XPerf
LanguagesSwift

swiftperf evidence policy

Gate

A rule may be treated as error only with evidence/<RULE>.md + .jsonl and an explicit policy decision. Measured JSONL alone does not auto-escalate. Batch B (2026-10-22) kept package defaults.

Current map

As of 2026-10-22, Batch B G4.

RuleSeverity (default)EvidenceSmell
sp-print-hotoffsp-print-hot.mdabout 10× versus a String, stdout at /dev/null; stays off
sp-array-no-reservehintsp-array-no-reserve.mdabout 5× on many small arrays; about 1.6× on one large fill; stay hint
sp-dateformatter-hothintJSONL present (sp-dateformatter-hot.md + .jsonl); stays hintdate formatter constructed in a loop; not corpus-triaged
sp-json-decode-hothintJSONL present (sp-json-decode-hot.md + .jsonl); stays hintJSON decoder or jsonObject in a loop; not corpus-triaged
sp-regex-hothintJSONL present (sp-regex-hot.md + .jsonl); stays hintregular expression constructed in a loop; not corpus-triaged
sp-file-read-hothintmeasured (sp-file-read-hot.md + sp-file-read-hot.jsonl); stay hintcontentsOfFile in a loop; same path only; not corpus-triaged
sp-clock-hothintmeasured (sp-clock-hot.md + sp-clock-hot.jsonl); stay hintDate() in a loop; not corpus-triaged
sp-value-for-key-hothintmeasured (sp-value-for-key-hot.md + sp-value-for-key-hot.jsonl); stay hintvalue(forKey:) in a loop; not corpus-triaged

The three rows added 2026-10-23 stay hint. Each now has a JSONL microbench. They are outside the 2026-10-22 corpus-ready verdict, and the ratio is one shape, not a general speedup.

On this page