Follow-up from Wave 3 / PR #126 (#116). The inlineParallelCapable fingerprint flag is wired through the AST detector path (where #116's acceptance criteria are met) and into core-scanner/EndpointRecord, but is not yet threaded into several detection-layer consumers that already handle batchCapable. None is a regression; the default AST path produces the correct "use the n/count parameter" suggestion.
Gaps (each parallels an existing batchCapable site):
Refs: #116, PR #126.
Follow-up from Wave 3 / PR #126 (#116). The
inlineParallelCapablefingerprint flag is wired through the AST detector path (where #116's acceptance criteria are met) and intocore-scanner/EndpointRecord, but is not yet threaded into several detection-layer consumers that already handlebatchCapable. None is a regression; the default AST path produces the correct "use the n/count parameter" suggestion.Gaps (each parallels an existing
batchCapablesite):src/scanner/patterns/openai-compatible.tsreadsreg?.batchCapablewith a hint-regex fallback but has noinlineParallelCapablepath;src/scanner/local-waste-detector.ts(the regex-only waste detector) has no inline-parallel finding. Net effect today: inRECOST_DISABLE_AST=1/ regex-only files, a DALL·E call in a loop emits no finding (previously a wrong "batch" finding — net-neutral, but the correct n/count suggestion isn't produced via this path).src/scanner/patterns/utils.tsdedup key includesbatchCapablebut notinlineParallelCapable; couple this with the pattern-path fix so distinct matches aren't collapsed.src/intelligence/types.tsApiCallNode+src/intelligence/builder.tscoercebatchCapablebut dropinlineParallelCapable, so AI-review context can't distinguish the two shapes.Refs: #116, PR #126.