| name | experiment-design |
| description | Design reproducible, well-controlled experiments to test a hypothesis. Use when planning data collection, choosing measures and controls, or structuring an analysis. |
Experiment Design
Turn a hypothesis into a concrete, reproducible experimental plan with unambiguous outcome measures.
Inputs
Require: one falsifiable hypothesis (from hypothesis-generation) and knowledge of the domain's measurement tools.
Workflow
-
Define the unit and population. What is the experimental unit (subject, sample, run, cell)? What population are you reasoning about, and how will units be obtained?
-
Fix the manipulation. Specify the independent variable, levels, and which levels are control vs treatment. Prefer minimal manipulation to isolate the mechanism.
-
Choose the outcome. Define the dependent variable operationally — precisely enough that two researchers would measure the same number. State the measurement instrument.
-
Control confounds. List plausible alternative explanations and how the design rules them out (randomization, blocking, blinding, negative/positive controls, counterbalancing).
-
Plan the sample. Decide the number of units, effect size you can detect, and analysis. State the statistical test before running it; pre-register the analysis plan.
-
Write the protocol. A step-by-step protocol that another lab could execute from prose alone, including data format and where results are stored.
Output
research/experiments/<name>/protocol.md — hypothesis, design, measures, controls, sample size & analysis plan.research/experiments/<name>/run.py (or .R/.ts) — analysis script producing results from raw data.research/experiments/<name>/results/ — data and outputs, reproducible on demand.
Rules
- Do not decide the analysis after seeing the data unless explicitly stated as exploratory.
- Distinguish exploratory analysis from confirmatory tests.
- If the experiment cannot distinguish the hypothesis from a plausible rival, say so and revise.
