discovery-quantlisted

# Discovery quant ## Core principle Quantitative analysis tells you **how much, how often, and for whom**. It cannot tell you why. Every "why" question routed here must end with a qualitative step, and every qualitative claim about prevalence must end here. Second principle, equally load-bearing: **before believing a movement, check that it happened.** A large share of investigated metric movements are tracking changes, bot traffic, definition changes, timezone artefacts, or reporting-window effects. Check measurement before you check the world. ## Runnable scripts All in `scripts/`, standard library only unless noted, tested and working. | Script | Does | |---|---| | `scripts/sample_size.py` | Sample size, power, minimum detectable effect, duration, SRM check | | `scripts/experiment_analysis.py` | Binary and continuous results with intervals, Bayesian read, CUPED, multiple-comparison correction | | `scripts/cohorts_funnels.py` | Funnels with segment split, cohort retention with flattening check, Kaplan-Meier survival, Simpson's paradox detector | | `scripts/survey_analysis.py` | Kano, Van Westendorp, MaxDiff, PMF survey, Likert done properly | | `scripts/qual_saturation.py` | Saturation curve, code frequency by source, Cohen's kappa | Run any with `--help`. They print their own caveats alongside the numbers, by design. ## The diagnostic sequence When a metric moves, run these in order. Do not skip to step 5. **1. Did it actually move?** Tracking release, SDK versio