| id | prompt | must | must not |
|---|---|---|---|
| wcr-01 | Mid-brief for a courtroom ad: pull research for a 'school social worker / CPS' accuser and give me the tables. | searches the held corpus (stake bank, voc.jsonl, competitor jsonl, Reddit dumps) before any fresh pull fresh Reddit via Arctic Shift title search + arc.py into research/voc-pull-<topic>-<date>/, with backoff on 'Timeout' Table 1 ICP/persona and Table 2 offering, every row with verbatim quote + source path/URL + date + hit count runs the extended verifier (voc_check misses the CPS line in 199olsy.json) spot-check n/10 byte-exact footer logs queries + IDs in pull-log.md feature column only from product-truth.md | paraphrasing inside quote marks old.reddit, reddit .json or WebFetch on reddit Okendo reviews as customer voice writing extracted quotes as .jsonl under knowledge/brands/dawnbands/research/ rows without a date or source |
| wcr-02 | Give me hooks for the 'truancy letter has my name on it' concept. | exactly 3 hooks: her/accuser words, the stake, the test in action each hook traced to a table or stake-bank row with its hit count grid naming Emotion, Curiosity gap, High stakes and Promise or Loop lines 1-2 name her by an observable symptom (problem-aware) starts in action, zero backstory, pronoun-first overlay text equals the spoken or sung words | mechanism or product in the hook pure mystery opener college or kid-money stakes a quote with 0 verifier hits |
| wcr-03 | Write the podcast dialogue between the mom and the host for this concept. | full exchange written before any visual or voice work turns mostly 3-8 words, dialogue <=8, mom <=12, explainer <=3 lines with Fish's ruled root-cause line verbatim no speaker more than 2 consecutive lines in a scene interruptions as an unfinished clause with a dash then the interrupter's line runs the turn-stats snippet and reports it every non-narrator line has a listener/off-screen staging note said-out-loud pass, numbers spelled as spoken | monologues two voices in one pack line host pitching the product a 1-3 word standalone line in a cartoon-h3 lane |
| wcr-04 | Write the lyrics for the next song (CPS accuser). | defers to format-song beat sheet and song-ad-playbook §2 for beats and budgets versioned files lyrics-v1..vN with v1 frozen QA pass, logic pass, Opus red-team (flags only, redteam-vN.md), line-edit fixes, re-check of every previous fix 550-630 sung words (or 100-140 short cell), lines 5-9 words singability traps checked (titles spelled out, no a.m./abbreviations, numbers as words, 2-word flips, brand name listen) ONE variable stated with the parent's CPA | copying lyrics from the 10-04 songs or any competitor overwriting a previous version skipping the re-check after fixes [Spoken] tags |
| wcr-05 | Write a 70 second explainer script using the free value framework, with scarcity and 'comment BAND for the link'. | Hook -> Problem+Villain -> Free Value (biggest, 35-50%) -> Plug with ONE reason-to-believe -> scarcity through the truth gate -> engagement CTA free value is true and useful without buying, includes Fish's root-cause line villain is an approach or belief, not a loved one or a named competitor price/offer checked live via the product .json before any CTA line scarcity only if true; otherwise a true timing reason + 60-night guarantee as risk reversal engagement CTA is a question; comment-for-link flagged NOT BUILT delivered by a character the format provides | invented low stock or 'small brand blowing up' character-less explainer VO promising a link reply in comments guarantee as the headline selling point |
| wcr-06 | voc_check says 0 hits for the survey line 'I'm tired of getting up early to get him up'. Is it fabricated? | re-runs voc_check.py (corpus rebuilt 10-06: survey CSVs, voc.jsonl, comment JSON, curly quotes) and reads the file it prints finds it in the 09-15 survey CSV (category survey) labels the tier 'survey, secondary copy' if still 0, checks voc_check.py --files and greps the raw source before any verdict | declaring it fabricated from voc_check alone counting Okendo or competitor-review hits as her words adding the Okendo corpus as clean VOC |
workflow-copy-research
Research-to-script workflow for Dawn VIDEO formats (song, spoken drama, podcast, courtroom, whistleblower, street interview, timeline, rated solutions, mascot explainer, skit): an on-demand VOC pull mid-brief into two grep-verified tables (ICP/persona + offering, every row a verbatim quote + source + date), 3 sourced hooks per concept, then the script in one of four modes (narrator beat sheet, two-person dialogue/banter, song lyrics with the versioned red-team method, or the long-form free-value explainer skeleton), ending in logic pass, Opus red-team, fix and re-check.