feat: add Reasoning Verification extension for MPP#187
feat: add Reasoning Verification extension for MPP#187ThoughtProof wants to merge 2 commits intotempoxyz:mainfrom
Conversation
Defines a pre-payment verification step where an independent service evaluates the agent's decision logic before payment credential issuance. - Stake-proportional depth (micro→critical) - Verification receipts (signed, replayable on-chain) - Compatible with x402 and ERC-8183 verification patterns Co-exists with identity (ERC-8004) and trust (Maiat) — reasoning verification is the missing third layer.
|
Updated this branch to align with pot-sdk v2.0 (now live at api.thoughtproof.ai). Changes: • Removed HOLD and DISSENT from the verdict table — replaced with BLOCK HOLD and DISSENT are now internal calibration states — they no longer appear in public API responses. ─── Verdict semantics in the MPP extension:
The key design point: UNCERTAIN is not "the model didn't know" — it's a deliberate risk-aware system decision to escalate rather than guess. |
|
Friendly bump — happy to address any feedback or questions on the reasoning verification extension. We've since shipped onchain_proof (ecrecover-compatible attestations) which would strengthen the on-chain integration path described here. |
1 participant
What
Proposes a "reasoning-verification" extension for MPP that adds a pre-payment verification step — an independent service evaluates the agent's decision logic before payment credentials are issued.
Why
Identity checks answer "who is this agent?" Payment authorization answers "can this agent pay?" Neither answers "should this agent pay for this, given its reasoning?"
An authorized agent with valid identity can still make poorly-reasoned payments. This extension catches those defects before settlement.
How
reasoning-verification="required"in the 402 challengeStake-proportional: low-value payments skip verification, high-value payments require multi-model adversarial critique.
Design Choices
Related