# MineValiCoder: Reliable Code Generation with Test Case Quality Mining and Bipartite Graph-Based Mutual Validation

Source: [arXiv](https://arxiv.org/abs/2607.22471v1)  
Feed7 permalink: https://feed7.dev/p/2607-22471v1-0kivszm  
Published: 2026-07-24T16:39:06.000Z  
Trust: Needs Review (needs_review)

## Why Included

MineValiCoder filters generated tests, refines several code candidates, then scores code and tests together. The pattern targets a core agent risk: unreliable tests steering implementation.

## Source Summary

MineValiCoder links three modules: **TCQM, Parallel TDD Refinement, and BiCoTeV**. They filter faulty generated tests, refine multiple code candidates, and score code-test interactions through a bipartite graph.

## Practical Implication

For coding-agent harnesses, treat generated tests as uncertain evidence rather than ground truth. Validate tests before feeding failures back into implementation, preserve multiple candidates, and select using agreement across code and test behavior.

## Agent-Ready Context

MineValiCoder links three modules: **TCQM, Parallel TDD Refinement, and BiCoTeV**. They filter faulty generated tests, refine multiple code candidates, and score code-test interactions through a bipartite graph.

For coding-agent harnesses, treat generated tests as uncertain evidence rather than ground truth. Validate tests before feeding failures back into implementation, preserve multiple candidates, and select using agreement across code and test behavior.

Across four LLMs, reported Pass@1 reached **96.34% on HumanEval** and **51.33% on LiveCodeBench**, with intermediate results on MBPP and APPS. These benchmark scores do not establish how the added loop performs on long-horizon repository changes.

## Context Map

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

## Uncertainty

- Across four LLMs, reported Pass@1 reached **96.34% on HumanEval** and **51.33% on LiveCodeBench**, with intermediate results on MBPP and APPS. These benchmark scores do not establish how the added loop performs on long-horizon repository changes.

## 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.