ACE v0.1.0 is now available on PyPI.
The core workflow is deliberately simple:
- preflight an experiment before results exist
- retain the trial evidence needed to review a result
- assess whether a claimed win is trustworthy, inconclusive, or unsupported
This release includes ace assess, which imports retained JSON or CSV trial outputs and checks them against the declared experiment contract. Decision packs fail closed when essential identity, provenance, split, seed, baseline, metric, failed-trial, or required-statistical evidence is missing or mismatched.
ACE validates supplied evidence. It does not run workloads, independently reproduce results, certify systems, or turn a benchmark contract into a performance claim.