# Falsification, Not Exposure: An Internally Preregistered Placebo-Controlled Decomposition of Self-Repair Feedback in Frozen Small Code Models

arXiv: https://arxiv.org/abs/2606.31511 · HTML: https://mlsanity.com/paper/2606.31511/

## Key facts

- Cited by 4 papers in the map (data through 14 Oct 2026).
- Builds on 39 of its 53 references that are in the map.
- First posted on arXiv 30 Jun 2026.

## Builds on (most cited first, 30 of 39)

- [Evaluating Large Language Models Trained on Code](https://mlsanity.com/paper/2107.03374/) (2021)
- [Reflexion: Language Agents with Verbal Reinforcement Learning](https://mlsanity.com/paper/2303.11366/) (2023)
- [Program Synthesis with Large Language Models](https://mlsanity.com/paper/2108.07732/) (2021)
- [Self-Refine: Iterative Refinement with Self-Feedback](https://mlsanity.com/paper/2303.17651/) (2023)
- [Scaling LLM Test-Time Compute Optimally can be More Effective than Scaling Model Parameters](https://mlsanity.com/paper/2408.03314/) (2024)
- [Is Your Code Generated by ChatGPT Really Correct? Rigorous Evaluation of Large Language Models for Code Generation](https://mlsanity.com/paper/2305.01210/) (2023)
- [DeepSeek-Coder: When the Large Language Model Meets Programming -- The Rise of Code Intelligence](https://mlsanity.com/paper/2401.14196/) (2024)
- [Large Language Models Cannot Self-Correct Reasoning Yet](https://mlsanity.com/paper/2310.01798/) (2023)
- [Large Language Monkeys: Scaling Inference Compute with Repeated Sampling](https://mlsanity.com/paper/2407.21787/) (2024)
- [Teaching Large Language Models to Self-Debug](https://mlsanity.com/paper/2304.05128/) (2023)
- [Quantifying Language Models' Sensitivity to Spurious Features in Prompt Design or: How I learned to start worrying about prompt formatting](https://mlsanity.com/paper/2310.11324/) (2023)
- [LLM Evaluators Recognize and Favor Their Own Generations](https://mlsanity.com/paper/2404.13076/) (2024)
- [OpenCodeInterpreter: Integrating Code Generation with Execution and Refinement](https://mlsanity.com/paper/2402.14658/) (2024)
- [Is Self-Repair a Silver Bullet for Code Generation?](https://mlsanity.com/paper/2306.09896/) (2023)
- [Software Testing with Large Language Models: Survey, Landscape, and Vision](https://mlsanity.com/paper/2307.07221/) (2023)
- [On the Self-Verification Limitations of Large Language Models on Reasoning and Planning Tasks](https://mlsanity.com/paper/2402.08115/) (2024)
- [Debug like a Human: A Large Language Model Debugger via Verifying Runtime Execution Step-by-step](https://mlsanity.com/paper/2402.16906/) (2024)
- [Can Large Language Models Really Improve by Self-critiquing Their Own Plans?](https://mlsanity.com/paper/2310.08118/) (2023)
- [Capturing Failures of Large Language Models via Human Cognitive Biases](https://mlsanity.com/paper/2202.12299/) (2022)
- [Inference-Aware Fine-Tuning for Best-of-N Sampling in Large Language Models](https://mlsanity.com/paper/2412.15287/) (2024)
- [Planning In Natural Language Improves LLM Search For Code Generation](https://mlsanity.com/paper/2409.03733/) (2024)
- [DebugBench: Evaluating Debugging Capability of Large Language Models](https://mlsanity.com/paper/2401.04621/) (2024)
- [Learning How Hard to Think: Input-Adaptive Allocation of LM Computation](https://mlsanity.com/paper/2410.04707/) (2024)
- [Towards an Understanding of Large Language Models in Software Engineering Tasks](https://mlsanity.com/paper/2308.11396/) (2023)
- [CYCLE: Learning to Self-Refine the Code Generation](https://mlsanity.com/paper/2403.18746/) (2024)
- [When To Solve, When To Verify: Compute-Optimal Problem Solving and Generative Verification for LLM Reasoning](https://mlsanity.com/paper/2504.01005/) (2025)
- [ALGO: Synthesizing Algorithmic Programs with LLM-Generated Oracle Verifiers](https://mlsanity.com/paper/2305.14591/) (2023)
- [LeDex: Training LLMs to Better Self-Debug and Explain Code](https://mlsanity.com/paper/2405.18649/) (2024)
- [Large Language Models of Code Fail at Completing Code with Potential Bugs](https://mlsanity.com/paper/2306.03438/) (2023)
- [What Makes Large Language Models Reason in (Multi-Turn) Code Generation?](https://mlsanity.com/paper/2410.08105/) (2024)

## Built on by (most cited first, 4 of 4)

- [Recursive Self-Improvement in AI: From Bounded Self-Refinement to Autonomous Research Loops](https://mlsanity.com/paper/2607.07663/) (2026)
- [Form, Not Content? A Preregistered, Placebo-Controlled Evaluation of Learned Error-Conditioned Self-Repair Through Prompts and Weights in Frozen Small Code Models](https://mlsanity.com/paper/2607.12962/) (2026)
- [Does Fault Localization Beat a Fresh Attempt? A Placebo-Controlled Study of Test-Guided Code Repair](https://mlsanity.com/paper/2609.00854/) (2026)
- [DENSE: Distilling Agent Trajectories into Evidence-Grounded Shortcut Trees for Self-Refinement](https://mlsanity.com/paper/2609.21423/) (2026)

## Abstract (arXiv)

In deployment settings where retraining is infeasible, small frozen code models are routinely asked to repair a failed program after seeing their own failing output, usually treated as a retry mechanism. From a Popperian view, a generated program is a conjecture and a test-execution violation is an oracle-relative, executable counterexample, so feedback's value should be attributed not to re-exposure to failing code but to whether the conjecture is opened to external, executable criticism. As the third stage of a falsification-centered measurement program, this study builds a placebo-controlled instrument that decomposes the feedback packet against a blind-resampling baseline at matched output-generation budget and against content-free, shape-matched placebos. The contribution is not a new repair algorithm but a reflexive methodology (packet decomposition, placebo mirroring, matched-budget discordant-pair tests, fresh-generation confirmation, executable audits) that makes both the model's program conjecture and the researcher's "feedback content works" claim falsifiable. Across six HumanEval+/MBPP+ cells with three 0.5B-1.5B frozen models, 290 dead task-cell units (no best-of-8 candidate passing the public tier) were evaluated; the main run produced 7,000 fresh generations and a preregistered follow-up 1,400 more. Blind resampling exceeded bare-code retry by +18 net unlocks (25/7, Holm p=0.0021). Code-plus-facts recovered +18 over bare code (21/3, p=0.00042) and +15 over a generic-bullet placebo (p=0.0041). An instruction-only effect was not distinguishable (+3, p=0.36). Code-plus-facts and blind resampling tied at 26 unlocks each (not equivalence). Six external-controller follow-ups tied a content-free shape placebo. In this regime, falsification helped not as vocabulary or self-critique, but as comparison with external, executable counterexamples.