# Your PR Says the Tests Passed. Prove It. - Stephen Chin & Cassandra Chin

Source: [AI Engineer](https://www.youtube.com/watch?v=gbeuC0ud1p4)  
Feed7 permalink: https://feed7.dev/p/your-pr-says-the-tests-passed-prove-it-stephen-chin-cassandra-chin-15wqjj5  
Published: 2026-11-10T17:57:21.000Z  
Trust: Source Linked (source_linked)

## Why Included

TraceTune combines coding-agent audit trails with Relate graph retrieval to produce evidence-backed PRs. Its early PR Bench comparison reports higher judge scores for Claude and Codex, but gives limited detail on the evaluation setup.

## Source Summary

TraceTune uses agent and code history to draft PRs with command context, test evidence, and linked facts. Its Relate component is an open source, in-memory graph database written in Rust; the workflow uses a stronger model for drafting and a cheaper one for mechanical review.

## Practical Implication

If agents write your PRs, keep their audit trail and tie claims like “tests passed” to commands run against the submitted change. Graph retrieval can assemble related evidence before drafting, reducing repeated agent research; TraceTune also checks claims against that evidence.

## Agent-Ready Context

TraceTune uses agent and code history to draft PRs with command context, test evidence, and linked facts. Its Relate component is an open source, in-memory graph database written in Rust; the workflow uses a stronger model for drafting and a cheaper one for mechanical review.

If agents write your PRs, keep their audit trail and tie claims like “tests passed” to commands run against the submitted change. Graph retrieval can assemble related evidence before drafting, reducing repeated agent research; TraceTune also checks claims against that evidence.

In the talk’s PR Bench comparison, judge scores rose from **74 to 76** for one of Claude or Codex and **82 to 86** for the other; the transcript does not clearly map each range to a model. Human review was also used for accuracy, but sample size and evaluation details are not given. Relate avoids LLM retrieval costs, while still requiring local compute and storage.

## Connected Context

Feed7 judgment across 1093 accumulated Signals:

TraceTune makes PR claims more auditable by retaining command context and linking claims to evidence, while graph retrieval can reduce repeated research during drafting. Its judge-score gains are modest and not clearly mapped to models, so they support a promising workflow rather than a general performance claim. Evidence checks show what was run and what supports a claim; they do not establish that the underlying claim is true.

- [Build for the Memo, Not the Demo — Shawn Chan, China Resources Holdings](https://feed7.dev/p/build-for-the-memo-not-the-demo-shawn-chan-china-resources-holdings-0i3s3oo) — Both treat provenance as part of the output: the finance candidate requires claim-level traceability and visible uncertainty, while TraceTune ties PR claims such as test results to commands and linked evidence.
- [How to build an AI-Native Health Company — Dan Feng, Maven Clinic](https://feed7.dev/p/how-to-build-an-ai-native-health-company-dan-feng-maven-clinic-0hr5j7m) — Maven’s repeated testing and compact PRs reinforce TraceTune’s focus on reviewable, verifiable changes; neither implies that automation removes the need for human review.

## Context Map

- Layer: agent
- Domains: coding
- Topics: harness-engineering, agent-reliability, retrieval

## Uncertainty

- In the talk’s PR Bench comparison, judge scores rose from **74 to 76** for one of Claude or Codex and **82 to 86** for the other; the transcript does not clearly map each range to a model. Human review was also used for accuracy, but sample size and evaluation details are not given. Relate avoids LLM retrieval costs, while still requiring local compute and storage.

## Agent Instruction

Use this item as source-backed context. Do not invent claims beyond the linked source. If this item conflicts with another source, call out the conflict.
