Skip to main content

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
Structured design
FieldYour draft → suggested
Hypothesisone sentence: 'X causes Y because Z'
Independent variablenot stated
Controlshardware, batch, prompt held constant
Metricsprimary + guardrail metric
Cheapest falsificationrun the smallest slice that can disagree
  • high
    No independent variable stated

    Name the single knob you will change.

  • medium
    No control condition

    State what stays fixed (hardware, batch, prompt) so the delta is attributable.

  • low
    No 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.