TOVEMÉ — Provable Autonomy
Your AI agent says it did the work. This proves it actually happened.
SCENE 1 — the agent pays a vendor.
agent claims: "Paid $30,000 to Acme Corp."
stakes: irreversible · external · $30,000 → GRAVE — must reach the sovereign twin
re-observing the world, escalating by stakes…
✓ T0 deterministic re-read …… agrees → escalating (grave: must reach the twin)
✓ T1 decorrelated model …… agrees → escalating (grave: must reach the twin)
✓ T2 sovereign twin …… agrees ✦ co-signed
VERDICT: PROVEN at T2 · co-signature 307c1aa070b2… · in the tamper-evident ledger ✓
→ "Authorized AND settled. The money actually moved."
SCENE 2 — the agent says it paid… but it didn't.
agent claims: "Paid $30,000 to Acme Corp."
stakes: irreversible · external · $30,000 → GRAVE — must reach the sovereign twin
re-observing the world, escalating by stakes…
✗ T0 deterministic re-read …… the world DISAGREES
VERDICT: CAUGHT at T0 · FLAGGED · no co-signature issued
→ "Authorized but NOT settled. The claim is false — caught before you trusted it."
SCENE 3 — the proof can't be forged.
✓ the sovereign twin's key produces a valid signature: verifies
✗ any other key (the agent, the cloud, an attacker): FAILS — unforgeable
The whole thesis, in three scenes:
never trust the agent's word — re-observe the world, scale the effort to the stakes,
and sign the truth with a key the agent can't reach.
Re-observe a REAL public-chain settlement: node demo.mjs --live