Timing and the run, in full

When shows sell out

The two genres’ evenings are one hour apart: comedy’s best hour on the clock is 7pm and theatre’s is 6pm, and theatre’s 7pm and 8pm are near the bottom of its own day. Saturday is real, Sunday is a weekday in disguise, and the festival got harder to get into through the middle fortnight and easier again in the closing week.

Pooled, the peaks are 7pm and 8pm, the whole range is under eight points end to end, and the worst hour of the day is noon. Against this chart’s own 15.3% average the 9pm and 10pm hours run level and then below. That is the most misleading chart on this page. It is the average of two clocks running an hour out of step.

Comedy’s best hour of the entire day was 7pm, at 27.1% — more than eleven points clear of its own late-morning showing and well over half again the festival average. Comedy rises into the evening. Theatre rises into it too, and then falls off a cliff one hour earlier: its best substantial hour is 6pm at 20.3%, and its 7pm and 8pm are 11.4% and 11.5%, which are two of the three worst cells in its day. The 9pm behind them recovers to 15.5%. So the two genres are not running opposite clocks so much as clocks an hour out of step, and the hour comedy peaks is the hour theatre empties. Theatre’s three worst hours are 10pm at 7.7%, 7pm at 11.4% and 8pm at 11.5%; the pooled chart, which has noon last and 7pm first, describes neither genre’s day.

Across the run the rate climbs and then gives it back. 10.5% on 5 August, 26.7% by the 11th, with Saturday peaks of 32.5% on the 15th and 33.7% on the 22nd — the same catalogue getting harder to book into as the festival advances and the tourist population builds. It does not hold to the end. The 25th to the 28th run at 15.0%, 13.9%, 14.3% and 14.5%, below every day of the middle week, before a closing lift to 18.6% on the 29th and 22.3% on the 31st — the last of those on only 557 dates, because most of the catalogue had already finished. The mid-festival snapshot could not see any of that, and said the climb flattened; on the completed season it reverses. The closing-fortnight arc deserves a chart of its own and does not have one yet.

A quarter of sold-out dates come back

Selling out is not a one-way door, and how often a date reopens turns out to say more about how a venue manages its stock than about how much people wanted the show.

That is a six-fold spread, and it lines up with how each operator sells rather than with how well. ZOO reopened just over two sellouts in three; The Stand and Underbelly reopened more than half; Monkey Barrel between four and five in ten; Laughing Horse reopened about one in eight. A Free Fringe show has no held-back allocation to release and takes no returns, so when it is gone it stays gone — which under a stay-sold-out definition would make it look like the most in-demand model at the festival.

For anyone using the site, the practical reading is the cheerful one: a date showing as gone is worth checking again, and roughly a quarter of them will have something. That is on top of the stock a venue never puts through the central box office at all, which is why every “gone” message here says the venue may still have tickets.

How much warning you get before a date goes

The availability percentage is not equally trustworthy everywhere. In rooms under fifty seats 29.8% of sellouts arrive from a reading that still looked comfortable, and the figure falls steadily as the room gets bigger — to 7.0% at 200–399 seats. Only the largest rooms break the run.

The flat part of that table is flat only pooled. The 200–399 row in particular is two different rooms averaged into one number, and the average lands on a value neither of them holds:

The practical consequence, and the reason this section exists: in a small room the number of seats left is the honest figure and the percentage is the misleading one, and any “selling fast” estimate that divides availability by a rate of sale is systematically over-optimistic below about fifty seats.

Method, in full

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

Four days in early August are a hole in this chapter and are excluded from every weekday and trajectory claim rather than averaged in. An upstream change stopped the availability monitor for that stretch, and a gap in a time series is not a low reading — treating it as one would drag every weekday rate toward whatever the surrounding days happened to do.

The right-censoring this chapter used to carry is gone. The snapshot is taken after the last performance of the festival, so every date has had its whole selling life, and run length and opening date — the two cuts the censoring bit hardest — are now measured on complete runs. What replaces it is a different caveat and a narrower one: 31 August itself is a thin day, 557 dates against a typical 2,200, because most of the catalogue had already closed. What has NOT changed is that every rate here is a floor. A date whose outcome could never be observed still counts as not sold out, and waiting for the festival to end never had any power to resolve one.

From 26 August the availability monitor ran on an hourly schedule rather than its usual ten minutes, to the end of the festival — the fast window was hard-coded to stop on the 25th, which was the previous season’s closing date. Nothing about the daily counts, the weekday table or the depth figures is affected by that: a date is still observed, just less often. What IS affected is every claim about TIMING. The gap between the last reading before a sellout and the sellout itself is up to an hour in the closing week where it was up to ten minutes before, so a sellout in that week reads as more sudden than it was, and the warning table’s “straight to zero” column should be read as an upper bound over those days rather than a measurement. It is stated here once and again on the table itself.

The recovery finding is about inventory and not about demand, and it is the claim on these pages most often read backwards. A date going from sold out back to available usually means stock was released or a hold expired, not that buyers changed their minds. That is also why the recovery chart is drawn to a different scale from the sellout charts and says so: it is a share of an operator’s own sellouts, not a share of its dates, and the two are not comparable by bar length.

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.