Compare, score, synthesize, retest, and plateau-stop candidate versions with source-grounded identity, target maps, calibrated objectives, validation-tier claim caps, evidence ledgers, recomputable score packets, artifact checksums, correction audits, and anti-compression safeguards.
