Learning the Loop, Not Just the Page: Execution-Grounded Loop Learning for Web Generation

Key points

• Functional Web generation is increasingly optimized with executable rewards, yet existing methods largely focus on the quality of the final page and leave the process of diagnosing and repairing imperfect implementations underexplored.1
• We identify a central challenge in this setting: the Generator and Refiner produce executable artifacts with direct environment rewards, whereas the intermediate Critic influences downstream behavior without a directly executable outcome.1
• WebLoop trains an execution-free Critic with complementary signals for requirement-level discriminability and downstream helpfulness, first establishing reliable diagnosis and then introducing consequence-aware credit, while all three roles are jointly optimized with group-relative policy learning.1
• With Qwen3.5-9B, WebLoop reaches 41.5 Overall on WebRise and 38.9% accuracy on WebGen-Bench, improving the base model by 11.3 and 15.4 points, respectively.1

Sources (1)

• [1]Learning the Loop, Not Just the Page: Execution-Grounded Loop Learning for Web Generation
  arXiv (AI, ML, NLP, CV, robotics, multi-agent) · Oct 8, 09:11 AM

  We introduce WebLoop, an execution-grounded framework that jointly learns generation, critique, and refinement within a shared policy.

  Functional Web generation is increasingly optimized with executable rewards, yet existing methods largely focus on the quality of the final page and leave the process of diagnosing and repairing imperfect implementations underexplored.

Extractive summary: sentences quoted from the sources.