The Tape Measure.

01Nominal length, and how far it strays

Sorted by median. The bar is where the middle 80% of performances land; the dot is the median; the thin line is every version we have. A song with a tight bar is a known quantity. A song with a bar three minutes wide is a decision the band makes on the night, and those are the ones that eat a set.

02The long ones close the sets

Set-closers run 12.2 min against 9.2 mid-set, which looks like a strong positional effect and isn’t one. Hold the song fixed and it evaporates: across the 22 songs played both ways, closing the set is worth -0.3 min, only 9 go longer, sign test p = 0.86. The gap was song identity all along: the closers are the jam vehicles. Same story for segues (-0.7 min, inconsistent).

That null is good news, because it makes the model simpler: length belongs to the song, not the slot. No position interaction to fit. What a set needs is a runtime budget, and a 78-minute set is four Rebubulas or twelve In Strides, which is exactly what a flat 30-song pool cannot tell apart.

03Every song we have a length for

The table view: same numbers, no color needed.

04Sandwiches: 87 nights where a song was played twice

moe. open a song, wander off through two or three others, and come back to finish it: Moth > Water > Yellow Tigers > Moth. Neither half is a performance of the song, any more than the first half of a sentence is a short sentence; the short halves were dragging every jam vehicle’s median down. We keep the longer part as the song’s length for that night and set the rest aside; the summed total is recorded too, because “how much of the night did this song eat” is the question a runtime budget will actually want.

Sorted with the most lopsided first: the ones where the band left the song hanging longest before coming back for it.