Certifying when decision-time information justifies adaptive experimentation

2026-07-30

AI Summary

This work addresses the risk that naive reliance on interim information in adaptive experimentation can lead to invalid or high-risk decisions. To mitigate this, we propose the OPAL framework, which employs pre-commitment contracts to authorize nontrivial adaptive measurements while ensuring target risk remains controlled and expected returns stay positive. We establish the first impossibility boundary for adaptive authorization and introduce a target-calibrated recovery method, positioning authorization as an independent safeguard layer for safe adaptive science. OPAL integrates pre-commitment contracts, conditional outcome shift analysis, target-calibrated estimation, and worst-case value validation. Evaluated on Cell Painting data from 11,265 compounds, OPAL identified 595 candidates, captured 384 valid opportunities with strictly positive returns, and maintained a false activation rate capped at 5.18%, significantly outperforming six baseline methods.

Abstract

Adaptive laboratories choose measurements during experiments, yet most methods begin after adaptation is permitted. We introduce Opportunity-aware Policy Authorization for Laboratories (\OPAL{}), a framework that decides whether adaptation should be enabled at all. \OPAL{} uses a precommitted contract to require non-trivial adaptation, controlled target risk and positive executed value after cost. We establish an impossibility boundary: source outcomes and unlabelled target covariates cannot uniformly support non-trivial authorization under unrestricted conditional outcome shift, and derive a target-calibrated recovery. Applied to an unseen 11,265-compound Cell Painting partition, the frozen gate selected 595 compounds, captured 384 positive opportunities and achieved strictly positive executed value under least-favourable completion; its 5.18\% false-activation upper bound remained below a 7.5\% limit. Among six methods, only \OPAL{} combined non-zero activation with this risk control. Locked pharmacogenomic and finite-campaign studies distinguish policy misalignment from non-certifiability, establishing authorization as a distinct layer for safe adaptive science.

Problem

• adaptive experimentation
• decision-time information
• policy authorization
• risk control
• non-trivial adaptation

Innovation

• adaptive experimentation
• policy authorization
• conditional outcome shift
• target calibration
• false-activation control