experiment-results-interpreter

# Experiment Results Interpreter This skill takes A/B test results — pasted from an analytics dashboard or described in plain text — and returns a plain-language significance assessment, a ship/rollback/extend recommendation with a documented rationale, and a ready-to-paste stakeholder summary. No statistics background or database access required. **Input:** - Test description: hypothesis, variant names, primary metric, test duration - Results: pre-computed (p-value or confidence interval + lift) or raw numbers (visitors and conversions per variant) - Optional: guardrail metrics (secondary metrics to protect) **Output:** - Test Summary, Results Interpretation, Recommendation with rationale, Draft Stakeholder Summary --- ## Language Detection Detect the user's language from their message: - If Russian (or contains Cyrillic): respond in Russian - If English (or other Latin-script language): respond in English - If ambiguous: respond in the language of the trigger phrase used --- ## Instructions ### Step 1: Validate Input 1. Check that the user has provided at minimum: - A primary metric (what was being measured) - At least one result value (conversion rate, lift, p-value, or raw visitor/conversion counts) 2. If the primary metric is missing: ask "What metric was this experiment measuring? (e.g., signup rate, checkout conversion, 7-day retention)" - Exception: if the user refers to "primary metric" or "main metric" without naming it but does provide result va...