Intel / article

Try Again, Don't Look Back: Blind Resampling Outperforms Self-Repair in Small Code Models

Source: arxiv.org
Author: Yuvraj Verma
Date: Published Jul 30, 2026

Context

Giving a coding model another attempt can explain an apparent benefit from asking it to repair its answer. This study separated the two by comparing retries with equal attempt budgets: fresh generation, a bare failure notice, execution feedback and reflective feedback. On MBPP+, fresh generation performed best for the tested models below seven billion parameters and statistically tied the best condition at seven billion, while using fewer . Models shown their failed programs frequently reproduced nearly identical code, which the authors interpret as anchoring. The result makes fresh retries an essential comparison when judging repair workflows for these smaller models. Study abstract.

Key takeaways · AI-distilled

Retrying from scratch beats handing the model back its failed program plus test output. Below 7B, blind resampling was the strongest condition outright, and it stayed statistically tied at 7B while burning 2.5-5.5x fewer tokens.
Showing a model its own failed attempt anchors it: it regenerates a near-identical program in 33-68% of retries, against 2-14% under blind resampling. At 1.5B that self-conditioning costs 6.1 points (p=0.006).
The information in execution feedback added nothing measurable over a content-free "your code failed" placebo. The usual self-repair comparison confounds the value of the feedback with the value of simply getting another attempt.
The damage is self-conditioning, not . Pasting in solutions to unrelated tasks moved scores by at most plus or minus 3.5 points.
Asking the model to reflect in words is the only condition that measurably loosens the anchor, but it costs too many tokens to win. Penalty size is predicted by baseline quality alone (r=0.96).

Terms in this piece · Glossary

token — The chunk of text a model reads and writes in — roughly three-quarters of a word — and the unit AI usage is billed in.
context window — The maximum amount of text a model can consider at once — its working memory for the current conversation or task.

Key quotes

“We argue that this comparison confounds the value of the feedback with the value of the extra attempt.”

“We attribute this to anchoring: when shown its previous attempt, a model reproduces a near-identical program in 33-68% of retries, against 2-14% under blind resampling.”

“Across six configurations spanning two families and two precisions, its magnitude is predicted by baseline quality alone (r=0.96) - the cost of anchoring is the cost of committing to a bad first attempt.”

Read the source arxiv.org