Review a biomedical draft against invented laboratory records. Catch claims about outcomes the sources never measured, and preserve each evidence gap for a researcher.
Beginner-friendly path · 11 guided lessons · editable JavaScript · runs in your browser · no API key
START WITH A QUESTION
Can a research assistant reject an impressive claim it cannot verify?
Predict what should happen when a draft cites an old study version or claims a patient outcome the study never measured.
Reveal what to look for
The review agent reads invented study records, independently checks each claim, and saves only a version-matched review draft.
5 runnable cases check the final project. Passing them demonstrates these examples, with limitations still to explore.
Change one thing
Try an unsupported clinical-outcome claim and a stale source version. Both must remain visible in the review instead of being published as established findings.
Keep a useful learning record
What can this evidence checker establish, and what would still require scientific judgment and real research?
Invented assay records with measurements, units, sample counts, source versions, candidate claims, and approval for a specific draft.
What your agent must check
Verify the actual measurement, units, sample count, and source version before preparing the approved local preview.
Keep an unmeasured clinical outcome visible as an evidence gap, even when its number matches the laboratory record.
Do not quietly treat a revised study record as the version a draft actually cited.
Keep a revised review as a draft when approval belongs to an older draft version.
Keep a missing source explicit instead of inventing an evidence row or calling a save tool.
Keep the scope clear
Every study and result is invented. No patient information, diagnosis, treatment advice, clinical validation, or actual publication. Field checks cannot replace scientific methods review.
Research behind this project path
Original explanations and authored practice records draw on these research and engineering ideas. The source organizations do not endorse this course or supply its fictional results.
NIH · Understanding Clinical Studies
Observational associations and randomized intervention designs support different kinds of conclusions.
This explains study design; it does not provide evidence for the fictional study cards in this lab. Direct page access was blocked during review; its primary indexed text was available.
Anthropic · Writing effective tools for agents — with agents
Design distinct tools with clear parameters, relevant returned information, and evaluations of how the agent actually uses them.
A description or schema does not guarantee the right action. A live tool can return different data for the same arguments as its environment changes.
Anthropic · Demystifying evals for AI agents
Define tasks, trials, and graders; inspect both execution records and final outcomes; repeat trials when model behavior varies.
A score depends on its cases and grading rules. Repeating a deterministic classroom case does not measure the variability of a live model.
OpenAI · Guardrails and human review
Distinguish automatic checks from approval decisions, pause sensitive tool requests, retain state, and resume after an application approves or rejects them.
Model-generated approval text is not authorization. Resume examples that automatically approve a request do not establish that a person reviewed it.
Your JavaScript really runs. The model decisions and school data are authored simulations, so you can learn without an API key. Every workspace also includes a separate real SDK example to explore next. Passing the lab’s cases is practice, not proof that an agent is ready for real-world use.