2.0 KiB
Reconciliation Efficiency
Observed Blocker
The current Draft review batch was waiting for model responses even though its two remaining records concerned diagnostic commands:
bed41451869b76dc7d8f: an extra closing parenthesis in an inline Node artifact inspection. Node rejected the expression during parsing; the CAD runtime was not executed.da106eb06026eb3077a7: a standalone unscopedgit status --shortcorrectly rejected by the inspection guard.
The first record has an explicit review backed by a published-artifact inspection. The second has a machine-checkable diagnostic disposition. Original failure records remain retained. Neither disposition claims a new CAD capability.
Measured Steps
| Step | Time | Model Required |
|---|---|---|
| Inspect Draft runtime, WASM and pthread worker; verify manifest hashes | 0.438 s | No |
| Recompute failure classification and validate current evidence | 0.3963 s | No |
| Reuse unchanged classification and revalidate evidence | 0.0735 s | No |
| Advance from the closed review to the next business batch | About 0.25 s | No |
The warm reconciliation measurement reused the same classification result and took 81.45% less time. The earlier model waits of approximately 90 and 45 seconds are separate observed waits, not directly comparable CPU work.
Result
The two blocking records are closed in the reconciliation ledger. Planning advanced to a TechDraw batch containing 13 desktop-command business records. These records remain implementation work, not completed migration.
Review attempts now yield unfinished records after one attempt without cooling the business module. Closed records are omitted from subsequent work packets. Evidence validity is still checked against the latest failure occurrence and current relevant sources.
Published-artifact inspection: docs/current/failure-evidence/draft-artifact-inspection-v13.json.
Refresh detailed timing with:
python3 tools/analyze_scheduler_timings.py --cycles 20 --report-prefix docs/current/unattended-timing