cs.MANov 16, 2026

Learning Design Taste through Verifiable Synthetic Environments and Agentic Reinforcement Learning

Authors: Sujay choubey, Siddhant Saxena, Sarthak Kapil, Suhas Motwani

Abstract

Code-generating language models produce user interfaces that build and run, but the pages are often weak in hierarchy, spacing, typography and composition. We study whether a small code model can learn design taste through reinforcement learning in verifiable synthetic environments. The method has three components. First, a layered verification framework splits design quality into hard gates, render-grounded metrics, rubric judges, within-group relative preference and a blinded expert panel. A signal enters the reward only if it passes an admission protocol that tests agreement with experts, causal sensitivity, resistance to hack traps and stability. Second, following SCALER [1], synthetic design environments are parameterised generators with measured difficulty ladders: base pages are written once by an agentic coder, and instances are produced by syntax-tree perturbation operators and constraint samplers, scheduled by an adaptive curriculum with a quarantine rule against reward over-optimisation. Third, a 4B code model is trained with agentic reinforcement learning over multi-turn episodes in which it edits and builds a repository and reads a textual report of its own render, while hidden graders compute the reward. The admission protocol admits 7 of 16 render signals and 6 of 10 rubric dimensions. On TD-Bench, agentic training raises the Taste Index from 0.630 to 0.795 in-family and from 0.598 to 0.700 on held-out families, and wins 0.600 of blinded expert comparisons against the base model and 0.569 against a single-turn control. Most of the advantage depends on the inference-time loop, and cross-brief distinctiveness falls slightly below that of the base model, the only success criterion not met.

Figures & tables

Explore similar work

Jul 22, 2026

Learning to Detect UI Principle Violations via Reinforcement Learning

Small language models and coding agents increasingly generate web front-end code, yet their outputs are typically evaluated primarily for functional correctness. A generated interface may compile, render, and pass unit tests while still violating established interface quality principles, including accessibility barriers, deceptive design patterns, poor visual hierarchy, and excessive decision complexity. Existing auditing approaches face a trade-off between cost, coverage, and scalability: expert human review provides rich judgment but is slow and expensive; frontier vision-language models offer broader reasoning capabilities but remain costly to deploy at scale; and rule-based tools such as axe-core and Lighthouse are inexpensive but primarily capture mechanically checkable accessibility issues. We investigate whether a lightweight vision-language model can serve as an effective critic for generated interfaces. We unify 19 interface-quality principles from three complementary sources of HCI knowledge: WCAG 2.2 accessibility standards, deceptive design taxonomies, and established theories of perception, cognition, and interaction. To train this critic, we construct a verified dataset of approximately 10,000 generated web pages by synthetically injecting known violations into clean, LLM-generated Tailwind pages. Continued reinforcement learning on a 4B vision-language model improves micro-F1 from 36% to 84%, with 13 of 19 principles exceeding 80% F1. The resulting critic can audit generated interfaces, filter low-quality interface training data, and provide a reward signal for design-aware code generation. We release our data-generation recipe and injection/verification prompts to support reproducible evaluation and future work on scalable interface-quality assessment.
Oct 13, 2026

Verifier-Grounded Agentic RL for Provably Correct Code

Reinforcement learning for code usually rewards passing tests, which a model can do while remaining wrong on the inputs the tests do not cover. A program verifier instead decides whether a program satisfies a formal specification for all inputs. Language models are almost always paired with verifiers at inference time only, and turning a verifier into a reward raises two difficulties. It reports only acceptance or rejection, which gives no gradient when every attempt fails, and a policy can cheat it, most simply by telling it to assume what the proof was supposed to establish. We build an agentic environment and reward that address both. The reward reads the verifier's diagnostics rather than its verdict, scoring a candidate by the weighted fraction of proof obligations it discharges, extended with hidden probes from a reference proof's annotations. An integrity audit outranks the verifier, rejecting any candidate that alters the specification or assumes what it was meant to prove. The model works as an agent against a live verifier over many turns, with a separately trained proof specialist as a frozen collaborator. Across Qwen3 models from 1.7B to 14B and a 35B Qwen3.6 mixture-of-experts model, held-out verification on code generation rises from 66.1% to 93.5% on DafnyBench and from 56.3% to 100% on SWE-Proof. This reward leads every other training signal we compare, including a binary pass-or-fail reward trained under otherwise identical settings, with the widest margin on code generation. When the agent must also deliver the repository patch, graded by each project's withheld tests, the fraction of held-out issues it resolves rises from 59.6% to 72.5%. Evaluated unchanged on three verification suites withheld from training, the checkpoints improve on all of them, including one whose language and verifier are absent from training.
Oct 26, 2026

Co-Design Bench: Multimodal Collaborative Coding Evaluation for Agentic Models

Evaluating human-AI co-design requires testing whether coding agents can translate real users' multimodal requests into working application revisions. We introduce Co-Design Bench, a reference-free benchmark of 273 tasks curated from 37,463 deduplicated queries across 26,314 real user coding sessions. Its two tracks, Multimodal Editing (ME) and Repository Editing (RE), cover localized edits with optional image, audio, video, or document inputs and coordinated repository revisions. Tasks span three capability dimensions: Grounded Understanding, UI Transformation, and Functional Implementation. The tasks pair normalized requests and runnable pre-edit applications with 2,331 hidden expert-authored acceptance criteria. Our judge-then-refute protocol grades criteria using source and browser evidence, challenges positive verdicts with a separate LLM verifier, and audits visual and interaction claims without gold code or target screenshots. Under a shared agentic harness, the strongest of ten model endpoints passes an average of 69.9% of criteria per task but fully satisfies only 26.4% of tasks. All ten endpoints achieve lower criterion completion on repository edits spanning three or four change types than on those spanning one or two. Co-Design Bench provides a testbed for studying the gap between partial requirement satisfaction and complete task success under agentic execution for real-user application revisions.