Codestin Search App

The Index — worst first

Loading live data…

Proof of integrity

Disclosure changes

Public training-data disclosures that changed after we first observed them — a license swapped, documentation removed, provenance tags deleted. Each change carries a cryptographic before/after fingerprint. Crovia records the fact; it does not interpret intent.

Loading…

Cite this index

Use it. Quote it. Link it.

Released under CC-BY-4.0. Every figure traces to the signed, append-only ledger anchored on Bitcoin. Cite the snapshot — not just the page — so your reference stays verifiable.

Loading citation…

Loading…

Methodology

The Crovia Transparency Index measures one thing: does a lab document what its models were trained on? It does not judge model quality, safety, or licensing — only the presence and completeness of training-data disclosure.

An autonomous observer network audits public model cards, dataset declarations and registry metadata across thousands of models.
Each observation becomes an Ed25519-signed claim in an append-only ledger — it cannot be silently altered after the fact.
Batches of claims are anchored to the Bitcoin blockchain via OpenTimestamps, proving the data existed at a given time without a trusted third party.
A lab's score (0–100) is the average transparency score across its audited models. Grades: A ≥90, B ≥80, C ≥70, D ≥60, E ≥50, F <50.
Labs that publish no open weights or model cards (e.g. closed-only providers) may not appear until public artifacts exist to audit. Absence is not a passing grade.

Full per-model data: registry.croviatrust.com/ranking · Raw signed feed: transparency_index.json