The four grades
Every game measured receives one of four grades, assigned from recorded sessions rather than from reading its options menu. The definitions are deliberately about the player's position rather than the game's technology.
| Grade | Definition | Count |
|---|---|---|
| Clean | A session can end at a natural boundary with no loss of any kind | 11 |
| Soft | Leaving costs a small amount of repeated play, under five minutes | 9 |
| Sticky | Progress commits only at a milestone whose arrival cannot be predicted | 5 |
| Hostile | Leaving destroys progress or triggers a penalty the game imposes deliberately | 6 |
The finding that surprised us
Nineteen of the thirty-one games autosave at least once a minute. Of those nineteen, seven are graded sticky or hostile. Frequent saving and stoppability turn out to be almost unrelated, and the reason is straightforward once stated: these games save your position, not a state you would want to resume from.
A tower defence game that autosaves every wave but restarts the map on quit has saved nothing useful. A card game that writes the board to disk continuously but only commits a case on completion has saved nothing useful either. In both cases the technology is present and the design ignores it.
Autosave answers the question "what happens if the power fails". Stop quality answers the question "what happens if I choose to go to bed". Almost no game treats those as the same problem.
What actually predicts a clean stop
Two variables between them account for nearly every grade in this log.
Unit length. The shorter the smallest complete unit of progress — a room, a puzzle, a delivery — the cleaner the stop. Eleven of the eleven clean games have a unit under six minutes. None of the hostile six has a unit under forty.
Reward visibility. Whether the player can see a reward that lies beyond the end of the current unit. This is the variable that separates soft from sticky, and it is entirely a presentation decision: the same progression, with the next unlock hidden until earned, produces measurably shorter sessions.
Distribution by genre
Genre correlates with grade more strongly than we would like, which limits how useful the finding is — a reader who already knows they cannot stop playing roguelikes has not learned much from being told that roguelikes are usually sticky.
| Genre | Measured | Clean | Hostile |
|---|---|---|---|
| Puzzle | 6 | 5 | 0 |
| Driving and flight | 5 | 3 | 0 |
| Roguelike | 5 | 1 | 1 |
| Strategy and defence | 7 | 0 | 4 |
| Cards | 4 | 1 | 1 |
| Simulation | 4 | 1 | 0 |
The single clean roguelike is the exception that made this log worth keeping. It demonstrates that the grade is a design choice rather than a property of the form, which is the entire argument this publication makes.
Where the grade changed after a patch
Three games have moved between grades since being measured. Two improved when updates added mid-unit saving; one moved from soft to sticky when a seasonal event introduced a timed bonus.
In each case the original figures were struck out and the game measured again over a fresh twenty evenings rather than adjusted. This is slow and it is the only method that produces figures we would defend.
What this log will do next
The current sample is too heavily weighted towards single-player games. Eight of the thirty-one involve other people in some way, and none of the four grades describes the situation where leaving costs somebody else something rather than costing you.
That is likely to require a fifth grade rather than a reinterpretation of the existing four, and it is the reason this log remains open.
Current position
Save frequency does not predict stoppability. Unit length and reward visibility do, and both are design decisions rather than technical constraints. Eleven of thirty-one games are cleanly leavable; six punish leaving outright.