Showing most recent matches first.
Or jump to a surprising concert
This is a statistical estimate from a logistic-regression model trained on show history (recency, tour rotation, song age) — not an announced setlist. It assumes the next show continues the current tour as-is. Full methodology in the project write-up (report.md / report.pdf).
Every night's setlist gets a "surprise" score — bits of information in the songs actually played, from a model that only knew what a typical show up to that point looked like. High = defied the rotation; low = thoroughly expected.
Where the surprising shows happen
Circle size = shows played there; color = surprise, empirical-Bayes shrunk (blue = predictable, red = surprising). Cities within 25km of each other (real distance, not state lines) pool as one scene — Boston/Cambridge/Somerville, Durham/Chapel Hill/Raleigh — so a thin sample benefits from genuine nearby history. Anywhere with no real neighbor (Watkins Glen, Pittsboro) keeps its own distinct signal rather than borrowing a same-state city's reputation.
Most surprising cities
Most predictable cities
Most surprising full concerts
Whole setlists that defied the rotation top to bottom (10+ songs, so a single guest cameo can't dominate).
Main set vs. encore
Rarely an encore
Encore leaders
Most surprising individual songs
Specific (show, song) moments the model gave almost no chance to — and got played anyway. 2023+ only (the model's held-out test era).