v0.1.0 readiness
The first maturity release remains a prerelease. It is not a v1.0 stability promise. Release only after every required check passes on the tagged commit; link that commit's workflow runs and artifacts in the GitHub release notes.
| Contract | Evidence |
|---|---|
| Authorization audit | #268 closed; Rust quality gate runs room authorization regressions and the route-wide semgrep rule |
| Private intake | Repository private vulnerability reporting enabled; SECURITY.md defines report classes and the supported line |
| State-resolution correctness | Daily Release readiness runs state_res_equivalence against ruma, preserving the retired merge's counterexamples, plus the exhaustive fork-window oracle; Matrix compliance runs contested-fork interoperability with Synapse |
| Federation adversity | #16 and #225 closed; Rust quality and Matrix compliance exercise fork recovery and interop |
| Real lifecycle boundary | Release readiness builds schema-2 commit 496a77a029026a3acd087b4851c42e746bd9b9db and the candidate; its release-readiness artifact records actual backup, restore, dry-run, schema-2→3 migration, media/state verification and old-backup rollback |
| Performance | #42 closed with the published client-workload comparison and counting discipline in benchmarks.md; Benchmarks publishes the tagged candidate's measurements |
| Operators | rate-limits.md inventories limits and remaining account-driven storage growth; operations.md specifies deployment controls and lifecycle.md the offline recovery procedure |
The Rust quality gate also runs side_stream_delivery: paused writers
cannot advance a sync token, limited to-device batches retain omitted
messages, and failed writers cannot stall the watermark. A drained-mailbox
restart test checks that future events remain above issued tokens after
queue rows disappear. Call-churn
measurements count unique senders and await background tasks.
Boundaries to disclose
The retired slot-wise merge is not equivalent to Matrix state resolution. Counterexamples are permanent tests; forks with different states use the room version's resolver. A release must not resurrect the retired merge or claim its disproved theorem. ADR 0005 and the source tests explain that decision.
The workload for first-sight keys regressed at 50,000 slots after the key intern change. benchmarks.md publishes the raw baseline/candidate measurements from alternate rounds and their controls. Extensions to the federated comparison remain in #647. The completed #42 client workload is the maturity gate's comparison, not universal parity.
Account-driven storage growth has no general per-account quota. Run closed registration, a bounded ingress, process memory limits and a monitored storage quota. Do not describe an unrestricted open-registration deployment as covered by this release's operator readiness.