Repository navigation
Playground: run your AI agent on a human study - #4
Merged
Merged
Conversation
The playground lets a researcher choose a study and a base model, design the prompt the agent takes into the experiment, and see how the run compares to the people in the original paper. The prompt designer offers the four shipped participant prompts, per-run demographic overrides, or a prompt written from scratch. Results arrive as summary tiles, the charts and reading produced for the run, a test-by-test table against the paper's findings, and the transcripts the agents actually produced. Runs are dispatched to HumanStudy-Bench and stored on a run branch of the private jobs repository, the same shape as Build Study; nothing about a run executes inside Vercel. Runs use the shared OpenRouter key with a small participant cap, and a researcher can supply their own key for a larger one. That key is sealed with AES-256-GCM before it is stored and never reaches a browser. The home page now explains what Datasets, Build a study, and Playground each do, and Agent Evaluations opens with a link into the playground. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
|
The latest updates on your projects. Learn more about Vercel for GitHub.
|
This branch was successfully deployed
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Adds
/playground, where a researcher picks a study and a base model, designs the prompt the agent takes into the experiment, and sees where the agent behaved like the original participants and where it did not.Pairs with HumanStudy-Hub/HumanStudy-Bench#14, which has to merge first — GitHub can only dispatch a workflow that exists on the default branch.
What a user does
How it runs
Dispatched to
run-playground.ymlin HumanStudy-Bench, with run state, progress, and results on aruns/<id>branch of the private jobs repository — the same shape as Build Study. Nothing about a run executes inside Vercel. The page polls/api/playground/runs/<id>.Researcher-supplied keys
Runs use the shared
OPENROUTER_API_KEYActions secret at 10 participants per condition. A researcher can paste their own OpenRouter key for 80. That key is sealed with AES-256-GCM underPLAYGROUND_KEY_SECRETbefore it is written anywhere, so the jobs repository only holds ciphertext, the sealed value is stripped inreadRunso it can never reach a browser, and the workflow drops it when the run ends.New environment variable:
PLAYGROUND_KEY_SECRET, which must match the Actions secret of the same name in HumanStudy-Bench. Without it the playground still runs on the shared key and only refuses researcher-supplied keys.Also in this PR
Verification
tsc --noEmit,eslint, andnext buildare clean. Pages were checked in a browser against a synthesized completed run: three charts drawing real data, the test-by-test table, correct summary tiles, and no horizontal overflow at 375px.Not yet exercised: a real dispatched run, which needs the bench PR merged and a deploy.
🤖 Generated with Claude Code