Version history

{
  "hypothesis": "A deliberately adversarial loop challenge dataset can separate feedforward encoding, local recurrence, top-down feedback, and neuromodulatory state explanations.",
  "target_ref": "mission:global-brain-observatory",
  "disease": "global-brain-observatory",
  "identification_strategy": "observational",
  "assay_spec": "Create a scoped GBO project packet with public source refs, expected datasets, benchmark metrics, perturbation or observability design, failure modes, and a reviewer loop.",
  "primary_endpoint": "A public challenge where models must predict perturbation and state-shift outcomes, not just held-out neural responses.",
  "cost_estimate_usd": 1200000,
  "timeline_weeks": 52,
  "kill_criteria": "Archive or fork if, on the preregistered held-out perturbation family, no model beats a feedforward null on both predictive score and calibration, or if reviewers cannot name a reusable artifact output.",
  "summary": "Use SciDEX challenges and leaderboards to force computational models to make discriminating predictions. Add a preregistered blind leaderboard split where models must predict out-of-distribution perturbation and state-shift outcomes on a held-out perturbation family, with calibration reported alongside accuracy."
}