Week 3: A perfect player describes nobody

A perfect bot wins 12 runs out of 15 and tells you nothing about a person. This week the simulated player gets a reaction time, and the wins go away.

9 March 2027

Is it me or is it the game? It was the player. Leave the game alone, make the player wait about 150 ms before turning and aim a little off, and the same seeds give 1 win.

Open the slides

Last week the player became a function. This week is about how good that function is allowed to be.

The claim#

A simulated player that reacts instantly and never misjudges a direction is measuring a game nobody will ever play. The numbers it produces are real and they describe nothing.

What happened to the specimen#

The first simulated player for Close Quarters re-decided its direction on every frame. Fifteen runs, ordinary upgrade choices: it won 12. On those numbers the game was too easy and wanted harder bosses.

Then the player was made to wait. It now re-decides about every 150 ms instead of every 16 ms, holds its last direction in between, and aims up to nine degrees off. Same fifteen seeds:

Nothing about the game changed between those rows. Every balance decision made against the first row would have been made for a player who does not exist.

Reading for weeks 3 to 5#

Aaron Isaksen, Dan Gopstein and Andy Nealen, Exploring Game Space Using Survival Analysis (Foundations of Digital Games, 2015).

They do to Flappy Bird what this course does to Close Quarters, and their player model is built from the same kind of limits: precision, reaction time, actions per second. Read the player model this week and the survival analysis next week.

What you leave with#

Two simulated players, one of them with a reaction time, and the win rate of each over the same fifteen seeds.