Contractor+ Jobber-style ads — preflight and next production step
2026-09-10. Project prepared; Justin has been emailed. Production remains on hold for missing facts/assets and actual visual/audio review. No paid media generation was submitted.
What is ready
- All five original items were recovered: complete session history, the individually linked failed ad, real Rands testimonial, Jobber reference and creative ZIP. Sizes, hashes and archive contents were checked. Source inventory.
- The supplied Jobber reference has the same encoded video and audio as our previous PromptLab project. Its earlier teardown can be reused, with its limitations preserved. Five historical analysis/log files were copied into preserved task snapshots and verified byte-for-byte.
- Our existing tools were actually exercised locally. PromptLab extracted 261 representative frames and three audio files from the three original videos and adapted the 27 prior scene descriptions into 162 prompt-dialect strings. Reverse Prompter's existing parser, text-removal and recipe-validation functions processed those 27 historical scene recipes. All 27 passed structural validation; none is approved for generation by that check. Execution receipt.
- Prior work across PromptLab, Tarot motion, Rogowo, Stół, Memoirs and OSHA4US was reviewed and converted into 12 project-specific checks. The checks include exact source/timecode identity, truthful claims, effective crop resolution, representative proof, motion continuity, type and logo correctness, separately composed ratios, natural sound, transition boundaries and versioned final-delivery review. Lessons audit.
- Questions were sent to Justin at
ceo@contractorplus.appin the original Gmail thread. Readback confirmed the correct recipient, thread, SENT state and exact reviewed body. Email · receipt.
Mistakes caught before production
| Finding | This project's response |
|---|---|
| Package README/index promises separate beats, 12 Veo plates, voice tracks and word timings that are not in the actual ZIP. | Ask for those source layers before paying to recreate them. We already have 13 brand PNG assets and the assembled edits. |
| The old five-film script still says “nineteen years” and includes an unsupported homeowner incident and five-star review, despite the later history saying the facts were corrected. | Start a clean current edit script from Scott's actual testimony. Keep all old scripts as historical drafts; do not trust their “verified” labels. |
| The 350% growth assertion has no underlying figures or date-range export in this package. | Ask for the metric, period and scope. Use a verified customer-quote treatment if the number cannot support the intended wording. |
| A portrait crop of the 1280×720 testimonial retains about 405×720 pixels, requiring roughly 2.67× enlargement for 1080×1920. | Test a deliberate documentary window/collage treatment or obtain better footage. A 4K generated insert cannot repair soft source coverage. |
| Old PromptLab prose calls the near-square source vertical; its sound interpretation partly relied on spectrogram inference. | Ground geometry in fresh probes and judge motion/audio directly before the edit is locked. Fresh extraction detected 31 reference shot candidates; those are detector candidates, not a director's final scene count. |
| The old match loop could lose prior baselines, and a prompt's model-dialect name could differ from the renderer actually called. Its previous match result cleared only 15/27 stills at the recorded threshold. | Preserve baselines; record actual model and job receipts. A matching still or a passing parser is not a successful ad. Avoid the default full render/match routes until their applicable defects are resolved. |
| Historical sound was described as accelerated dry narration, and picture sync was mistaken for editorial rhythm. | Fit copy to a natural performance, design sound early, and review a representative combined sequence before scaling. These remain historical diagnoses until directly checked against the actual moving images and final mix. |
The next creative step
Proposed pilot: one Rands customer-story ad, about 30 seconds, with separately designed square and vertical treatments; the August package additionally specifies 4:5. Confirm current placements and series priority with Justin. Preserve Contractor+'s actual brand and facts while borrowing the reference's early proof, complementary picture/text tracks and varied motion.
Build a short representative sequence before a whole batch: real customer footage, a truthful proof element, brand typography, natural voice and an intentional sound mix. Test it at phone viewing size in the intended ratios, with sound and muted. If the real source cannot carry that approach, use an editorial recut, change the framing or request a small pickup shoot. A new generator is not automatically the answer.
The older calm property-film pacing rules do not apply to this fast case-study ad. The client reference determines rhythm. The transferable lessons are deliberate timing, clean transitions, readable type, source fidelity and review of the delivered result.
What still blocks a production-ready decision
Client answers: first story/series priorities, placements and review date; missing source layers and current actual UI footage; supporting growth figures and customer reuse/story approval; current offer and CTA destination; pilot budget. All were asked in one email. Existing logo files and earlier voice preferences were acknowledged rather than requested again.
Internal review: fresh full moving-picture and listening review is still open. No fresh vision model was called on these client assets. The current local classification route suppresses thinking and an optional external support trial exists; neither was silently used against this task's constraints. The extracted frames/audio make that next review concrete. Correct or explicitly adapt the applicable model/effort and project assumptions before a full semantic reverse-prompting pass. A caller's setting alone is not observed execution proof.
Actual tool scope: local extraction, prompt adapters and Reverse Prompter contract validation were completed. A new reverse_recipe() visual analysis, full semantic/audio interpretation, rendering, video similarity scoring and creative acceptance were not completed. This is a preflight packet, not a finished video or a fully validated production pipeline.
Money: zero paid image/video/audio generation calls. A read-only current Veo catalog/cost check is saved under tools-preflight-v1/; no generation job resulted. A fresh production quote, concrete shot needs and an authorized ceiling are still required before spend. Justin's answers provide client facts and preferences; they do not alone authorize spending from Jeremy's accounts.
Evidence and ownership
Project home: BRIEF.md · stages · claim/asset ledger · tool preflight.
Requested review models were explicitly Astra/ultra. Exact root and delegated runtime effort was not exposed to this task, so it remains unverified in RUN-SETTINGS.json. Local component execution involved no extra model-helper calls. The saved Plan is contractorplus-jobber-preflight-0910; its automatic event notification was refused, so no background watcher or automatic response handling is claimed. The saved artifacts and sent email are the verified handoff.