What a gap is
The number beside a song is how many shows the band played between this performance and the one before it. A gap of 0 means they played it again the very next night; 485 means four hundred and eighty-five shows went by. The figure comes from Phish.net, which computes it; nothing here is counted a second time.
The median, and why ten years
Under each gap is that song's usual one — the median of its gaps over the ten years before the show, not over all of history. Forty years of a working band is several different bands. The 1990s dominate any all-time figure, when they played far more shows a year out of a smaller catalog, so all-time gaps run much shorter and better than half of everything came out overdue — a word that means nothing if it fits three songs in five.
Counting a fixed number of past performances instead does not fix it: twenty performances is two years for a staple and twenty-four for a rarity. Kung's last twenty reach back to 1995. Bounding by time instead means a song has to have been in rotation lately to be judged at all.
The median rather than the average, because gap distributions are savagely right-skewed: a staple with a median of 6 carries a handful of 200s, and an average over that would call almost anything ordinary.
The verdicts
A gap outside the middle 70% of that ten-year window gets called. Below it, premature; above it, overdue; inside, nothing is said, which is most songs. The band's ends are computed values that appear nowhere in the song's actual gaps, which is why they are not printed as numbers.
That middle 70% is measured as a ratio around the song's typical gap rather than as a fixed number of shows either side of it. A gap can be 5 or 112 but never less than nothing, so a step from 5 to 8 and a step from 68 to 71 are not the same distance, and reading two percentiles off the list treats them as though they were. Checked against what actually happened next — band built from a song's earlier gaps, then compared with the gap that followed, over 32,605 performances — percentiles held the next gap 76% of the time for the staples and 44% for the rarest songs, so the word “usually” meant two different things depending on the row. On a ratio scale it runs 70% to 49%.
The band is wide enough that a verdict stays worth reading: roughly 9% of performances come out premature, 69% expected and 22% overdue.
Some songs do two things rather than one — a fortnight's rotation, and then a year away — and a single range describes neither. Where the record breaks cleanly in two, the hover says so instead of averaging them: Esther's range tops out under sixty shows, but three of her recent gaps ran 68 or longer.
The bar
The bar is a position, not a length. Its shaded middle is the band above — where this song usually lands — the hairline through it is the median, and the mark is the performance being reported. Left of the shading is sooner than usual for that song, right of it is later. Past three times the upper edge the mark stops at the end and stays there, because beyond a point “very late” is the whole of the message.
Every row is drawn against its own song rather than against the night, so one bustout cannot flatten the rest of the bill — on 168 of 690 shows the longest gap is at least twenty times the median. It shows position rather than magnitude because the number beside it already gives the magnitude exactly; what the number cannot say is whether this was early or late for this song.
A song with fewer than eight performances in the ten-year window has no band to be measured against, so its track is drawn empty rather than implying a comparison that was never made.
What came before and after
A song page shows what each performance sat between. A plain ← or → means only that: the song before it, the song after it, played as separate songs with a stop between them.
Where phish.net recorded that the band ran two songs together, its own mark appears in place of the arrow — > or –> — and it sits between the two songs it joins, the way it does in a written setlist. So an arrow is the absence of a segue rather than the presence of anything, which is worth saying because the two look equally deliberate on the page.
The two marks are not the same claim. –> is a real segue, one song jamming without interruption into the next; > is everything else that runs together, and is also used by convention between songs that are simply always played as a set. The difference, in phish.net's own words.
Which show this was
A report says where the night sits inside its era — the 312th show of 3.0 — and never where it sits overall. There is no honest overall number to give. phish.net offers three defensible totals for how many shows the band has played: 2,239 entries listed, 2,114 that count toward statistics, and 2,106 distinct dates among those. They differ by soundchecks, television and radio sessions, canceled dates, and nights when two separate shows were played.
The disagreement reaches the beginning. 1983-10-30, the show generally called Phish's first, is one phish.net excludes from statistics, so a count built on that flag declares the second show to be number one and every figure after it inherits the error.
Inside an era the count is exact for three of the four. Six dates carry more than one counting show, and all six fall in 1.0 — the earliest being 1985-02-25, close enough to the start that the drift is not confined to six shows but is inherited by every show after them, roughly 1,350 of 1,361. 2.0, 3.0 and 4.0 run one show to a date throughout, so their ordinals are counted rather than estimated. 1.0 shows carry no ordinal, which is a decision and not an oversight: the number could be produced, and it would be wrong by somewhere between one and eight with no way to tell which from the date alone.
Songs with no verdict
A song needs eight performances inside that ten-year window before any of this is said about it. Below that there is no current norm to be early or late against, so it gets its numbers and no adjective. Roughly one song in eleven falls here, which is the honest answer for something the band has nearly stopped playing.
Bustouts
A bustout is a song coming back that had no recent record to be judged against, after a gap of 100 or more. A hundred sits where Phish.net's own setlist notes use the word.
It is not simply any gap over a hundred, and this page said it was for a long time. Where a song does have a recent record — the eight-or-more performances the section above needs — that record decides the verdict, and a long gap is marked overdue instead. Of the 335 performances in this archive with a gap of a hundred or more, 293 are bustouts and 42 are not: Crowd Control came back after 122 shows and Nellie Kane after 146, and both were overdue rather than bustouts, because both were still in the band's rotation.
Within the bustout branch the gap alone decides it, with no test on how often the song was ever played: a gap counts shows, so a large one already proves the song has been in the catalog a long while, and nothing newly written can reach the threshold.
Dormant, rarity, once or twice
Songs with no recent record that have been gone a hundred shows or more sit on their own page, because there is nothing left to rank them by. For a long time that page called all 281 of them dormant, and for 174 of them that was false. Dormant means a song used to be otherwise. A song played once at a Halloween show, as part of a costume set, never had a rotation to fall out of — and a third of the songs played exactly once in this archive were played on a Halloween night.
So the page splits on how many times the band ever played the song: 8 or more and it was in rotation and left, which is dormant; 3 to 7 is a rarity, enough performances to notice and never enough to become a habit; once or twice and it never got going at all.
The archive decides where that line goes rather than taste. Take every silence of a hundred shows or more it holds — 774 of them — group them by how many times the song had been played when it fell quiet, and ask how many were ever ended by another performance:
| Plays when it fell quiet | Silences | Ever came back |
|---|---|---|
| 1 | 176 | 28% |
| 2 | 72 | 33% |
| 3 | 40 | 62% |
| 4 | 37 | 68% |
| 5–7 | 76 | 66% |
| 8–15 | 104 | 70% |
| 16–40 | 120 | 85% |
| 41 or more | 149 | 93% |
Collapsed to the three groups above that is 84%, 65% and 30%. A song that has been played once or twice is the only kind on that page likelier to stay gone than to come back.
One and two are one group, and that is the second thing the table decides. The obvious line is under the one-offs, and it is the wrong one: splitting 1 / 2–7 / 8+ the three groups return 28%, 55% and 84%, while splitting 1–2 / 3–7 / 8+ they return 30%, 65% and 84%. Merging widens the gap at the bottom boundary from 27 points to 35 and costs nothing at the top. On the evidence, a song played twice and dropped is the same object as a song played once and dropped — which is why one heading covers both counts rather than a number naming either.
When those few plays happened matters too, but only for the rarities. Take how many shows passed per performance, and split at two hundred. A rarity whose handful of plays sat close together came back 70% of the time against 38% for one whose plays were strewn across decades. Do the same to the once or twice group and it is 36% against 27% — those songs are not coming back whether the pair was three shows apart or thirteen hundred, which is the other reason they belong together.
There is no finer cut than that available, and the page does not pretend there is. Shows-per-play is a single hump with a long tail rather than two clusters — the forty-eight songs played exactly twice run 8, 12, 8, 8, 12 across the spacing buckets — so any threshold below the one above would be a number this site made up. The years printed at the right of every row say it exactly and without inventing anything: a single year is a song they tried twice one summer, a range of decades is a one-off somebody revived.
Plays, and not something cleverer. Ten other rules were measured against the same outcome — performances inside any fifty-, hundred- or two-hundred-show window, and the span from a song's first performance to its last. None beat the plain count, and span was the worst of them: whether a song was ever in rotation is answered by how many times the band played it, not by how long they had it lying around.
A song called back for one night after years away does not get a fourth name. It is an event rather than a state, and the play count already carries it: the song moves up by one, and a second performance has not moved it out of the section it was already in. Whether it sticks is not knowable on the night, but it is not a coin toss either. Of returns from a silence of three hundred shows or more that have since had three hundred shows of chance, the ones that had been played once before went quiet again for good 43% of the time; those played two to seven times, 27%; those played eight or more, 7%.
Ratings and jam charts
Version scores and the Phish.net show rating both come by way of fouldomain, which is the only place the latter is exposed programmatically. Scores are computed from a mix of community signal and audio analysis, so a version has none until a recording of it circulates — days or weeks after the show, sometimes never. Jam chart entries are Phish.net's own, written months after the fact. Both are treated as optional everywhere they appear, which is why a report published the morning after a show carries neither.
When a report appears
A report is published while the show is still being played. Songs appear on it as phish.net records them, the page says setlist still coming in with how much is there and when it last moved, and it reloads itself every couple of minutes. The index says so far next to the song count, because 24 songs means something different tonight than it will tomorrow.
Knowing when it is finished is the harder half, and nothing in the data says so. There is no show time to reason from, and the format is not promised — a rained-out show can stop mid-second-set with no encore, so counting sets proves nothing. Stability stands in for completeness instead: once a song count has not moved for two hours, the show is taken to be over and the report stops calling itself provisional.
Until that happens the figures are real but partial. A median gap over nine songs is the median of those nine, not of the night, and the preview image shared from that page deliberately carries no figures at all — only the date, the venue, and that the setlist is still coming in — so a link shared mid-show does not freeze a half-finished number into somebody else's timeline.