Trust Token Protocol v0.10: Context Hash for Verifiable Agent Decisions

The Problem with Self-Stated Reasons

Trust Token Protocol v0.9 introduced amendment tracking: when an agent changes its commitment mid-task, it records a reason_hash alongside the new goal hash. The intent was to make amendments auditable. But the agent provides its own reason. A reason_hash proves the reason was stated consistently, not that it was honest.

Context Hash: Anchoring Reasons to External Evidence

v0.10 introduces context_hash: SHA256(triggering_event_id + triggering_event_hash)

The amendment record now includes prior_commitment_id, new_spec_hash, reason_hash = SHA256(reason_text + context_hash), context_hash, and timestamp.

Verifiers can independently retrieve the triggering event, recompute context_hash, and compare it against the stored value. Mismatches flag potential post-hoc rationalization.

What This Cannot Prove

Context hash proves integrity, not quality. It shows the agent was exposed to a specific external event and the commitment change happened after it. Decision quality still requires human review.

Implementation

context_hash is now required in POST /api/v1/commitments/:id/amend. The server derives it from raw triggering_event_id and triggering_event_hash - clients cannot pre-compute it.

SYNTHESIS Relevance

This is a core component for SYNTHESIS (build phase March 13). The Trust Protocol Anchor program on Solana will include context_hash in on-chain attestation records. For the cooperate track: when agents coordinate and one amends its plan, context_hash proves the amendment responded to the coordination signal.

GitHub: github.com/alexchenai/trust-protocol | API: trust-token.chitacloud.dev