Decision Assurance Infrastructure

Summit Cognitive Documentation

Summit Cognitive builds the assurance layer for consequential, increasingly automated decisions. These docs cover the Decision Receipt API — the live service that evaluates an autonomous agent's action for admissibility and returns a cryptographically signed, independently verifiable receipt.

An agent merges a pull request, changes infrastructure, grants access, ships a deployment. Each of those is a decision. The question this platform answers is not only was it authorized? but can it be shown to have been right? — reconstructed from its evidence, checked against policy, replayed deterministically, and contested by anyone who cares to. A Decision Receipt is the record that makes that possible.

Start here

Understand the model

Four short concept pages explain the receipt and the evidence model behind it. Read them in order, or jump to whatever you're wiring up.

Build something

How a receipt is made

Every receipt comes out of the same four-part evaluation. The anatomy page covers each in detail; in brief:

1. Evidence. You submit a claim with one or more typed, provenanced sources — CI results, code review, test output, human approval, model traces, and more.
2. Policy. The claim is judged against a policy — minimum sources, distinct source types, confidence, provenance, replay, and risk scope — producing a verdict of ALLOWED, ESCALATED, or BLOCKED.
3. Replay. The decision is checked for deterministic reproducibility; a non-deterministic result is recorded as such rather than quietly passed.
4. Attestation. The result is sealed with an Ed25519 signature and chained to the previous receipt, so the record is tamper-evident and independently verifiable.