Performance evidence

Release measurements apply to one exact executable, source revision, harness, and runner. A later documentation commit records that evidence without rebuilding or relabeling the released artifact.

• Quirl 0.5.1 native release evidence: all four published native executables passing on their first attempt, with the recording and screenshots made from the macOS ARM64 executable.
• Quirl 0.5.0 native release evidence: all four published native executables, unchanged reports, and a preserved macOS x86_64 first attempt that a local A/B traced to runner variance.
• Quirl 0.4.0 native release evidence: all four published native executables, unchanged reports, preserved failed candidates, and the Spotlight indexing finding on the macOS ARM64 runner.
• Quirl 0.3.0 native release evidence: all four published native executables, unchanged reports, preserved failed candidates, and exact release identities.
• Quirl 0.2.0 native release evidence: historical measurements of the published 0.2.0 executables.
• Earlier release performance research: historical candidate measurements; see the recorded artifact scope.
• Preview benchmarks and project discovery: scoped engineering experiments.
• Embedded-language selection and Steel/Lua/Fennel comparison: runtime research, separate from product release gates.

The project records executable size and advisory warnings without a default size ceiling. Latency, artifact identity, sample completeness, bounded retention, and process cleanup remain release requirements. Hosted PTY results and automated recordings do not substitute for human or physical-terminal review. The release checklist explains the candidate and evidence-commit distinction.