Skip to main content

COMPEL Glossary / replicability

Replicability

The property that an independent team reproduces the qualitative conclusions of an experiment using different data, tooling, or implementation.

What this means in practice

A stronger claim than reproducibility; key to scientific trust in published model-performance numbers but frequently unmet in applied ML due to proprietary data and configuration.

Synonyms

independent replication , experimental replicability

See also

  • Reproducibility — The property that re-running an experiment with the same code, data, and configuration produces the same results within declared tolerance.
  • Experiment tracking — The infrastructure and practice of recording artifacts, metrics, parameters, environment, and lineage for every experiment run — enabling later reproduction, comparison across runs, and audit.
  • Benchmark contamination — The presence of benchmark test data in foundation-model training corpora — whether through web crawling or deliberate inclusion — inflating reported benchmark scores and breaking the comparability of benchmark results across models.

Related articles in the Body of Knowledge