> ## Documentation Index
> Fetch the complete documentation index at: https://docs.chance.cc/llms.txt
> Use this file to discover all available pages before exploring further.

# The verification harness for AI agents

> Chance is the independent judge between an agent's intent and its execution — every verdict recorded, attested, and provable.

Chance sits between what an AI agent *intends* to do and what it's about to *execute*. An agent is useful because it reasons on its own — and risky for the same reason: when it's wrong, real funds move, and nothing normally checks the action against what you actually asked for.

Chance checks the proposed action against the intent **before** it goes through, returns a verdict with its reasoning, and leaves a tamper-proof, independently verifiable record of the decision.

<CardGroup cols={2}>
  <Card title="Quickstart" icon="bolt" href="/quickstart">
    Get a provable verdict from the API in five minutes.
  </Card>

  <Card title="How it works" icon="scale-balanced" href="/concepts/harness">
    PROPOSE → VERIFY → SETTLE and the ALLOW / BLOCK / ESCALATE verdict.
  </Card>

  <Card title="Provable by anyone" icon="shield-check" href="/concepts/proofs">
    Hash-chained, signed in hardware, anchored onchain.
  </Card>

  <Card title="API reference" icon="code" href="/api-reference/introduction">
    `POST /api/v1/intent` — verify an action against an intent.
  </Card>
</CardGroup>

## The model

<Steps>
  <Step title="PROPOSE">
    Your agent builds an action from your intent.
  </Step>

  <Step title="VERIFY">
    Chance checks the action against the intent and returns a verdict — with reasoning — before anything executes.
  </Step>

  <Step title="SETTLE">
    Only a passing verdict clears. Fail-closed: a failed check executes nothing.
  </Step>
</Steps>

This documentation covers the **VERIFY** step — and how every verdict is made provable.

## Why it's different

Any model can say "looks fine." Chance produces a verdict you can **prove**:

* **Tamper-evident** — the full decision is hash-chained; change one byte and it shows.
* **Attested** — signed by a key that only exists inside hardware-isolated, attested code.
* **Permanent** — anchored to a public blockchain; the record outlives our servers.
* **Independently verifiable** — anyone can recompute every check themselves, trusting no one.