Just Over Half Of Second Phases Throw Away Everything The Player Spent The First Phase Learning

A second phase is a promise about what the first phase was for. If the opening moveset carries forward, the forty minutes you spent learning it were an investment. If it is replaced, they were a toll.

We had assumed most fights sat somewhere in between, adding several attacks and retiring one or two. Almost none do. Plotting retained attacks as a proportion of the first phase's moveset produces two clusters with very little between them: a group above seventy per cent and a group below thirty.

The two clusters

Note that attempts rise as retention falls, which on its own might look like evidence that replacing phases are harder and therefore more demanding. The named-cause column argues otherwise: those additional attempts are being spent on deaths the player cannot account for.

Why replacement feels like a different game

Because it is one. A boss whose second phase shares nothing with its first is two encounters sharing a health bar, and the player meets the second of them at whatever resources the first left them with, having had no opportunity to learn it on fresh footing.

This is worse than simply presenting two bosses in sequence, because a sequence lets you fail the second without re-earning the first. A replacing second phase means every attempt at the new material costs a full run through material you have already mastered — which is exactly the arithmetic that produces our highest attempt counts and our lowest explanation rates.

If your second phase discards the first, you have not made the fight harder. You have made a new fight and charged the player forty minutes to look at it.

The additive version, done well

Crowned Ash is the cleanest example in the archive. Eleven of its nineteen bosses have second phases and all eleven retain the entire opening moveset while adding to it. The Warden adds two attacks and extends two existing chains; nothing is retired.

The consequence is measurable rather than aesthetic. That game's named-cause rate is ninety-seven per cent, the highest we have logged, and its attempts-to-first-clean-phase figure is nine against an archive median of fourteen. Players learn faster and understand more, in a game nobody would call easy.

The special case of timed transitions

Twenty-two encounters in the sample transition on a timer rather than on damage, Tidewake among them. These sit at the bottom of the named-cause distribution regardless of retention, at fifty-four per cent median.

The reason appears to be that a timed transition removes the player's ability to control when the new material arrives. In a damage-driven fight, a competent player meets phase two early, with resources; a struggling player meets it late, having had more practice. A timer gives everyone the same moment and reads their preparation as irrelevant.

What about escalation that shortens tells

Thirty-one encounters combine replacement with an animation-rate increase, and they are the worst-performing group in the archive at fifty-one per cent named cause. Canyon Shell is the extreme case, at three and a quarter times.

This combination is worth naming separately because it fails twice: the player's learned vocabulary is discarded, and the new vocabulary is delivered too quickly to be learned in its place. Our note on tell length covers the second half of that problem.

The finding, stated for reuse

• Moveset retention across phase transitions is bimodal: mostly above seventy per cent or below thirty.
• Fifty-four per cent of phased encounters discard most of the first phase.
• Named-cause death rates fall from ninety-one per cent to fifty-eight across that range.
• Attempt counts rise as retention falls, but on unexplained deaths.
• Timed transitions perform worst regardless of retention, at fifty-four per cent.

Status: published. The effect survives removing any single game and holds within each genre band we have enough sample to split.