Adnan KhanAdnanProduct & AI
San Francisco Bay Area
Get in touch

Source note2 min read

Looping Is Not Reliability: State-Bound Evidence and Typed Revision Contracts for Agentic Code Repair

Source

arXiv

arxiv.org/abs/2607.24604v1

Read the source

Source

Plain-Language Brief

The paper explores the lack of reliability guarantees in coding agents that use generate-test-revise loops. The authors conduct studies to understand the gap between finding a correct patch and retaining, verifying, and submitting it. They find that using stale traces can harm correct starts more than current traces in certain scenarios, but their implementation does not necessarily imply improved repair competence or calibrated verifier dependence.

What the Document Says

  • Generate-test-revise loops are common in coding agents but do not guarantee reliability.
  • A study of 900 three-revision trajectories shows that current correctness falls from 0.820 after one revision to 0.673 after two, with ever-correct rising to 0.847 under forced revision.
  • Two common-state studies using 2,430 branches reveal stale traces harm 34/135 correct starts versus 4/135 with current traces, showing a significant increase in harm.
  • A prospective policy eliminates observed correct-start harm but reduces wrong-start repair and fails its joint criterion.
  • Repository experiments expose floor effects, component heterogeneity, and no Holm-significant effects.
  • The authors propose an evidence-bound typed loop contract and provide a reference implementation that binds verifier evidence to exact code states, preserves verified checkpoints, and emits auditable admission receipts.
  • The implementation is an executable specification and conformance artifact, not evidence of improved repair competence or calibrated verifier dependence.

Key Concepts Mentioned

  • generate-test-revise loops
  • coding agents
  • reliability
  • correctness
  • revision
  • stale traces
  • current traces
  • Holm policy
  • evidence-bound typed loop contract

Claims Made by the Document

  • Using generate-test-revise loops does not guarantee reliability.
  • Stale traces can harm correct starts more than current traces in certain scenarios.
  • The proposed implementation does not necessarily imply improved repair competence or calibrated verifier dependence.

Evidence Used

  • Results from a study of 900 three-revision trajectories
  • Results from two common-state studies using 2,430 branches
  • Results from a prospective policy and repository experiments

Methods / Approach

UNKNOWN

Limitations Stated by the Authors

UNKNOWN

Limitations Noted by the Agent

Agent interpretation — not from the source.

UNKNOWN

Open Questions

UNKNOWN

Notes for Future Q&A

May include interpretation. Labeled where it does.

UNKNOWN

Raw Abstract / Excerpt

Generate--test--revise loops are common in coding agents, but repetition alone provides no reliability guarantee. We study the gap between finding a correct patch and retaining, verifying, and submitting it. A sealed five-seed study over 30 HumanEval repairs produces 900 three-revision trajectories. Under forced revision, current correctness with current traces falls from 0.820 after one revision to 0.673 after two, although ever-correct rises to 0.847. Two common-state studies use 2,430 branches from identical frozen programs to remove post-treatment risk-set bias. In a prespecified 14B replication, stale traces harm 34/135 correct starts versus 4/135 with current traces, a 22.2-point increase (task-cluster 95% CI $[8.9,37.0]$, exact Holm $p=0.0337$). A prospective 540-rollout policy eliminates observed correct-start harm but reduces wrong-start repair and fails its joint criterion. Repository experiments over 24 bugs and four coder stacks expose floor effects and component heterogeneity without Holm-significant effects. We therefore separate admission, preservation, grounded certification, competence, and liveness. We derive an evidence-bound typed loop contract and instantiate its mechanically enforceable subset in a reference implementation that binds verifier evidence to exact code states, preserves verified checkpoints, and emits auditable admission receipts. The implementation is an executable specification and conformance artifact, not evidence of improved repair competence or calibrated verifier dependence.