Receipts for Wet-Lab Claims: Proof-Carrying Experiment Logs for AI-Designed Antibiotics

AI systems can propose antibiotic candidates faster than labs can close claims about them. The bottleneck is not generation; it is scientific closure that survives travel: across labs, platforms, instruments, and incentives, without collapsing into narrative disputes. We use antibiotics as a stress-case: the contribution is a receipts-first closure architecture applicable to any health wet-lab pipeline under drift. This paper proposes Proof-Carrying Experiment Logs (PCEL): a receipt-first evidence architecture that turns each wet-lab claim into a contestable, auditable artifact rather than a story attached to a figure. A receipt (ρ) binds provenance, model context, synthesis trace, protocol versions, instrument calibration, controls, raw data, statistical decisions, and cryptographic commitments into a single unit that a third party can verify under a stated verifier budget. We define transportability (τ) as a pass/conditional/fail decision under explicit context shifts and admissible drift, and we define closure gates that output ALLOW / HOLD / ESCALATE / REJECT as first-class outcomes. The result is a pipeline where evidence “travels” with bounded revalidation cost, negative results remain visible, and overclaiming becomes measurable. We outline an evaluation plan from bench replication to platform portability, emphasizing cross-lab verification time, drift stress tests, and reductions in downstream failure rates. PCEL is a governance layer for translational biology that scales throughput without sacrificing integrity.