Do Code Language Models Use Tests? A Behavioral and Representational Study of Test-Driven Code Generation

• URL: http://arxiv.org/abs/2607.26244v1
• Date: Tue, 28 Jul 2026 20:25:42 GMT
• Title: Do Code Language Models Use Tests? A Behavioral and Representational Study of Test-Driven Code Generation
• Abstract summary: We study test-driven code generation on HumanEval+, MBPP+, and recent LiveCodeBench tasks.<n>We compare natural-language-only prompts with relevant visible tests, shuffled outputs, irrelevant tests, assertion-only tests, and stronger-model-generated synthetic tests.
• Score: 14.44743150931186
• License: http://arxiv.org/licenses/nonexclusive-distrib/1.0/
• Abstract: Public tests are widely used to guide large language model code generation, but whether models treat them as executable specifications or merely as extra prompt context remains unclear. We study test-driven code generation on HumanEval+, MBPP+, and recent LiveCodeBench tasks using Qwen2.5-Coder-7B and Qwen3.6-27B. We compare natural-language-only prompts with relevant visible tests, shuffled outputs, irrelevant tests, assertion-only tests, and stronger-model-generated synthetic tests. Evaluation combines hidden or private test pass rates with task-level behavior flips, linear probes, and layer-wise hidden-state shifts. Visible tests substantially improve Qwen2.5 performance on MBPP+ but have little or unstable effect on HumanEval+ and LiveCodeBench. For Qwen3.6 on LiveCodeBench, the natural-language-only pass rate rises from 13.1% to 39.4%, yet relevant original tests add only 2.9 percentage points (p = .458), and synthetic high5 tests add 1.7 points over their matched baseline (p = .701). More tests produce larger representation shifts, but accuracy gains quickly saturate or disappear; assertion-only and irrelevant-test prompts can strongly alter hidden states without improving correctness. After controlling for prompt length, task-level shifts are associated with both beneficial and harmful flips, with a substantially stronger association for harms. These results show that tests influence code models through both semantic guidance and prompt-context perturbation, and that representational change alone does not demonstrate effective test utilization.

Related papers

• How effective are traditional test criteria at detecting bugs in large language models generated code? [5.2762490553598065]
  Most faults introduced by Large Language Models are relatively trivial to catch.<n>Fault detection rates remain extremely low, often near zero.<n> prompt-aware oracles can improve fault detection, but their overall effectiveness remains limited.
  arXiv  Detail & Related papers  (2026-09-08T18:04:21Z)
• ExecCritic: Learn to Test, Test to Improve for Coding Agents [68.42713285082912]
  Execution feedback can guide coding agents toward correct repository repairs, but only when the tests capture the behavior requested by the issue.<n>We introduce ExecCritic, combining a test--verify--revise scaffold with a role-specific reinforcement learning recipe for training agents within it.<n>The scaffold separates test construction from source-code repair: a Test agent independently generates repository-native tests, a fail-closed harness qualifies and freezes them, and a Repair agent revises source code from their execution feedback without changing the tests.
  arXiv  Detail & Related papers  (2026-09-08T17:53:37Z)
• Breaking, Stale, or Missing? Benchmarking Coding Agents on Project-Level Test Evolution [20.606877071567958]
  We present TEBench, the first project-level benchmark for test evolution.<n>We construct TEBench through a four-stage pipeline over Defects4J projects.<n>We evaluate seven configurations spanning three industrial agent frameworks.
  arXiv  Detail & Related papers  (2026-05-07T12:31:09Z)
• Are Benchmark Tests Strong Enough? Mutation-Guided Diagnosis and Augmentation of Regression Suites [49.16055123488827]
  Insufficiently strong test suites can admit plausible yet semantically incorrect patches, inflating reported success rates.<n>We introduce STING, a framework for targeted test augmentation that uses semantically altered program variants as diagnostic stressors.<n>STING produces 1,014 validated tests spanning 211 instances and increases patch-region line and branch coverage by 10.8% and 9.5%.
  arXiv  Detail & Related papers  (2026-04-02T01:13:40Z)
• Code-A1: Adversarial Evolving of Code LLM and Test LLM via Reinforcement Learning [54.95476453942411]
  Reinforcement learning for code generation relies on verifiable rewards from unit test pass rates.<n>Recent self-play methods unify code and test generation in a single model.<n>Code-A1 achieves code generation performance matching or exceeding models trained on human-annotated tests.
  arXiv  Detail & Related papers  (2026-03-16T17:58:13Z)
• Consistency Meets Verification: Enhancing Test Generation Quality in Large Language Models Without Ground-Truth Solutions [1.9196411948992402]
  We present ConVerTest, a novel two-stage pipeline for synthesizing reliable tests without requiring prior code implementations.<n>Experiments on BIGCODEBENCH and LESS BASIC PYTHON PROBLEMS benchmarks demonstrate that ConVerTest improves test validity, line coverage, and mutation scores by up to 39%, 28%, and 18% respectively.
  arXiv  Detail & Related papers  (2026-02-11T04:40:38Z)
• CodeContests-O: Powering LLMs via Feedback-Driven Iterative Test Case Generation [71.42965967582147]
  Existing approaches attempt to synthesize test cases using Large Language Models (LLMs)<n>We propose a $textbfFeedback-Bench Iterative Framework$ for comprehensive test case construction.<n>Our dataset achieves an average True Positive Rate (TPR) of $89.37%$ and True Negative Rate (TNR) of $90.89%$, significantly outperforming the CodeContests and CodeContests+ by margins of $4.32%$ and $9.37%$, respectively.
  arXiv  Detail & Related papers  (2026-01-20T07:32:44Z)
• Assertion-Aware Test Code Summarization with Large Language Models [0.0]
  Unit tests often lack concise summaries that convey test intent.<n>This paper presents a new benchmark of 91 real-world Java test cases paired with developer-written summaries.
  arXiv  Detail & Related papers  (2025-11-09T04:58:32Z)
• AugmenTest: Enhancing Tests with LLM-Driven Oracles [2.159639193866661]
  AugmenTest is an approach leveraging Large Language Models to infer correct test oracles based on available documentation of the software under test.<n>AugmenTest includes four variants: Simple Prompt, Extended Prompt, RAG with a generic prompt (without the context of class or method under test), and RAG with Simple Prompt, each offering different levels of contextual information to the LLMs.<n>Results show that in the most conservative scenario, AugmenTest's Extended Prompt consistently outperformed the Simple Prompt, achieving a success rate of 30% for generating correct assertions.
  arXiv  Detail & Related papers  (2025-01-29T07:45:41Z)
• Effective Test Generation Using Pre-trained Large Language Models and Mutation Testing [13.743062498008555]
  We introduce MuTAP for improving the effectiveness of test cases generated by Large Language Models (LLMs) in terms of revealing bugs. MuTAP is capable of generating effective test cases in the absence of natural language descriptions of the Program Under Test (PUTs) Our results show that our proposed method is able to detect up to 28% more faulty human-written code snippets.
  arXiv  Detail & Related papers  (2023-08-31T08:48:31Z)
• Position: AI Evaluation Should Learn from How We Test Humans [65.36614996495983]
  We argue that psychometrics, a theory originating in the 20th century for human assessment, could be a powerful solution to the challenges in today's AI evaluations.
  arXiv  Detail & Related papers  (2023-06-18T09:54:33Z)
• CodeT: Code Generation with Generated Tests [49.622590050797236]
  We explore the use of pre-trained language models to automatically generate test cases. CodeT executes the code solutions using the generated test cases, and then chooses the best solution. We evaluate CodeT on five different pre-trained models with both HumanEval and MBPP benchmarks.
  arXiv  Detail & Related papers  (2022-07-21T10:18:37Z)
• PRover: Proof Generation for Interpretable Reasoning over Rules [81.40404921232192]
  We propose a transformer-based model that answers binary questions over rule-bases and generates the corresponding proofs. Our model learns to predict nodes and edges corresponding to proof graphs in an efficient constrained training paradigm. We conduct experiments on synthetic, hand-authored, and human-paraphrased rule-bases to show promising results for QA and proof generation.
  arXiv  Detail & Related papers  (2020-10-06T15:47:53Z)

This list is automatically generated from the titles and abstracts of the papers in this site.

This site does not guarantee the quality of this site (including all information) and is not responsible for any consequences.