Tests that can
say no.

REAL SQLITE / REAL CHILD PROCESSES

Beyond one recovered claim.

Two simultaneous claimants, interruption inside a transaction and after commit, lease renewal, inclusive expiry and old generation rejection. Ownership checks are inside the new candidate methods.

TANG VU’S ADDITIONAL PUBLIC REPOSITORIES

Broader inputs. Honest scope.

FROZEN DIRECT / FEEDBACK STUDY

Measure the feedback.

The amended comparison uses three tasks, two repetitions and two requests per arm. It preserves the original failed setup as a separate artifact. Inspect both versions, every proposal and the final held-out results below.

Twelve planned arms were attempted once. After the original studies stopped, a separately frozen post-hoc continuation collected only the seven previously unrun arms. No failed attempt was retried. Continuation artifact · Protocol and deviation ↗

Amended attempts · Frozen protocol ↗ · Original failed setup · Full planned matrix / summary

Recompute locally.

npm ci --ignore-scripts --no-audit --no-fund
npm run prepare:tend
npm run prepare:repositories
npm test
npm run evaluate:repositories
npm run evaluate:lifecycle
npm run evaluate:comparison
npm run evaluate:comparison:v2
npm run evaluate:comparison:continuation

These commands make no model calls. Repository validation is not human feedback or productivity evidence. The original TEND investigation remains unchanged.