Weight And Momentum
How movement feels, measured
Read this before disputing a figure. Most disagreements about frame counts turn out to be disagreements about where a frame count begins.
Every number in this publication was taken by hand from high-frame-rate capture under the conditions set out below. Where a figure could not be established reliably, the review says so rather than estimating.
Fifteen hours minimum before a review is written, and in practice more: issue seven totals a hundred and nineteen hours across six games. The minimum exists because movement systems are learned, and a system measured by somebody who has not yet internalised it will be described as unresponsive when it is merely unfamiliar. The mech game in this issue takes about two hours to become legible at all.
Measurement happens in the second half of the playing time, never the first. A figure taken during the learning period is a figure about the reader, not the game.
| Element | Setting | Reason |
|---|---|---|
| Capture rate | 240fps | Four samples per game frame |
| Display refresh | 120Hz, fixed | Removes variable latency |
| Frame rate cap | Locked at 60 | Frame figures require a fixed rate |
| Input | Wired, in a jig | Wireless adds variable latency |
| Assists | Off where optional | Stated in review where not |
| Difficulty | Default | Stated where it alters movement |
Frame rate is locked at sixty even where the game supports more, because a frame count is meaningless without a stated rate and because several systems we have measured behave differently at higher rates. Where a game's movement changes with frame rate we treat that as a finding and report it; it happened once, in issue five, and it was the most interesting thing in that issue.
Acceleration is counted from the first frame of input to the first frame at maximum velocity. Deceleration is counted from input release to the first frame at zero velocity. Anticipation animation is excluded from both and recorded separately, because a player experiences it as input cost rather than as motion.
Commitment is counted as the number of frames during which no alternative input is accepted. This is the figure we consider most important and the one least often published elsewhere: it is what distinguishes a heavy character from a slow one.
The first bench note sets out the full boundary table and the procedure for coyote windows and input buffers. The second covers how scale is established for any figure quoted in metres.
Ten repetitions per configuration for frame figures, and a published value only when all ten agree exactly — which is usually the case, because these systems are integer frame counts. Disagreement between repetitions means something in the setup is varying, and the correct response is to find it rather than to average it.
For probabilistic behaviours — the assistance rates in review five, the drift outcomes in review four — we log a minimum of ninety events and report the count alongside the percentage, so a reader can see the sample rather than trusting the fraction.
Whether the movement is good. Eleven frames of acceleration is excellent in one game and ruinous in another, and no figure in this publication should be read as a score.
The numbers exist to make the argument specific. The argument is still an argument, and where the three of us disagree the review prints the disagreement rather than resolving it editorially.
Feel. There is no composite score, no weighted index and no attempt to produce a single number representing movement quality, because any such number would be arithmetic performed on incommensurable things. Two frames of acceleration and a 1.4-second step are both correct and cannot be ranked against each other.
Also refused: input latency in milliseconds attributed to the game. Total latency includes display, input device, operating system and game, and separating the game's contribution requires equipment we do not have. We measure in frames from capture, which is a measure of the game's response and nothing else.
Every game is bought at retail with our own money. We take no review copies, no early access and no travel, and we have no relationship with any studio. This is partly principle and partly schedule: a quarterly that publishes months after release is of no use to anyone's marketing plan.
Figures are corrected in place with a note at the top of the affected review. Two corrections exist across seven issues, both described on the bench page, and one of them changed a review's conclusion. Readers who repeat a measurement and get a different answer are the reason both were found.