How well does the check work?

These are historical results from our own test on 18 October 2026, using the rules as they were then. The check found 62 of the 70 faults we had planted in 21 made-up plans (89%). It also flagged 5 of 23 decoys (deliberate near-misses that are not faults), and raised 22 findings on 9 of the 21 plans before any faults were planted. The plans, faults and decoys were written by AI models working blind to the rules. This is a small test, it does not measure later versions of the rules, and it is not a guarantee for your film. The professional-cut control (one scene) raised no findings.

The checks cover editing craft and filming feasibility, not whether a film is good. No AI model runs. A plan without enough detail is not assessed.

How we tested it

An AI model (Claude), acting as a film director and editor, wrote the 21 plans without seeing the readiness rules. A separate AI agent, also without seeing the rules, planted and labelled the faults and decoys. The rules were developed on separate development sets, without access to this test set.

We ran our benchmark tool (dpl bench) on 21 plans (test set v3) on 18 October 2026, before and after our latest round of rule changes (rules change E10: before it, dpl-eil main 0686a6ff; after it, the E10 rules at ca487129). Each changed plan was scored against the findings on its own original. A further 11 faults the check is not designed to find were left out of the headline.

Test sets rotate: once failures are read or used to tune the rules, that set becomes a development set. That has now happened to v3, so these are historical blind-test results, not a fresh score for later rules.

Limits and false alarms

Of the 70 planted faults, 62 were found under the expected finding, 1 was flagged under a different finding and 7 were missed. The weakest areas were short holds (2 of 5 found) and speaker faults (3 of 6 found).

Before any faults were planted, the 21 plans raised 22 findings across 9 of them. Decoy alarms alone do not describe every false alarm. This small, constructed test is not a guarantee for your film.

The clean professional control is our own written description of one dialogue scene from Tears of Steel (2012) by the Blender Foundation, used under CC BY 3.0. Credit: (CC) Blender Foundation | mango.blender.org. The Blender Foundation has not reviewed or endorsed this test. No images, audio or subtitle text from the film are reproduced here.

Full blind-test results · How the set was made · Figures as JSON