It is the room, not the venue

It is the room, not the venue

The league table above treats an operator as one thing. Inside a single building, rooms of the same size at the same price sell out nearly forty points apart — and the ranking holds up on a completely different set of shows.

Pleasance Courtyard is one operator at one address. Its Bunker One sold out 63.5% of its dates and its Below 27.5%. The two rooms hold 56 and 55 seats, average £13 a ticket, sit in the same building, and finished the season 36.0 points apart. That gap is not an oddity of one venue: across the festival the median operator spans 32.9 points from its best room to its worst, against 40.3 points across the whole ranked operator table from Monkey Barrel down to C venues. Those two numbers are the same order, not equal: a max-minus-min over four to twenty-four rooms of 61–371 dates carries sampling noise the operator range does not, and shuffling the labels inside each chain — rooms kept, dates kept, only which room a date sits in randomised — still produces a median spread of 7.7 points (6.3 to 9.1 over 500 shuffles). The observed 32.9 is far clear of that floor, but roughly a quarter of it is floor. The room is a dimension of the same size as the venue chain the page has been publishing.

The obvious objection is that a room is just where the good shows get put. So rank every room on a random half of the shows that played it, and score it on the other, disjoint half — ranked inside its own operator, so a good chain cannot masquerade as a good room.

In full

The material the scroll has no room for.

The findings page draws the readable cut of this chart: buildings with at least 4 measurable rooms, which is 21 buildings and 139 rooms. It is a sample of the finding, not the finding. Drop that one threshold — keep every per-room floor exactly where it was — and the same query returns 70 buildings and 224 rooms, 49 buildings and 85 rooms more. That fuller chart replaces the sampled one here rather than sitting beside it, so no building is drawn twice on this page.

Method, in full

What this chapter measures, what it excludes, and where it is weakest.

A room’s rate pools whatever that room programmed, and rooms are a long way from genre-pure. Take every room in the comparative population with 60 dates or more — not the drill’s cut, which also wants 3 distinct shows and a rankable operator, but all 247 of them — and only 25% played one genre 90% of the time while 51% are under 70%, the median room running its commonest genre 68.9% of the time. Assembly’s Studio Two ran 51 theatre dates, 37 music and 36 opera, and Underbelly’s Udderbelly split 48 circus, 48 theatre and 46 children’s shows. Since the next chapter of the scroll is this page’s sharpest genre reversal — capacity runs one way for comedy and the other for theatre — a room league table that had never been checked against genre was the obvious hole in the room finding. Adding genre to the stratification moves the held-out odds ratio from 1.32 to 1.27, so the hole is not one. The per-genre gaps are on the scroll.

Both cuts are one query with one set of floors. A room is drawn if it has at least 60 dates and at least 3 distinct shows in this snapshot; a building is drawn if it has at least one such room here, and at least 4 on the scroll. The 21 buildings the scroll draws appear here with identical rooms and identical rates, and a test fails the build if they ever stop matching — the deep cut is an extension of the scroll’s chart, never a second opinion about it.

The 3-show floor is the one doing the most work, and it was added after the fact. Without it a single listing that played one room every day of the festival becomes that room’s entire record, and Underbelly Cowgate acquired an 84-point internal spread out of one show. A room bar is meant to describe a room, so a room whose record is one act does not get one.

Operators that sell every date through the Fringe’s central box office — Braw Venues, Laughing Horse, The Stand, Traverse — were excluded from this page until 2 September, and the reason was sound while these charts ranked the sellout label: that label is reported when it happens and never disconfirmed, so ranking one of their rooms against a room whose misses are observed means inventing the missing bound. What the charts rank now is how much of the room sold, read off the availability gauge, which works the same at those venues as anywhere else — so their rooms are back, ordered on the quantity that is measurable at them. The rate each bar draws is still the binary one, and at those operators it is still a floor, so it is drawn with a ≥ exactly as in the operator league table; the best-to-worst spread, which is a difference between two floors and therefore not a bound in either direction, is not reported for them at all.

Population is the driver set rather than the ticketed floor the rest of the analysis publishes, and since 1 September the twelve shape-detected timed-entry listings are dropped on top of that. One such listing running dozens of slots a day moves a small room’s rate by twenty points: the genre exclusion the driver set already applies catches the ones filed as exhibitions and events, and the shape rule — five or more performances on one calendar date — catches the rest. Hoot 1 was reading 29.4% on 252 dates with one of them in it and reads 40.3% on 149 without. It is the one population difference between this page and the scroll’s headline figures, and it is why a room rate here will not reconcile against a venue rate elsewhere to the decimal.

The held-out test in the section above is the only thing on any of these pages built on a random split. The split is seeded and the seed is fixed, so re-running it does not reshuffle until the data itself moves; the whole split is repeated ten times and the gap never changes sign. Rooms are ranked inside their own operator, so a good chain cannot enter the test disguised as a good room, and the interval resamples shows rather than dates because dates inside one run are not independent draws.

The population, the definition of “sold out” and the two bounds every rate sits between are the same across every page here, and are set out once in how this was measured and what these numbers cannot tell you.