Photoreal vox-pop / street interview ad, 30-60 s. An off-camera interviewer asks moms one question ("Who gets your kid out of bed on school mornings?"), 6-8 different moms each answer in ONE 4-8 s take with real VOC lines, and the band is the answer, shown in a product-gated b-roll insert (real footage as fallback).

Lane: — · not built yet: 2 (see below) · source: /Users/ayden/.openclaw/workspace/skills/format-street-interview

Street interview: beat sheet, take math, line rules

Runtime target 30-60 s (≈ 45 s: 2 interviewer lines ≈ 3 s, 7-8 answers ≈ 33-38 s, end card (optional, --end-card; off by default per Fish 10-06) overlays the last 2.5-3 s). Percentages are of runtime. Every answer row = one mom = one take = one lines[] entry.

Take math (Gemini Omni via ugc-omni)

Bucket	Words that fit at 2.7 wps (omni.py `DEFAULT_WPS`; `plan_line` needs words/wps + 0.5 s ≤ bucket; `lock` re-measures)	Use for
4 s ($0.315)	≤9	hook answers, chorus counts, payoff
6 s ($0.42)	10-14	stakes answers
8 s ($0.525)	15-20	the answer mom (root cause + mechanism + product)
10 s ($0.63)	21-25	only if the answer line can't be split; never two takes of one mom
`plan_line` picks the smallest bucket for the measured wps and pads filler, and `verify` trims the filler off. Re-check the table after `lock --approve` stores the real wps.

Beats

#	Beat	%	Moms	Job	Line rules
1	Question + hook answer	0-12	Interviewer (off camera) + mom 1	Name her symptom, stop the scroll	Question ≤8 words, names the symptom (her as the alarm, his alarms failing). Shown as on-screen text from frame 0. Mom 1 starts speaking ≤1.5 s in, with the highest-emotion VOC line; 3 variants = the 3 hooks
2	Chorus	12-45	moms 2-4	Prove every mom here is the alarm	Counts, times, trips, who wakes up instead of him. Concrete numbers are fine (from VOC). Failed solutions are generic (louder alarms, phone across the room, lights on), never brand names, never bed shakers. No product
3	Stakes	45-65	moms 5-6	The cost lands on HER	Stake-bank lines only: legal (letters in her name), violence (he swings at her), her breakdown (cried in the car), family (the house fights). The mornings caused it. No college/kid-money stakes
4	Turn	65-70	Interviewer	Pivot to the answer	≤6 words, off camera: "So what finally worked?"
5	Answer	70-88	mom 7 (the answer mom)	Root cause → mechanism → product	One take, ≤10 s (≤25 words at 2.7 wps). Fish's ruled lines: sound doesn't reach a deep-sleeping brain; touch gets through; silent to everyone else; keeps going until he turns it off. Product named once ("a band", "the Dawn band"). The band insert (`broll:band`: product-gated omni b-roll, or real footage) overlays the second half
6	Payoff	88-100	mom 8	Independence / calm mornings	Kid-independence result on her; ≤9 words (4 s) or ≤14 (6 s); the end card (optional, `--end-card`; off by default per Fish 10-06) covers its last 2.5-3 s, so the key words come first

Answer rules (all moms)

Every answer = a real VOC line (owned survey, stake bank, customer-evidence Q-quotes, competitor-review q-IDs) used verbatim or trimmed for speech: cut words, split a run-on, turn "my 14 yr old" into "my fourteen-year-old". No new facts, numbers or claims. The src ID goes on the line. Exceptions: the answer mom (Fish's ruled mechanism lines, claims-register.md §mechanism) and the payoff (a desire line turned into a result, checked against product-truth; flag it at red-team).
Said-out-loud test on every line. Pronoun-first, fragments OK ("Every day. Every single day."). Real people answering a stranger don't make speeches.
Max one sentence of setup per answer. Each answer has to stand alone: the viewer doesn't know these women.
Numbers spelled for speech ("fifty", "six thirty"); acronyms via job.say (ADHD → A-D-H-D).
No mom praises a product except the answer mom. No names, cities or "verified buyer" captions: these are AI actors saying real parents' words, not named customers.
Never quote dialogue inside a visual/action description (the video model speaks quoted text; yapper 10-05).

Structure example (ILLUSTRATIVE: shows the shape only, not shipping copy; real lines go through red-team and Fish)

Q (off camera, banner): "Who gets your kid up for school?" 1. [m1, H1, src stake bank #11 / survey #10232] "Me. Fifty trips every morning. And it ends in yelling." 2. [m2, src survey S05] "He sleeps through multiple alarms. I'm up early to get him up." 3. [m3, src Q04] "His alarms wake the whole house up before they wake him." 4. [m4, src survey #9099] "Everyone was up because of her alarm. Except her." 5. [m5, src stake bank #6, reddit 1rfciee] "You start getting letters in the mail when your kid has too many absences." 6. [m6, src stake bank #25, reddit 1ow28ub] "I screamed and cried the whole way to school." Q2 (off camera): "So what finally worked?" 7. [m7, answer mom, broll:band, src Fish's ruled lines] "Sound doesn't reach him that deep asleep. Touch does. It's a band. It buzzes his wrist till he turns it off. Nobody else hears it." (25 words: the 10 s ceiling at 2.7 wps) 8. [m8, src survey S01 desire → result] "He gets himself up now. Nobody else in the house wakes up." End card (optional, --end-card; off by default per Fish 10-06) (endcard.py, 2.5 s): product cutout + "60 nights to try it."