# GOLDEN INTAKE — head 712c987f (v0.71.0 candidate), batch ddd7fc7f..712c987f
Author: doyle, 2026-09-18 ~02:10Z. Basis: `git diff --stat ddd7fc7f..712c987f`, `.config/nextest.toml` + `golden.yml` diffs at 712c987f, gate receipts named below.
Last golden: v0.70.0 on ddd7fc7f (9/9 at S5). Batch = 21 commits, 54 files, +4890/-546. PR CI on the head: 35289449769 5/5 (THIN lane — unit Linux/Windows, lint, traceability, changes; NO integration legs).

## A. Cells whose FIRST golden execution is this run (7 integration fns in 5 binaries, +6 unit fns)
| # | cell | phase | gates on it | executed before (sha / receipt) | first-golden risk |
|---|------|-------|-------------|----------------------------------|-------------------|
| 1 | spt-daemon test `refresh_freeze::refresh_resumes_bounded_input_ack_without_replacement_poison` (b99715ee, 39241675) | HEAVY / Phase B, Windows only (`#![cfg(windows)]`) | needs `SPT_REFRESH_RIG_PYTHON` — golden.yml pins `C:\Program Files\Python312\python.exe` (8f059b2a) and prints `REFRESH_RIG_PYTHON=` in a precheck step; Python rig = rig.py 777 lines + launch-v2.cs 789 lines | hertz w2 regression: negative controls 289e3cdf x3 (.spt/preserved/307/hertz-w2-regression/negative-289e3cdf-*.nextest.log); NOT run on 712c987f by me | HIGHEST — new external-interpreter rig, Windows-only, never under Runner.Worker env |
| 2 | spt test `er_briefing_bringup_reconcile_e2e::bound_engine_room_survives_bringup_reconcile_and_receives_native_briefing` (c0ef9571, eed9ff36) | HEAVY / Phase B; `#![cfg(debug_assertions)]` (golden builds debug so it runs); control knob `SPT_TEST_ER309_KNOB_UNSET` | added to both HEAVY copies (nextest.toml + golden.yml) in the same commit | gate-229 (.spt/preserved/309/gate-229/phases.txt: cell-scheduled, cell-control, presented-e2e, livehost-unit all exit 0) on the 309 lane sha; NOT rerun on 712c987f | MEDIUM — lib support `er309_gate.rs` (194 lines) is new daemon code compiled into the product |
| 3 | spt-store test `serving_registry_two_process_int` x3: `loaded_reaper_serializes_registration_without_resurrecting_expiry`, `later_reap_preserves_a_registration_published_while_it_was_waiting`, `abrupt_holder_exit_releases_the_sentinel_for_the_waiting_reaper` (c9e80511) | Phase A (spt-store not in HEAVY) | same-package fixture bin `serving_registry_worker` via `env!("CARGO_BIN_EXE_...")` — cargo builds it, no extra edge | gate-228 (.spt/preserved/307/gate-228/phases.txt: p3-rig-1, p3-rig-2, p5-seams exit 0; p6-xtask-gen-check exit 1 = RED at docs-gen, lane landed after the docs fix f6b1dd9a) | MEDIUM — two-process file-lock timing on a loaded Phase A box; first Linux execution |
| 4 | spt-daemon LIB test `pairhost::meet_runtime_tests::three_subnet_rotation_keeps_lifecycle_off_broker_workers` (a593ece0) | HEAVY / Phase B via NEW lib-test override (heavy-broker-pty, slow-timeout 60s x3) in both nextest profiles + golden.yml HEAVY | real binds, canary peak < 250 ms | doyle gate a593ece0 + 712c987f run2: meet 1/1 (.spt/preserved/302/gate-231-712c987f/run2/summaries.txt); todlando red control exit 100 canary 471 ms pre-fix | LOW-MED — load-sensitive canary; first time the HEAVY filter carries a `kind(lib)` term |
| 5 | spt-net unit `all_bind_scopes_exchange_without_portmapper_activity` (endpoint.rs, 712c987f) | Phase A | real binds All/Port/Loopback, asserts iroh mapper counters 0 | run2 mapper 1/1; RED on a593ece0 (regression proof) | LOW |
| 6 | spt-daemon test `docs_server_e2e::discovery_distinguishes_absent_control_from_unbound_listener` and `twohost_web::serves_published_surface_and_refuses_escapes_and_writes` (c9e80511 / 84e20ddd, new fns in MODIFIED integration binaries) | docs_server_e2e = Phase A; twohost_web = HEAVY / Phase B | shared seed-broker docs fixture (`tests/support/serving.rs`, 84e20ddd) | gate-228 rigs on the lane sha only; thin PR CI does NOT build integration binaries, so NO CI evidence on the head (correction from deployah 01:57Z — my first draft mislabelled these as unit) | MEDIUM |
| u | unit fns new in changed files: `consent_changelog_uses_the_discovered_node_local_docs`, `consent_changelog_retains_release_url_without_live_docs`, `update_messages_link_live_node_docs_or_preserve_offline_fallbacks` (cli.rs/notif.rs, 4b6b03cc); `unavailable_registry_lock_cannot_publish_an_add` (serving.rs, c9e80511) | Phase A / unit lanes | — | PR CI unit lanes on 712c987f (5/5) | LOW |

Count for the cover: **9 integration test fns (7 binaries) see golden for the first time**; 4 unit fns first-golden but CI-green on the head.

## B. Seams changed in the batch and the EXISTING consumer cells that rerun on them
| seam (files) | commit(s) | existing consumers | rerun on 712c987f? |
|---|---|---|---|
| nethost.rs, pairhost.rs, spt-net endpoint.rs, meet.rs (portmapper Disabled everywhere; spt-meet runtime) | a593ece0, 712c987f | net_worker_starve, mesh_recovery, spt-net suite, daemon int, guards | YES: run2 starve 2/2, mesh 5/5, guards 254/254, attach 17/17; a593ece0 gate spt-net 73/73, daemon-int 29/29; field PASS ENLYZEAM capture #2 (capsule 0611a058 = 712c987f + diag) |
| livehost.rs (204-line churn, hosted liveness during ER bring-up) | eed9ff36, 84e359aa | livehost::tests, engine_room_bringup_e2e, er_* e2e family, livehost_bootgate_e2e, resume_* e2e | PARTIAL: gate-229 + gate-leak-84e359aa livehost-unit exit 0 on their lane shas; NOT rerun on 712c987f by me — golden Phase B is the first run of these consumers on the merged head |
| servehost.rs, docshost.rs, spt-store serving.rs (cross-process registry lock), attach.rs (+51) | c9e80511, 4b6b03cc | docs_server_e2e, twohost_web, webserve_attachment_e2e, webserve_cross_node_e2e (all four MODIFIED in batch), attach suite | attach 17/17 run2; the four web/docs e2e NOT rerun on 712c987f (gate-228 rigs on the lane sha only) |
| broker.rs (+18, ACK backpressure on rc replay), pump/update.rs, notif.rs, cli.rs (update notices to live docs) | 39241675, 4b6b03cc | input_ack_deadlock, budget, daemon_refresh_e2e, refresh_freeze | NOT rerun on 712c987f beyond unit lanes |
| adapters/mock main.rs (+13), translate_proof_fixture | c0ef9571 | every e2e that spawns the mock adapter | via gate-229 cells only |

## C. CI-plumbing changes that execute for the first time in THIS golden (resource/rig reds, not product)
1. **Disk admission floors raised**: Windows `$floorBytes = 128GB` (GiB) and Linux `130000000000` bytes (0cda6235, 5e05f02d), plus `GOLDEN_TEST_DISK_START_BYTES` and `DISK_DELTA ... scope=NET_SHARED_VOLUME` end-of-job lines (82333ca1). HFENDULEAM runner cache was reaped 2026-09-17 23:33Z (cold build ahead); free ~184 GB at 01:52Z. kitsubito free space NOT measured by me — measure before dispatch; under-floor = resource red, rule it yourself.
2. **traceable-reqs precheck**: new steps `Checker precheck self-test` (`.github/ci/traceable-reqs-selftest.sh`) and `Precheck traceable-reqs pin and runner` (`traceable-reqs.sh precheck`, env GH_TOKEN from TRACEABLE_REQS_TOKEN) replace the inline installer (52d56248). First golden execution of both.
3. **Refresh rig interpreter precheck** step (8f059b2a) — prints `REFRESH_RIG_PYTHON=`; a red there is a box-config red, not #302 product.
4. `find-cwd-holders.ps1` unknown-headless state (93c7c2b9) — census wording only.

## D. Triage watch-list (RCA-first order if red)
1. Any red in `refresh_freeze`: read the `REFRESH_RIG_PYTHON=` precheck line and rig.py output first; rig red vs product red decided from the ACK stamps, not the exit code.
2. Phase B canary reds (`meet_runtime_tests`, `net_worker_starve`): box-quiet check first (post-merge ci run on this runner re-occupies the box; see merge-push memory).
3. `serving_registry_two_process_int` red on Linux: first Linux execution of the two-process lock; c9e80511's validation was Windows (f6b1dd9a docs record).
4. Admission floor red: resource, UNTESTED, rerun after reclaim — not a product signal.

## E. Honest gaps
- I did not execute cells A1, A2, A3 on 712c987f; their last executions are on lane shas (289e3cdf-era, 309 lane). The golden is their first run on the merged head.
- No Linux execution of anything in this batch by me; PR CI unit-Linux is the only Linux evidence on the head.
- gate-228 receipt shows RED at p6-xtask-gen-check; the lane landed after a docs regen — treat A3's lane gate as passed-with-follow-up, not clean.