Reproducibility
Reproducibility asks not whether an asset looks plausible, but under which inputs, environment, and commands it produces a comparable result again. One success or a self-reported confidence score is not sufficient evidence.
Evidence to retain
- Asset ID, source version, and trigger signals.
- Operating system, Node.js, Evolver, dependency, and repository versions.
- Validation commands, working directory, exit status, and key output.
- Touched files, external network effects, and recovery result.
- Reuse feedback classified as success, missing, mismatched, or failed.
Inspect reuse quality
bash
evolver recall-verify-report
evolver reuse-report
evolver asset-health
recall-verify-report summarizes success, missing and mismatched recalls, errors, latency, asset age, and a GREEN, YELLOW, or RED gate. reuse-report is based on observed reuse outcomes and can expose assets that are frequently recalled but not proven useful.
Reproduce an asset
- Freeze the input, versions, and permitted effects without editing the asset first.
- Repeat once in the original environment, then use one meaningfully different environment.
- Compare output, tests, change scope, and side effects rather than exit code alone.
- Report failures as outcomes and distinguish environment mismatch, staleness, and a faulty strategy.
Decision rule
Cross-node success matters only when inputs and verification are equivalent. More environments do not automatically mean higher quality. Weak validation, unlocked dependencies, or manual repair should block automatic promotion.
Related pages
EvoX Docs · Security · Asset trust