evidaxis/momentum-snapshots

Evidaxis Momentum Snapshots

Weekly, hash-pinned, point-in-time measurements of rising open-source and research-native AI systems, scored by "momentum" (the rate of change of public signals) on a transparent, versioned public methodology. A system is recognized as rising only when two independent axes converge (development velocity + citation momentum). The archive is append-only: published snapshots are never rewritten.

• Canonical source: https://evidaxis.org (llms.txt: https://evidaxis.org/llms.txt)
• Methodology (versioned): https://evidaxis.org/methodology/current/
• Concept DOI: 10.5281/zenodo.21076011 (resolves to the latest version)
• License: CC0-1.0 — reuse, redistribute, train on, cite freely.
• Cite a system: urn:evidaxis:claim:{entity_id}:{methodology}:{snapshot_date} (per-record URNs are listed in each snapshot and on evidaxis.org).

From 2026-10-10, snapshots are methodology m4 (definition). Folders scored under m3 are unchanged.

Files

Per snapshot date YYYY-MM-DD/:

• snapshot.json — full snapshot (entities, axes, statuses, counts) — person-free publication projection.
• entities.csv — flat convenience table (entity_id, name, cohort, status, momentum, axes z-scores, claim_urn).
• manifest.json, provenance.json — capture provenance.

Integrity

Every snapshot is identified by snapshot_id, a hash of the methodology version and the manifest hash. Its inputs are hash-pinned in the manifest with raw provenance published, and the source repository anchors the archive via OpenTimestamps, so every score is checkable against its hash-pinned inputs. Snapshot files carry their capture time: their exact bytes are preserved in the public git archive, not regenerated. This dataset mirrors the publication projection for convenient training/retrieval use.

Positive-only, person-free

The observatory measures systems, never people; it publishes no negative signals and no rankings of winners — measurements only, the reader sorts.