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

Source

Source name: arXiv
Source kind: paper
Authors / organization: Xueping Gao, Jianwei Yang, Qiang Yang
Venue: arXiv: cs.CL
Published: 2026-07-27
Collected: 2026-07-28
URL: http://arxiv.org/abs/2607.24604v1
PDF: https://arxiv.org/pdf/2607.24604v1
DOI: UNKNOWN

Plain-Language Brief

The paper discusses the unreliability of looping in coding agents and the challenge of ensuring that correct patches are retained, verified, and submitted. The authors conducted several studies to understand this gap and found that using stale traces can harm repairs, while a prospective policy eliminates observed correct-start harm but fails its joint criterion. They propose a typed loop contract that can be mechanically enforced in a reference implementation.

What the Document Says

Looping is common in coding agents, but it doesn't guarantee reliability.
A study of 900 three-revision trajectories showed that current correctness with current traces falls from 0.820 after one revision to 0.673 after two, although ever-correct rises to 0.847 under forced revision.
Two common-state studies using 2,430 branches found that stale traces harm repairs more than current traces.
A prospective 540-rollout policy eliminates observed correct-start harm but reduces wrong-start repair and fails its joint criterion.
The paper proposes a typed loop contract that can be mechanically enforced in a reference implementation.
The implementation is an executable specification and conformance artifact, not evidence of improved repair competence or calibrated verifier dependence.

Key Concepts Mentioned

looping
coding agents
reliability
correct patches
revision
stale traces
current traces
typed loop contract

Claims Made by the Document

Looping in coding agents doesn't guarantee reliability.
Stale traces harm repairs more than current traces.
A prospective policy eliminates observed correct-start harm but reduces wrong-start repair and fails its joint criterion.

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 540-rollout policy.

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.