Experiment Design
What it does
Turns a research goal into a protocol a bench scientist can execute, and, once you have data, takes it through statistics, plotting, and an analysis report.
How it works
Its process discipline is stricter than the other agents’:
- Brief — write the goal, hypothesis, and measurable endpoints down first, and confirm them with you.
- Multiple candidates — propose several viable designs rather than jumping to one.
- Critique — run a structural check and an adversarial review to surface design flaws that would invalidate the conclusions.
- Revise — incorporate the critique into a final design; the structure check is a hard gate.
- Sample size and power — quantitative, not “n = 3 per group”.
Once data exists, it continues: ingest → statistics → charts → report.
What you get
A protocol with groups and controls, sample size and its justification, readouts and methods, a timeline, and risks and failure modes — plus statistics, charts, and an analysis report when data is involved.
Capabilities it draws on
Compound and physicochemical lookups, target and expression data, public dataset search, sample-size and power calculation, assay guardrails, statistics and plotting, bibliography assembly, delivery audit.
This agent defaults to the Pro tier — its candidate-generation and critique stages need the extra reasoning budget.
Orphic