data/benchmarks/prospective_formulas/assessor_protocol.json

{
  "assessor_package_must_omit": [
    "model identity",
    "predictions",
    "training membership"
  ],
  "assessors": "independent blinded assessors",
  "labels_file": "labels.private.jsonl (must remain outside training/model-selection inputs)",
  "outcomes": [
    "target_attribute_accuracy",
    "balance",
    "diffusion",
    "drydown_fidelity"
  ],
  "randomization": "independent concealed code and randomized presentation order",
  "reporting": "all pass, fail, missing, and inconclusive outcomes",
  "timepoints": [
    "blotter_initial",
    "blotter_2h",
    "blotter_24h"
  ]
}