Decision Receipt by Summit Cognitive

Evidence for governed software decisions

Decision Receipt records policy evaluation results and evidence for software-change decisions. The production ledger currently contains PR-derived records; a receipt records an evaluation and does not itself prove that a merge, deployment, or autonomous execution occurred.

Get API Key — Free Check a Receipt API Reference Deploy Guide

You do not have to take our word for any of it. Check a receipt against our published key on your own machine with the verify kit, or re-execute the decision and reproduce its hash with the replay kit. Around 30–45 KB each, no install, no account, no network. Both ship adversarial fixtures and a self-test, because a tool that cannot demonstrate a failure is indistinguishable from one that always says yes.

13205

Production records

13177

Accepted

Blocked

Escalated

Production-scoped ledger snapshot measured 2026-10-09T07:19:27.825Z. Counts are receipt records, not autonomous executions, merges, or deployments.

Public audit: degraded. Production verdict: blocking — ARTIFACT_MISSING_UNDISPOSITIONED. Chain continuity: valid; artifact integrity: degraded; 7 reported error(s). Of those, 7 record(s) are missing their stored content, 0 could not be read, and 0 disagree with the content hash held in the ledger. Of the 7 missing, 5 are synthetic evaluation probes (the rule is published in /v1/chain/audit as artifact_availability.demo_probe_rule) and 2 are production records, 2 of which have no accepted disposition yet. No receipt is known to have been altered: every artifact still present hashes to its ledger record, and the chain links are intact. Server signatures are verified against the key the published registry (/.well-known/summit-decrec-keys.json) scopes to each receipt's chain sequence: 13011 of 13205 receipts verified, 1397 of them with the key retired at chain sequence 1463; 187 carry no signature at all and 0 signed receipts have no published key. Checked 2026-10-09T07:05:23.586Z.

The problem

Software-change decisions need durable evidence

Automated and human-associated repository events can create provenance gaps when evaluation inputs, policy results, and identifiers are not retained together. Decision Receipt creates a reviewable record; enforcement remains the responsibility of the integrating system.

Cognition (Devin) optimizes for autonomous capability.
Summit optimizes for admissible authority.
Those are not the same market.

How it works

Three steps to admissibility

For submitted evaluations, the service records supplied evidence, applies configured policy checks, and returns a receipt disposition.

Submit Evidence

Submit a claim and supporting evidence such as CI results or code review metadata. Provenance fields reflect what the caller supplied or an integration captured.

Evaluate Policy

The configured policy checks the submitted record and produces an evaluation result. Which checks run depends on the endpoint and policy version.

Receive Receipt

The response records evidence hashes, policy results, identifiers, and an ACCEPTED, BLOCKED, or ESCALATED disposition. Integrators decide how that result affects their workflow.

Integration points

Where Decision Receipt sits

Above, beside, underneath, or between autonomous agents and production systems.

GitHub PR Records

Webhook integrations can derive receipt records from pull-request events. The current public production ledger is primarily this kind of record.

CI/CD Integration

A workflow can consume an evaluation result and retain the receipt beside test results. Blocking behavior must be configured and verified in that workflow.

Orchestration Records

Callers can submit evidence about handoffs or tool activity. A receipt reflects the submitted evaluation, not independent observation of every action.

Runtime Integration

Applications may consult a receipt disposition before an operation. This site does not claim that current production records enforce infrastructure or deployment changes.

Audit Review

The public audit checks ledger-chain continuity and canonical receipt-content hashes. Historical signature verification is not included in that audit.

Regulated Environments

Defense, finance, healthcare, and critical infrastructure. Configurable policy rules. Sovereign deployment ready. Air-gapped fixture mode.

Try it now

Try a synthetic, non-production evaluation

The public demo endpoint accepts fictional sample input and returns a non-production demo receipt. It does not authorize, merge, deploy, or attest that the described action occurred.

curl -X POST https://decrec.summitcognitive.ai/v1/demo/evaluate \
  -H "Content-Type: application/json" \
  -d '{
    "claim_id": "my-first-receipt",
    "entity": "my-org/my-repo",
    "claim": "AI agent requests merge authority for PR #42",
    "sources": [
      {"id": "ci", "type": "ci", "uri": "https://ci/run/1",
       "confidence": 0.9, "content": "All tests pass"},
      {"id": "review", "type": "code_review", "uri": "https://pr/1",
       "confidence": 0.85, "content": "Approved by maintainer"}
    ]
  }'

The response is explicitly marked environment: nonproduction and demo: true. Production evaluation requires authentication. See full API docs →

Enforcement flow

How enforcement works

A bot-authored PR triggers a deterministic pipeline — every evaluation is recorded, signed, and checkable offline with a pinned public key.

PR Opened

Agent opens a pull request on GitHub

→

Webhook Fires

GitHub sends the event to Decision Receipt

→

Evidence Collected

CI status, review state, labels, PR metadata gathered as sources

→

9 Rules Evaluated

Deny-by-default policy engine runs all 9 rules

→

Receipt Signed

Ed25519 signature + hash chain entry

→

Verdict Posted

GitHub status check + PR comment with full policy breakdown

✕ Blocked Agent sees specific rule failures. Improves evidence. Resubmits.

✓ Allowed On repositories where the admissibility check is required by branch protection (Summit-Cognitive/summit since 2026-08-24), a PR without an ACCEPTED receipt cannot merge. Each verdict’s receipt can be checked offline. Receipts issued before a repository’s enforcement date are evidentiary records, not enforcement records.

Pricing

Free verification. Predictable governance pricing.

The primary unit is the governed agent, not the receipt. Normal receipt volume is included so customers are not rewarded for under-instrumenting a consequential workflow.

Verifier

per month

→ Offline verifier, no account

→ No telemetry or rate limit

→ Public format and criteria

→ Developer API key included

→ 1 demo agent

→ Best-effort hosted issuance

Get API Key

Team

$1,200

per month · annual agreement

→ 3 governed agents

→ Normal receipt volume included

→ Retention stated in the order form

→ Buyer-named witness option

→ Inclusion proofs + evidence packs

→ $1,450 month-to-month

Start Team

Enterprise

From $48,000

annual contract

→ 10 governed agents

→ Unlimited receipts

→ Custom retention

→ Self-hosted deployment

→ Bring your own key custody

→ Contracted service level

Receipt Readiness Assessment — $7,500. A fixed 30-day engagement with a named deliverable, limited to three starts per quarter and fully credited against the first annual Team or Enterprise agreement. Summit Attested will be available only after an independent examination is completed; no examination or opinion is implied today.

Why Decision Receipt

What makes this different

Most AI governance tools are retrospective dashboards. Decision Receipt produces the evidence an enforcement decision can be checked against later.

Not observability

We don't watch AI act and report afterward. Evidence is evaluated against policy at the moment of the decision, and the receipt records that evaluation. Applying the verdict is the integrating system's job; the receipt is what makes it checkable afterward.

Not another model

We don't compete with Devin, Codex, or Copilot. We verify whether their output is admissible. The more autonomous they get, the more we matter.

Not compliance theater

Receipts are cryptographically signed, chain-linked, and offline-verifiable with a key established through the relying party's trust channel. Replay the decision path. Audit the evidence. The public audit states what verification does and does not cover.

30-second setup

One webhook URL. First receipt in under a minute. No agents to install, no config files to manage, no infrastructure to provision.

Protocol, not product

Decision Receipt is a verifiable standard. Receipts travel. Third parties can verify. The spec is public. We're building infrastructure, not a dashboard.

Real production data

Not a demo. 13205 receipt records derived from real GitHub PRs — 13177 accepted, 28 blocked, 0 escalated. These counts are the ledger's own; this page publishes no summary rate above them. Agents tracked: dependabot, Devin, Codex, Jules, human.

Get started

Get your API key

Enter your email to receive an API key instantly. No credit card required.

Live ledger

Recent receipts

Summit-Cognitive/summit — 14-day receipt activity

Receipt	Verdict	Agent	Repository	PR
rcpt_gh-Summit-Cognitive-sum...	ALLOWED	BrianCLong	Summit-Cognitive/summit	#41786
rcpt_gh-Summit-Cognitive-sum...	ALLOWED	BrianCLong	Summit-Cognitive/summit	#41785
rcpt_gh-Summit-Cognitive-sum...	ALLOWED	[redacted]	Summit-Cognitive/summit	#41770
rcpt_gh-Summit-Cognitive-sum...	ALLOWED	BrianCLong	Summit-Cognitive/summit	#41757
rcpt_gh-Summit-Cognitive-sum...	ALLOWED	BrianCLong	Summit-Cognitive/summit	#41753
rcpt_gh-Summit-Cognitive-sum...	ALLOWED	BrianCLong	Summit-Cognitive/summit	#41752
rcpt_gh-Summit-Cognitive-sum...	ALLOWED	[redacted]	Summit-Cognitive/summit	#41765
rcpt_gh-Summit-Cognitive-sum...	ALLOWED	[redacted]	Summit-Cognitive/summit	#41764
rcpt_gh-Summit-Cognitive-sum...	ALLOWED	[redacted]	Summit-Cognitive/summit	#41763
rcpt_gh-Summit-Cognitive-sum...	ALLOWED	[redacted]	Summit-Cognitive/summit	#41766

View full ledger (JSON) → | Timeline →

Repository and actor inventories are available to authenticated operational reviewers.