better compute, artifacts, and UI

[Aayam Bansal (aayambansal)]

Summary

• make Compute a project-scoped live ledger with active kernels, commands, Modal, and GPU jobs only
• add meaningful notebook and R action titles, compact cell inspectors, and durable Generated artifact galleries that open in Files
• keep failed scientific calls compact while promoting successful cells and results
• redesign Customize as one calm 12px system across every lazy-loaded panel, with consistent materials, spacing, pressed feedback, focus states, and accessibility fallbacks
• refine the home project surface into a quieter compact workspace with factual project metadata
• remove editorial Quick Models descriptions and show only capability, context window, and provider
• preserve model-specific ChatGPT and Codex context windows instead of flattening the entire family to 272k; flagship models now show 1m and mini models 400k
• record the Claude Science behavior comparison in docs/notes/claude-science-ui-behavior-audit.md

Verification

• backend bun test: 1,867 pass, 2 skip, 0 fail
• focused frontend suites: 32 pass, 0 fail
• focused Codex provider suite: 82 pass, 0 fail
• frontend and backend typechecks pass
• pre-push monorepo typecheck: 7 of 7 packages successful
• production build passes
• Prettier and git diff check pass
• browser-verified the rebuilt app at http://localhost:4097: all 11 Customize panels have no visible text above 12px and no horizontal overflow; Quick Models shows factual metadata; OpenAI Codex models show model-specific windows; home, Compute, and artifact surfaces remain compact

Context

Successor to merged PR 276; this PR contains the Claude Science workflow audit and implementation.

[vercel]

The latest updates on your projects. Learn more about Vercel for GitHub.

Project	Deployment	Actions	Updated (UTC)
openscience	Ready	Preview	Aug 9, 2026 6:58am