Human preference voting measures popularity, not phenomenology or what mode a model was in when producing output.

{
  "doc": "/Users/antonborzov/Documents/Claude/work/AboutBlank/strategy/2026-05-12_room-to-play-in-eval-cohort.md",
  "claim": "Human preference voting measures popularity, not phenomenology or what mode a model was in when producing output.",
  "vector": "ai-phenomenology-first-person-reports",
  "section": "The cohort's collective blind spot",
  "extracted_at": 1778899589,
  "extracted_by": "haiku-4-5",
  "supporting_refs": [
    "Wei-Lin Chiang",
    "Arena"
  ]
}