Like in most calculators, digits are entered the TOS (top of stack) register from right, pushing all other digits up for 4 bits, with MSN (most significant nibble) lost.

There are many ways to accomplish this, but the approach here is a bit complicated because the optimization was done to minimize microcode needed:

• When entering digits, the input register ("instruction register") contains the ASCII code of the digit
• First, R12 and R13 are loaded (these are parallel loadable) - R12 with bit reverse of lower 4 bits of ASCII code (e.g. A = 0x41 = 01000001 so with 1000) and R13 with a correction value which is 1101 (.data value alias)
• R12 and R13 are connected to inputs of the adder, and TOS as output, carry is cleared, nop1() matrix is set up for rest of registers
• 4 shifts down are made for all registers except TOS which is shifted up - this way the 4 bits of correct value appear in LSN of TOS (opr = m2_d2_d2)
• bitcnt is now loaded with n - 4
• all registers (except TOS which is not shifted) are shifted down until bitcnt is 0. This means that they are shifted n times, so they remain with same value (opr = np_d2_d2)

...
        .map 'a';
        .map 'A';
        .map 'e';
        .map 'E';
        data 0b1101, goto hexchar;    // correction from reverse ASCII is 11
...
 hexchar:    load_bitcnt 3, c_flag <= zero, nuke, matrix_nop1();
        connect row_const to col_adc1;
        connect row_direct to col_adc2;
        connect row_sum to col_tos;
        STATUS = busy_using_mt, opr = m2_d2_d2, bitcnt <= dec, if bitcnt_is_zero then next else repeat;
        disconnect row_sum from col_tos, bitcnt <= max;
        connect row_tos to col_tos, bitcnt <= dec;
        bitcnt <= dec;
        bitcnt <= dec;
        bitcnt <= dec;
        STATUS = busy_using_mt, opr = np_d2_d2, bitcnt <= dec, if bitcnt_is_zero then nextchar else repeat;

 State before shifts:

State after the shifts (digit A was entered, switch is position 0, 0 could be stayed off):