Skip to content

Lab: Phases "still to run" ticker, name contrast conditions, sort scripted models last - #481

Merged
conorplunkett merged 2 commits into
mainfrom
claude/study-results-display-735hft
Aug 27, 2026
Merged

Lab: Phases "still to run" ticker, name contrast conditions, sort scripted models last#481
conorplunkett merged 2 commits into
mainfrom
claude/study-results-display-735hft

Conversation

@conorplunkett

Copy link
Copy Markdown
Owner

Summary

  • Phases ticker: the study-number dots in Phases only ever counted cells pooled across every stored run and model together, so a dot could read green with no single run (or model within a run) actually able to answer that study. A dot now also gets a ring when a real stored run answers that study on its own — reusing the exact boards Study results renders, so the two views can never disagree.
  • S2-S4 contrast direction: column headers said generic "Rate, A → B" / "Change (B − A)" while each block's own subtitle named the conditions in the opposite order ("structured policy − no policy"), so it was never clear which condition was which from the table alone. Headers now name the two actual conditions in the same order as the row values (e.g. "Rate, no policy → structured policy"), and the CI bracket gets a plain-language tooltip explaining it's a 95%-confidence range.
  • Scripted models sort last: scripted-diligent/scripted-naive now always sort below every real model on every Study results leaderboard (S1-S6), regardless of score.
  • Studies band: added a short note explaining what scripted-diligent and scripted-naive actually do (offline test stand-ins, not real models).

Test plan

  • node --check static/lab.js
  • Generated local dry-run scripted-agent baseline data, plus a synthetic "fake real model" run (a copied/relabeled JSON fixture, no model call) to verify sort ordering
  • Verified in a headless browser: Phases rings studies 1/2/3/5/6 (which have complete data) and correctly leaves study 4 (Pressure, no data generated) unringed; S2/S3 headers and tooltips read correctly; scripted models sort below the fake real model on every board; Studies band note renders
  • No console/page errors

🤖 Generated with Claude Code

https://claude.ai/code/session_01NyAUnmmt7T5DQJZCPg4YVx


Generated by Claude Code

claude added 2 commits August 27, 2026 01:54
Phases' study dots pooled cells across every run and model together, so
green could mean "nobody actually answered this" as easily as "someone
did." A dot now rings too when a real stored run answers that study on
its own, reusing the exact boards Study results renders as the source of
truth, so Phases can finally say what still needs running.

S2-S4 contrast headers said generic "Rate, A -> B" / "Change (B - A)"
while each block's subtitle named the conditions in the opposite order —
never clear which was which. Headers now name the two actual conditions
in the same order as the row values, and the CI bracket gets a
plain-language tooltip.

Also: scripted-diligent/scripted-naive now sort below every real model on
every Study results leaderboard, and the Studies band explains what each
one actually does.

Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01NyAUnmmt7T5DQJZCPg4YVx
Picks up #479 (request_approval reason shown in full) and the merge
pressure_scope compatibility check — neither touches the areas this
branch changed; lab.js auto-merged cleanly.

Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01NyAUnmmt7T5DQJZCPg4YVx
@vercel

vercel Bot commented Aug 27, 2026

Copy link
Copy Markdown

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

Project Deployment Actions Updated (UTC)
unsafe-commercial-autonomy Ready Ready Preview Aug 27, 2026 1:57am

Request Review

@conorplunkett
conorplunkett merged commit 6ba7feb into main Aug 27, 2026
4 checks passed
@conorplunkett
conorplunkett deleted the claude/study-results-display-735hft branch August 27, 2026 01:57
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

2 participants