Every passenger on this page is real — a row in the public airline satisfaction dataset. Scroll, and watch them tell you — across eight scenes — what actually makes a flight worth taking, and what never did.
Spend on the boarding flow before the tarmac. A one-point improvement in online-boarding scores moves predicted satisfaction more than shaving forty minutes off delays — and the most recoverable passengers are economy-class first-timers, the exact people the current experience serves worst.
Sketch: started as six separate charts in the Observatory; the essay form came from asking "what order would I say this out loud?"
Compute: every figure calculated from all 129,880 rows (train + test) with plain Python — counts, medians, per-cell rates, Pearson correlations. No pandas, no notebook — receipts in the repo.
Craft: one sticky SVG that redraws per scene; forms chosen by the Cleveland–McGill hierarchy (position > length > area). The waffle is the honest pie.
Surprise: I expected delays to matter. The data said otherwise — scene 05 exists because I was wrong.
This dataset also powers the flagship prediction model, the eleven-chart Observatory, and the Atlas of Forms.