For what it's worth…
if I understood what you said in global chat one evening a few weeks
ago, the current calculation appears to be a linear interpolation based
on min, max and median values,
If the set of values used were, for instance
{0,1,4,3,7,6,1,0,3,9,6,62,4,36,4,3,6,5},
and ignoring significant place errors...
mean = 8.89, min = 0, max = 62
Thus, the interpolation would skew the coloration towards the two
high values, via deviant maximum and median values, resulting in
coloration showing primarily green (and perhaps light green), plus a
pair of very angrily red residues. I based this conclusion on observed
behavior in the program (before the last update). It behaved in exactly
this way in that form, and "normal" colorations were absent until such
time as the player significantly reduced the instability of the high
valued chains.
If, however, the two "hotspot" values above were isolated, ie,
tossed during calculation of the score coloration function, then the
bulk of the points would have "normal" coloration....then iff values
outside the coloration range were given a final color...say white, or
light purple, or whatever, then both objectives would be met: normal
energy gradations would show for all of the side chains, AND hotspots
would be clearly identifiable from more "normal"
maximums...viola...problem solved (coloration would no longer be skewed
until the hotspots are dealt with), and useful feature gained,
ie,hotspots can now be clearly delineated from high "normal" values...
{0,1,4,3,7,6,1,0,3,9,6,4,4,3,6,5} (toss, 62,36)
mean1 = 3.88, min1 = 0, max1 = 9
then If x> max1, Color(x) = ex:white, else Color(x) = f(x),
where f(x) represents the trimmed coloration function
--- If I'm not mistaken, the old method was similar to what I
propose, except that the threshold is stated as "for all values EQUAL
TO or above the median-based max, let color = bright red" ---
In mine, it would read "for all values above the median based max, a let color = (ex: white)
and the median would be calculated by first removing the "offending"
points, themselves determined by a measure of either standard deviation
or variance, whichever is computationally less intensive. this is where
I suppose it gets tricky, but I hope you see where I'm going. In my
ideal, the hotspots wouldn't be colored red, but they would have a
value different from others, hence white, bright yellow, magenta, or
something else deviating from the normal span
fwiw, I used such a technique to isolate mountainous terrain from
"lowland" terrain in a color coding of a fractal surface generator many
years ago, and while the requirements were slightly different, the
global objective is identical...increased sensitivity in low magnitude
areas and coalescence it in high magnitude areas. One could carry the
idea even further, but it is probably sufficient for our need to simply
delineate the hotspots from the high "normal" values by separating
their colors by more than just a shade
In any case, I'm glad you reverted the color code for the interim,
since I now have a better idea of what is and isn't in need of work 
