Track record
Every prediction is checked against the real setlist once the show is over. Here is how they have done, good nights and bad.
Live scorecards
Live grading starts after the first show of the fall tour, on Fri Oct 2, 2026. The morning after each show, its predictions are checked against the real setlist and posted here.
Backtest
A backtest over the 200 most recent past shows. The model was retrained at the start of each tour on earlier shows only, then asked to predict that tour.
Of the songs Phish actually played, the share that were in our top 20. A baseline built like Phish.net’s Trey’s Notebook, the most-played songs of the past year, got 31%.
How far the percentages were from what happened, across every song. Lower is better; the baseline scored 0.075.
Openers are the hardest slot: the pool of songs that open a show is large.
How often the song that opened set 2 was one of our five likeliest.
Encores draw from a much smaller pool, so they are easier to call.
Read these as optimistic. The training window (shows from 2009 on) was chosen by scoring it on these same 200 shows, so the setup was tuned against the answers it is graded on. The live scorecards are the honest measure.