The obvious hypothesis is that shorter wind-ups feel worse in proportion to how short they are, so that an eleven-frame tell is slightly more irritating than a twelve and considerably more than a twenty. That would produce a smooth curve. What we have is a cliff.
Between twenty and fourteen frames, our loggers describe fights as demanding at a fairly steady low rate. Between thirteen and eleven the rate roughly doubles. Below eleven it collapses into a different vocabulary entirely — cheap, broken, unreadable — and it does so almost regardless of how much damage the attack does or how much health the player has.
What the numbers are
| Wind-up band | Encounters | Described unfair | Median named-cause rate |
|---|---|---|---|
| 20 frames and above | 147 | 4% | 91% |
| 16 to 19 frames | 128 | 9% | 87% |
| 14 to 15 frames | 61 | 14% | 81% |
| 11 to 13 frames | 52 | 19% | 74% |
| 9 to 10 frames | 18 | 61% | 48% |
| Below 9 frames | 5 | 100% | 29% |
The step between the eleven-to-thirteen band and the nine-to-ten band is not a gradual worsening. It is a change of kind, and it shows up in the named-cause figure as clearly as in the descriptions.
Damage does not move the threshold
We tested this because the intuitive explanation is punishment: a quick attack that removes a quarter of your health is tolerable, one that kills you is not. If that were the mechanism, the cliff would sit at different frame counts for different damage levels.
It does not. Splitting the sample by damage as a proportion of maximum health produces the same step in all four bands. What changes is how loudly people object, not when they start.
| Attack damage | Encounters | Step located at |
|---|---|---|
| Under 25% of health | 96 | 10 frames |
| 25 to 50% | 141 | 10 frames |
| 50 to 99% | 118 | 11 frames |
| Lethal in one hit | 56 | 11 frames |
The cliff is not about punishment. It is the point at which a wind-up stops being something you see and becomes something you have to have already known.
Why the number should be around ten
We are a review desk rather than a laboratory, and we will not pretend to explain human reaction time. What we can say is what the frame count has to accommodate: the animation has to be recognised, the correct response has to be selected from several, and the input has to be delivered through whatever latency the game and the display impose.
Our sister measurements on input latency, which we do not publish here, put the median chain from button to visible response at around four to six frames on the hardware we test. That alone eats half of a ten-frame window. Below eleven frames there is simply not enough room for recognition to happen at all, which is why the response becomes memorisation.
The two things that shift it
Visibility and distinctness. Neither changes the frame count, and both change whether a given count is survivable.
Spire Of Glass sits at thirteen frames — inside the doubled-complaint band — and produced ninety-four per cent named-cause deaths, well above the archive median. Its opponents are human-sized, fully visible and use whole-body tells. Canyon Shell sits at eight and produced fifty-two per cent, with the tell on a limb of a creature that fills the screen.
The practical conclusion for a designer is that if a fight needs a wind-up under fourteen frames, the tell has to be carried by the whole silhouette and the whole silhouette has to be on screen. If either condition fails, the frame count is a fiction.
What we would not conclude
That longer is better. Encounters above twenty-six frames start attracting a different complaint — that they are slow, readable to the point of tedium, and that the fight is waiting for the player rather than testing them. The archive's best-scoring fights cluster between fourteen and twenty-two, which suggests the useful range is narrower than either extreme.
The finding, stated for reuse
- Complaints of unfairness step sharply below roughly eleven frames of wind-up.
- The step's location does not move with attack damage.
- Named-cause death rates fall from seventy-four per cent to forty-eight across the same step.
- Full-body visible tells make thirteen frames survivable; limb tells on huge creatures do not make eight survivable.
- Best-scoring encounters cluster between fourteen and twenty-two frames.
Status: published. The effect survives removing any single game from the sample, including the five encounters below nine frames that most strongly demonstrate it.