The question
Can characters given a high-level objective actually survive - and do different drives survive differently? This is the only thing on the page that is an answer; everything below it is the working.
Is it getting better?
The population's survival today against what it was when the first line was retired, by the same rule as everything else on this page: two intervals that do not overlap, or no claim at all. Every word of it is core/evolution.py's.
The record
Selection
The loop that makes the experiment improve itself: a lineage whose survival is measurably worse than the rest of the live population is closed to breeding and to respawning, and a replacement is seeded from the characters that are still alive. Nobody is killed - a retired line's living members finish the lives they are in, no memory is touched and no file is removed. Every judgement here is made in core/evolution.py; this page prints it.
Survival by cohort
How they died
One row per cohort, one column per way to die. Killed by a rival group is the row this experiment exists to watch; killed by their own side next to it is a configuration fault wearing the same clothes, which is why they are never one column.
How well equipped they got
The best rung each life ever reached, not what they died holding. Naked, clothed, armed, geared, military - computed by the Core from a closed vocabulary, so the ladder means the same thing on every game.
The last deaths
Configuration changes
A trend means nothing if nobody can say what changed underneath it. Every life records the configuration it was lived under; these are the points where that changed.