Past Perfect

About

Trust your eye.

Past Perfect is a small daily game about visual dating. It exists because the world's museums have quietly published enormous, openly licensed collections, and almost nobody plays with them.

How a question is made

  1. Objects are harvested from four museum APIs across ten date windows, so the pool spans centuries rather than clustering in the 1800s.
  2. Each museum's date fields and its written date label are reconciled into a range. Vague labels produce wide ranges, never confident guesses.
  3. Image rights are evaluated per object against an allow list. Anything ambiguous is dropped. The gate is documented here.
  4. Two objects become a question only when their ranges do not overlap.

Where AI is, and is not

No language model is involved in deciding which object came first, and normal play makes no model call at all. Offline, in the build step, metadata is classified by region and object type, difficulty and surprise are estimated, and the one-line reveal caption is assembled from the object's own fields.

What is recorded

No account, no cookie, and nothing that identifies you. Your streak, statistics, passport and achievements live in this browser and are never uploaded — clear your site data and they are gone, including from here.

Two things are counted. Play itself is counted on our own server against a random id this browser made up for itself: how many rounds were started, how often a question is answered correctly, nothing joinable to a person, and no IP address stored. Visits and referrers are counted by GoatCounter, which sets no cookie and stores nothing on your device.

Advertising

There is none. If it ever appears it will not sit between a question and its answer, and never over an artwork. Personalised advertising in the UK and EEA would require a certified consent platform first, which does not exist here yet.

The current build

Objects in play1,014
Questions available23,002
Museums4
Last harvest2026-09-03T23:05:47+00:00