[NoahHendrickson]

Problem

UsageService.test.ts > reports saved usage of a removed directory only for windows it reaches writes a transcript dated 2026-08-01 10:00 UTC, and the usage cache keeps files for 92 days (CACHE_RETENTION_DAYS). As of 2026-11-01 10:00 UTC the saved usage is past retention, so the test fails on every branch (it's failing Test Server 7 on #260).

Change

Cherry-picks upstream's fix, 1280ccd (pingdotgg#28603), which pins the test clock near the write.

Verification

• The cherry-pick applies cleanly.
• Locally the test passes with it and fails without it. Two other cases in the file fail on this machine with or without the fix and pass in CI.

🤖 Generated with Claude Code

[NoahHendrickson]

Code quality review of 1d204b6bc3c67cfe80a47c6e5b56f1693cdcdacf: no major structural issues found.

The diff matches upstream's 1280ccd line for line, and it uses the same TestClock.setTime + Effect.provide(TestClock.layer()) pattern this file already uses for the Cursor case (UsageService.test.ts:1255/1284). Pinning the clock is the right fix. The retention cutoff comes from Clock.currentTimeMillis (UsageService.ts:664, :678), so the test now controls it, and the wall-clock time bomb is gone without touching production code. No new helpers, branches, or file growth.

The file's other hard-coded 2026-08 timestamps read live transcripts, not saved usage past retention, so they aren't affected by this time bomb.

[NoahHendrickson]

Reviewed commit 1d204b6bc3c67cfe80a47c6e5b56f1693cdcdacf against base 1cbd9f2bab5d10ae81dc40297fbf92f23ba10136.

No actionable findings. The test-scoped TestClock pins the retention cutoff within 92 days of the fixture's file timestamp and preserves the assertions for both query windows. Production code is unchanged.

Validation: vp test run apps/server/src/usage/UsageService.test.ts -t "reports saved usage of a removed directory only for windows it reaches" passed (1 passed, 39 skipped).

[NoahHendrickson]

Code review — commit 1d204b6

No issues found.

• Matches upstream. The diff is the same as upstream 1280ccd (pingdotgg#28603). Only the line offsets differ.

• The fix works. UsageService gets "now" from Clock.currentTimeMillis (UsageService.ts:664), and the retention cutoff is derived from it (:678). Wrapping the case in TestClock.layer() and setting the clock to 2026-08-15 puts the 2026-08-01 transcript inside the 92-day window no matter what the real date is. The file already uses this pattern for the Cursor case (UsageService.test.ts:1255/:1284).

• Placement is fine. Effect.provide(TestClock.layer()) wraps the whole generator, so the clock is frozen for setup and UsageService.make too. Nothing in this path sleeps on the clock, and the case passes.

• Local run (macOS, this commit): the target case passes. In the full file, 38 of 40 tests pass. The two failures don't involve dates:

  • discovers Pi history…: a /private/var vs /var tmpdir realpath mismatch.
  • retains Codex reconciliation…: also a path comparison (path === transcript, :2007).

  Both fail the same way without this change and are macOS-specific.

Not blocking: other cases still hardcode 2026-08 fixtures (BEFORE_NARROW_WINDOW, the created/timestamp literals), but they don't depend on cache retention, so they shouldn't break with time. CI was still pending when this review was posted.

Reviewed by Claude Opus 5.5 (Claude Code).

[github-actions]

Thread transfer impact

✅ Thread transfer remains within every enforced ceiling.

Provider	Metric	Main baseline	This PR	Impact	PR ceiling
Codex	Total thread wire	5.0 KiB	5.0 KiB	+6 B (+0.1%)	6.8 KiB	✅
Codex	Thread snapshot wire	3.8 KiB	3.8 KiB	0 B (0.0%)	4.9 KiB	✅
Codex	Live turn WebSocket wire	1.2 KiB	1.3 KiB	+6 B (+0.5%)	2.0 KiB	✅
Codex	Live turn WebSocket decoded	21.6 KiB	21.6 KiB	0 B (0.0%)	24.4 KiB	✅
Codex	Live turn messages	2	2	0 (0.0%)	8	✅
Claude	Total thread wire	5.1 KiB	5.1 KiB	0 B (0.0%)	6.8 KiB	✅
Claude	Thread snapshot wire	3.8 KiB	3.8 KiB	0 B (0.0%)	4.9 KiB	✅
Claude	Live turn WebSocket wire	1.3 KiB	1.3 KiB	0 B (0.0%)	2.0 KiB	✅
Claude	Live turn WebSocket decoded	21.9 KiB	21.9 KiB	0 B (0.0%)	24.4 KiB	✅
Claude	Live turn messages	2	2	0 (0.0%)	8	✅

Baseline: 1cbd9f2 · PR result: 1d204b6 · Source CI: success

Scenario and decoded snapshot size

10 historical turns, 5 command tools per turn, 878.9 KiB retained MCP result per historical turn, and a 1.05 MiB retained result in the measured turn.

• Codex decoded thread snapshot: 108.5 KiB
• Claude decoded thread snapshot: 108.8 KiB

Updated in place by a trusted workflow. PR artifacts are strictly validated and never executed.