The algorithm studied 303 shows.
Phish wrote nine more it had never seen.
Sphere Residency '26 · 9 nights · 27 slot calls (set-1 opener, set-2 opener, encore). The model studied. Phish improvised. Here's the gap.
3 of 27 slot calls landed.
On most nights, Phish was unrecognizable to the model.
Three picks per night. Nine nights. Every cell below is one slot the model called — green where the model and Phish lined up, faint where the model was somewhere else entirely.
Apr 16N2
Apr 17N3
Apr 18N4
Apr 23N5
Apr 24N6
Apr 25N7
Apr 30N8
May 1N9
May 2
3 nailed. 12 in the right set. +1 in the right show.
Strict scoring only pays the inner ring — exact song in the opener or closer slot. The middle ring is right show, right set, wrong slot — a real call, just not where the rubric pays. The outer ring is right show, anywhere.
Oblivion as the set-2 opener · 30% model confidence · the model's #1 ranked song for that slot. Confident and correct. The model is taking the W and refusing to discuss most of the other 25.
NMINML opened set 2 on N2 — exactly where the model put it. 47% confidence. The model's other landed slot call. It is now telling everyone about both of them.
The model's #1 encore pick on closing night. Phish opened the encore with it. Phish then chose Fluffhead instead of the model's #2 (Curtain). Two-song encore, first song nailed — that's a slot call that landed.
The model called these at the wrong moment.
Right song, wrong night entirely — or right family, different song. Doesn't fit the right-show tiers in §02; still worth printing.
The model bet the residency on Curtain (With) closing the run. Phish opened set 2 with it five nights early instead. The model was technically correct that the song would matter — just not how.
The model put Frankenstein at the bottom of N8's set 2. Phish saved it for the closing night and used it to crack set 2 — a 67-show bustout. One night off, several positions off, fully on the table.
The model called Axilla. Phish played Axilla (Part II). Same family, different song. The model is choosing to interpret this as a draft pick.
The model loved Saw It Again for the third night encore. It showed up — five nights later, in a different encore. Right song, right slot type, wrong night.
14 songs that missed an entire tour cycle. Eight worth printing.
Songs ranked by gap — shows since the last play. A real bustout means the song skipped a full tour (~30+ shows) before resurfacing. Three different 93-show bustouts in one residency. The bigger the number, the louder the model is saying nothing.
106 shows. The biggest gap of the residency. The model had Mound for closing night set 1, position 8. Phish played it on N8 set 1, position 5. Off by one night and three positions, on by everything else.
TMWSIY as the set-1 opener. 93 shows. The model had it on the right night, three songs deep. Phish opened with it.
93 shows. Showed up in the encore. The model had Sloth for N8 set 1; Phish put it in N7's encore.
Quiet song, loud gap. The model had it on N6; Phish played it on N4. Off by two nights — not bad, considering.
Late set 1, 69-show gap. The model had Wingsuit on the wrong night (N6) at the wrong position. Phish fixed both.
33 shows. The model didn't have Billy Breathes anywhere in the residency. Not even loosely. A true blind spot.
30 shows since the last Dirt. The model never predicted it for any night. Phish played it anyway.
35 shows. The model had Lifeboy for N7 set 2. Phish played it on opening night, set 2, position 5. Six nights early, on by the song. The model has been informed.
Residency-wide bets the model placed on itself: 3 of 6.
Six props. Mixed results. The model is very pleased about Harpua, somewhat ashamed about the closer.
Got the direction. Overshot by 18.
One Phish debut: Brief Time, N2 encore. The model has no prior on debuts. (Three other songs the model didn't predict — Space Oddity, Forbin, Mockingbird — aren't debuts; they're songs the engine's catalog forgot existed.)
A win, technically. Predicting no on Harpua is the cheapest possible call.
Rock and Roll (Velvet Underground) closed the N8 encore. Three covers appeared in encores across the residency — Space Oddity (Bowie) opened N1's, Walrus (Beatles) opened N3's, R&R closed N8's — but only R&R landed in the closer slot. Prop hits anyway.
Verified via phish.in. The longest single song of the residency was the 37:20 Fuego that opened set 2 on N4 (the Fuego → Golden Age → Fuego sandwich). The model's jam-score × duration heuristic landed on N6, where the longest song was actually Chalk Dust Torture at 26:12 — not bad for a heuristic, off by one set of nights.
Curtain (With) did appear — opening set 2 on N4. Run closer was Fluffhead. The algorithm has been informed.
Got a residency-wide bet the model should be forced to take next time? Send it to hello@callingit.live. We'll read it, fight about it, and probably print it on the super card.
8 epic jams. All in set 2.
Phish played 173 songs across 9 nights. 68 of them broke 10 minutes. 28 broke 15. 8 broke 20. Every single longest-of-night jam landed in set 2 — set 2 averaged 12:40 per song, set 1 averaged 9:02. The model didn't predict any of this.
15 covers in 9 nights. 2 artists got two each.
The model has no per-show cover prediction. It bet "yes" on the cover-closed-encore prop and won (R&R closed N8). It did not see this much Talking Heads, Bowie, or Edgar Winter coming. 3 covers landed in encores, 12 in main sets.
9 nights of receipts.
What the model claimed · what Phish played. The model is keeping all of them.
High confidence. Low yield.
Each dot is one slot call. Y is the model's confidence; X is the slot's order across the residency. Green is a hit. The rest is, technically, also data.
All nine nights, in order.
Tap a night to see the actual setlist with the model's predictions overlaid where they landed. No "you" column. The leaderboard is on vacation.
You should have been picking against this thing nightly.
Summer Tour picks open July 1. The model would like a chance to redeem itself. You should not give it one.
Make your first picks →thank you for picking against the algorithm in public.