This piece exists because "it got frustrating" is not a claim anybody can check. We wanted a way to describe the point at which a difficult game stops being instructive and starts merely consuming time, and the only honest route was to count.
The method is set out on the method page. In summary: one attempt is one try from a checkpoint to death or success, we record the best progress reached in each attempt as a percentage of the obstacle, and we log every attempt at the time in a notebook rather than reconstructing from memory afterwards.
The shape of the curve
Aggregating all 31 games and normalising each obstacle to a hundred-point progress scale produces a consistent picture. The first band is steep, the second is shallower, the third is nearly flat, and a break resets the curve to something close to its early slope.
| Attempt band | Median progress gain | Share of total attempts | Character |
|---|---|---|---|
| 1–5 | +34 points | 9% | Learning the basic shape |
| 6–14 | +22 points | 14% | Rapid, genuine improvement |
| 15–25 | +9 points | 17% | Slowing markedly |
| 26–40 | +4 points | 21% | Marginal |
| 41–70 | +1 point | 24% | Statistically indistinguishable from none |
| 71+ | 0 points | 15% | No measurable learning at all |
Thirty-nine per cent of all attempts we have ever logged fall in bands where progress is one point or zero. That is nearly two thousand attempts spent not improving, and it happened across every genre we tested.
What a break recovers
The most useful finding is also the most actionable. In seventeen cases we deliberately stopped after a flat band and returned after a gap, and in fifteen of those the first band after the break behaved like an early band rather than a late one.
| Break length | Cases | Recovered early-band slope | Median gain after break |
|---|---|---|---|
| Under 1 hour | 4 | 1 of 4 | +3 points |
| 1–4 hours | 5 | 4 of 5 | +14 points |
| Overnight | 6 | 6 of 6 | +26 points |
| Over 24 hours | 2 | 2 of 2 | +21 points |
An overnight break was worth roughly twenty-six points of progress in our data, which is comparable to the entire first band of attempts. The Crowned Ash cathedral fight is the clearest single case: sixty attempts before the break produced two points, and forty attempts afterwards cleared two full phases.
Sleeping on a boss fight was worth more than sixty consecutive attempts. Nothing in any game we tested tells you this.
Per-game figures
The six games with full reviews are listed below with their individual plateau points. The variation is real but narrower than we expected, and only Spire Of Glass departs substantially from the pattern.
| Game | Attempts logged | Plateau begins | Retry loop | Mark |
|---|---|---|---|---|
| Spire Of Glass | 1,340 | Attempt 45 | 0.4s | 9.2 |
| Crowned Ash | 1,104 | Attempt 80 | 6.1s | 8.9 |
| Sigil Run | 486 runs | Run 30 | 2.2s | 7.4 |
| Bastion Line | 212 | Attempt 26 | 11s | 7.9 |
| Lantern Watch | 163 | Attempt 18 | 14s | 5.8 |
| Winterhold Run | 94 | Attempt 22 | 38s | 6.1 |
Note the relationship between plateau point and retry loop. The two games with the fastest retry loops sustain learning far longer, and Lantern Watch plateaus at attempt eighteen partly because its inconsistent rules make learning impossible in principle rather than merely difficult.
Bullet density and pattern recall
A related finding from 600 attempts in a bullet-pattern shooter, which we file here rather than as a separate review because the game itself was unremarkable. Above roughly forty simultaneous projectiles, our performance stopped correlating with reaction speed and started correlating with how many times we had seen the pattern.
| Projectile density | Attempts | Clears on first sight | Clears after 10 exposures |
|---|---|---|---|
| Under 20 | 140 | 71% | 94% |
| 20–40 | 180 | 38% | 86% |
| 40–70 | 190 | 6% | 72% |
| Over 70 | 90 | 0% | 61% |
Nobody clears a pattern of over seventy projectiles on first sight, which means those sections are not tests of skill in any ordinary sense; they are memorisation exercises with a reflex component. That is a legitimate design choice, but reviews describing it as demanding reflexes are describing something that is not happening.
What we would like games to do with this
Nothing elaborate. A visible per-obstacle attempt counter, as Spire Of Glass already provides, lets a player notice their own plateau. Beyond that, a game could reasonably suggest a break after thirty consecutive failures without progress, in the same neutral tone it uses to suggest anything else. Neither of those changes the challenge.
What we would not want is automatic difficulty reduction. Our data says the plateau is a fatigue effect rather than a skill ceiling, and the correct response to fatigue is rest, not a weaker opponent.