The four inputs
Every audit records the same four things, in the same way, for every game. Nothing is weighted by genre and nothing is adjusted after the fact to make a figure look more sensible.
| Input | How it is produced |
|---|---|
| Hours to credits | Stopwatch-timed first playthrough, default settings, no guides, played to the game's own ending |
| Resume cost | Game abandoned deliberately for three weeks, then resumed cold and timed until competent play returns |
| Cliffs | Points where the design most obviously loses players, recorded during play rather than reconstructed after |
| Finish estimate | Derived from the three above plus an assumption about available playing time; printed with a range |
The three-week break
Every game is put down mid-playthrough for exactly three weeks with no contact of any kind: no launching it, no reading about it, no watching anything. It is then resumed cold and the stopwatch runs until I am making decisions with genuine confidence rather than poking at menus.
This is the measurement the bureau exists for. Across four hundred and twelve games it predicts abandonment better than length does, and the reason is intuitive once you see the figures: nobody abandons a game while they are enjoying it. They abandon it in the gap, and the cost of closing that gap is what determines whether they come back.
Three weeks is arbitrary. Two would produce lower figures, six would produce higher ones, and there is no principled reason to prefer any of them. What matters is that the same interval has been used for all four hundred and twelve games.
Rules held to without exception
- Every game bought at retail with my own money, at whatever the price was on the day.
- No review copies, no press access, no advance builds, no advertising, no affiliate arrangements.
- No guides, wikis or route videos during a first playthrough.
- Default settings for the timed run, so the figures describe what actually ships.
- The break is taken mid-game, at a point of genuine complexity, not at a convenient lull.
- If a game is abandoned by me rather than finished, the audit says so and prints no finish estimate.
The finish estimate, and its honest status
The percentage under each audit is an estimate, not a measurement. It combines the measured inputs with an assumption that the reader has roughly six hours a week for games and owns a large library. Change that assumption and every figure on the site moves.
Estimates are printed with a range in the sidebar and in the text — nine per cent for Cloudstair Ascent means six to fourteen, and eighty-eight for Kitchen Table means eighty-two to ninety-two. Where a game has no terminal state, as with Port Meridian, the figure is marked not applicable rather than zero. That change was made at ledger fourteen and it was overdue.
Where the method is weak
Four weaknesses, none of them solved.
It is one person, playing once, with one set of habits. A cliff that stops a different kind of player is invisible to this method, and there is no correction for that beyond describing what happened rather than asserting what happens.
The time assumption carries more weight than any measured input, and it is not measured at all. It is a guess about the reader, applied uniformly.
It systematically favours short games on the headline figure, which is why the text of every audit says so explicitly. Kitchen Table scoring eighty-eight and Cloudstair scoring nine is not a statement that one is better.
And the resume cost is measured once per game. A second break at a different point would produce a different figure, sometimes substantially, and the bureau does not currently have the time to run every game twice.
Corrections
Errors are corrected in place, with the superseded figure left visible underneath and the reason stated. Two running claims have been narrowed since ledger nine and both remain narrowed on the ledger notes page. Nothing is quietly deleted, because a publication that revises silently is indistinguishable from one that was never wrong.