CroviaTrust/tacet-disclosure-ledger
Viewer • Updated • 41.6k • 3.16k
AI & ML interests
AI provenance, verifiable AI evidence, cryptographic receipts, audit trails, model transparency, AI governance, continuity monitoring, reproducible research, machine-readable trust infrastructure, AI agent accountability, public verification.
Recent Activity
View all activity
Organization Card
∴ Crovia Trust
Crovia proves what did not happen.
Every hour, Crovia fetches the public model cards of AI models on this Hub, asks one published question — does this card disclose its training data? — and signs what it saw. The hour is opened by a public randomness beacon (drand) and closed by a Bitcoin anchor. Anyone can verify the record without an account and without a Crovia server.
Live from the log · updated 2026-10-05 04:06 UTC · latest.json
370
current epoch · since 2026-09-19
34,919
signed observations
368
hours anchored in Bitcoin · last block 969,934
6,332
models in the target list
17,934
observations that found no disclosure
19
featured silence proofs
Dataset · updated hourly
TACET disclosure ledger
Every observation since 2026-09-19, byte for byte as served at croviatrust.com: the signed rows, the hourly epoch sheets, the map changes, the OpenTimestamps anchors and the featured proofs. Three tables in the viewer (observations, epochs, targets), CC-BY-4.0.
Verify a proof yourself
pip install crovia-tacet-operator tacet-operator verify proofs/<slug>.seal.json \ --operator-pubkey <operator key from trust_root.json>
Checks signatures, chaining, non-inclusion paths, snapshot hashes; drand rounds and Bitcoin blocks against public relays.
Weekly report · open method
Survival Report (causari)
How much AI-tagged code is still at HEAD, repository by repository, against the same repository's untagged code of the
same age. Built with causari, one open-source Rust binary, no cloud: counts, not grades, and every number in the report
can be recomputed with the command the report prints and verified offline.
Latest
Survival Report #3 · 2026-09-28 · 43 repositories · 39,706 AI-tagged commits · method v3
npx causari audit <owner/repo> pipx run causari audit <owner/repo>
Featured proofs · 19 model cards, silence documented hour by hour
A silence is a run of anchored hours in which every observation of a model card found no training-data disclosure on that surface. It is a statement about a web page over time — not about fraud, bad faith, or what the provider disclosed elsewhere.
| Model card | Hours observed | Silence (days) | Proof |
|---|---|---|---|
| mistralai/Mistral-7B-v0.1 | 318 | 13.25 | proof |
| mistralai/Mistral-Small-3.2-24B-Instruct-2506 | 318 | 13.25 | proof |
| mistralai/Mixtral-8x7B-v0.1 | 318 | 13.25 | proof |
| Qwen/Qwen2.5-7B | 318 | 13.25 | proof |
| Qwen/Qwen2.5-Coder-32B-Instruct | 318 | 13.25 | proof |
| Qwen/Qwen3-32B | 318 | 13.25 | proof |
| black-forest-labs/FLUX.1-dev | 317 | 13.20 | proof |
| black-forest-labs/FLUX.1-schnell | 317 | 13.20 | proof |
and 11 more in proofs/index.json
TACET specification (Internet-Draft draft-crovia-tacet) ·
LACUNA ·
Crovia Seal ·
github.com/croviatrust ·
causari.dev ·
info@croviatrust.com
Data CC-BY-4.0, code Apache-2.0. Hours not observed count toward nothing. Corrections are made by revision, never by rewriting an anchored epoch. This card is regenerated hourly by the same script that publishes the dataset.
Data CC-BY-4.0, code Apache-2.0. Hours not observed count toward nothing. Corrections are made by revision, never by rewriting an anchored epoch. This card is regenerated hourly by the same script that publishes the dataset.
models 0
None public yet