Rubric-to-Code Credit Assignment for Reinforcement Learning

arXiv:2608.27906v2 Announce Type: replace Abstract: Interactive web application generation requires models to produce usable HTML, CSS, and JavaScript applications from natural language requests. Unlike conventional code generation, application quality depends on multiple user-facing functional requirements, each often tied to localized code regions such as event handlers, state updates, DOM fragments, or CSS selectors. Standard GRPO collapses these structured outcomes into a single sequence-level reward and applies the resulting advantage uniformly to all tokens, weakening credit assignment. We propose extbf{Rubric-to-Code Credit Assignment} (RCCA), a reinforcement learning framework that converts rubric-level functional feedback into localized optimization signals over generated code. RCCA builds training tasks around explicit functional rubrics, uses a hierarchical reward to separate format, source-code, runtime, and functional failures, and aligns evaluator-generated textual attributions with responsible code spans and generated tokens. The resulting model, extbf{Ling-RCCA-Flash}, scores 41.25 on MiniAppBench, improving Ling-3.0-Flash by 32.20 points and slightly surpassing Claude Opus 4.5. It also reaches 76.19 on ArtifactsBench, improving the SFT model by 4.48 points and establishing a new top score under the official ArtifactsBench leaderboard setting by surpassing the GPT-5 score by 3.64 points, suggesting transferable implementation-level gains.

Related

• ICA: Information-Aware Credit Assignment for Visually Grounded Long-Horizon Information-Seeking Agents
• Let Credit Follow Computation: Architecture-Aware Credit Transport for Large Language Model Reinforcement Learning
• RISE-RL: Rubric-Informed Selective Exploration for Open-Ended Reinforcement Learning

Source: arXiv cs.AI | 2026-09-01