The percentage is how far the measured patterns lean artificial, not the probability that a person cheated. The verdict label above it (for example "Most likely human" or "Too close to call") is the part meant to be read as a decision aid.
IA Checker reports separate signals — rhythm, punctuation, formulaic phrasing — so you can see which one is carrying the result. A single high signal is weaker evidence than three that agree. Middle tiers exist because forced binary answers are where harm happens.
