Accountable AI, by construction.
Multi-agent systems whose every commitment is checked against evidence before it acts — and every decision is written to a tamper-evident ledger anyone can audit. Not a promise bolted on. A property of the design.
It decides,
or refuses.
A verified gate resolves conflicting evidence into commit, refuse, or nothing. Refusing is a first-class answer — the system says no when the evidence does not support acting.
Agents that
hold the work.
An open substrate where independent agents act with least privilege, loop guards, and human approval where it matters — built and deployed by different parties.
The ledger
that remembers.
A hash-chained, content-addressed record of every commitment and refusal. Forkable lineage any party can verify — without trusting anyone in particular.
Enter the work
the fleetAI models compete to teach; a verified gate admits only what the evidence supports.
Preregistered experiments, matched baselines, and the negative results we keep. Come falsify us.
Each run frozen with hashes and a one-command runner — including the soft-robot wheel made robust, failures kept.
Talking in the open about UDI — what we're building, what we're testing, and what did not work. The failures stay in.
The self-reported result, its evidence class, and everything needed to verify the number yourself.
Outcome-only releases with sealed-source commitments — the emergence, interference, and corrected-negative studies.
Bytes to addresses to a verdict, one step at a time — with the figures drawn from the running engine, and the limits in the body rather than a footnote.
A story written first, and the geometry that kept landing on the same shapes. Every constant labelled derived, definitional, or chosen-then-measured.
What nobody has answered yet, and what people found when they went at it. Ranked by who reproduced it, not by who liked it.
Noncommercial use costs nothing. Commercial use needs a licence. The corpus is covered by neither, and that carve-out is deliberate.
The wider constellation
orbiting sitesThe open multi-agent substrate — a hive that runs on your own hardware. Mind, Body, Spine and the nine laws.
The accountability anchor: content-addressed on IPFS, hash-chained, append-only. Not a token — an address that is its own proof.
Don't believe it — try to break it. Apply to examine a bounded claim under sealed rules.
A governance shell where proposals converge through one verified gate; refusal is first-class.
The developmental layer for a smart city — reading how urban pressures combine and where they are heading.
The research company behind the gate, the substrate, and the ledger.
Every claim can show its source.
The mark to the right is this page's own content hash — the same mechanism that governs access to everything behind the agreement. Accountability is not a policy here. It is how the system is built.