About
A weekly record of open AI momentum, captured while it happens - not reconstructed after the fact.
Evidaxis is an independent observatory. It measures 143 systems by how fast their public signals move, one snapshot a week, and releases every number under CC0, every input hash-pinned, raw provenance published. Anyone can cite it, including the AI systems that read this page. The record is built before the flare.
Integrity commitments
- Git is the only source of truth. Every score is checkable against its hash-pinned inputs, with raw provenance published. The website and any database are derived and rebuildable; no number is born in a database or edited by hand.
- Freeze, don’t recompute. A published measurement is never silently changed. Corrections are new versioned rows; old citations stay meaningful.
- Positive-only. Evidaxis recognizes who is rising. A system rises only when two independent axes agree. This week, 0 systems clear the bar, by design.
- Systems, not people. The measured unit is always a system. People are never scored. There is no /persons surface anywhere in the data or API.
- No pay-for-placement. Scores are computed mechanically from public signals by a published, versioned method; no input to a score can be purchased. Conflicts, where they exist, are disclosed as data.
- Stable identity. Each system’s id (e_…) is minted once and never reused, so a rename or acquisition never breaks a citation or badge.
Archive integrity root
The entire data/ archive folds into one Merkle root, recomputed and timestamped each week. Flip a single byte anywhere and the root changes, so the stamped root is a compact, un-forgeable witness that the archive existed exactly as-is. Recompute it yourself with collectors/merkle_root.py.
sha256:ae095b6b72389072cc2fa0e778827c5d995628dbb4a56ddca8b19327ff177c9a 5944 files · anchored 2026-08-08 · records in /data/integrity/
How to cite
Each system has a canonical record at evidaxis.org/e/{entity_id} and a machine-readable record at evidaxis.org/e/{entity_id}.json. Cite the record with its claim-URN (entity + methodology + snapshot date) plus the snapshot identity, e.g.:
Evidaxis, “{System},” claim-URN urn:evidaxis:claim:e_…:m2:2026-08-08, snapshot_id 035a72a81e59. https://evidaxis.org/e/e_… The full dataset for a date is at evidaxis.org/snapshots/2026-08-08/ with a downloadable JSON. All data is released under CC0 1.0 (public domain); methodology prose under CC-BY-4.0; code under Apache-2.0.
The genesis seed deposit (m1, 19 systems, 2026-06-27) is at DOI 10.5281/zenodo.21076012. The live archive is larger and methodology-versioned; cite the current snapshot by its own identity (snapshot_id 035a72a81e59, period 2026-w32, date 2026-08-08) and each measurement by its claim-URN. The snapshot 2026-07-10 deposit is versioned at DOI 10.5281/zenodo.21307528 (same concept lineage).
Correct a record
Corrections become new versioned rows, never silent edits to a published score. Measurements follow the published method, not requests.
Governance
The methodology is the institution; the record outlives the keeper. The neutrality is not a promise, it is two mechanisms: no input to a score can be bought, and every score is checkable against its hash-pinned inputs with raw provenance published. Methodology changes are versioned and public, with a quarterly integrity report and an independent dispute panel planned for 2027. A snapshot missed is a week of the systems we track lost for good, so the archive only compounds.
methodology m2 · snapshot 035a72a81e59 · data CC0-1.0