chore(release): bump version set to 10.0.12

[Jurij89]

Summary

Release-prep bump for 10.0.12, per RELEASE_PROCESS.md §3. Moves the root package and all 20 packages/* workspaces to 10.0.12 in lockstep and converts [Unreleased] into the 10.0.12 CHANGELOG entry.

Last gate before the comprehensive devnet run (§4) and the signed tag (§5). No functional changes — all 10.0.12 code landed via #2072.

Changes

• Version set → 10.0.12 (21 package.json files). Internal deps use workspace:*, so a partial bump would ship a skewed dependency graph.
• CHANGELOG.md: new ## [10.0.12] section with an ### Upgrading from 10.0.11 matrix. [Unreleased] is now empty.

Version-only: pnpm-lock.yaml untouched. engines.node ranges from 10.0.10 preserved.

Why the CHANGELOG was rewritten rather than extended

The prior [Unreleased] body documented #2006 only — 8 of the 9 PRs in the release had no entry — in a single 2,834-character bullet, under a ### Operator knobs heading that appears nowhere else in the file. Since §6c copies this section verbatim into the GitHub Release, that would have shipped 1-of-9 coverage. The new section covers all nine PRs with the #2006 headline cut to house length (1,043 chars).

Why an upgrade matrix

§10 requires guidance for a breaking or builder-impacting release. Five independently builder-impacting changes, each verified in the shipped code:

• A public export was removed from a published package — CATCHUP_BACKPRESSURE_RETRY_DELAYS_MS is in packages/agent/src/index.ts at v10.0.11 and absent at f9922f262. retryDelaysMs is now typed never (sync/catchup-policy.ts:145) and rejected at runtime (:245). No node is affected — the CLI depends on the agent as workspace:* — but external consumers of @origintrail-official/dkg-agent break at compile time and runtime.
• New 503 on subscribe (CATCHUP_SHUTTING_DOWN).
• DKGAgent.stop() now rejects on a drain timeout where 10.0.11 warned and continued.
• sync-global operation becomes <work class>:<source> — dashboards grouping on it need updating.
• Write routes validate a qualified contextGraphId at the point instead of falling back to a catalog scan; a qualified ID for an unsynced graph now gets a fast 400.

⚠️ One defect ships knowingly

Recorded under ### Known issues, not silently:

On sync-global, per-lane state in GET /api/diagnostics/backpressure cannot reflect a full global queue. packages/agent/src/sync/priority-admission-queue.ts:177-180 calls updatePressureCapacity({ queueLimit, inflightLimit }) with no lanes map, while packages/storage/src/store-priority-scheduler.ts:313 does pass one; packages/core/src/backpressure-observability.ts:361-362 reads lane limits only from capacity.lanes?.[lane], so both the queue-full and utilization branches are dead for every sync lane. A saturated sync-global scheduler can report every lane healthy until queue age or a rejection moves it.

Scope is the per-lane state field only — the scheduler rollup that /api/status publishes is computed from totals.queueLimit and is correct, and the store scheduler is unaffected. All 18 review threads on #2003 remain unresolved. Worth deciding whether this blocks the tag or ships as documented.

Prior-review follow-through

Concerns raised on earlier reviews of this work, checked against the shipped code:

Concern	PR	Verdict
Retry loop's updated_at touch defeats the TTL prune, removing a self-heal	#2002	RESOLVED — prune unchanged; a bounded 7-day retry budget now retires a row to REJECTED
CAS after apply ⇒ double-write	#2002	NOT REPRODUCED — ordering unchanged by this PR
Jammed sync lane reads healthy	#2003	NOT RESOLVED — see above
Gates added by W2a are green-and-inert	#2051/#2059	RESOLVED — mutation-tested; the abandoned form survived the same mutation
Catch-up early stop must be curator-only, empty must be whole-round	#2007	RESOLVED — verified at catchup-runner-worker-impl.ts:490 and catchup-runner.ts:802-841

Green-and-inert gates: none found. Every newly added gate in the range was probed for whether it can actually fail.

Open before tagging

• §4 comprehensive devnet gate has no recorded artifact for f9922f262. CI green there (59 checks, 0 failures) is not that gate.
• feat(sync): W1 measurement contract — source-attributed sync instruments (I1–I9) #2033's W1 testnet merge gate ran and passed, but the report lives in a git-ignored run-reports/ directory, so it is not auditable from the repo. Its own caveats: M6/I9 blocked on Windows, M3 and I6 not exercised.
• fix(agent): relieve repeated sync pressure #2053's post-merge Blackbox conductor run (its PR checklist line 163) is unchecked and was overtaken by the promotion.

Test Plan

• pnpm release:verify-versions --version 10.0.12 → passes, 21/21
• Version-only diff; pnpm-lock.yaml untouched; engines.node preserved
• Removed-export and lane-attribution claims verified directly against both refs
• CI green
• §4 devnet gate on the resulting main commit

Related

All 10.0.12 code landed via #2072. #1985 and #1971 remain open and are not in this release.

🤖 Generated with Claude Code

[otReviewAgent]

Review Agent completed this review and found no issues.