How the forecast works

Every percentage on an artist's page is one number: how often the band has played that song across the nights of their current tour, counting the most recent nights more heavily. There is no editorial judgement in it, and no model — just their own recent setlists, arithmetic, and the cut-offs below.

The window

We fetch up to the last 50 shows an artist has on setlist.fm, then find where the current tour starts: walking back from the most recent night, the tour ends at the first gap of more than 42 days between consecutive shows. Only the nights inside that window are forecast from. Everything older is used for one thing only — the "Not on this tour" list.

We only forecast an active tour: the most recent night has to be within 45 days, and there have to be at least 2 nights in it. Below that we show what the band plays live over the years instead, because a one-night extrapolation is a guess dressed up as a forecast.

We ignore setlist.fm's own "tour" field. It is too inconsistently filled in to trust, so the gap between dates is what draws the line.

The weighting

A percentage is a recency-weighted frequency, not a plain count. Each night in the window carries a weight of 0.9 to the power of how many shows back it is, so the latest night counts fully and a show's influence roughly halves every six or seven nights. Prog setlists rotate mid-tour; a song added three nights ago should outrank one dropped three weeks ago, even when the raw counts match.

That is why two songs with the same count can show different percentages, and why each row prints its evidence as two figures — the full-window count, and how many of the last 10 nights it appeared in. The gap between those two figures is the weighting, made visible.

One other adjustment: a night much shorter than the window's median set — under 60% of it, which usually means a festival or a support slot — is down-weighted to 0.4 rather than dropped. It still carries signal (the essentials get played), it just should not dominate. Because the threshold is relative to the band's own median, a band that always plays short sets is never penalised.

The tiers

The cut-offs are fixed, and these are the real ones:

Locks — 85% or higher.
Rotations — 55% up to 85%.
Wildcards — played at least once this tour, but below 55%.
Not on this tour — absent from the window entirely, but played at some point in the history we fetched.
Interludes — set entries that are not songs: jams, guitar doodles, drum solos, tuning, intro and outro tape. setlist.fm has a flag for these and we honour it, but it depends on a crowd remembering to tick it, so we also match on the entry's name. They are grouped and collapsed rather than deleted, because a name filter will eventually be wrong about a real song — and when it is, the song is still on the page.

What it can't tell you

It is a frequency, not a calibrated probability. "85%" means "played that often, weighted for recency" — not "we have checked and it lands 85% of the time". Calibrating against real outcomes needs the hit-and-miss data the attended-shows accuracy card is starting to build.
Legs change. If the night you are looking at is far after the nights we read, or on the other side of an ocean, the set may have been rebuilt in between. The artist page says so on the page when it can tell.
The data is crowd-sourced. setlist.fm is written by fans. A missed song, a mistyped title or a night nobody logged all move the numbers, and nothing here can detect that.
Position doesn't change the number. Whether a song opens or encores generates the "Opener" / "Encore" note beside it, but frequency alone sets the percentage.
Upcoming dates are Ticketmaster-only. Shows ticketed elsewhere, and most festivals, simply will not appear.

Setlist history comes from setlist.fm. ConcertPrep is a non-commercial project — see the terms and privacy policy.