I thought I’d celebrate our 10th full data point by looking over the numbers and seeing if there’s anything taking shape. That said, 10 points does not a data set make, so anything I think I see could very well be due to chance.

Cred gain is, as far as I can tell, the strongest predictor of the total value of deliveries. Looking down the list (sorted by cred gain) there are however a few sudden drops and glitchy looking numbers, due to differences in stats. There’s a nice little collection of 4 runs, all with similar cred-gain (40-50K) but widely varying Luck/Charisma and Total Deliveries Value.

I had the computer draw some graphs of Luck, Charisma, Luck+Charisma and Luck*Charisma for those 4 runs, to compare to a graph of the resulting delivery value, see if any of the curves seemed to match. They did not. From looking at the numbers it seemed like Charisma was the more important stat, so I experimented with weighting the sum of the 2 (instead of Luck+Charisma, Luck+ n*Charisma) and found that making Charisma 3 or 4 times as valuable as Luck got a reasonably shaped curve.

Did some more fiddling with the numbers to standardise the effect of cred-gain, then tried applying it to the other data. The resulting formula was V = (L+4C)*c/80, where V = Delivery value, L = Luck, C = Charisma and c=cred-gain. That did quite well for most of the data (within a few percent of the real delivery value), but gave absurdly large values (up to 40 times the actual value) for the 2 runs on record with very high cred-gain (10 times the amount of cred of other data points). Needs more work, but it’s a starting point.

I suspect there’s a diminishing returns thing going on with cred-gain, so the model described only works well for a while before it starts drastically over-predicting deliveries because it hasn’t devalued the higher cred-gain appropriately. Or it might all be statistical noise Log in to see images!

Log in to see images!