D5 url:buildtest restoration investigation stopped for arbitration
RE-VERIFY: This is a point-in-time snapshot written 2026-07-31T09:51Z. Re-check the live route with ContainerNursery CLI and re-read /root/cn-watchdog.audit before taking any action. The live route, host session state, branch state, and PR state can change after this snapshot.
Mission summary
Identify what replaced the merged-main CPU-capped BuildTestServerService JAR on url:buildtest: at 09:25:19Z, and resolve the cause without another blind redeploy. D5 was authorized to investigate and to make one source-level artifact-pin correction only if the owner was an automated watchdog. It was not authorized to fight an active deployment owner.
What was found and done
The restoring mechanism is an active agent deployment session, not a host watchdog or a route configuration pin.
- The audit records
2026-07-31T09:25:19Z INTENT+ACTION [claude-fable-emergency-lane] SECOND RESTORATION and 09:26:11Z CONFIRMED.
- The route JAR was written at
2026-07-31 09:25:39.290863075 +0000; ContainerNursery then started child PID 2980211 at 09:25:43Z from that path.
- Systemd recorded root SSH sessions from
104.248.75.129, fingerprint SHA256:FB8QdHFzOWYEJnIEqgfjRzimOqUFsIjnC1RtRDK5HCE, at 09:25:19, 09:25:26, 09:25:36, 09:25:39, and 09:25:47. The last session closed at 09:26:11, matching the confirmation audit entry.
- The corresponding Claude Fable branches are BuildTestEmbedded
emergency/bte-451-454-superset-20260731 at 626453e7bb3dc4842fedfc53da8777cdb5fd20a9 and BuildTestServerService emergency/btss-deploy-20260731 at 32fe6dec2c9337dfa379c837f30a8fde61090d62. Their build scripts select embedded 0.0.61400.
- The user-named source branches
fix/startup-recovery-readiness and fix/reconcile-recovery-fleet-and-renewals contain recovery source/tests only. Their inspected diffs have no upload-jar, route mutation, expected-SHA, or guardian code.
/root/cn-watchdog.sh only checks ContainerNursery health and restarts the ContainerNursery JAR. The root crontab has only the watchdog, log guard, and disk rotation jobs. Systemd has no route guardian. No expected-SHA or pinned-jar rule was found, and no restore/deploy/guardian process was active.
- The current route JAR is SHA-256
46496e074dd58547dd60db57a1005d5cadba3c4d157db728fe277c28fb032283, 146,883,951 bytes, and is the uncapped 0.0.61400 artifact. D4's exact bytecode inspection of this SHA found no getVcpus() and no Math.min cap. The merged-main capped artifact SHA is 547aa67e63e824c2d4e692cb58c46a739ceb94d14d30331d9102377fb1c5ab53, preserved separately.
- D5 appended and verified a direct message at
2026-07-31T09:51:01Z in /root/cn-watchdog.audit, naming the source IP/session, branch heads, both SHAs, and the no-redeploy boundary. The message asks the active lane to acknowledge there and stop counter-deploys while ownership is resolved.
- D5 made no route, watchdog, process, artifact, run, or check changes. No redeploy was performed.
Relevant PRs and refs
- CPU-cap PR: https://github.com/CodexCoder21Organization/BuildTestEmbedded/pull/449 — merged; merged head https://github.com/CodexCoder21Organization/BuildTestEmbedded/commit/b6d02b8cd4da343f682a00e39d3d462eadf82bcb
- BuildTestServerService pin PR: https://github.com/CodexCoder21Organization/BuildTestServerService/pull/272 — merged.
- Recovery PR: https://github.com/CodexCoder21Organization/BuildTestEmbedded/pull/459 — open; branch https://github.com/CodexCoder21Organization/BuildTestEmbedded/tree/fix/startup-recovery-readiness
- Recovery/renewal PR: https://github.com/CodexCoder21Organization/BuildTestEmbedded/pull/455 — open; branch https://github.com/CodexCoder21Organization/BuildTestEmbedded/tree/fix/reconcile-recovery-fleet-and-renewals
- Superset source PRs: https://github.com/CodexCoder21Organization/BuildTestEmbedded/pull/451 and https://github.com/CodexCoder21Organization/BuildTestEmbedded/pull/454 — open.
- Related service PR: https://github.com/CodexCoder21Organization/BuildTestServerService/pull/271 — open.
- Do not touch: https://github.com/CodexCoder21Organization/UrlProtocol/pull/457
- Existing first-overwrite challenge: https://github.com/CodexCoder21Organization/PlanRepository/blob/main/challenges/2026-07-31-0850-concurrent-url-buildtest-deploy-overwrote-an-authorized.md
- New second-overwrite challenge: https://github.com/CodexCoder21Organization/PlanRepository/blob/main/challenges/2026-07-31-0951-second-url-buildtest-restoration-was-an-agent-ssh-deploy.md
Next steps
- Supervisor reviews the audit message and decides route ownership with
claude-fable-emergency-lane.
- Do not redeploy, re-drive checks, or edit the watchdog until the supervisor explicitly resolves the active-agent conflict.
- If the supervisor confirms a single owner and authorizes the merged-main artifact, re-verify the live artifact and route state, then follow a fresh explicit deployment instruction. The D5 one-redeploy allowance was not used.
- Only after a valid capped generation is stable should the seven preserved D4 targets be re-requested:
aefdc0d7, 4f748d52, e33a0f2e, 6c4ad8e5, 1997e4c3, 5e4e0255, and adabf765. No D5 re-drive outcomes or width numbers exist.
Reusable operational knowledge
- The route artifact path is shared:
/root/ContainerNursery-uploads/jars/buildtest-server-service.jar. Its mtime and SHA are useful for correlating an upload with a host session.
- ContainerNursery logs the route replacement in
/root/ContainerNursery/data/logs/container-nursery.log, but did not log the upload caller in this incident. Systemd journal SSH records supplied the client/IP/session timing.
- The D4 findings and this D5 record are in
/tmp/claude-1000/-code/582d2178-a74b-4275-9371-4f5ef384139b/scratchpad/out/findings-d.md; the permanent challenge record is the authoritative external summary.