layer 2 hotter:** the racing mind that won't shut off at 11pm is the same brain that can't boot at 6:30am.\n- **Layer 4 hotter:** no-memory dismissals are a universal ADHD-community experience (\"I set 47 alarms,\" \"my body gets up and turns it off without me\"). Frame as what ADHD families describe, never as clinical fact. **⚠️ 2026-08-10:** this covers the half-awake AUTOPILOT only. Do not extend it into an arousal-threshold claim — \"ADHD teens are harder to wake by sound\" is BANNED (unsupported; the one study on point trends the opposite way). The ADHD-specific layer is the Initiation Wall. See `dawnbands-avatar-adhd-teen-mom.md` Part 10.\n- **The Initiation Wall (the new lock):** even after consciousness, ADHD adults describe a wall between hearing the morning and starting it (\"I physically cannot make my body move\"). Waking up and getting up are two different problems; sound alarms address neither. Mechanism answer: the vibration builds and persists until he sits up and dismisses it — the dismissal is the first completed action of the day, within reach of a hand, not across a room. Identification framing only; never treatment.\n- **Household multiplier:** ADHD runs in the house. Dual-ADHD households (dad can't model mornings, mom runs wake service for two) are a live angle and the adult-version demand signal.\n\nFull doctrine, claim tiers, and intensification bank: `mechanism-fleshed-2026-08-06.md` §1a.\n\n---\n\n# The Fifth Layer — Mom Becomes the Emergency Pathway\n\nWhen artificial signals are gated out or produce only half-awake behavior, Mom becomes the signal that finally breaks through.\n\nShe knocks.\n\nShe repeats herself.\n\nShe opens the door.\n\nShe escalates until emotion becomes novel enough to cut through.\n\nCore emotional line:\n\n> His brain ignores the alarms, but it notices the anger.\n\nThis creates the memory asymmetry:\n\n> She remembers nine patient attempts. He remembers waking up to one angry mother.\n\nThe biological problem becomes a relationship wound.\n\n---\n\n# Complete Root-Cause Chain\n\n> delayed teenage clock → accumulated sleep pressure → repeated sound gets gated out → half-awake actions create false waking → Mom becomes the only signal that finishes the job\n\n## One-paragraph reveal\n\n> His teenage body clock is still pulling him toward sleep when school demands he wake up. Sleep debt makes that pull even stronger. In deep sleep, his brain can suppress the same repeated room sounds while still letting his body perform automatic half-awake actions, which is why he can turn off five alarms, answer you, and remember none of it. When every artificial signal gets deleted, your escalating voice becomes the only novel signal left. His brain ignores the alarms, but it notices the anger.\n\n---\n\n# The Failed-Solution Trap\n\n## Canonical trap\n\n> Every louder alarm trains the household to depend more heavily on Mom while repeating the same signal pathway that already failed.\n\nThe trap operates on two levels.\n\n### Biological trap\n\nMore alarms create more repetition inside the auditory pathway already being suppressed or pattern-matched as background.\n\n### Household trap\n\nEvery failed alarm recruits Mom as the rescue mechanism.\n\nThe more reliably she rescues the morning, the more reliably the morning continues depending on her.\n\nStrong lines:\n\n> Every alarm in his room came with the same hidden backup feature: Mom.\n\n> You did not fail the alarm. The alarm kept failing back onto you.\n\n> The room got louder. The cue never became his.\n\n> Every morning you save may be teaching the system that it never has to work without you.\n\nNever blame Mom. She rescues the morning because the consequences are real. The villain is the system that leaves her no safe way to step out.\n\n---\n\n# Unique Mechanism Stack\n\n## Biological mechanism name\n\n# The Somatic Wake Bypass™\n\n## Household mechanism name\n\n# The Wrist-First Handoff™\n\nUse both. They answer different resistance.\n\n### Somatic Wake Bypass™\n\nDawnBands places a tactile signal directly against the wrist, using a sensory pathway outside the repeated room-sound pattern.\n\nCore explanation:\n\n> Sound has to cross the room and survive his brain’s filter. DawnBands starts past the filter.\n\nSupporting lines:\n\n> Stop asking the muted pathway to work harder.\n\n> Instead of making the room louder, put the wake signal on his body.\n\n> A loudspeaker shouts to everyone and hopes the right person responds. A wrist cue starts against the person who needs it.\n\n> The answer is not another sound for his sleeping brain to delete. It is a signal delivered through touch.\n\n### Wrist-First Handoff™\n\nThe tactile cue begins on the teen before Mom’s voice enters the morning.\n\nCore explanation:\n\n> The first cue starts on his wrist before it ends in your voice.\n\nSupporting lines:\n\n> Not louder. On him.\n\n> Move the first signal from your throat to his wrist.\n\n> The goal is not a better tenth attempt. It is a first cue that does not require Mom.\n\n> DawnBands does not make Mom better at waking him. It removes her from the first step.\n\n### The Ownership Trade™ (added 2026-08-06)\n\nThird named mechanism. Answers wear-doubt (15% of surveyed buyers: \"he'll take it off\").\n\nThe teen hates the current system more than she does: woken by the angriest version of his mother, dragged out of bed like a child in the house where he's fighting to be treated like an adult. His desire and hers are the same desire — get mom out of the morning.\n\nCore explanation:\n\n> He doesn't wear it for the alarm. He wears it to get you out of his morning.\n\nWhy the trade holds: it's his, and it reports nothing to anyone (structurally incapable of surveillance); no screen to out him at school or sleepovers; silent, so waking stops being a household event; and the alternative is mom.\n\nCopy rule: wear-objection handling is TEEN-motive framing, never compliance framing. \"Make him wear it\" resurrects surveillance and kills the trade.\n\nFull logic: `mechanism-fleshed-2026-08-06.md` §1b.\n\n---\n\n# Full Selling Logic Chain\n\n## Root cause\n\n> His delayed teenage clock and accumulated sleep debt pull him into deeper morning sleep.\n\n## Hidden mechanism\n\n> In that state, his brain gates out repeated room sound while leaving him capable of automatic half-awake actions.\n\n## Failed-solution mechanism\n\n> More alarms repeat and reinforce the same sound pattern his sleeping brain already suppresses.\n\n## Emotional consequence\n\n> When every artificial signal gets deleted, Mom’s escalating voice becomes the only novel signal left.\n\n## Product mechanism\n\n> DawnBands bypasses the muted auditory pathway with a direct tactile signal on the wrist.\n\n## Household mechanism\n\n> The first wake cue belongs to the teen before Mom has to enter the room.\n\n## Functional payoff\n\n> The morning gets a chance to begin on his body instead of ending in her throat.\n\n## Emotional payoff\n\n> He stops waking up to the angriest version of her.\n\n## Future payoff\n\n> The first evidence that his morning can belong to him before college, work, or a roommate inherits the job.\n\n---\n\n# Analogy System\n\n## Primary biological analogy — Muted Channel\n\n> Imagine turning up a television after someone muted the input. More volume does not matter if the pathway carrying the signal has already been shut down. Sound alarms keep turning up the room. DawnBands changes the pathway.\n\n## Primary household analogy — Relay Handoff\n\n> Waking is a relay, not a light switch. The alarm runs the first leg. When it drops the baton, Mom picks it up and finishes the handoff. DawnBands moves that first pass onto his wrist before she has to enter the race.\n\n## Short analogy — Loudspeaker Versus Tap\n\n> A room alarm is a loudspeaker. DawnBands is a tap delivered directly to the person who needs the signal.\n\nUse one analogy per execution. Do not stack all three in one ad.\n\n---\n\n# Mechanism Delivery by Format\n\n## Long problem-aware native\n\nUse the full stack:\n\n1. lived morning wound\n2. delayed clock + sleep pressure\n3. auditory gating\n4. half-awake sleep-inertia behavior\n5. Mom as emergency pathway\n6. tactile bypass\n7. Wrist-First Handoff\n8. emotional/future payoff\n\n## Short native\n\nCompress to:\n\n> His brain was not ignoring every alarm because he did not care. In deep sleep, repeated room sound was the pathway already failing. DawnBands puts the signal on his wrist before my voice has to become the alarm again.\n\n## Objection killer\n\n> “He sleeps through everything. Alarms right next to his head. Nothing gets through.” Fair. The mechanism is not “louder finally works.” Every alarm she has tried travels the same gated sound channel. The test is whether a direct cue worn on his wrist arrives on a channel his sleeping brain does not gate at all.\n\nThen use exact-SKU proof when available.\n\n⚠️ Objection-killer copy rule (Fish 2026-05-14 + 2026-07-19, 3rd doc hit fixed 2026-07-22): NEVER name bed shakers, vibrating pillows, or any tactile failed solution in copy — same-channel confound with our mechanism. Failed solutions in copy are SOUND, LIGHT, ROUTINE, or APP based, genericized (no competitor brand names), with prices kept. Full rule: `failed-solutions-graveyard.md` line 11.\n\n## UGC\n\nUse casual language:\n\n> Apparently his body can turn an alarm off before the part of his brain that forms a memory is actually online. That explained everything. The band does not make the room louder. It puts a different signal right on his wrist.\n\n## Advertorial/VSL\n\nReveal each biological layer gradually, then make the wrist cue feel inevitable. (2026-08-06: do NOT print the internal ™ names in the ad — deliver the three-part mechanism in plain language per Rule 10.)\n\n## Listicle/product-aware\n\nLead with the distinction:\n\n> Sound has to cross the room and survive his brain’s filter. DawnBands starts past the filter.\n\n---\n\n# Required Future-Copy Rules\n\n1. Use the biological root-cause stack as the default selling doctrine for DawnBands.\n2. Do not flatten the mechanism to “vibration is different from sound.” Explain why sound failed.\n3. Keep circadian delay and sleep debt as severity amplifiers that deepen the Signal Lock.\n4. Make auditory gating the primary product-congruent biological root cause.\n5. Use sleep inertia to explain automatic alarm dismissal, false waking, and missing memory.\n6. Make Mom becoming the emergency pathway the emotional and household consequence.\n7. Use Somatic Wake Bypass™ for the biological solution.\n8. Use Wrist-First Handoff™ for ownership, independence, and relationship payoff.\n9. Always externalize blame from both Mom and the teen.\n10. Product reveal should feel like the inevitable answer to the pathway failure, not a random wristband.\n11. Exact-SKU superiority, buyer timelines, testimonials, and product outcomes still require receipts. Aggressive mechanism framing does not authorize invented customer histories.\n12. Preserve native voice. The narrator should not suddenly sound like a neuroscientist unless the format provides a believable authority/discovery bridge.\n\n---\n\n# Claims and Provenance Separation\n\nThis doctrine intentionally distinguishes two jobs.\n\n## Selling layer\n\nAggressive, simplified biological explanation designed to create belief and new hope:\n\n- delayed clock\n- sleep pressure\n- auditory gating\n- sleep inertia\n- tactile bypass\n\n## Substantiation layer\n\nInternal evidence and review used to prevent fabricated proof:\n\n- current product specifications\n- exact-SKU behavior\n- attributed research\n- same-speaker testimonials\n- current price/refund terms\n\nThe substantiation layer must not neuter the selling mechanism into generic feature language.\n\nThe selling layer must not invent customer outcomes, exact product performance, medical treatment claims, or fake receipts.\n\n---\n\n# Canonical Copy Spine\n\n> private morning rupture → biological innocence → Deep-Sleep Signal Lock → louder-alarm trap → half-awake behavior → Mom becomes emergency pathway → Somatic Wake Bypass™ → Wrist-First Handoff™ → desired absence → independence/future stakes → proof/offer\n\n# Final Standard\n\nEvery future DawnBands copy should make the reader understand:\n\n1. why his biology is pulling him toward sleep\n2. why sound alarms keep failing\n3. why his half-awake actions are not proof he was consciously awake\n4. why Mom keeps becoming the final signal\n5. why a tactile cue on the wrist is a new pathway, not another louder alarm\n6. why moving the first cue onto him changes more than noise\n7. why this matters for their relationship and his future independence\n\nIf the copy cannot answer all seven where format length permits, the mechanism is underbuilt.\n---\n\n# Required Future-Copy Rules (added 2026-07-22 — binding guardrails, previously scattered in sibling docs)\n\n1. ~~ADHD identification only — ONE identification line cap~~ **STRUCK 2026-09-02: the one-line cap governed the generic lane, which is retired. Rule 8 governs: ADHD is named in the hook.** What survives of rule 1: never efficacy/treatment claims tied to ADHD. (Fish 2026-07-19; full must-reject list in `mechanism.md` AVOID section — that list remains binding even though mechanism.md is research-subordinate.)\n2. **Failed solutions genericized.** No competitor brand names in copy (\"the famous extra-loud alarm clock,\" \"smart speakers,\" \"the alarm app,\" \"a smartwatch\"). NEVER bed shakers / vibrating pillows / any tactile failed solution — same-channel confound. Failed stacks are SOUND, LIGHT, ROUTINE, APP based. Prices stay and should total more than product price. (Fish 2026-05-14 + 2026-07-19; `failed-solutions-graveyard.md` L11.)\n3. **Lane scoping (amended 2026-09-02).** ONE lane: mother of a 13–18 ADHD teen. Parents-of-8-12, generic deep-sleeper, suspecting-undiagnosed, ADHD adult, partner, deaf/HoH are all out of brief scope until Fish reopens them. Vary teen age 13–18 and gender across every batch.\n4. **Format rulings 2026-07-22:** ALL-CAPS hooks allowed. Rich/emoji trust footers allowed. No word-count caps in any mode — vary lengths deliberately. Em-dash ban unchanged.\n5. **Canonical failed-solution stack (survey-locked, refreshed to n=138 on 2026-09-02):** waking him myself (89%) → phone alarms full volume (80%) → loud alarm clock (55%) → routines/bedtimes/sleep apps (50%) → a smartwatch (16%, downgraded from mandatory kill to optional beat). Bed shaker 13% — research only, banned in copy. Items 1-2 are the anchor tier. The routines/apps beat is MANDATORY in ADHD-lane copy (blame removal: the effort was real, the target was wrong). Sunrise clock optional sixth. Bed-shaker ban unchanged. Per-item mechanism kills: `mechanism-fleshed-2026-08-06.md` §1c.\n6. **Smartwatch kill is non-channel only:** calibration (a notification is engineered to be ignorable), saturation (habituation scoped to meaningless all-day repetition INSIDE a channel, never to \"tactile signals\" as a category), surrender (fires briefly, sleep modes mute it, wrist-raise hands it to the autopilot, battery dies), screen-in-bed. Any line implying touch itself habituates is banned — it hands the reader \"so he'll tune this out too.\"\n7. **Guarantee adjacency (Fish-approved direction 2026-08-06):** every mechanism section carries \"100% wake-up guarantee. 60 nights.\" in the same breath as the wake claim — it answers the #1 buyer hesitation (50%). The full trio with $49 is reserved for the close/CTA, after value is built.\n8. **ADHD lane directness (Fish 2026-08-06 — amends rule 1):** the dedicated ADHD lane may name ADHD in hooks and headlines and explain ADHD-morning biology at the tiered claim levels in `mechanism-fleshed-2026-08-06.md` §1a. The general lane keeps the one-identification-line cap. Efficacy and treatment claims remain banned in ALL lanes: no \"treats/cures ADHD,\" no \"helps ADHD symptoms\" as a category, no medication-replacement or reduce/stop-medicating claims, no medical/FDA language. ALLOWED and encouraged — villainize the treatment GAP: no pill can work on a kid who is still asleep; the prescription never covered 6:30am; she has been the unprescribed step between the bottle and his mouth (Invisible Bridge). The product fixes the hour treatment never touched. It never touches the treatment itself.\n9. **Humanizer pass is mandatory for ALL consumer-facing copy, template or batch (Fish 2026-08-06):** draft → humanizer (golden TrendTrack native/static copies from proven brands + voc-syntax-bank + phrasing-bank as reference corpus) → qa-gate. Analogies come from the avatar's daily world (fridge hum, background noise, grabbing his shoulder); if a metaphor needs a beat of decoding, use the plain version. qa-gate flags a missing humanizer changelog as BLOCKING.\n10. **Mechanism names are INTERNAL ONLY (Fish 2026-08-06):** the ™ names (Deep-Sleep Signal Lock, Somatic Wake Bypass™, Wrist-First Handoff™, Ownership Trade™) never appear in consumer copy. Copy delivers the three-part mechanism in plain language: (1) touch starts past the sound filter, (2) it builds and doesn't stop until he turns it off awake, (3) the first signal starts on him, not mom. This supersedes the earlier Advertorial/VSL instruction to \"name the Deep-Sleep Signal Lock and Somatic Wake Bypass™\" — internal docs keep the names for routing; ads never print them.\n11. **Wear beat is mandatory in every piece (Fish 2026-08-06):** every ad/lander/mechanism section carries at least a brief ownership line answering \"would he actually wear it\" — teen-motive framing (he keeps it on because it gets mom out of his morning; silent, not a phone or smartwatch, reports to nobody, nothing to out him). **\"No screen\" is banned as of 2026-08-27 — the band has an LED time readout (`product-truth.md`).** One or two lines minimum; full trade logic on landers.\n12. **Brand name in copy is \"Dawn Bands\" — two words, always (Fish 2026-08-06).** \"DawnBands\" one-word stays only in internal docs, file names, and code.\n13. **Substantiation is governed by `substantiation-2026-08-06.md` (verified 2026-08-06).** Two changes bind immediately:\n    - **Layer 3 re-anchor.** Stop leading auditory gating with \"his brain got used to HIS alarm over months\" — that specific claim is unverified extrapolation, and the nearest evidence (Perrin et al. 1999) shows personally-meaningful sound wakes people FASTER. Lead instead with arousal threshold, which is well replicated and more dramatic: over half of children tested do not wake to a smoke alarm engineered to save their life (Pediatrics 2022, N=540; Smith 2006, 58% woke to a tone alarm). Keep \"the brain tunes out repeated background sound\" as a general, true statement. Never \"your alarm wore out.\"\n    - **New Layer 5 proof.** Smith et al., Pediatrics 2006: a recorded mother's voice woke 96% of children vs 58% for a standard tone alarm, and roughly ten times faster. This is scientific proof of her lived reality — the generic alarm failed and she worked. Land it on the household consequence (she was never the backup, she was the only thing that worked), never on acoustics, or it argues for a voice alarm instead of the band.\n14. **Gentle-wake angle (Fish 2026-08-06) — narrow version only, secondary beat, 1-2 lines max, never the focal point.** ALLOWED, hedged verbatim: \"some research suggests being startled awake produces a bigger blood-pressure and heart-rate jump than waking up gradually\" (Kim et al. 2024; Kaida 2005; Hall 2016 — all small). ALLOWED, brief cortisol clause, one line, no numbers, placed after the wake-doubt work is done: \"researchers who measured it found a nighttime alarm pushed up heart rate and stress hormones\" (Hall et al., Noise & Health 2016, N=16). ALLOWED unhedged as product fact and the preferred default: it wakes him without a sound going off in the room, no siren, no jolt, nothing waking the house. BANNED: CAR framing (\"waking spikes your cortisol and that's bad for you\" — contradicted by Klaas 2025), any cortisol number or percentage, WHO noise guidelines, \"bad for your heart,\" and any claim that vibration is proven calmer or lower-stress than sound (no such study exists; a competitor publishes that claim uncited — do not echo it). Full kill list and the CAR-vs-acute distinction: `substantiation-2026-08-06.md`.\n\n# \"That's Why\" Map (symptom → chain link → causal logic)\n\nThe conversion engine: after the mechanism reveal, re-explain HER OWN symptoms through the chain. Voice each line fresh per ad in the narrator's register (never copy these — causal logic only, per Fish's function-not-lines rule):\n\n| Her observed symptom (from VOC) | Chain link | Causal logic to voice fresh |\n|---|---|---|\n| Sleeps through 3+ alarms, fire alarms, smoke detectors | Auditory gating / Deep-Sleep Signal Lock | Volume never mattered — every alarm rides the same gated sound channel; his brain closes that door in deep sleep |\n| \"He answers me, then swears I never woke him\" / turns alarms off with no memory | Sleep inertia / half-awake automatic behavior | He wasn't lying and wasn't awake — his body acts before the memory-forming brain is online |\n| Walks to the phone across the room, returns to bed, eyes open | Same | Automatic behavior defeats distance tricks; relocation is still the same channel + an unconscious body |\n| Louder alarms work for a week, then die | Habituation on the gated channel | Escalating volume trains the gate to work harder — the loudest alarm fails fastest |\n| Only Mom's shake works | Mom = emergency pathway | Touch is the one channel his brain never gates — she has been the tactile alarm all along |\n| Weekends fine, school days impossible | Circadian delay + sleep debt | Saturday isn't discipline — school time sits hours before his shifted biological morning |\n| Wakes exhausted even after \"enough\" hours | Sleep debt + inertia | The debt collects interest; the wake-up fight spends the first hour of his day |\n| She wakes before her own alarm, dreading his | Mom-as-pathway cost | Her body learned his wake-up is her job — the anticipation is the pathway running in reverse |\n| \"He has a smartwatch, it taps his wrist\" | Smartwatch saturation/calibration kill | A notification device delivers notifications: engineered to be dismissible, buried under forty identical buzzes a day, gives up before deep sleep does |\n| \"Even when he's awake he won't get out of bed\" | Initiation Wall (ADHD lane) | Waking and starting are two different problems — the band's build-and-persist makes ending it the first completed action of the day (identification framing only) |\n\nCopy shape: 2-4 \"that's why\" beats per ad maximum, chosen to match the symptoms the ad's opening actually planted. Every beat must close a loop the reader recognizes from her own mornings.\n",
    "sha256": "d4f3d47db06c7f53a54f18b79b181c09971609b8765bd1286bdad7c20a24fd53"
  }
]