Hero at 5* which could have XP at 3000 pt, how the XP tiny bar would useful? I think you aren’t using 1* to feed them up because AFAIK the limitation of training heroes remains 10.

The answer is saturation logarithmic scale. Let me explain with an example.

*** NEAR-LOGARITHMIC XP SCALE ***

The common hero at 2* would be trained with heroes at 1* which add 130 or 150 depending the match of the colours. The 5* hero trained with same colour 1* heroes would require 20 to fill an entire level passing through.

In a 5* hero scenario the difference between (A) 3000-150=2850 and (B) 3000-300=2700 would be 1/20 of the XP tiny scale the same difference between 18 and 19. Very difficult to see. You are right.

Semi logarithmic scale would help us to highlight that the heroe (A) requires less effort to pass.

LOGARITHMIC BAR

Imagine a bar of 10 pieces stick together.

The first piece would be 2px long, the second piece would be 2² = 4px long and the 10th piece would be 1024px long. The whole bar would be 2047px long. On this scale the difference between the last piece are half of the bar.

NEAR LOGARITHMIC BAR

Like above but the pieces length follow a different rule like this one:

L(n) = (2^n) / n

With a bar of 10 pieces the last piece would be long 102px and the whole bar would be long

2 + 2 + 3 + 4 + 6 + 11 + 18 + 32 + 57 + 102 = 235

How to draw the (A) and (B) heroes on this near logarithmic scale? Each piece account for 3000 /10 = 300 XP so the (A) and the (B) are in the beginning and in the middle of the 10th segment. Which means (B) is at 57% of the bar and (A) is at the 78%. The (F) is the full scale:

(A) ###########

(B) ################

(F) ####################

On the linear scale it would have been

(A) #################

(B) ###################

(F) ####################

*** CONCLUSION ***

At the beginning the near-logarithmic scale is pretty linear. So for a hero of 1200 xp the last segment is 2x longer than the first segment. While for an hero of 3000 xp the last segment is 50x longer than the 1st segment.