Evidence
Records you can hand to someone who has no reason to trust you — or us. We seal what happened into an append-only tree, publish its root once a day, and give you a file the other side checks on their own machine.
The problem with a log is that you control it. Anyone can keep one, and anyone who keeps one can edit it afterwards; a signature proves who wrote a line, not that the line was there yesterday. What changes the argument is a structure where inserting, removing or reordering anything breaks a proof that somebody else already holds. That is RFC 6962 — the mechanism behind Certificate Transparency — and it is what these products are built on.
What you can prove
One seal with its inclusion proof, the daily anchor and the instructions, in one self-contained file.
Both inclusion proofs and the consistency proof between them, so the order is checked and not asserted.
Every line sealed as it arrives, so an audit trail can be checked line by line afterwards.
Manifests, datasets, weights and runs, linked into a graph you can walk backwards.
Independent rebuild claims, compared, with REPRODUCIBLE and DIVERGENT verdicts.
If you are shipping AI in the EU
The AI Act expects a provider of a high-risk system to keep technical documentation and automatically generated records, and to be able to produce them on request. The hard part in practice is not writing them — it is showing, later, that what you produce now is what existed then. A provenance graph whose every node carries a dated inclusion proof is a direct answer to that question.
We are not selling you compliance, and nobody honestly can. Whether your system is in scope, what you must retain and for how long is between you, your lawyers and your notified body. What we sell is the records being checkable by someone who was not in the room.
What else you could use, and when we are not the answer
Sigstore and Rekor are excellent and free, and Rekor is a transparency log in the same family as this. The trade is that it is public: what you log, everyone reads. Here the entries stay yours and only the root is published, so you can prove an entry without publishing the entry. If your artifacts are open source, Rekor is probably all you need and we would rather you used it.
in-toto describes a supply chain end to end, and rebuilderd runs rebuilds at scale for a distribution. They are the right tools if you want to model your whole pipeline or rebuild a distro. We do one narrow thing next to them: hold the claims, compare them, and keep a record neither side can edit afterwards. For media, C2PA signs the asset itself — that is a different layer, and it pairs with this rather than competing.
Do not buy any of this if a log you keep yourself is enough. It usually is, right up until somebody disputes it.
How it holds up
Every entry gets an inclusion proof: a short list of hashes that lets anyone recompute the tree root from your entry alone. Once a day the root of every tree is published in a manifest, and that manifest is counter-signed into a separate append-only chain, so we cannot quietly rewrite yesterday either. When the tree has grown since, a consistency proof shows today's tree extends the anchored one and nothing was removed. All of it is RFC 6962, and the verifier is published — it reads the file and talks to nobody.
You can exercise the whole thing in the sandbox first, on test networks, before spending anything. The API reference has the exact steps, the error codes and the limits.
Test networks only, simulated confirmation — no coins move.