What four hundred and twelve games say when you stop looking at them one at a time

The resume cost, across every game logged

The single most useful number the bureau produces is not length. It is the time it takes to become competent again in a game you last touched three weeks ago. Across four hundred and twelve games the figure runs from zero to fifty-two minutes, and it predicts abandonment more reliably than length, genre, price or difficulty.

The distribution is not smooth. There is a large cluster under five minutes, a thin middle, and a long tail above twenty minutes populated almost entirely by role-playing games, grand strategy and detective games. Nothing in the tail has a finish estimate above thirty per cent.

Three design properties account for most of the variation, and all three are cheap. Whether the current objective is stated in plain sentences rather than quest titles. Whether the player's configuration is visible without reconstruction. Whether the game keeps its context on screen instead of in the player's head — the property Kitchen Table demonstrates at one extreme and Cloudstair Ascent at the other.

What bundles actually buy

Of the four hundred and twelve games owned, one hundred and eighty-seven arrived in bundles. Of those one hundred and eighty-seven, twenty-two have been launched and nine finished. The remaining one hundred and sixty-five have never been opened.

The interesting part is not the waste, which is obvious. It is what happens to the rest of the library. In the four weeks following a bundle purchase, the number of hours logged in games already owned falls measurably, and the number of distinct games launched rises. A bundle does not add playing time. It fragments the playing time that already existed.

Superseded claim, ledger nine: "bundles are the primary cause of large unplayed libraries." That was too strong. Bundles are the primary cause of large unplayed counts. The unplayed hours were mostly bought at full price, one game at a time, with real intent.

Against our own numbers

The finish estimate is the weakest figure this publication prints, and this note exists so that nobody has to discover that on their own.

It is derived from three measured inputs — length, resume cost, and the position of the observable cliffs — and one unmeasurable assumption about how much time a reader has. That assumption is doing more work than the bureau would like. Change it from six hours a week to twelve and the estimate for Cloudstair Ascent roughly doubles.

Three further weaknesses, stated plainly. The cliffs are identified by one person playing once, so a cliff that affects a different kind of player is invisible. The three-week break is arbitrary; two weeks or six would produce a different resume cost and there is no principled reason to prefer any of them. And games without a terminal state cannot be scored at all, which Port Meridian demonstrated at the cost of a rewrite.

Why it is published anyway

• The inputs are measured even if the output is an estimate, and they are all listed on each audit.
• Every estimate is printed with an honest range in the sidebar figure and in the text.
• The alternative — publishing a two-hundred-and-ten hour length with no comment on its consequence — misleads more than an estimate with visible error bars.
• Where the estimate has been wrong, the correction stays on the page with the old figure visible.

Two claims have been narrowed since ledger nine and both remain narrowed. If a third has to go the same way, it will be recorded here rather than quietly deleted.