LOADING
Free toolLLMs.txt Generator — help ChatGPT and Claude understand, index and cite your website.Try it free →
LOADING
Make the question of whether the test can be trusted a computed answer, the same way for every experiment and independent of the pressure to ship.
Skill packs from ₹999 for up to 5 skills.
Computes: (1) sample ratio check, a chi-squared test of observed assignment counts against the planned split at a configured threshold; (2) sample size and duration reached against plan, including whole weekly cycles where the plan requires them and whether the test stopped early; (3) missing or duplicated exposure and assignment records; (4) overlap with other experiments on the same population; (5) the primary metric result with the plan's stated test and interval; (6) each guardrail against its stated tolerance; (7) the count of metrics and segments examined against those pre-declared, flagging any unplanned ones as exploratory. All thresholds come from the readout standard. The skill reports numbers and flags; it does not interpret them.
Works in the AI tool you already use. Pick yours for the exact steps — full guide.
Which AI tool do you use?
If you already use ChatGPT on the web or the app.
Download your skill (from the skill's page or My orders) and unzip it.
Open ChatGPT. In the left sidebar click New project and name it after the skill.
Open the project's Instructions, paste in everything from SKILL.md, and save. If the folder also has a knowledge folder, add those files to the project too.
Start a new chat inside that project and ask for the job in your own words, for example:
Type something like
Use the experiment-integrity-check skill. Here is what I need: <describe your task and paste any details, like a list or a document>
You're done when: ChatGPT asks for anything it's missing, then works through the job step by step. Every new chat in this project uses the skill.
Want to share it with your team? Create a custom GPT instead: paste SKILL.md into Configure → Instructions and add any extra files under Knowledge.
Input
{ plan: ExperimentPlan, aggregates: ExperimentAggregates, otherExperiments?: ExperimentRef[], standardId: string }Output
{ integrity: {check, status: "passed"/"flagged"/"failed", value, threshold, detail}[], results: {primary: {estimate, interval, test, p}, guardrails: {metric, estimate, tolerance, breached}[], examined: {predeclared, unplanned}}, decisionGrade: boolean }Yes. Verify your email and the Experiment Integrity Check Skill downloads straight away — no card, no trial.
Any assistant that accepts files or custom instructions: Claude (Code, desktop and web), ChatGPT, Gemini, GitHub Copilot, Cursor, Windsurf, Codex and Gemini CLI, or any model through its API. The download includes HOW-TO-USE.md with the steps for each.
Unzip it, then add SKILL.md to your AI tool: a skills folder in Claude, a Project or custom GPT in ChatGPT, a Gem in Gemini, a rules file in Cursor or Copilot — the how-to-use guide has the exact steps.
Nobody trusts our numbers or our knowledge base. Explain metric moves, reconcile dashboards, read experiments, synthesise feedback and research, and keep knowledge fresh.
