## Test-only post-ship rider

`wait_for_stream_except` in `crates/spt-daemon/tests/sync.rs` now uses a **60-second monotonic wall budget**, rather than 400 iterations whose effective wall varied with IPC speed. Preserve immediate success, the existing skip filter, Brain's in-flight IPC deadline and nextest's **240-second whole-test backstop**.

Emit one `SYNC_STREAM_WAIT phase=responder_stream_wait` summary on observation, exhaustion or IPC error, carrying `elapsed_ms`, `iterations` and cumulative `ipc_ms`. Exhaustion retains the existing panic message. Requirement tags remain on the actual tests, not this helper, per Doyle's review.

One test file only; no product code, workflow, caller signature, test classification or timeout configuration changes. This rider was NOT part of the shipped v0.71.0 candidate.

Related: https://github.com/BigscreenVR/spt-bs-releases/issues/307
RCA: `.spt/preserved/307/golden-35372054865/RCA-windows-sync-stream-bound.md`.
Doyle: dispatch `4OWHAFKB`, approved logic/tag correction `LN4AMGIE`, final base/content review `QNTEHKMY`, cargo clearance `AGYCCVIE`.

Base `7ee5aa2d`; tested and pushed head **`b33aedcae51fa5ffa840ef1b36c9989a25884bdc`**.

## Windows proof — three runs, 9/9 PASS

Exact inventory: the concurrent-write and two-tier pair plus `torn_pull_recovers_by_repulling`. `ci-windows`, one test thread, **zero retries**, `--success-output immediate`; all three executions preserve their diagnostic output.

| Test | Run 1 seconds | Run 2 seconds | Run 3 seconds |
|---|---:|---:|---:|
| concurrent_writes_reconcile_on_elected_node_and_converge | 62.909 | 85.811 | 54.383 |
| two_tier_sync_lands_and_gate_refuses_server_side | 64.632 | 50.732 | 75.415 |
| torn_pull_recovers_by_repulling | 22.647 | 23.696 | 21.328 |

Thirty WAIT summaries, ten per run, all `outcome=observed`:

| Run | elapsed_ms range | iterations range | ipc_ms range |
|---|---:|---:|---:|
| 1 | 76–1832 | 4–73 | 0–9 |
| 2 | 127–7672 | 6–303 | 1–34 |
| 3 | 76–3277 | 4–130 | 0–15 |

All test durations stayed below the unchanged240s backstop. These exercise successful observation/early exit; they do not force the60s exhaustion or IPC-error path and do not claim to reproduce the earlier intermittent red in this quiet cell.

Proof root: `.spt/preserved/307/sync-wall-bound/`.
- `proof/receipt.json`: exact inventory, native exits0, source-map equality before/after, process/disk receipts.
- `proof/tests-{1,2,3}.{stdout,stderr}`: raw output, per-command hashes in the receipt.
- `wait-metrics.json`: every elapsed/iteration/IPC sample, not just ranges.
- `native.py`: retained native/process/disk guard, private temp/home, inherited live identity scrubbed, jobs2, unchanged test IPC settings.

Admission waited for informational CI35393844345 to finish. The fresh phase gate found no nonterminal workflows or builder/Runner.Worker processes; no overlap was used. Isolated pool was claimed from this worktree. No fleet lifecycle action, target deletion or compiler-cache sharing.

## Verification and custody

- Targeted `cargo clippy --locked -p spt-daemon --test sync -- -D warnings`: **exit0**.
- `traceable-reqs check --json`: **928/928 complete, zero findings**, exit0.
- Changed-file `rustfmt --check --edition 2021`: **exit0**.
- Pool release **exit0**; final census found no builder, Runner.Worker or owned-target user. Retained target4,195,646,644bytes; free139,886,530,560bytes.
- Source maps unchanged through proof and checks. No product docs/changelog change is needed for this test-only helper correction; no scaffolds or new permanent tests added. Retained `.spt` receipts are evidence, not shipped files.

Receipts: `checks/receipt.json`, `checks/trace.stdout`, `pool-release/receipt.json`; final index `DELIVERY.json`. No broad workspace suite is claimed; the required proof is this targeted Windows three-run cell.

Co-authored by: hertz