Reel Engine Review Board

Three reels were rendered from the 2026-08-24 jobs through the approved pipeline design: flat brand fields, lower-third caption plates, canonical outro, silent output. Every mechanical gate passed (65 tests, dimensions, silence, no banned content). You rejected them on quality:

What was actually wrong, named honestly: chronological ordering (slow establishing footage first), long bottom-anchored sentences instead of a muted-legible hook, landscape sources letterboxed into a mostly empty vertical frame, and 2.6 second static holds.

The vault already held four video-craft capabilities that would have raised the bar (cinematic vertical ad craft, post-processing, real-footage gates, video scoring). None were consulted, because the session's capability gate had been legitimately bypassed for an earlier one-paragraph edit and the bypass silently persisted into the build. Classified as a reachability failure, not a missing-knowledge failure.

• Incident written: reel quality rejection, capability gate bypass (2026-08-26).
• Gate-leak fix queued so a quick-edit bypass can never again cover a build.
• No reel SOP existed at all. That gap became Stage 4.

Before re-touching anything creative, a dated research pass (social listening plus the current platform guides) produced the craft brief. The findings that now bind every reel:

• The first frame decides distribution: strong 3 second holds outperform weak ones 5 to 10x in reach. Open ON the payoff.
• Result-first ordering beats chronology; transformation content earns about 3x the interaction.
• Design muted-first: about 40 percent of viewing is silent. Hook of 6 to 8 words, bold, centred, readable in one second.
• Hook classes by measured retention: pattern interrupt beats curiosity gap beats question beats plain problem label.
• Pacing over polish: holds of 1.5 to 2.2 seconds, more cuts, under 30 seconds total (about 72 percent completion).
• Fill the vertical frame; letterboxing reads as repurposed content.
• Before/after pairs shot from identical angles convert best. This is a capture-side lever for the techs.

The standard you asked for now exists, is registered vault-wide, and auto-binds any future reel task on every runtime. Its four gates:

• New hook beat: opens on the money frame with a large centred plate, hard-capped at 8 words.
• Crop-to-fill vertical framing; the flat brand field is now the fallback, never the default.
• Real bug found and fixed: photos were being aspect-distorted up to 25 percent by the zoom filter. Proven fixed by a square-stays-square test.
• Low resolution CCTV now gets upscale plus sharpening.
• 70 automated tests green.

All three reels re-rendered under the new rules. You approved these three, then ruled the bar still is not met, so the approvals are recorded but held and nothing publishes.

Two quality rejections in a row triggered the two-strike rule: no third blind attempt. Instead, the same Greystanes job cut three ways so you can lock the house style. Real-world anchor for A is the Drain Addict channel (781K subscribers, top clears over 16 million views); B is the 2026 creator grammar; C is the branded-ad look.

• Style lock: A, B, C, a mix, or a link to any reel you want ours to look like. The SOP gets tuned to your pick and all reels re-render in that style.
• Sound: reels are deliberately silent (raw job-site audio is stripped; feeds autoplay muted). A licensed or generated music bed on the social cuts is possible if you want it.
• Capture brief: the biggest remaining quality lever is what the techs film. A 30 second brief (vertical before clip, the action, the clear, same-angle after) transforms the source material. Say go and it ships to the team.