Talk:Enemy Level Scaling/@comment-7645613-20141125113846/@comment-20753550-20141125182458

Author here, your argument is valid. I was concerned about this myself actually, I intended to write an article in the user research on how the formulae can be re-verified, then link it to the "Currentness of the Formula" chapter here who is still a stub and more like a placeholder for that purpose. I will probably find the time to do so next weekend. But I will explain some things here.

I got the health, shield, armor and affinity formula from the official forums, they should still be there, though currentness can be considered in question. I verified the armor formula, shooting Acrid into cloned flesh+alloy enemies (no damage modifiers so changes to damage formula couldn't interfere with armor scaling formula), Elite Lancers, at different levels, no difference between measured value and function value. I checked the hp formula with a Lato, counting shots needed to kill and applying damage modifiers. Did not check shield formula, but it seems alright from the development of the shielding ratio that is visualized in the hitpoint bar in the HUD when looking at a shielded enemy (relative portion of health to shield increases with level, but seems to converge which indicates identical exponent). I did not check the affinity formula although this is actually the easiest one to check, but the values seem to fit.

The damage scaling formula I got through original research, I explained my method and listed the data here.

As I said in the beginning, I intend to write an article on how the coefficients and exponents can be obtained to make them more replicable.

Apart from making the methodology comprehensible, about the actual data, how do you want this, Darthmufin? I see a problem that we have no secure method of authentication at our disposal, google spreadsheet values can be faked about as easily as short youtube videos (switching weapon mod loadouts, using a Corrosive Projection, stuff like that), if someone maliciously wants to feed us illegit data there is not much we can do about it, except replicating the experimental setup and checking our own results, which is pretty easy. Google spreadsheets that anyone could edit also bare the risk of someone changing the listed data or messing with the formulae, though there is probably something like a file version history (not much experience with google docs), but it would take time and effort to detect vandalism in these. So, should we present the data (if so in which format? YT-vid, google spreadsheet, image file?) or do you consider it sufficient to have an article or blog post explaining the methods to the point that anyone can replicate the results?

As it still will be a few days until I can actually write that article/blog post, a short outline. While the structure of the formula has to be guessed initially, we know that the one known to us can be used to produce accurate results, so there is no indication that they changed the structure in some way. So assuming the structure of the formula is correct, you need to collect some data. The way this can be done is different for each stat, check my reddit post on the damage scaling research for an example, important is just that you get several couples of values for enemy level and the value of the stat at that level. You also need to know the base level of that enemy. If you have the value and the level at which it is taken for at least two different levels, you can insert all this information into the equation made from the formula of the assumed structure. As you have two different points of data, you get two of these equations, and all variables have fixed values in these except for the exponent and coefficient. You then have to solve this as a system of equations, transforming two equations with two variables into one equation with one variable (equivalent transformations to get the formula to the structure coefficient = ... and then equating the two, so you merge coefficient = ...1 and coefficient = ...2 into ...1 = ...2 ), then solving this one equation for the exponent (this requires to take the logarithm at some point), and then inserting the value you got for the exponent into either one of the starting equations and solving it for the coefficient. You now have all values for the function, apply the function to independent data points as a control. The function with your values for coefficient and exponent should at any level give the same value for the examined stat as you can actually measure in the game, if there is a deviation either you did something wrong (in the math or in the experimental conception of acquiring the data) or the formula structure doesn't apply.

This isn't 100% true, I did not mention the influence of rounding. Values for hit-, damage or affinity points shown ingame are always rounded, and the rounding error will impact your results, even though slightly. For the exponent for damage, I got something like 1.60023... if I remember correctly, for example. Now, when you get these results, as humans made set these numbers you can assume them to take a "nice" value, not something with a whole lot of decimal digits, so if there are many 0s or 9s in sequence you can assume the actual working value is the round one. Use the value you think is right for the equation, and see if the results you get for any value round to the whole number actually shown ingame. To reduce the impact of rounding on your results in the first place, try to get the data with really large numbers, as the relative error from +- .5 is smaller for larger numbers. If you get reasonably close to a "nice" value, and using that nice value creates consistently correct results after rounding, it can be considered correct (until data is gathered that suggests otherwise).

This was probably hard to follow, I wrote it in a hurry without much structure, as I said, will explain more thorougly and easier to follow at the weekend and link the articles, so everyone can easily review the validity of these formulae, and that we get to know as quick as possible when they will have been changed from more people re-verifying them occasionally.

Until then, a clear announcement in which form(at) references/data should be presented (in this case, and in general) would be nice.