experiment-plan
/experiment-plan
Plans the smallest test of a hypothesis: method, sample, duration, cost, success and stop criteria, and who decides. Use when a hypothesis card exists and before building, buying or rolling out the full solution to find out whether it works.
- Author
- SHOTClubhouse ↗
- Repository
- shotos ↗
- Licence
- MIT
- Last repository change
- 10.11.2026
Assessment and security
- Files in this package
- 3
- Scripts
- 0
- HTML comments
- 0
- URLs in SKILL.md
- 0
No signals detected. Repository changed on 10.11.2026, without references to retired models or APIs.
MIT, open: use and modify while retaining author attribution.
What it does and when to use it
Plans the smallest test of a hypothesis: method, sample, duration, cost, success and stop criteria, and who decides. Use when a hypothesis card exists and before building, buying or rolling out the full solution to find out whether it works.
Original author description
Plans the smallest test of a hypothesis: method, sample, duration, cost, success and stop criteria, and who decides. Use when a hypothesis card exists and before building, buying or rolling out the full solution to find out whether it works.
How it works
- Context and handoff
- When not to use this
- Steps
- Output
- Checks
SKILL.md preview
--- name: experiment-plan description: "Plans the smallest test of a hypothesis: method, sample, duration, cost, success and stop criteria, and who decides. Use when a hypothesis card exists and before building, buying or rolling out the full solution to find out whether it works." --- # Plan the smallest test The question is not "how do we build this" but "what is the cheapest thing we can do to find out whether it is worth building". Plan that test, and agree how it will be judged before it runs. **Stage:** Hypothesise. **Rules:** 2 Agree acceptance first; 1 Stop at the boundaries. ## Context and handoff - **Business question:** what is the cheapest test that could show this bet is wrong? - **Use this when** a hypothesis card exists and someone is about to build, buy or roll out. If there is no card, write one with `hypothesis-card`; if a lookup or short chat answers it, skip this. - **Minimum context:** the card's belief, measured baseline, guard measure, the authorised population and the owner's limits on spend and contact. - **Decision boundary:** the plan is agreed before the test runs and its criteria do not change afterwards. Money, outside contact and live systems need the owner's yes. - **Next step:** `pilot-before-scale` when the test is a limited live run; otherwise run it and read the result back against the criteria, then the decider continues, changes or stops. Save expected evidence, stop conditions and owner with the plan. ## When not to use this Skip it when the hypothesis can be answered by a lookup or a conversation that costs minutes, or when the decision is already made and the work is required. Do not turn a plan into a research project. ## Steps 1. **Start from the hypothesis card.** Copy the belief, measure, baseline and stop result. If there is no card, write one first with `hypothesis-card`. 2. **Find the riskiest assumption.** Pick the one assumption that, if false, makes the whole idea fail. The test is aimed at that one. 3. **Choose the smallest method that tests it.** Options, from cheapest: ask five people directly; look at existing data; a mock-up or manual version run by a person (the "concierge" test); a link or button that counts clicks before anything is built; a limited pilot with real users; a split between two groups. Choose the cheapest one that could actually show the belief is wrong. 4. **Set the sample.** Say who is included, how many and how they are chosen. For small numbers, say what the number can and cannot show, and do not claim a statistical result from it. 5. **Set the duration and cost.** Give a start and end date, staff hours and money. Money, outside contact or live systems need the owner's yes before the plan starts. 6. **Write success and stop criteria before the test.** - Success: the result that would justify the next step, as a number or a plain yes/no. - Stop: the result that ends the test early, including any harm to the guard measure. - In between: say what you will do if the result is unclear (usually: run once more with a changed design, or stop). 7. **Name who decides** and when they will look at the result. Agree that the criteria will not change after the test starts. 8. **Plan the read-back.** Say where each number will be read from and who records it. Plan to report failures as plainly as successes. 9. **Check the harm.** List anyone who could be affected by the test, and what they will be told. Anything that touches customers, children's data or money has to be shown to the owner first. ## Output Use [template.md](template.md). Save it with the work, for example `.work/<task>/experiment-plan.md`. ## Checks - The test targets the riskiest assumption, not the whole idea. - Success and stop criteria were written before the test. - Duration, cost and sample are stated, with the limits of the sample. - One person decides, and the owner has said yes to any spend or outside contact.
Files in this package (3)
- README.md 3 KB
- SKILL.md 4 KB
- template.md 1 KB
Versions
| Repository | Score | Licence |
|---|---|---|
| This page · SHOTClubhouse/shotos | 100 | MIT |
| jiangfeng-strathclyde/Athena | 100 | MIT |
| AfonsoZhang/auto-research-system | 100 | MIT |
| zjunlp/mechanist | 100 | MIT |
Database snapshot: 13.11.2026 · package version 7835a3f10a63 (SHA-256 digest of the entire package)
Downloads in the last 30 days: 0 · ranking
Report a problem with this skill: harmful content, incorrect descriptions or an author’s removal request.
Similar skills
- 90create-experiment-design
Design a rigorous A/B test or experiment when the user asks to create an experiment, design an A/B test, or validate a hypothesis - 100pm-ab-test
Design rigorous A/B tests with hypothesis formulation, sample size calculation, success criteria, guardrail metrics, and rollout planning. Includes Bayesian vs frequentist guidance and compliance-aware staged rollout for ERP features. Use when someone says "A/B test",… - 100validation-plan
Use when a recommendation, hypothesis, action plan, or business decision needs explicit evidence, tests, decision gates, and continue-adjust-stop criteria. Use when applying validation planning, experiment design, hypothesis test design, decision gates, or learning plan. - 100cheapest-experiment
Turns a recommendation into its cheapest decisive test — the smallest bounded pilot, its cost and duration, a pass/fail rule fixed before the run, and what each outcome would change — in the spirit of Ries's Lean Startup build-measure-learn loop and MVP. Use when a brief ends in… - 100hypothesis
Write testable product hypotheses with clear success metrics, baselines, targets, and timeframes. Produces a structured hypothesis statement, supporting evidence, a validation plan with method and sample size, and a decision framework for what to do if validated or invalidated.… - 100ab-test-setup
Design, plan, and analyze A/B tests with statistical rigor. Use when the user asks about A/B testing, split testing, experiment design, statistical significance, sample size calculation, test duration, multivariate testing, or conversion experiments. Trigger phrases include "A/B…