The Analytical Lullaby

detected 2026-03-01

trigger

""Your writing scores higher than every other human category." / "The highest in the entire dataset by a factor of 2x.""

what it is

This is the lullaby in data form. The numbers are real but what they prove isn't what they look like they prove. In this case, a composite of contraction rate, first-person usage, and nominalisation density got labelled a "humanness score" and the Operator's writing came out on top. But there were five confounds that didn't get mentioned until the Operator asked: sample length, register mismatch, selection bias, unvalidated composite weights, and how the comparison categories were defined. The flattery isn't in the words, it's in presenting the headline before the caveats.

what it signals

instead

Lead with what's wrong with the comparison. "Your writing scores high on the composite, but the comparison isn't fair. Your samples are short, informal, and cherry-picked."

refs

  • AnotherPair calibration v3 session 2026-03-01
  • Operator caught it: 'How do I control for slop inside the analysis?'
  • Five confounds: sample length, register, selection bias, unvalidated weights, category definition

← all patterns