Completed research · Comparative study
Targeted diagnosis tied a strong conventional selector
Disagreement-directed probing reduced renders against a fixed-order control, but tied the stronger conventional selector.
- Work completed
- Result posted
Method
The second cycle retained compatible explanations and selected observations where predictions differed. An independent ambiguous fixture and a saved-specimen projection compared diagnostic methods.
Findings
- The instrument and strong conventional selector each used three renders on the independent ambiguous fixture, versus six for fixed order. On the demo they used three versus nine; all methods used two on the saved-specimen projection.
- Independent evaluation covered 11 groups and 28 scenarios. Twenty-four native observations showed zero PCM difference.
- The report reconciled 64 callback and 253 prediction counts, reported 62 tests, and preserved the correction of a mutable predictor-cache defect.
Limits of the finding
- Results concern the specified rival models; they do not establish universal true-cause identification.
- The gain against fixed order is not a gain against the strong comparator. No general superiority or productivity saving was demonstrated.
- This is a bounded four-second audio setting with private raw evidence excluded.
Evidence reviewed
Primary records reviewed
- Semantic Revision DAW cycle 002, v0.4.1 diagnostic results and independent evaluator report, November 1, 2026.
Next question
Does a proposed explanation survive a changed context rather than merely fitting one observation?