What we buy and when we play it
Every game is purchased at retail with our own funds on the day it becomes available, on the platform named in the review. We do not accept review copies, pre-release builds or press access of any kind, and we do not publish until the build we are playing is the one available to buyers.
This costs us the ability to publish on launch day and gains us the ability to describe the game people actually own. Where a game has patched significantly since our audit, the review carries a revision line and we re-log the affected encounters.
How an attempt is defined
An attempt begins when control of the player character is returned inside the encounter boundary and ends when the character dies, the encounter resolves, or the player leaves the boundary voluntarily. Retries from a checkpoint count individually. Attempts abandoned to a crash, an obvious bug or an interruption outside the game are discarded and noted.
We publish no encounter with fewer than forty logged attempts. Where an encounter resolves faster than that — the two unfightable encounters in Hollow Choir are the obvious case — we log what we have and say plainly that the figure is not comparable.
Naming a cause of death
After each failed attempt the logger writes, before retrying, one sentence stating what killed them. If they cannot, the attempt is recorded as having no nameable cause. The sentence is written from memory rather than from footage, deliberately: the figure is about whether the fight communicated at the time, not whether the information existed somewhere in the frame.
Two loggers minimum per encounter, working independently and without seeing each other's sheets until the audit closes. Where their named-cause rates differ by more than fifteen percentage points we log a third and note the disagreement.
| Category | Counts as named | Example |
|---|---|---|
| Misread tell | Yes | Read the sweep as the thrust |
| Greed | Yes | Took a third swing instead of retreating |
| Mistimed resource | Yes | Healed inside a recovery window |
| Camera or geometry | Yes, and separately | Lock-on dropped during a transformation |
| No account | No | Died, do not know to what |
Timing a wind-up
Capture at sixty frames per second, stepped by hand. The wind-up is counted from the first frame in which the animation is distinguishable from the creature's idle or from any other attack it possesses, to the first frame in which damage can be applied. Attacks with a variable hold are recorded as the post-hold count with the hold noted separately, which is why Crowned Ash's delayed thrust appears as sixteen frames and is nonetheless the easiest attack in the fight.
Each encounter's published figure is its shortest measured wind-up, not its average. A fight is judged by the narrowest window it demands, because that is the window that kills people.
Arena and reach
Arena width is measured along the axis the boss uses most, in metres, from in-game traversal timing against a known walk speed. Widest reach is the longest lateral coverage of any single attack. The proportion between them is provisional as a published metric — see the open note in encounter notes — but we report the raw figures in reviews now because they are simple facts about the room.
What a score is
A score on this site answers one question: were these encounters worth the hours we spent on them. It is not an average of the four figures. It is a judgement informed by them, argued out between the loggers, and it can disagree with the numbers where we can say why.
Hollow Choir scores above Sable Ring despite worse conventional fights, because two of its encounters do something none of the others attempt. Hangar Cluster scores below Canyon Shell despite better creatures, because five of its six rooms are the wrong size. Both of those are arguments, not calculations, and the reviews make them in full.
Numbers are how we stop ourselves being persuaded by spectacle. They are not how we decide.
When we withdraw a figure
If a patch changes an encounter's tells, phase structure or arena, the affected figures are marked superseded until we re-log. If a note's effect disappears as its sample grows, the note is marked withdrawn and left published with the withdrawal stated at the top. We have withdrawn one note to date, on checkpoint distance, after the effect proved to be an artefact of which games we happened to have logged.
Standing commitments
- Retail builds, bought with our own funds, no exceptions.
- Forty attempts minimum before an encounter is published.
- Cause of death written before the retry, from memory.
- Wind-ups stepped by hand at sixty frames per second.
- Shortest wind-up published, not the average.
- Scores are arguments; the numbers exist to discipline them.