Ok, new influx of data into the system… sorted it out to move the runs that are pressed against 40k/forum off to one side, then extended my analysis over all of the data. Current setup has a column for the prediction, given different values fed into the basic formula that it seems to follow, then a column that divides the actual value by the prediction (so if that’s less than one, the run was worth less than predicted, over 1 it got more deliveries than predicted), then a column for the difference between that division and 1, taking the absolute value to get a measure of how far wrong the prediction is for that data point.

On the tops of the columns I have a cell each for the median of the actual/prediction column, and the maximum and average differences from 1.

So my method at the moment is to pick a pair of weightings for Luck and Charisma, then adjust the factor that credgain is divided by until the median is on 1, then see what the max/average error is. I’d like to point out at this point that I’m not a statistician, so there’s probably a better method, but trial and error for the win. Log in to see images!

Current best: Weight Charisma at 3 times luck, divide cred-gain by 69. Biggest error came in at 0.7 times the prediction, and the average deviation is 0.06, it also has the plus point that 69 is the kind of number someone programming this thing might decide to use…

[tl;dr: Current best formula is Delivery Value = (Luck + 3*Charisma) * (credgain/69), capped at 40,000 per unique forum pwned]