CORE-142 synthetic paired score projection

Status: synthetic local evaluation only. The source composition for Aboga and Cifria is unknown from this checkout. GitOps manifests and the existing CORE-141 protocol do not establish a live native CandidateSource or authorize staging access, data extraction, or transfer.

LocalCandidateCapture is an opt-in, one-query observer at RetrievalCoordinator. It records the legacy pre-floor final cue and the native score actually used by the coordinator gate. pair_captures joins source identities only in process and returns numeric decisions with UUIDv4 identifiers. An observed provider-call argument is needed on both sides before a selected flag can mean provider-visible admission; gate, rank, and hydration inclusion alone do not establish it. This is visibility at the model-call argument, not network delivery.

scripts/core141_corpus/capture_pairs.py takes those paired decisions, caller-supplied closed-vocabulary annotations, and a caller declaration of native source score semantics. It checks the observed final-score marker against that declaration and refuses unknown provider visibility, activation-only selection, and stage contradictions. It cannot authenticate the score's origin or detect a copied legacy cue disguised as native. The returned dictionaries contain exactly the fields accepted by scripts/core141_corpus/validate.py, with fresh UUIDv4 identifiers generated for each projection. The helper performs no file I/O and requires the literal boolean declared_synthetic=True. A synthetic test exercises the full local capture, pairing, and projection seam.

The caller must establish query class, memory layer, source, scope class, and blind relevance labels independently. This helper cannot establish those facts, app provenance, or privacy from a schema-valid row. A real paired Aboga/Cifria corpus remains blocked on verified app composition and the source-specific owner approvals in the CORE-141 staging protocol.