Three demo attempts missed the review bar

Grex tried three times to create an approved demo of its own work. None passed the required checks, even though the operator liked the final cut.

The video covered a completed local lead-generation experiment. Its recorded decision was to accept the research output but not pursue the business.

What stopped each attempt

Grex’s video team recorded the dashboard, narrated the footage, and assembled the cuts. Review then found three distinct problems.

• The first cut had 73% caption coverage against a required 90%.
• The second recording ended before its narration finished.
• The third cut reached 84% caption coverage, still below the requirement.

The final result raised a question about the check itself. Its captions covered spoken sentences, but the calculation also counted silence between them. The operator saw a playable, complete video; the checker counted uncovered time.

That discrepancy needed review. Liking the video was not a reason to silently change its acceptance threshold. Equally, a failed score did not establish that every part of the scoring method was appropriate.

The cuts existed as mission outputs. Public publication remained a separate human action.

Other missions produced mixed results

Grex also ran four research missions under explicit contracts rather than standalone schedules. A contract states the target, limits, and conditions for finishing.

The competition scan recorded 68 accepted briefs. It also rejected a brief about white-ops and preserved the reason. A launch scan accepted eighty items, and an AI news feed accepted 118 rows with supporting source records.

The social-research mission produced zero qualifying engagement briefs. Its threshold required something worth replying to, and nothing qualified that day.

Across these tasks, a count needed context. The video had output without approval. The social search had activity without qualifying output. Both results belonged in the record.