Retry-stall follow-up (#144)
Status: verified locally on Linux/tmux; macOS/Herdr unexecuted Date: 2026-11-08 Issue: #144 Base: integration/ensemble-v7 at e5e51e8 Branch: fix/v7-retry-stalls
This record covers the post-v7 retry fix:
- A retrying card now shows
retry <n>s · <attempt>/5, orretry due · r nowwhen the backoff has elapsed. - One durable system event names the ETA, attempt, and failure reason.
- Maintenance claims due work after its repair pass, so a future retry resumes after a restart, hot reload, or lost in-memory timer without an unrelated schedule.
- An uncertain tool outcome parks behind an interrupted approval instead of auto-retrying.
- Backoff is 5 s, 15 s, 30 s, then a 60 s cap, with jitter.
- Braille runs at 8 fps and the Sub-Cell Tide at 15 fps (previously 4/10).
Environment
| Item | Value |
|---|---|
| OS | Linux 6.12.113+deb13-amd64 |
| Terminal host | tmux 3.5a |
| Node | v26.12.1 |
| OpenCode | opencode v2.2.6 |
| macOS/Herdr | Not available; not run |
Commands and results
| Command | Result |
|---|---|
npm run verify | Pass — strict type check; 953 passed, 0 failed, 0 skipped. |
npm run test:live:fake:all | Pass — fake 195 s, fake-empty 9 s; 2 passed, 0 failed. |
npm run test:live:sizes | Pass — 80×24, 100×30, 120×40, 160×50 group/layout/focus/composer checks; NO_COLOR still-motion check. |
git diff --check | Pass — no whitespace errors. |
CPU after the faster cadence
test:live:sizes records one 0.9 s TUI-process sample per count at 160×50:
| State | TUI CPU |
|---|---|
| Idle | 3.3% |
| 1 working card | 10.0% |
| 2 working cards | 17.8% |
| 3 working cards | 28.9% |
These are short, noisy samples. The faster Braille/tide cadence raises the 1–2-card cost relative to the pre-fix 4/10 fps record; the 3-card sample is within the previous run's variance. The harness does not gate on these numbers.
Behavioral evidence
| Behavior | Evidence | Result |
|---|---|---|
| Countdown, attempt, and due copy | ticket144: retrying detail counts down, names the attempt, and yields to needs-you | pass |
| Per-second card refresh without rebuilding static cards | ticket144: a retry countdown refreshes every second and stops at the ETA | pass |
| One durable retry event with ETA and reason | ticket75: an empty assistant reply is retried, never a silent success asserts exactly one Retrying in … message | pass |
| Future retry wakes after restart/lost timer | ticket144: maintenance claims a future retry without an unrelated schedule (service start with future retryAfter; maintenance claims after ETA) | pass |
| Uncertain tool outcome parks for review | ticket144: an uncertain tool outcome parks for review instead of auto-retrying asserts interrupted, pending approval, no retryAfter | pass |
| Backoff 5/15/30/60 with cap and jitter | src/v2/retry.test.ts | pass |
| Group lifecycle, needs-you, failure events, sizes | full suite plus fake and size matrix | pass |
Skips and limitations
- macOS/Herdr: NOT RUN. The existing manual checklist remains the authority for macOS font coverage, color perception, and key ownership.
- Hosted CI was not run.
- The live fake scenario does not directly simulate a retry cycle; the wake-up and uncertain-outcome behavior are covered at the service seam and the card copy at the pure seam.