Swift implementation status

Swift examples and independent CodeQL and Joern execution evidence are available in the repository. They have not yet entered a published release freeze. The current snapshot remains authoritative for published benchmark participation.

Analyzer language breadth, implemented fixtures, completed native queries and qualified benchmark outcomes are separate claims. In particular, Bifrost Swift support remains unsupported. Native path observations do not establish qualified correctness or semantic completeness.

• Browse the Swift examples.
• Inspect the historical v1 registry reconciliation, including the original unresolved families and links to retained evidence.
• Inspect the v2 fixture additions, with separate native and Result validation boundaries.
• Read the Swift applicability contract.
• CodeQL activation and execution.
• Joern activation and execution.

The v2 population adds 12 platform-native examples and two Result extension examples to the original 90. Native examples are compile-only; the Result pair has bounded source-dependence controls. Independent analyzer qualification and execution for these additions remain pending. Two opaque modeling identities remain unresolved, with no accepted exclusion or completion of the Swift epic. Real-project work requires its own selection and review; reserved tracks remain inactive.

A future snapshot must follow the normal pin review, evidence, immutable freeze and result-generation process. Inconclusive, unsupported and runner-error outcomes remain separate from clean negatives. A partition with no definitive outcomes has no correctness rate; publishing honest coverage does not require inventing one.

The prospective native integration plan tracks the next CodeQL qualification dependencies with separate vendor-native and adapter-assisted lanes. The candidate configuration remains pending, and Joern’s external-property binding gap remains blocked. These diagnostics add no published results.

The canonical source–sink qualification retains a separating positive/negative observation for the adapter-assisted CodeQL lane. It remains non-scored; vendor-native observations and the pending activation state are reported separately.

The canonical entrypoint qualification treats the command-line count guard and indexed source separately, retaining vendor-native and adapter-assisted observations without scored promotion.

The independent append control is blocked by a retained false flow through a constant-body lookalike. Native path and shipped heuristic evidence explain the precision gap; the canonical propagator pair remains unrun.

The independent integer sanitizer control is also blocked: a string-holding local numeric lookalike is admitted as a barrier, and both local wrapper flow controls are missing. These observations remain separate; the canonical sanitizer pair has not run.

The independent Data roundtrip control is blocked by a false flow through constant-body local lookalikes. Resolved identities, attached ports, simple transfer edges and OptionalSome stores are retained separately; the canonical summary pair remains unrun.

The independent UserDefaults control is blocked by missing same-key flow and scoped model engagement despite resolved calls and literal keys. The direct baseline flows; the canonical persistence pair remains unrun.