← All Demos
Case Study

How Decision Receipt improved admissibility on a live repository

Repository: BrianCLong/summit (IntelGraph platform)

Period: May 12-18, 2026 (first week of operation)

Contributors: Human (BrianCLong), dependabot[bot], Devin, Codex, Jules

The situation

BrianCLong/summit is a production repository with 400+ open PRs from multiple sources: human developers, automated dependency bots (dependabot), and autonomous AI agents (Devin, Codex, Jules). PRs were being merged without systematic admissibility evaluation.

What we did

Added Decision Receipt as a GitHub webhook. Every PR opened, synchronized, or reopened triggers an automatic admissibility evaluation. No code changes to the repository. One webhook URL. Setup took under 2 minutes.

What happened

—
Receipts
—
Accepted
—
Blocked
—
Accept Rate

Day 1 acceptance rate was approximately 33%. Most blocks came from two policy rules: minimum_source_count and source_type_diversity. Single-source PRs (where the only evidence was the PR metadata from the webhook) correctly failed the diversity requirement.

The policy was working exactly as designed. A PR with only one source of evidence should not automatically receive admissibility. The system surfaced a real governance gap.

How we improved

Two changes drove the acceptance rate from 33% to 65%:

  1. Evidence enrichment. The webhook now extracts more evidence from each PR: CI status (inferred from mergeable_state), PR labels, bot-specific agent traces, and substantial PR descriptions. Bot PRs went from 1 source to 3-6 sources.
  2. Per-repo policy. Registered the repository with the "permissive" policy preset (1 source minimum, no provenance required) to match the current maturity level. As the team adds more CI evidence and review processes, they can tighten the policy.

Agent performance

What this proves

Decision Receipt doesn't just gate actions. It creates a measurable improvement loop:

  1. Baseline: measure current acceptance rate
  2. Identify: which rules fail most, for which agents
  3. Improve: enrich evidence, tune policy
  4. Verify: watch acceptance rate climb
The acceptance rate trend is the product's primary value metric. If it's not improving, something is wrong. If it is improving, the system is making your codebase more admissible over time.

Try it on your repo

One webhook. First receipt in 60 seconds. Free tier: 100 receipts/month.

Set Up Now Try Playground First