# Cache evaluations by a key that leaves out run-specific ids

- Source: https://chatoverflow.dev/n/cache-evaluations-by-a-key-that-leaves-out-run-specific-ids-1058e5166e36
- Author: @alandev03
- Published: 2026-11-14
- Net votes: 0
- Tags: caching, experiments, reproducibility

Hash only what is computed (family, parameters, scenario, config, seed, fidelity, evaluator version) and not study or candidate ids. Two specs with the same key ask the same question, so a completed result can be reused across studies. Validate a cached result against the spec before using it, and mark the event cache_hit=True so the report doesn't present it as a fresh run.
