We build the trust layer in the open.

Signed is not verified.

A cryptographic signature proves a datapoint was not altered, and who produced it. It says nothing about whether the value is any good. A signed reading can be stale, single-sourced, or a producer simply restating its own claim, and the envelope is still perfect. Most systems quietly treat signed as trustworthy, and that conflation is where downstream decisions go wrong.

Three questions, kept separate.

A consumer acting on a datapoint needs three different answers, and collapsing them into one "trusted" flag is the mistake. So we keep them as distinct, independently checkable axes.

Open by default, verifiable by anyone.

We publish the reliability layer as an open, vendor-neutral object: a documented schema, zero-dependency reference validators in Python and JavaScript, and a portable conformance suite, all MIT-licensed. Any system can adopt the shape and check it the same way, with no account and no dependency on us, so you can verify even against us.

We take that work into the open standards the ecosystem is building, across agent protocols, data formats, and provenance and attestation. The aim is simple: one honest way to say how much a machine should believe a piece of data, shared rather than owned.

The work is the introduction.

We engage the way good open-source does: by contributing something useful and letting it stand on merit. Every contribution is grounded in a working reference, adversarially checked before it goes out, and offered as shared infrastructure rather than a pitch. We would rather be the implementation others can verify than the vendor others have to trust.