Simulation run on this device. Scenario values are not sent anywhere — the engine is fully local.
research studio · pro
Experiment Design Copilot
Structure variables, controls, and readouts before GPU-hours burn.
One variable at a time — does your design have it?
Engine 1.0.0 · studio
- Independent variables
- 0
- Controls stated
- no
- Hypothesis stated
- no
- Metrics named
- 0
| Field | Your draft → suggested |
|---|---|
| Hypothesis | one sentence: 'X causes Y because Z' |
| Independent variable | not stated |
| Controls | hardware, batch, prompt held constant |
| Metrics | primary + guardrail metric |
| Cheapest falsification | run the smallest slice that can disagree |
- highNo independent variable stated
Name the single knob you will change.
- mediumNo control condition
State what stays fixed (hardware, batch, prompt) so the delta is attributable.
- lowNo directional hypothesis
Write the one-sentence prediction the experiment can falsify.
One variable, everything else pinned, and a falsification step — that is the whole design.
Method
- Extracts independent-variable candidates, stated controls, and metrics with phrase patterns, then flags more-than-one-variable designs.
- The output is the one-variable template: hypothesis, IV, controls, metrics, cheapest falsification.