Quirl 0.4.0 native release evidence

Recorded 2026-09-30 for the published v0.4.0 release. All four native artifacts passed the enforcing release benchmark with 101 successful PTY samples each. These results measure the exact published executables on the hosted runners below; they do not establish universal latency guarantees.

Release identity

• Candidate and resolved tag commit: 595f1536e908227a1076d274ea5107c7dfefda7f.
• Native build and performance run: 36668853231, attempt 1.
• GitHub release ID: 399726821; published at 2026-09-30T04:49:11Z as a nondraft, non-prerelease, immutable release.
• Release manifest SHA-256: 91e3b71bedfc9a0cd9769a22ebea635655186fc6c4714c7c48c9992e76343b36.
• Every entry in the bundle's SHA256SUMS verified before publication was approved, and the four formula hashes in tap PR 28 matched the archives byte for byte.

Each report records the candidate as its clean source revision, with artifact digest, profile, source, and harness identity checks passing.

Native results

All values below are milliseconds except executable bytes. Limits were startup P50 ≤25 ms, first-prompt paint P95 ≤21 ms, and keystroke-to-frame P95 ≤8 ms. Each row completed 101 of 101 requested PTY samples with no recorded sample failure.

Size follows ADR 0036: there is no default executable-size ceiling. All four sizes exceeded the advisory 8 MiB threshold and retained warnings; macOS ARM64 grew by 16,704 bytes over 0.3.0.

Runners and toolchain

All builds used Rust 1.97.1 (8bab26f4f, 2026-07-14), LLVM 22.1.6, and Cargo 1.97.1 (c980f4866, 2026-06-30) with the official release profile: optimization z, fat LTO, one codegen unit, stripped symbols, and panic = "unwind".

Exact report files

These are byte-for-byte copies of the schema 8 reports uploaded by the native jobs.

Published executable hashes

Each digest below was computed from the bin/quirl bytes inside the published archive.

Preserved earlier attempts

Earlier candidates did not publish a tag or release:

• Run 36657374419, candidate 75c53fe595b6ab5322b07d4db64b9981ac1e4bc5, passed three targets but missed the macOS ARM64 first-prompt P95 on all three bounded attempts: 24.994959 ms (attempt 1), 26.093459 ms (attempt 2), and 23.983458 ms (attempt 3), against the unchanged 21 ms limit. Their report SHA-256 values are 03cc21293e4d3777922b51cc9cdcb96befcc77d27af7fe8dfae7d9605acc6803, d1b611ba9edb2398c4cbb45ad14aafc23e25d1482f23db01bc2630cfb0622525, and 6af92fc56da70607116e29f81e2958dd8be0cb698307be9942fc215cb724375e.
• Run 36665750746, candidate 56debdea32e10c153c0368a10a7ec3763e2a0969, added a bounded wait for the load average to settle; the runner stayed at load 7.61 on 3 CPUs after four minutes and missed again (28.21 ms).

To separate a code regression from the environment, temporary A/B runs on the same runner type built the published v0.3.0 source and the 0.4.0 candidate and measured them interleaved. Straight after compilation both missed (v0.3.0 23.25 ms, 0.4.0 23.79 ms). After the load settled both passed (v0.3.0 12.65-18.42 ms, 0.4.0 12.17-14.54 ms). A diagnostic run then showed mds, mds_stores, and mdworker using over 200% CPU while Spotlight indexed the checkout and the fresh build; with mdutil -a -i off the load average fell from about 10 to 1.4 within 90 seconds. The release workflow now turns off Spotlight indexing on macOS runners before building and waits for a settled load before measuring, as documented in the release procedure. No size or latency budget was relaxed.

Method and limitations

The PTY harness uses a private, initially empty home, configuration, history, catalog, project database, and XDG directories shared across its samples. Each sample starts a fresh native process in a 120×40 terminal, answers cursor-position queries, reconstructs the screen, checks the prompt and exact binary identity, then proves editability. Percentiles use nearest-rank selection. Hosted scheduling, virtualization, CPU frequency, other machine load, and filesystem caches are not controlled beyond the documented indexing and settling steps. The reconstructed PTY frame does not measure physical terminal-emulator scheduling or display. This automated evidence does not cover every terminal, visual accessibility scenario, clipboard, IME, or sustained human session, and no human checklist item is inferred from it.

The website recording still shows the v0.3.0 release package; it was not re-recorded for 0.4.0.

Historical 0.3.0 measurements retain their original artifact identities, methodology, limits, and outcomes.