[bewaterhere-coder]

Summary

• Scan every declared source path for a Semantic Rule instead of returning after the first existing file.
• Aggregate token evidence across the bounded source set; a token found in any declared readable file is recorded.
• Keep not_evaluated when no declared source can be read.
• Add regression tests for evidence split across files and for a later file being ignored after an earlier file exists.

Why

semantic_rules[].paths accepts multiple paths, but the evaluator previously returned after the first existing readable file. Tokens located in later declared files were ignored, causing false review_needed results and incomplete evidence.

Boundaries

• The scan remains bounded by the existing per-file size limit and declared path list.
• Comment filtering remains conservative and is not a language parser.
• pattern_observed_control_flow_unverified still does not prove control-flow or runtime correctness.
• No API or schema version change; no new gate.

Validation

Python 3.10–3.13, MCP stdio E2E and Sample Project Advisory.