Completed research · Comparative study
A real integrity repair exposed a shared upgrade blind spot
Ordinary and composed approaches both solved the frozen repair checks. All four proposals missed the same upgrade case, which required a separate correction.
- Work completed
- Result posted
Method
A real document-management defect was tested using a frozen 27-case suite. Two ordinary Astra attempts and two composed-team attempts received matched source, task, settings and public feedback. Compatibility review followed the primary comparison.
Findings
- The original system passed 1,048 existing tests but failed 18 of the 27 new checks. All four final candidate repairs passed all 27.
- Review found that all four proposals missed historical successful mappings lacking the new index. Six separate upgrade checks were added; the selected candidate initially passed five of six.
- A separate repair completed the upgrade behavior. The final local suite passed 1,081 tests, including 33 new regression cases. No added correctness benefit from the composed team was demonstrated.
Limits of the finding
- One chosen problem and two attempts per approach cannot establish broad invention quality or comparative equivalence.
- The upgrade checks and later repair are not part of the frozen primary comparison. Local test success does not establish a production rollout or crash-atomic behavior across external records.
- Raw task fixtures, document records and private research custody are excluded from this editorial summary.
Evidence reviewed
Primary records reviewed
- Docman Occurrence Integrity study: frozen comparison, separate upgrade review and local verification report, November 9, 2026.
Next question
Can a reusable compatibility replay instrument catch failures involving older records before review?