We are currently tracking opponent quality, but not using this information at all. We might want some sort of 'adjusted' run differential that takes into account the quality of recent opponents during the rolling window. Since wins/losses as unexpected outcomes might be more impressive/unimpressive, we could reward/penalise that.
This should be factored in before the run differential differential is calculated.
The tuning process would be something like
Identify a sample of training games (maybe some section of 2023?)
Adjust each game's run differential result by the expected quality beforehand (e.g. a prior rolling un-adjusted run differential)
We are currently tracking opponent quality, but not using this information at all. We might want some sort of 'adjusted' run differential that takes into account the quality of recent opponents during the rolling window. Since wins/losses as unexpected outcomes might be more impressive/unimpressive, we could reward/penalise that.
This should be factored in before the run differential differential is calculated.
The tuning process would be something like