Week 6: A dead option looks like a live one

A choice that does nothing looks exactly like a choice that matters. This week you measure what each of the specimen's seven upgrades is worth, by taking it away.

30 March 2027

Is it me or is it the game? Ban pickup reach, the card that looks least like a weapon, and wins fall from 12 of 51 to 1. What a card is worth comes from the game's rules.

Open the slides

The specimen offers three upgrade cards every level. An earlier build of it had seven cards in the pool and three of them did nothing.

The claim#

An option that does nothing is indistinguishable, to the player choosing it, from one that matters. Nobody finds a dead option by playing, because nothing happens when you pick it, and nothing happening is not an event.

How the dead cards were found, and why they were dead#

Damage, attack rate and pickup reach each measured within 1.5 s of taking no upgrade at all, over 25 runs (4b36c32). The cause was one line of arithmetic. Damage was dealt per second of contact, so a weapon that spun faster touched each enemy more often for proportionally less time, and the attack-rate card was worth exactly nothing.

Two ways to ask, and they disagree#

Both questions below are put to the same player, the 100 ms one, over the same 51 seeds.

Take only this. Give the player several levels of one card and refuse everything else. For movement speed, median survival reads 66.0 s with no speed and with one, two or three levels, then 55.4 s with four. Taken at its word, speed is worthless and then harmful.

Do not take it at its word. 66 s is the first boss’s deadline: the boss arrives at 40 s, and the run ends if it is still alive 26 s later. With no speed, 36 of the 51 runs end at exactly that moment. The median is reading the deadline, not the player. Count the runs that kill the first boss instead, and the same five builds read 0, 0, 0, 8, 18. Week 9 has a name for a reading like 66.0 s.

Speed is not the only thing changing, either. The boss is sized by how many upgrades the player holds when it arrives, counting the starting weapon (week 7), and with everything else refused, each level of speed is one more upgrade held. With none and with one level the player meets the same 884 hp boss; from two levels on, each meets a smaller one than the last. A player choosing normally arrives holding 7 upgrades.

Never take this. Let the player choose normally but ban one card, and count what is lost. Baseline is 12 wins out of 51. This is restricted play, the method in this week’s reading, below. The player still takes a card at every level, so whichever card is banned, they reach the first boss holding a median of 6 to 8 upgrades.

Asked this way, speed is worth 5 wins, and the most valuable card in the game is the one that looks least like a weapon. “Take only this” is a rigged question for any card whose value is in what it lets you do next. In this game it is rigged twice, because refusing every other card changes the boss as well.

Reading for weeks 6 to 8#

Alexander Jaffe and colleagues, Evaluating Competitive Game Balance with Restricted Play (AIIDE, 2012).

Their games are two-player and ours is not. Take the method and leave the setting: to find out what something is worth, play without it.

What you leave with#

A table like the one above from your own simulated player, and the name of the card you were most wrong about.