Setlist prediction β 2026-07-22
Madison Square Garden Β· New York Β· based on 1991 prior shows Β· 48 recently-played songs rotation-suppressed Β· model v3.2.0 Β· actual setlist known
Predicted vs. actual
Predicted
Set 1: Cars Trucks Buses [5:37] Β· AC/DC Bag [9:02] β 34% Β· Foam [8:03] Β· Wolfman's Brother [12:48] β 32% Β· Tube [8:01] β 23% Β· Bathtub Gin [14:56] β 46% Β· Reba [12:40] β 16% Β· Stash [12:07] β 30% Β· David Bowie [11:47] β 19%
Set 2: Wilson [4:23] β 19% Β· Down with Disease [16:11] β 36% Β· Simple [13:00] β 50% Β· Tweezer [16:41] β 55% Β· Golden Age [13:00] β 41% Β· A Life Beyond The Dream [7:34] Β· Harry Hood [13:36] β 26%
Encore: Show of Life [5:42] Β· Tweezer Reprise [3:20]
Actual β MADISON SQUARE GARDEN
Set 1: Glide [4:03], Runaway Jim [6:57], Foam [8:10], Axilla [3:28], Stash [12:58], Cold as Ice [0:34] > Love You [3:05] > Hold Your Head Up [1:47] > Sparkle [3:46], AC/DC Bag [11:55] > Reba [11:51] > Possum [8:31] > David Bowie -> The Vibration of Life [2:54] -> David Bowie
Set 2: Mike's Song [15:16] > I Am Hydrogen [3:05] > Weekapaug Groove [7:07], Bouncing Around the Room [17:43] -> My Friend, My Friend [11:38] -> Split Open and Melt [12:52], The Lizards [9:21], Harry Hood [12:01]
Encore: The Horse [1:19] > Silent in the Morning [4:55], Big Black Furry Creature from Mars [4:23], Run Like an Antelope [10:33]
Score vs. what they actually played
Each of these is rare on its own β the percentage is measured over the last 60 shows, so it says how unusual the call was rather than just congratulating you.
The raw score is graded against a perfect setlist call, which never happens. Phish rotate deliberately and go out of their way to be unpredictable, so 100 is unreachable β over the last 60 shows the score runs from about 2 to 42.09, with a median near 19.3. That is why the headline ranks this night against the ones actually achieved rather than against an imaginary ceiling. The number that means something is Skill: how much of the headroom above a baseline that just guesses the most-played songs this prediction captured. 0% is βno better than guessing the hitsβ; anything above it is real signal. On this night it captured 11.2% of that headroom. Difficulty is the average rarity of the songs they actually played (4.55 bits) β a night of deep cuts scores lower for the same skill, which is why the headline is normalised against a perfect prediction of this show.
Sub-scores
Each is out of 100 and rarity-weighted: picking the song at all earns the most, putting it in the right set earns more, landing it near the right slot earns the rest.
Precision & recall
| Metric | Value | What it means | |
|---|---|---|---|
| precision | 33.3% | Of what it predicted, how much got played | |
| recall | 23.1% | Of what got played, how much it predicted | |
| f1 | 27.3% | Harmonic mean of the two | |
| precision (rarity-weighted) | 32.1% | Precision, weighted by how rare each song is | |
| recall (rarity-weighted) | 15.6% | Recall, weighted by how rare each song is | |
| top5 hit rate | 20.0% | Of the top-5 shortlist, how much got played |
18 songs predicted, 26 played. The rarity-weighted variants are the interesting ones: calling a warhorse is easy, calling a bustout is not, and the weighted numbers pay out accordingly.
Sequence
| Adjacent pairs called right | 0 of 15 predicted |
| Adjacent pairs actually played | 24 |
| Longest correct run | 1 song |
Getting two songs adjacent in the right order is much harder than getting both somewhere in the show.
Structure
| Set | Predicted | Actual |
|---|---|---|
| Set 1 | 9 | 15 |
| Set 2 | 7 | 8 |
| Encore | 2 | 4 |
| Total | 18 | 27 |
Biggest win
The rarest song it called correctly β 4.369 bits (uncommon). Right set, too.
Biggest miss
The rarest song it didn't see coming β 8.844 bits (bustout).
By rarity
How it did across rarity bands. The warhorses are supposed to be easy; the deep cuts are where a model earns its keep.
| Tier | Played | Called | Hit rate | |
|---|---|---|---|---|
| staple | 2 | 1 | 50.0% |
|
| common | 11 | 4 | 36.4% |
|
| uncommon | 3 | 1 | 33.3% |
|
| rare | 6 | 0 | 0.0% |
|
| bustout | 4 | 0 | 0.0% |
|
Song by song
Everything they played, rarest first.
| Song | Rarity | Tier | Called? | Right set? | Credit |
|---|---|---|---|---|---|
| Cold as Ice | 8.844 bits | bustout | miss | β | 0.0 |
| The Vibration of Life | 8.844 bits | bustout | miss | β | 0.0 |
| Love You | 8.815 bits | bustout | miss | β | 0.0 |
| Glide | 8.225 bits | bustout | miss | β | 0.0 |
| Hold Your Head Up | 6.531 bits | rare | miss | β | 0.0 |
| Big Black Furry Creature from Mars | 6.523 bits | rare | miss | β | 0.0 |
| Axilla | 6.049 bits | rare | miss | β | 0.0 |
| Sparkle | 5.791 bits | rare | miss | β | 0.0 |
| The Horse | 5.157 bits | rare | miss | β | 0.0 |
| Silent in the Morning | 5.107 bits | rare | miss | β | 0.0 |
| Foam | 4.369 bits | uncommon | hit | yes | 0.818 |
| I Am Hydrogen | 3.976 bits | uncommon | miss | β | 0.0 |
| The Lizards | 3.583 bits | uncommon | miss | β | 0.0 |
| David Bowie | 3.275 bits | common | hit | yes | 1.0 |
| Runaway Jim | 3.086 bits | common | miss | β | 0.0 |
| AC/DC Bag | 3.081 bits | common | hit | yes | 0.75 |
| Bouncing Around the Room | 2.959 bits | common | miss | β | 0.0 |
| Reba | 2.953 bits | common | hit | yes | 0.839 |
| Split Open and Melt | 2.95 bits | common | miss | β | 0.0 |
| My Friend, My Friend | 2.832 bits | common | miss | β | 0.0 |
| Mike's Song | 2.687 bits | common | miss | β | 0.0 |
| Weekapaug Groove | 2.671 bits | common | miss | β | 0.0 |
| Run Like an Antelope | 2.623 bits | common | miss | β | 0.0 |
| Stash | 2.601 bits | common | hit | yes | 0.75 |
| Possum | 2.478 bits | staple | miss | β | 0.0 |
| Harry Hood | 2.203 bits | staple | hit | yes | 1.0 |
Predicted but not played
Reported, but never subtracted from the headline: penalising bold calls by rarity would teach the model that rare songs are never worth predicting, which is the opposite of what this is for.
Predicted by model v3.2.0 Β· scored by v3.2.0 rules.
Top 5 most likely songs
- Down with Diseasein heavy recent rotation; due back by gap; gap 8 (usual ~5)
- Tweezer + Tweezer Reprisein heavy recent rotation; due back by gap; gap 7 (usual ~5)
- Stashin heavy recent rotation; due back by gap; gap 6 (usual ~5)
- Wolfman's Brotherin heavy recent rotation; due back by gap; gap 8 (usual ~4)
- Bathtub Ginin heavy recent rotation; due back by gap; gap 6 (usual ~4)
Predicted full setlist
Set 1: Cars Trucks Buses [5:37] Β· AC/DC Bag [9:02] Β· Foam [8:03] Β· Wolfman's Brother [12:48] Β· Tube [8:01] Β· Bathtub Gin [14:56] Β· Reba [12:40] Β· Stash [12:07] Β· David Bowie [11:47]
Set 2: Wilson [4:23] Β· Down with Disease [16:11] Β· Simple [13:00] Β· Tweezer [16:41] Β· Golden Age [13:00] Β· A Life Beyond The Dream [7:34] Β· Harry Hood [13:36]
Encore: Show of Life [5:42] Β· Tweezer Reprise [3:20]
Ordering is heuristic (openers/encore placed by role); treat it as a likely song pool with structure, not an exact sequence.
Predicted lengths & jam odds
| Song | Set | Likely length | Range (q10βq90) | P(jam chart) |
|---|---|---|---|---|
| Cars Trucks Buses | Set 1 | 5:37 | 5:29 β 6:09 |
5.3%
|
| AC/DC Bag | Set 1 | 9:02 | 5:41 β 10:29 |
34.1%
|
| Foam | Set 1 | 8:03 | 7:27 β 8:23 |
9.3%
|
| Wolfman's Brother | Set 1 | 12:48 | 9:58 β 15:26 |
31.5%
|
| Tube | Set 1 | 8:01 | 6:11 β 12:01 |
22.7%
|
| Bathtub Gin | Set 1 | 14:56 | 11:42 β 20:41 |
45.6%
|
| Reba | Set 1 | 12:40 | 11:32 β 13:45 |
16.1%
|
| Stash | Set 1 | 12:07 | 10:18 β 15:30 |
30.4%
|
| David Bowie | Set 1 | 11:47 | 9:15 β 14:41 |
18.7%
|
| Wilson | Set 2 | 4:23 | 3:46 β 4:49 |
19.2%
|
| Down with Disease | Set 2 | 16:11 | 7:49 β 24:54 |
35.9%
|
| Simple | Set 2 | 13:00 | 1:49 β 35:35 |
49.8%
|
| Tweezer | Set 2 | 16:41 | 2:37 β 32:05 |
54.9%
|
| Golden Age | Set 2 | 13:00 | 8:54 β 18:39 |
40.7%
|
| A Life Beyond The Dream | Set 2 | 7:34 | 7:15 β 9:31 |
7.9%
|
| Harry Hood | Set 2 | 13:36 | 10:54 β 17:58 |
25.8%
|
| Show of Life | Encore | 5:42 | 5:27 β 6:21 |
1.9%
|
| Tweezer Reprise | Encore | 3:20 | 3:01 β 5:33 |
3.6%
|
Lengths are recency-weighted medians conditioned on the set β the q10βq90 band is where the honesty lives. P(jam chart) is the chance curators later flag this version; it is deliberately estimated from curated history only, since the last year of shows hasn't been written up yet.
Bustout watch
Model backtest: top-5 hit rate 39.5%, setlist recall 31.4% over 40 shows. Full scoring history β