I haven't found a single case where 3% would apply. If you want to say 3% +/- 0.3% (that's 3% with a 10% confidence), I've found two cases - out of 506 cars. And, lest we forget, some cars actually gain power.
You cannot say that cars lose 3% of power and torque between the dealer (either UCD or Premium) and the garage because that simply never occurs (on any occasion I have found so far). As an average it's not particularly close either, even if you want to exclude "Gentleman's Agreement" cars.
Let's have ten seconds of meditation while considering how ironic it is to see a value found by empirical analysis challenged by someone looking at the very same datas the value is born from.
The way you consider errors margin as a % has no meaning here.
You have to apply the 97% "rule", then round the value, then consider an error margin of +/- 1 unit. It would lead to high percent error on 50 HP low end car. Then you'll find plenty of Premium cars that match this, and then you'll find exceptions, as i mention in my very first post about that matter (exposing one or more of them won't help here, neither atlking about cars from UCD).
In addition, we must keep in mind that the 97% is likely to be itself an approximative value resulting of two different calculations :
- Car Dealership stat = ROUND(Base stat / constant A)
- Car Just Delivered in garage stat = ROUND(Base stat / constant B)
Base stat would be OC car, and maybe after run-in (i didn't dig much the run-in factor yet), or maybe a never shown value (as the game use for used car's internal price value)
If you "work" with HP, not even mentionning KW, instead of PS/CV/CH , approximation that results of rounding are even higher.
This is why i'm currently using this "rule" as a one way state check among others (that works very well, by the way).
It's a rule of GT games - if it can be made more complex, it will be made more complex. There's no single multiplier that can be invoked to generate expected garage stats from observed dealer stats. For an extreme example, check the Amuse S2000 GT1 '04...
Incidentally, check your weights. Some cars change weight from dealer (UCD or Premium) to garage and, at present, they are not reflected on your site.
And yet, dealership datas are usefull and matches garage data so often that they allow to build what i'll call a good "canvas".
The database has been built with the best data as possible at given time, some from dealers, some from garage, and is then now in "phase 2", opened to users contribution through a form on each car page. If a few weight are wrong (i never run into error on weight data myself yet, so i assume there are few), let's hope some helpfull users will share the correction with everyone.