A not entirely inaccurate oracle for whether or not Windows XP will recognize a given string as being unicode (ref. IsTextUnicode).

censor_oracle.py

#!/usr/bin/env python3

import sys

# Please excuse the awfully formatted code, I did not take the time to make it look nice.

def main():
    print()
    if len(sys.argv) != 2:
        print(f'     Usage: {sys.argv[0]}')

    string_to_check = sys.argv[1]
    if len(string_to_check) % 2 == 1:
        print(f'     ERROR: String is of ODD length ({len(string_to_check)}). Strings of ODD length are never recognized as Unicode.')
        return

    print(f'                 String: {string_to_check}')
    print()

    values = [ord(c) for c in string_to_check]

    last_lower = 0
    last_higher = 0

    lower_difference = 0
    higher_difference = 0

    max_len = 9
    print('                                    _LowD HighD _LowS HighS Char')
    print('            Lucky ASCII:', end= '')
    for i in range(0, len(values), 2):
        ascii_val = f'[{values[i]} {values[i+1]}]'
        padding = max_len - len(ascii_val)
        if i == 0:
            print(f' [{values[i]} {values[i+1]}]', end='')
        else:
            print(f'                         [{values[i]} {values[i+1]}]', end='')
        print(' ' * padding, end = '')

        lower = values[i]
        higher = values[i+1]

        lower_difference_curr = max(lower, last_lower) - min(lower, last_lower)
        higher_difference_curr = max(higher, last_higher) - min(higher, last_higher)
        low_padding = 3 - len(f'{lower_difference_curr}')
        high_padding = 3 - len(f'{higher_difference_curr}')
        print(f'   {" " * low_padding}{lower_difference_curr}   {" " * high_padding}{higher_difference_curr}', end='')

        lower_difference += max(lower, last_lower) - min(lower, last_lower)
        higher_difference += max(higher, last_higher) - min(higher, last_higher)
        low_padding = 3 - len(f'{lower_difference}')
        high_padding = 3 - len(f'{higher_difference}')
        print(f'   {" " * low_padding}{lower_difference}   {" " * high_padding}{higher_difference}', end='')

        print(f'   {string_to_check[i]}{string_to_check[i+1]}')

        last_lower = lower
        last_higher = higher

    print()
    print(f'             Lower Diff: {lower_difference}')
    print(f'            Higher Diff: {higher_difference}')
    print(f'        2 * Higher Diff: {2 * higher_difference}')
    print(f'        3 * Higher Diff: {3 * higher_difference}')
    print(f'3 * Higher Diff - Lower: {3 * higher_difference - lower_difference}')
    print()
    print(f'         higher < lower: {"YES" if 1 * higher_difference < lower_difference else "NO"}')
    print(f'     2 * higher < lower: {"YES" if 2 * higher_difference < lower_difference else "NO"}')
    print(f'     3 * higher < lower: {"YES" if 3 * higher_difference < lower_difference else "NO"}')

    print()
    print(f'         >>>>> CENSORED: {"YES" if 3 * higher_difference < lower_difference else "NO"} <<<<<')

if __name__ == '__main__':
    main()

Did the developers of this algorithm not consider Chinese or any UTF-8 language at all? 教你的技術人員如何飛行。 reports as ANSI, for example.

i'm pretty sure this is made for ANSI strings to be recognized as Unicode

There are other components to IsTextUnicode, if you look at the MSDN page and scroll down to the big list of flags: https://learn.microsoft.com/en-US/windows/win32/api/winbase/nf-winbase-istextunicode.
I am a bit too lazy to check what "this algorithm" meant in the comment -- does 教你的技術人員如何飛行。 or \u6559\u4F60\u7684\u6280\u8853\u4EBA\u54E1\u5982\u4F55\u98DB\u884C\u3002 triggers just this Python thing or the whole IsTextUnicode function? If it works on the whole function, this would mean there's now an inverse "bush hid the facts" (causes the bug in the other direction... and a meaningful sentence) after all these years.

There are other components to IsTextUnicode, if you look at the MSDN page and scroll down to the big list of flags: https://learn.microsoft.com/en-US/windows/win32/api/winbase/nf-winbase-istextunicode. I am a bit too lazy to check what "this algorithm" meant in the comment -- does 教你的技術人員如何飛行。 or \u6559\u4F60\u7684\u6280\u8853\u4EBA\u54E1\u5982\u4F55\u98DB\u884C\u3002 triggers just this Python thing or the whole IsTextUnicode function? If it works on the whole function, this would mean there's now an inverse "bush hid the facts" (causes the bug in the other direction... and a meaningful sentence) after all these years.

@Artoria2e5
This script only checks the "char sum imbalance". If this check says NO, that doesn't mean the string is classified as non-unicode. You can try using the other code I posted that simply calls IsTextUnicode on Windows (or, if you're at that point, just try it in an XP VM).