This weekly makes claims about minutes lost and endings counted, and those claims are only worth anything if the definitions behind them are public. What follows is precise enough that a reader can disagree with our conclusions using our own figures.
For every game we identify the longest stretch of ordinary play in which no checkpoint is reachable and no deliberate save is permitted. We then force a quit inside that stretch, at the least convenient moment we can find, and record precisely what the game returns on reload.
The quit is a genuine one: closing the process rather than using an in-game exit, since an in-game exit frequently triggers a save that a crash or a closed laptop would not. We repeat the test at least three times per game at different points and report the worst outcome, because the worst outcome is the one that determines whether a player can safely stop playing.
The gap is measured in minutes of ordinary play at a moderate pace, not speedrun pace and not exhaustive-exploration pace. We time it with a stopwatch across at least two runs and take the longer figure, because the point of the number is to describe a realistic worst case rather than an optimal one.
Where a game permits manual saving anywhere, the gap is recorded as effectively zero and the note explains that any loss is the player's own decision. Where saving is confined to rest points, the gap is the longest distance between two of them along a route a player would actually take.
An ending is distinct if it differs from every other ending by more than a single closing scene. Something must happen differently, not merely be described differently. An epilogue paragraph that names a different character is a variant, and we file it as such.
We reach this judgement by playing every ending to completion and recording the final sequences in full, then comparing them against each other. Where our grouping is arguable we publish the groups and the divergence points, as in our Nine Doors review, so that a reader can regroup them.
Branch counts advertised by a publisher are never taken on trust. Nineteen games have now been audited this way; eleven claims were accurate, five overstated by a factor of two or more, and three described something that was not branching in any meaningful sense.
We record separately whether a save captures the whole game state or only part of it. Position without inventory, world state without held items, and chapter progress without side objectives are all partial, and we treat partial state as more serious than infrequent saving.
The reasoning is behavioural rather than technical. A game that never saves teaches a player to save deliberately. A game that saves incompletely teaches a player to relax, and then charges them for it.
Every review answers whether a second playthrough is worth the hours it asks for. A replay must offer material a first run could not reach: new mechanics, new locations, or a reframing substantial enough to change what earlier scenes meant. Seeing a different closing paragraph is not a replay, and hearing the same story with a different companion rarely is either.
A game that exhausts its variations honestly within a single run scores better here than one which withholds a portion to be found on a second. Length is not value, and asking for twenty hours to deliver four is a structural failure regardless of how good those four are.
One score out of ten to one decimal place, describing how a game treats the hours a player puts into it. Technical quality, art and writing are noted where relevant but they are not what the number measures.
| Band | Meaning |
|---|---|
| 9.0–10 | Loses effectively nothing, and any replay it invites is genuinely different |
| 8.0–8.9 | Generous with progress; a replay adds little but nothing is wasted |
| 7.0–7.9 | Workable overall with one badly handled stretch or an incomplete save |
| 6.0–6.9 | Overstates its branching or requires far more hours than its content supports |
| 5.0–5.9 | Destroys work the player earned, or saves partially enough to mislead |
| Below 5 | Progress is unsafe; we would not recommend starting |
Roughly thirty-five per cent of a score reflects what a forced quit costs. Twenty-five per cent covers checkpoint spacing and save availability. Twenty per cent is the honesty of any branching claim. The remaining twenty per cent addresses whether the game's overall structure respects the time it asks for, including the replay question.
All audits are conducted on default settings, on the build available at the time, with no mods and no external save-management tools except where a game's own limitations force us to copy files in order to compare endings. Where we have done that, the review says so plainly.
Measurement errors are corrected in place with a note, and the original figure remains visible. A patch that changes something we measured — a checkpoint added, a save slot introduced, partial state fixed — will produce a re-test and a revised score, with the previous figure retained for comparison. Nothing is revised because opinion has drifted.
No advertising, no sponsorship, no affiliate arrangements and no commercial relationship with any publisher or platform. Games are bought at our own expense. There is no correspondence route on this site, and no facility anywhere on it for submitting anything to us.
Performance benchmarks, hardware comparisons and art criticism are outside our scope. Other publications do that work properly, and adding a thin version of it here would only dilute the single question this weekly exists to answer.
