A 30-person Beijing production company was running 8–12 ad films and short-video projects every month with six to eight core producers. The work itself was not the problem. The coordination was. Across an 8-week baseline window covering 22 projects, producers were spending roughly 45–60 hours per week — call it ~50 hours — restating requirements, re-sequencing schedules, confirming asset versions, and fielding questions that had already been answered. The production cycle from client-confirmed brief to first deliverable plan, storyboard, and budget package ran 7 working days.
We started by reading project logs, not running workshops. The signal was in the artifacts: where did rounds of confirmation cluster, and why? Two patterns held across every project.
Two invariants, one clear boundary
The first invariant: any script change exceeding 30% of the original scope must return to the client for confirmation before production continues. It had always been true in practice. Nobody had written it down. The second: before a single shooting day is booked, five items must be complete — scene confirmed, crew confirmed, props confirmed, budget signed, risk flagged. Both rules were being reconstructed, project by project, from producer memory.
The harness encodes them. A redacted production decision trace — brief to storyboard to budget to risk flag to human confirmation — shows exactly where each rule fires and where it waits. The judgment calls that remain (creative direction, client relationship, appetite for schedule risk) stay with the producers. The harness does not touch them.
“What Specability gave us wasn’t just efficiency — it lifted the ceiling on our revenue.”
The result
Over the 8-week pilot and into Q2 stabilization, the cycle from confirmed brief to first deliverable package dropped from 7 working days to 2. Requirement-response satisfaction, measured by post-project client evaluation, moved from 65 out of 100 to 87. The ~50 recovered hours per week did not go to rest — they went to taking on more projects. That is what the founder meant. The invariants were never the interesting part of the work. Encoding them made room for the part that is.