From Guessing to Placeholding: A Cost-Theoretic Framework for Uncertainty-Aware Code Completion

2026-04-02

AI Summary

This work addresses the limitations of conventional code completion models, which often generate concrete code under insufficient context, leading to high editing or rejection rates. The authors propose Adaptive Placeholder Completion (APC), a framework that formulates code completion as a cost-minimization problem under uncertainty, wherein explicit placeholders are emitted at high-entropy positions for user intervention. They provide the first theoretical justification from a cost perspective, proving the existence of an entropy threshold beyond which placeholder insertion yields lower expected cost than forced generation. An end-to-end, uncertainty-aware pipeline is developed using real-world edit logs for training and a cost-aware reinforcement learning reward function integrated with large language models. Experiments across models ranging from 1.5B to 14B parameters demonstrate a 19%–50% reduction in expected editing cost while preserving standard completion performance.

Abstract

While Large Language Models (LLMs) have demonstrated exceptional proficiency in code completion, they typically adhere to a Hard Completion (HC) paradigm, compelling the generation of fully concrete code even amidst insufficient context. Our analysis of 3 million real-world interactions exposes the limitations of this strategy: 61% of the generated suggestions were either edited after acceptance or rejected despite exhibiting over 80% similarity to the user's subsequent code, suggesting that models frequently make erroneous predictions at specific token positions. Motivated by this observation, we propose Adaptive Placeholder Completion (APC), a collaborative framework that extends HC by strategically outputting explicit placeholders at high-entropy positions, allowing users to fill directly via IDE navigation. Theoretically, we formulate code completion as a cost-minimization problem under uncertainty. Premised on the observation that filling placeholders incurs lower cost than correcting errors, we prove the existence of a critical entropy threshold above which APC achieves strictly lower expected cost than HC. We instantiate this framework by constructing training data from filtered real-world edit logs and design a cost-based reward function for reinforcement learning. Extensive evaluations across 1.5B--14B parameter models demonstrate that APC reduces expected editing costs from 19% to 50% while preserving standard HC performance. Our work provides both a theoretical foundation and a practical training framework for uncertainty-aware code completion, demonstrating that adaptive abstention can be learned end-to-end without sacrificing conventional completion quality.

Problem

• code completion
• uncertainty
• placeholder
• large language models
• editing cost

Innovation

• Adaptive Placeholder Completion
• uncertainty-aware code completion
• cost-theoretic framework
• entropy threshold
• reinforcement learning