## Instrument handoff — committed; no product remedy

Rig commit: **`c0ef957189edea03842afc4fa3ee0ce2ab769544`**, branch **`test/309-er-briefing-regression`**. Parent/baseline: **`39241675642648a21a3e5ceca0ca95bf6e988155`** (latest ruled main).

One nextest cell: `spt::er_briefing_bringup_reconcile_e2e::bound_engine_room_survives_bringup_reconcile_and_receives_native_briefing`.

### Matched native proof
- **Scheduled RED, exit 100:** real bind established online/controllable/ready/no-relay eligibility; broker returned `sessions=[]`, `bringing_up=[engine-room]`. The admitted real livehost pass returned `offlined=[engine-room]` and wrote offline. PTY retained the same live birth identity. Exactly one briefing remained undelivered, with no native-inject claim. The test failed its product assertion, not a setup/teardown assertion.
- **Knob-unset GREEN, exit 0:** same real first-seat lifecycle with `SPT_TEST_ER309_DIR` absent. No hook observation/error files. ER remained eligible; exactly one briefing was audited delivered/native-inject; the real harness read `XLATE_OK\r\n` from its PTY before idle.
- Each mode ran one cell, retries zero. Both inner populations had no survivors, unreadable processes, or lost scope. Both native jobs reported `coverage=COMPLETE`, `termination=CONFIRMED_GONE`, `active_processes=0`.

Receipts under `.spt/preserved/309/` (each includes raw-evidence, source-input, binary-input and native-containment hashes):

| Mode | Receipt | SHA-256 |
|---|---|---|
| RED | `39241675-plain-red/receipt.json` | `0a246aa5cc9e154b0f90eb6c10d4bacb7beba0cdb4eee955f7236a9c27438439` |
| Unset GREEN | `39241675-plain-control/receipt.json` | `a7e7f0f609f611a190e3ccc2f6016784be61c64d146ca9d2de48ce4ed5b3497c` |

### Instrument boundaries
Publication is held after translation setup but before HostedSession insertion; the first actual reconciliation is admitted explicitly. Both one-shot barriers are `cfg(debug_assertions)`, bounded to 30 seconds, and scoped to an explicit isolated non-resident home. No liveness policy, status mutation, rebind/idle healing, or product presentation-deadline change.

Fixture-only options: mock-session `--input-log` records real consumed input; translation fixture `plain` mode emits text/Enter/commit without unrelated stash/restore chords. Both defaults remain unchanged. Existing attach/rc test sources are unchanged; both HEAVY classifications include the new cell.

Earlier evidence is retained separately: the original native RED on `4b6b03cc`; the first unset oracle wrongly expected dummy stdout echo; the next choreography probe captured `LATE_OK` rather than the complete marker. Neither fixture-oracle failure is represented as a product failure or as the final control.

### Reproduce / remedy gate
Build `mock-session`, `spt`, and `translate_proof_fixture`, then:
```
cargo nextest run -p spt --test er_briefing_bringup_reconcile_e2e --retries 0
```
For the inert control, set `SPT_TEST_ER309_KNOB_UNSET=1` for that same command. The normal scheduled cell is intentionally RED on this instrument-only branch; the remedy must make it GREEN without weakening its oracle.

`traceable-reqs check`: **924 complete, 0 incomplete, 0 findings**. Owned build pool released; target/evidence retained. No resident lifecycle operation. Co-author trailer: `Co-authored by: hertz`.