Docs / Evaluation

A result needs its method.

Ax0s does not publish an evaluation value here without a current method artifact. Previously published values remain withheld on this surface while rerun and publication gates are open.

Re-run pending.

The retrieval recall@10 evaluation is withheld until a fresh run and dated methodology artifact are published. The consolidation precision evaluation is withheld until a fresh run and dated methodology artifact are published.

This page intentionally contains no benchmark value, rank, verification stamp, or synthetic result.

What must travel with a future result.

Ax0s intends to publish methodology, null results, and killed ideas as part of its evidence policy.

  • A dated methodology artifact.
  • The evaluation setting and dataset variant.
  • The retrieval cutoff and scoring definition.
  • The answer and judge configuration when applicable.
  • Limitations, nulls, and disconfirmed approaches.
  • A current rerun linked to the method.

Interface checks are not benchmark claims.

A recorded 2026-07-14 fresh-store round-trip receipt covers write, recall, human-confirmed supersession, current recall, and history recall in one stated environment. It does not establish retrieval quality or a current full-suite run.

See the quickstart receipt boundary.

Read the surrounding evidence.

Open the claim registry, read the retrieval semantics under evaluation, or check the published operations state.