One run tells you nothing#
SLOP2875 · Week 4
Swipe, or press →
Seed 97 ends at 69.3 s#
That sentence is true.
It is an anecdote.
The claim#
A single run is one draw from a distribution.
The distribution is the thing you are trying to talk about.
How balance arguments go wrong#
Quote one run. Or the best run. Or the run you happened to watch.
Everybody in the argument is being honest.
How wide it is#
The 150 ms player. Ordinary upgrade choices. 501 different seeds.
Shortest run: 29.8 s.
Longest run: the full 120 s.
Nothing differs between them except the dice.
The median as the runs grow#
| Runs | Median survival |
|---|---|
| 1 | 69.3 s |
| 5 | 69.3 s |
| 15 | 81.5 s |
| 51 | 78.2 s |
| 101 | 78.0 s |
| 501 | 73.7 s |
It never stops moving completely#
Fifteen runs put the median 12 s above what one run said.
Five hundred and one pull it back down by 8 s.
What you decide is how much movement the decision in front of you can tolerate.
51#
Why 51#
It is odd, so the median is a run that actually happened.
51 runs of the specimen cost 166 ms.
There is no excuse for quoting fewer.
Thursday#
The run count you will quote from now on.
The smallest difference you will claim with it: how far a second set of seeds moves your median.
Week 4, in full#
The page for this week has the numbers and where they came from.