Lets say we represent .4389

3n   value   Cumulative Total Balance Percentage Error
-1 1/3 .33333333 + .333333 -.105567  
-2 1/9 .11111111 + .444444 +.005544  
-3 1/27 .03703703 0 .444444    
-4 1/81 .01234567 0 .444444    
-5 1/243 .00411226 - .440329 +.001429  
-6 1/729 .00137174 - .438957 +.000057  
-7 1/2187 .00045724 0 .438957    
-8 1/6561 .00015241 0 .438957    
-9 1/19683 .00005080 - .4389062 +.0000062  

.4389 = 0.++00--00-


.25483 =

3n   value 1/6 value   Cumulative Total Balance Percentage Error
-1 1/3 .33333333 .0555555 + .333333 -.078503 31%
-2 1/9 .11111111 .0185185 - .222222 +.032608 12.79%
-3 1/27 .03703703 .0061728 + .259259 -.004429 1.73%
-4 1/81 .01234567 .00205761 0      
-5 1/243 .00411226 .00068587 - .255144 -.000314 .12%
-6 1/729 .00137174 .00022862 0      
-7 1/2187 .00045724 .000076207 - .2546868 +.0001432 .056%
-8 1/6561 .00015241 .000025402 + .25483921 -.00000921 .0036%
-9 1/19683 .00005080 .000008467        

 .25483 =   0.+-+0-0-+ 


Lets say the number 0.1415936

 

3n   value   Cumulative Total Balance Percentage Error
-1 1/3 .33333333 0 0 -.1415936 100%
-2 1/9 .11111111 + .11111111 -.0304825 21.52%
-3 1/27 .03703703 + .14814814 +.0065545 4.6%
-4 1/81 .01234567 - .13580246 -.0057911 4.08%
-5 1/243 .00411226 + .13991769 -.0016759 1.18%
-6 1/729 .00137174 + .14128943 -.0003041 0.21%
-7 1/2187 .00045724 + .14174668 +.00015308 0.10%
-8 1/6561 .00015241 - .14159426 -.00000066 <.0001%
-9 1/19683 .00005080 0 .14159426 -.00000066  

Therefore the value of .1415936 = 0.0++-+++-0


Lets say the number is 0.7182818

This number is larger than .5, so we shall find the 1s complement of it. (subtract from 1) which is equal to 1 - 0.7182818 = 0.2817182

.2817182 = 

An interesting quirk was noticed while resolving this number manually and under the algorithm. There was a difference. The difference showed up in the 15th and 16th place. The same number seemed to resolve to two different digits. The difference was startling at first but then has been understood. There is no overlap.

Manual resolution.

3n   value 1/6 value   Cumulative Total Balance Remarks
-1 1/3 .33333333 .0555555 + .33333333 +.0516151  
-2 1/9 .11111111 .0185185 0      
-3 1/27 .03703703 .0061728 - .29629629 +.0145780  
-4 1/81 .01234567 .00205761 - .28395061 +.00223238  
-5 1/243 .00411226 .00068587 - .27983539 -.00801719  
-6 1/729 .00137174 .00022862 + .28120713 -.00051107  
-7 1/2187 .00045724 .000076207 + .28166438 -.00005382  
-8 1/6561 .00015241 .000025402 0      
-9 1/19683 .00005080 .000008467 + .28171518 -.00000302  
-10 1/59049 .00001693508 .0000028225 0      
-11 1/177147 .000005645029 .00000094083 + .28172083 +.00000263  
-12 1/531441 .000001881676 .00000031361 - .281718949 +.000000749  
-13 1/1594323 .0000006272254 .00000010453 - .2817183217 +.000000121 violation of rule here
-14 1/4782969 .0000002090751 .000000034845 0      
-15 1/14348907 .0000000696917 .000000011615 - .2817182520    

We had resolved that number completely by the 15th place. 
.2817182 = 0.+0---++0+0+--0-

However, when we use the algorithm, the number resolved upto the 16th place like this.

.2817182 = 

3n   value 1/6 value*   Cumulative Total Balance Percentage Error
-1 1/3 .33333333 .0555555 + .33333333 +.0516151 18%
-2 1/9 .11111111 .0185185 0      
-3 1/27 .03703703 .0061728 - .29629629 +.0145780 5%
-4 1/81 .01234567 .00205761 - .28395061 +.00223238 .79%
-5 1/243 .00411522 .00068587 - .27983539 -.00188281 .66%
-6 1/729 .00137174 .00022862 + .28120713 -.00051107 .18%
-7 1/2187 .00045724 .000076207 + .28166438 -.00005382 .019%
-8 1/6561 .00015241 .000025402 0      
-9 1/19683 .00005080 .000008467 + .28171518 -.00000302 .001%
-10 1/59049 .00001693508 .0000028225 0      
-11 1/177147 .000005645029 .00000094083 + .28172083 +.00000263 .0009%
-12 1/531441 .000001881676 .00000031361 - .281718949 +.000000749 .0002%
-13 1/1594323 .0000006272254 .00000010453 - .2817183218 +.000000121 .00004%
-14 1/4782969 .0000002090751 .000000034845 - .28171811274 -.0000000872 .000030%
-15 1/14348907 .0000000696917 .000000011615 + .28171818243 -.0000000175 .0000062%
-16 1/43046721 .0000000232305 .0000000038717 + .28171820566 +.0000000056 .0000019%
 

.2817182 = .+0---++0+0+---++

There is a difference between the two.

.2817182 = 0.+0---++0+0+--0-        (Manual)
.2817182 = 0.+0---++0+0+---++      (Algorithm)

It appears that the manual value resolved completely within 15th place whereas the algorithm went on to 16th.

Actually
The Algorithmic resolution is correct.
Even at the 15th place its value is closer to the target value than the manual result.

This quirk appeared because we are working with decimal numbers.
And we are ignoring the fact that .2817182 is really also .28171820

There is however no overlap between the two resolutions - they are entirely different numbers.