A Closing Date That Moves Both Ways, and Who Actually Moved It

An exhibition date is a reading, not a fact

An exhibition date is a claim about the future made by a museum's web page, and this archive stores the date it read along with the day it read it. That second half is what makes the first half checkable. A record that holds only the current date can never tell you whether that date has always been the date, and a calendar that cannot say when it last looked is a calendar that quietly rewrites its own history every time it updates.

The corpus at issue holds 937 exhibitions, gathered mostly from museums' own listing pages: 450 currently running, 305 closed and 182 still to open. Every one carries a first_seen date, a last_confirmed date and a change log. The log is the interesting part, because it turns out to say more about the reader than about the museums.

1,605 log entries, and only 54 of them are a date moving

Across the whole corpus the change log holds 1,605 entries, and the overwhelming majority of them record nothing changing. A daily pass walks the list and stamps each show as closed, current or upcoming, which accounts for 1,407 entries on its own. A further 140 are the note written when a show is first listed. That leaves 54 entries in which a date genuinely moved, and four in which the archive corrected itself.

Fifty-four movements out of 1,605 entries is a reassuring number at first glance. Museums mostly do what they said they would do. The trouble is where those 54 sit.

Seven exhibitions carry every date move in the corpus

All 54 movements fall on seven exhibitions out of 937, and two of those seven moved exactly once each, which is what a real change of plan looks like. The remaining five carry 52 movements between them. One show at the Acropolis Museum in Athens accounts for 31.

A single show generating 31 date movements in about two weeks is not a museum being indecisive. It is a signal that something in the reading is unstable, and the log is detailed enough to show which.

Thirty-one moves between exactly two dates

The Acropolis Museum's 31 movements run between exactly two closing dates, 30 August 2026 and 6 September 2026, sixteen times in one direction and fifteen times back. They begin on 24 August and continue through early September, sometimes twice on the same day, and the source is one URL throughout. A museum that had genuinely revised its closing date sixteen times would have said so once, in a sentence, somewhere on the page.

Harewood House behaves the same way and does it on both ends. Its Cecil Beaton show flips its opening date between 10 August and 2 September and its closing date between 4 October and 18 October, ten movements in total, the two pairs moving in step. When both ends of a run swap together, the honest reading is that two different versions of the same listing are being seen in alternation, not that a house is shuffling a show back and forth by three weeks.

Two shows at one museum exchanged each other's closing dates

MACBA in Barcelona is the clearest case, because two of its exhibitions oscillate between the same pair of dates. Anna Moreno. The Third Twist flips between 28 September 2026 and 17 January 2027 seven times over nine days, twice on 16 September, twice on 18 September and twice on 23 September. Prisoners of Love, a separate show at the same museum, moves twice, both times from 17 January 2027 to 28 September 2026.

The two shows' settled dates are 17 January 2027 and 28 September 2026 respectively, which is to say each one keeps being handed the other one's closing date and then handed its own back. Anna Moreno's source is the museum's general exhibitions index; Prisoners of Love has its own page. A listing index that carries several shows at once is exactly the kind of page on which one show's dates can be read against another show's title, and the log is what makes that visible.

For two of the seven, the archive names itself as the cause

Four log entries in the whole corpus are the archive correcting itself, and they are the most valuable four entries in it. They fall on the Naples Parthenope show and the Harewood Cecil Beaton show, and they say plainly what happened: a stale staged file that had never been archived was replayed into the live data by a merge run without a scope, resetting both shows to dates that had already been superseded.

One of those entries records the trigger exactly. The merge script was called with --help, which it does not recognise as a flag, so instead of printing usage it ran the full unscoped merge. The entry notes this as the fourth occurrence of the same cause. Another names a stray duplicate slug, reintroduced by the same mechanism, as the source of a wrong closing date that was then read back as a change.

So for two of the seven exhibitions the movement has a documented internal cause and it is not the museum. For the Acropolis and MACBA oscillations the log says only that the date moved per the source we read. Those are consistent with the same mechanism and this archive does not claim more than that, because the log does not record it.

The same ambiguity sits on a single object's page

None of this needs a scraper to go wrong. Read on 24 September 2026, the Mauritshuis page for Girl with a Pearl Earring carried two separate absences: the painting is not on display from 15 August to 4 October 2026, and the museum is closed to the public from 24 August to 20 September 2026 for maintenance. Both statements are true, they overlap without matching, and the painting is away for two weeks after the building reopens.

A visitor asking when they can see the picture has to notice that the shorter, more prominent closure is not the answer to their question. A database that stores one date for this object will store the wrong one about half the time, and it will not know which half. This is the same failure as the oscillating listing, without any software involved: one page, two valid readings, and nothing on the page ranking them.

What a change log has to record to be worth keeping

A log that records only the new value is worse than no log, because it lends the authority of a timestamp to a reading that may be an artefact. The entries that earned their place here are the ones carrying three things: the old value, the new value and the reason, including when the reason was our own tooling. Four such entries turned 52 apparently mysterious movements into a diagnosis.

The cheap improvements follow from that. A move that reverses a move made within the last few days should be held rather than written, and flagged for a human. A show whose dates come from a shared index page should be re-read against its own page before the index is believed. And a staged file that has been merged should be archived the moment it is merged, because a file that can be replayed will be replayed.

How to read an exhibition date

Treat a published run date as the answer a page gave on a particular day, and ask for that day. Where this archive shows a date it shows when it was last confirmed, and a date confirmed three weeks ago is a weaker claim than the same date confirmed yesterday, even though the two look identical on the page.

Where you have a choice of source, prefer the exhibition's own page to a listing index, and prefer either to an aggregator. Where a date matters enough to travel for, the museum's ticketing page is firmer than its editorial one, because a museum that is wrong about its own opening hours sells tickets it has to refund. And treat any date that has recently moved as unsettled until it has held still for a while, which is advice this archive had to learn about its own data first.

Why this archive is free to read

We do this for everyone who loves art. The people who have studied it for years, and the people who just stopped to look. Iconotheca exists so that what we learn belongs to all of them.

Nobody pays us for this. No ads, no sponsors, no commercial interest. These works belong to all of us, and we would rather share them with as many people as we can than keep them to ourselves.

If it gave you something today, tell us to keep going. Follow us, leave a like, or write a positive comment. We read every one, and they are what keeps us going.