0.1.0 exact-candidate performance record

Current exact-candidate evidence: This record measures candidate 23fd5d36907fc816bdafd9aa3c2dcb3afb69feb5 and artifact 9a893a5f1a0b49d62712f331c88966113d910d94efa9651dc4feffe9fd55b637. It is the current exact-candidate automated performance record. Human Linux/macOS terminal, remote-PTY, and real-terminal demo review remain incomplete.

Current exact-candidate evidence. This automated measurement record applies only to candidate 23fd5d36907fc816bdafd9aa3c2dcb3afb69feb5 and its named artifact. Evidence commit B records candidate A's results; it is not the measured binary source. Every automated merge-relevant gate passed. This is not a claim that human terminal, Linux, remote-PTY, or candidate-demo review occurred.

Measured: 16 August 2026 at 23:27:15 UTC

Candidate A: 23fd5d36907fc816bdafd9aa3c2dcb3afb69feb5 (tree 613b948814d2fecd91fbe170c5b13b699e89ff59, clean)

Artifact SHA-256: 9a893a5f1a0b49d62712f331c88966113d910d94efa9651dc4feffe9fd55b637

Measurement platform scope: macos-15.7.9-24g830-apple-m2-pro-aarch64-apple-darwin-automated-pty

Automated performance gate: passed all release budgets

Overall release readiness: blocked pending human and platform review. Named human terminal review on Linux and macOS, remote-PTY review, and human review of a candidate-derived real-terminal demo remain unavailable. No tag, release, upload, deployment, publication, or remote operation was performed.

This record measures the exact quirl 0.1.0 artifact named above. Report schema v7 rejected tracked or untracked source changes, required an independently supplied SHA-256, copied the artifact into a private read-only staging directory, and verified the staged copy before execution. It also verified the artifact's embedded source revision, profile, optimization level, panic strategy, operating system, architecture, and the independently embedded source identity of quirl-bench.

The harness ran the actual release binary in a fresh pseudo-terminal for every end-to-end sample and reconstructed terminal frames from the PTY byte stream. It did not measure terminal-emulator scheduling, GPU composition, monitor scanout, or human usability.

Source, toolchain, and artifact identity

The digest was computed independently with shasum -a 256 and supplied to both enforcing harness entry points. The canonical JSON run reported artifact_digest_verified, artifact_profile_verified, artifact_source_verified, and harness_source_verified as true. The digest printed by the reproduction, the artifact's hidden build report, the harness source identity, and the independently recorded candidate commit all matched.

Measurement environment

These measurements ran on actual local macOS hardware. They are automated PTY evidence only, not a named human macOS terminal signoff. No actual Linux hardware, named Linux terminal reviewer, SSH session, or remote PTY was used.

End-to-end release budgets

Times are wall-clock milliseconds. Percentiles use nearest rank over 101 independent fresh processes. All samples completed successfully before the phase timeout, and every enforcing release budget passed on the single canonical run.

The first-prompt gate uses P95. Binary units are MiB, where one MiB is exactly 1,048,576 bytes. This artifact is 566,608 bytes below the 5 MiB ideal limit, 3,712,336 bytes below the 8 MiB soft cap, and 5,809,488 bytes below the 10 MiB hard ceiling.

The one independent cargo xtask release-gate reproduction also passed all budgets: editable-prompt P50 21.704 ms, keystroke P95 2.364 ms, and first-prompt P95 19.311 ms, each with 101/101 valid samples. That reproduction confirms the gate result; it does not replace or average the canonical JSON measurement.

Stream retention evidence

LiveBuffer received 100,000 bounded typed samples at each supported window capacity. Retained samples equal min(input, capacity) and dropped samples equal input - capacity; this demonstrates O(window) sample retention. It is not an RSS, allocator, record-size, or producer-backpressure claim.

The production buffer invariant was valid at every tested capacity and the enforcing harness accepted the stream-retention gate.

Supplementary diagnostics

These probes are exact raw aggregate values from the canonical enforcing run. They are diagnostic only and cannot satisfy an end-to-end release gate.

Commands and outcomes

Every duration below is the observed /usr/bin/time -p real duration. The single canonical JSON benchmark and the single independent reproduction both returned zero; neither was retried.

The post-generation idempotence checks and post-commit semantic attribution and website results are reported with evidence commit B.

Missing evidence and blockers

All automated merge-relevant gates passed. Overall release readiness remains blocked on:

• a named human reviewer completing the real-terminal checklist in an actual macOS terminal application, including resize, narrow, custom-theme, config web, foreground-terminal, and remote-PTY observations;
• the same named real-terminal and process/job-control review on actual Linux hardware;
• remote-PTY observations on candidate A; and
• human review of a candidate-derived real-terminal demo before any README or release asset is published.

The automated real-PTY harness passed all 11 scenarios against the candidate source: rich editing, deferred catalog admission, catalog-failure terminal restoration, completion, integrated runtime, rich regressions, native job control, noninteractive dialect islands, suspend/resume, fallback selection, and NO_COLOR semantic hints. That is supporting automated evidence, not human or cross-platform signoff.

Immutable historical P20 evidence

P20 remains historical evidence only for corrected candidate 631d7849a22d04a028cbe90315dd63b8a526b03a, tree 25508427d228209ff026fdcd89f62e30ee73703c, and artifact 6ab5d8c5069956b5b5fe2d0b37b50e7d3c2a287df33a36ae75b88c87f900d8a0. Evidence commit 31e9b9b9c6ee8b0b7fdcc8f4ebc7aabbdb006bf8 preserves its full measurement record, environment, commands, and aggregate results. It measured on 16 August 2026 at 21:54:29 UTC with the same automated-macOS-PTY platform scope. Its enforcing run had 101/101 valid samples but failed the first-prompt budget at P95 21.140250 ms, a 0.140250 ms miss; editable-prompt P50 was 23.557333 ms and keystroke P95 was 2.229959 ms. Its independent reproduction also failed only first-prompt P95 at 21.463 ms. None of those timings, that failure state, its environment, or its artifact identity is attributed to candidate A in the current record.

Immutable historical P14 evidence

P14 remains immutable historical evidence for candidate 7bf188344ca61798a3cd8657787eacb8ec26ef84, tree 60df4fe8309f624642273ca8981a43d95217191e, and artifact 81cd33388cf610a7aac23a9781dbf2771b5dfb6b01b17522c2257cd3676d0ae6. Evidence commit 05df4b09349394cfd34f24d514e0e005365d0ed8 preserves its full measurement record, environment, commands, and aggregate results. It measured on 16 August 2026 at 18:51:07 UTC with platform scope macos-15.7.9-24g830-apple-m2-pro-aarch64-apple-darwin-automated-pty and passed the budgets then observed. None of its timings, pass state, environment, or artifact identity is attributed to candidate A in the current record.

An earlier record for commit c5a8d757a35a92a9a269686a1cd166c5a486e2b3 and artifact 12989bce23eccda4330cd815f5b234fb7c59e5c5cc23f5f93987592af3f0341e is also historical only. None of its measurements or pass state is copied into this candidate's evidence.