I think that this must be an AoD or AoD version issue:
Experimented on upgrade costs:
1. AoD 1.10, GER, November 1937, -8% hawk lobby, +20% army slider, +15% central planning, no relevant ministers or ideas, one INT-2 to INT-3. 9.58 IC/day (9.78 after day 57) x 65 days = 624.3 icd.
2. All same except July 1938, -12% hawk lobby, +20% army, +20% central planning. 11.64 IC/day x 52 days = 605.3 icd.
The gross change is -3.0%versus -5.1% (from 0.92*0.92*1.35 to 0.88*0.88*1.40) predicted by the sliders. And the two components separately are way off.
Also have no clue why the IC/day changed partway through case 1. (In my first trial of case 2 I forgot to spend anything, and the IC/day increased by +0.20 at the same point and +0.19 more after 70-odd days).
Is there a formula or file that would explain?
EDIT: I did a 3rd trial starting 7/15/1938. This one started at 11.64 but escalated to 12.03 (+about 0.39, as with trial 2 except that it happened in one step and after just 10 days). 53 days = 633.7 icd, or 4.7% than #2. (Counting days is hard b/c of the 1 day delay in these numbers showing on the production screen. So if trials 2 and 3 actual took the same number of days the difference in cost was closer to 2.8%. But why should there be any? The only thing I can think of is that #3 took place in a difference provint with higher infra, but why should that raise cost?