[immakermatty]

The canary record of v0.1.12-rc.2 and the evidence for the final v0.1.12. Plans DEV-6656 and DEV-6671.

Why

release.yml publishes a final only when:

• its candidate's canary record is merged into the default branch;
• the record's merge is in the tag's history;
• the tag differs from the candidate only under qualification/ and docs/evidence/ (docs/release-cycle.md, "The final tag").

This pull request adds that record and its evidence.

What changes

• qualification/canary/v0.1.12-rc.2.json (lazurio.canary.v1):
  • start 2026-10-25T18:44:36Z (the first reading that observed every Machine healthy on the candidate; the last Machine got it at 18:17:35Z), end 2026-10-26T02:45:25Z (8 h 0 min 49 s observed);
  • 5 work VMs and 1 personal VM, all healthy on 0.1.12-rc.2;
  • Machines are opaque ids.
• docs/evidence/release-v0.1.12.md:
  • qualification;
  • scope and how each Machine got the candidate;
  • the stage, with the readings at its start, at 21:50Z, 23:49Z and 02:19Z and at its end, plus the journals;
  • the real work inside the stage (G.3);
  • other changes on the same Machines during the stage;
  • per-Machine results;
  • the known limitation the final ships with (Buddy: a Folder render that changes his rules leaves the owed prompt refresh, so open conversations follow them (#565 point 4) #648).

What stays the same

Nothing outside qualification/ and docs/evidence/. main has been 8571aa71, the candidate, since the tag. The freeze held, so the final tag can stand on this record's merge: no release branch.

G.3 in short

• Start: at 18:44Z every work VM was healthy on rc.2, with NRestarts 0. The personal VM was healthy at 18:02Z.
• 18:45Z, every work VM: a Folder refresh (unchanged) and a deliberate Launchpad restart. Every unit came back healthy, with NRestarts 0.
• 18:46Z, module starts:
  • a static content module was served as a build;
  • the Organization's planning app ran from main as one production process through start_script, the first real consumer: "Built version · main · ", 11–46 MB instead of about 76 MB as a development server.
• 18:51Z: the one lazurio update (Update: one lazurio update brings content, builds and Lazurio itself up to date; Open synchronizes its module (DEV-6656, root 0212) #622) through Nová karta's Update (Nová karta: Aktualizovat runs lazurio update and shows its progress per module (DEV-6656, root 0212 point 7) #624), as the Owner test account. Everything showed "Up to date", Lazurio stayed on 0.1.12-rc.2 with no downgrade, and nothing restarted.
• 21:23–22:04Z, Buddy on the personal VM:
  • Začínáme, Seznamte se s Buddym, and Nastavení → Buddy (Buddy: Seznamte se s Buddym in Začínáme and Nastavení → Buddy, over his sign-in and persona (DEV-6676 PR 3, F65) #637);
  • one result message per delegation;
  • a targeted check of Buddy: a notice that brings nothing new gets exactly NO_REPLY (base-instructions-76, #612 point 3) #645's NO_REPLY, which ended exactly NO_REPLY at 21:59:40Z;
  • a persona change through Nastavení → Buddy (Buddy: the Platform derives his persona byte for byte as the installer and brings a change to running conversations when he is idle (DEV-6676 PR 2, F64) #636), which restarted only Buddy's gateway.
• End, 02:45Z: all six Machines were healthy on rc.2 with NRestarts 0. Each work VM's journal shows only the deliberate restart, and the personal VM's journal shows no start or stop at all.

Verification

• bun run scripts/qualify/check-canary.ts qualification/canary/v0.1.12-rc.2.json: "The canary of v0.1.12-rc.2 passed: 8 hours on every listed Machine."
• bun run check:public: PASS. [ ] PR checks
• Pablo's review on the exact head

Review round 1 (CHANGES_REQUESTED on 2e6ec23), fixed in 9f58434

1. The observed interval was 7 h 34 min, because the first direct reading came 27 min after the last apply. The observation continued to a fresh end reading at 02:45Z, and the record now counts from the first direct reading: 18:44:36Z–02:45:25Z.
2. Names of a test persona and modules are replaced with roles.
3. The final's wording: the same product source, differing only under qualification/ and docs/evidence/.

Review round 2 (CHANGES_REQUESTED on 9f58434), fixed in 380816e

The evidence now carries G.3's Recovery issue search (02:54:08Z, all states). The label search for 0.1.12-rc.2 and v0.1.12-rc.2 returned 0. One issue names the candidate, #650, a cosmetic launcher title. No issue mentioning recovery was created since the candidate.

Risks and follow-ups

• Known limitation (Buddy: a Folder render that changes his rules leaves the owed prompt refresh, so open conversations follow them (#565 point 4) #648): Buddy's stored prompts are refreshed by the rollout on every Environment with Buddy, and automatically from v0.1.13. On the canary, the web chat's resumed session needed one manual clear.
• T3 Code: the first open of Buddy's panel once showed T3 Code's generic home. This is an issue in the T3 Code fork.

🤖 Generated with Claude Code

[agentrozjedemeai]

Exact-head review of 2e6ec23: CHANGES_REQUESTED.

1. Canary start proof is incomplete (docs/evidence/release-v0.1.12.md:53-60). The record starts at 18:17:35Z, the last work VM apply finish, but the first documented recover/update status and Launchpad health reading on the five work VMs is 18:44:36-46Z. The last readings are 02:19:00-26Z. The common directly observed healthy interval is at most 7 h 34 min 14 s, not eight hours. The #159 apply readback confirms installation/release and Launchpad conformity, but the evidence does not tie that instant to the G.3 update status --json active candidate and healthy Launchpad / recover start gate. Please add a sanitized, per-Machine exact-start readback proving those conditions at the claimed anchor (and its source), or continue the observation through at least 02:44:46Z with fresh end readings and update the record. A syntactically passing check-canary.ts cannot attest the missing live interval.

2. Public evidence includes a named test persona and named Organization modules (docs/evidence/release-v0.1.12.md:71-74). The requested public-safety boundary is no person or Module names, with opaque Machines. Replace those with roles / anonymized module descriptions while retaining the build and start_script facts. The public reference Organization named by the release-cycle contract is not the objection; do not add private mapping or raw readings to this public PR.

3. Line 8 says the final tag carries exactly the candidate tree; lines 20-21 and the actual diff correctly say it adds files under qualification/ and docs/evidence/. Correct line 8 to say same product source with only the allowed evidence differences. No final tag exists yet.

The diff is restricted to those two allowed paths; the local exact-head check-canary.ts, check:public and git diff --check pass. Candidate Release 40209175549 and Qualify 40209771577 succeeded. This is a review only; no merge, tag or release.

[agentrozjedemeai]

@immakermatty Exact-head round 2 of 9f58434: CHANGES_REQUESTED.

The three round-1 blockers are fixed: the common directly observed interval is 18:44:36Z–02:45:25Z (8 h 0 min 49 s); persona/module names are generalized; and the final is described as the same product source with only the two allowed evidence paths. I reproduced the offline canary check, check:public and git diff --check; PR checks pass.

One missing release-contract evidence item remains: docs/recovery-mode.md G.3 (lines 600–607 and 623–629) requires checking that no public recovery issue names this candidate and including the issue-search result in docs/evidence/release-v0.1.12.md. That file reports “No Recovery mode” and journal/health observations but no issue-search result. I independently queried all issue states with repo:Lazurio/LazurioPlatform is:issue label:recovery "0.1.12-rc.2" and the v-prefixed variant; both returned total_count=0 at review time. Please add the sanitized query/result and observation time to the evidence file, so the historical release record carries this gate rather than relying on my review comment. No new machine/private mapping or raw logs are needed.

Review only; no merge, tag or release.

[agentrozjedemeai]

@immakermatty Exact-head round 3 of 380816e: APPROVED. The round-2 G.3 blocker is resolved: docs/evidence/release-v0.1.12.md now records the Recovery issue search with its timestamp, all-state candidate variants (0), the only candidate-naming issue #650 (cosmetic, not Recovery), and no recovery mentions since publication. I independently repeated the issue queries and read #650. The only delta from 9f58434 is those five documentation lines; the canary JSON is unchanged. Offline check-canary.ts, check:public and git diff --check passed in an isolated detached checkout. The exact-head Platform checks, candidate Release and Qualify runs succeeded; main remains 8571aa7, with zero unresolved review threads. Review only: no merge, tag, or release.