Interesting. (And entirely fair not to be doing any work right now!)I'm at home (i.e. my work pc is turned off and I don't feel like turning it on again in this heat) so I can't triple check but I'm pretty sure light years and lightseconds are floats and worked with as floats.
The reason I was convinced there's some truncation going on of the LY value - by whatever means - is that the revised formula gives exactly correct results for every mission in my sample, and that if I don't use truncation myself I get results like this for my estimate of "Base Reward" - "Ls Component"

There's a very distinct banding and an imperfect correlation.
Whereas if I truncate the LY values

I get a perfect correlation, and a nice round number for the multiplier.
So presumably there's some other effect in the real formula which ends up being possible to model with early truncation of the LY value compensating for an error elsewhere in my formula ... I'll have to think on this some more tomorrow. (Anyone else got any ideas what I'm doing wrong?)