[JustYannicc]

Problem

A checkpoint check arriving while the worker waits for truncation acknowledgement is dropped. Its caller then keeps a check marked in flight forever; later checkpoints stop and write admission can eventually hang. This remains in the merged #1000 code.

Change

Backport pingdotgg/t3code@7e69ce7d032badca71e8071c0ccbe31101943e12. Answer the check with its request ID, the pending generation and blocked: true. This releases the in-flight request without accepting an older snapshot as current. Barrier handling stays intact.

Scope and approval

Follow-up to #1000 and #779, explicitly requested after the upstream review found the race. The two adapter files match the corrected upstream source. When syncing pingdotgg/t3code#21037, take the upstream version and reconcile this carry.

Verification

• The upstream deterministic worker regression fails before the correction and passes after it. The Signalbox base has identical adapter source to that failing upstream base.
• vp test run packages/shared/src/nodeSqliteClient.test.ts -t 'answers checks received while the worker waits for truncation acknowledgement' --maxWorkers=1: 1 passed. It queues a check during held truncation, verifies its own reply and the old generation, then proves a later write completes.
• Shared-package typecheck, two-file lint and formatting checks pass.
• The same correction and regression pass in the patched app source under Electron's bundled Node runtime. No live database or running app was used.

Independent review found no issues against the exact two-file backport; both file blobs match upstream and the corrected patched-app source.

Independent review: GPT 6.1 Sol xhigh (Opus unavailable: rate-limited at 2026-10-17 15:47:05 UTC).

Implementation: GPT 6.1 Sol through the Codex harness in T3 Code; backported from the upstream fix.