{
  "schema_version": "1.0",
  "review_pack_version": "v0.1",
  "recorded_date": "2026-10-22",
  "status": "proposal_for_independent_review",
  "approved": false,
  "locked": false,
  "scoring_eligible": false,
  "reviewers": [],
  "review_responses": [],
  "sources": [
    "https://airc.nist.gov/airmf-resources/airmf/5-sec-core/",
    "https://www.aisi.gov.uk/blog/our-approach-to-evaluations"
  ],
  "notice": "These are proposed operational tests, NOT independently endorsed criteria. Unresolved questions prevent prospective locking. They do not retroactively alter the existing v0.1 forecast probabilities.",
  "forecast_id": "F-002",
  "record_path": "/data/forecasts/F-002/v0.1.json",
  "candidate_question": "By 2040-12-31 23:59 UTC, has at least one AI system independently demonstrated broad artificial superintelligence?",
  "candidate_definition": "A stable AI system reliably and substantially outperforming leading qualified human expert teams across diverse novel cognitive tasks, not merely a single narrow benchmark.",
  "candidate_domains": [
    "Advanced mathematical and scientific discovery",
    "Systems engineering and coding",
    "Multidisciplinary reasoning and strategic planning",
    "Complex scientific investigation and experimentation",
    "Novel problem solving and transfer",
    "Extended research projects with independently evaluated outcomes"
  ],
  "proposed_thresholds": [
    "Independent preregistered test batteries in all six domain families, with multiple novel high-difficulty tasks",
    "Human comparator is a preregistered expert team, not an average individual; qualifications, available tools and evaluation time must be stated",
    "Candidate margin for review: at least 25% advantage on a pre-specified comparable aggregate performance metric in at least five of six domains and no meaningful regression in the remaining domain",
    "Independent assessment of sustained reliability, reproducibility, tool/environment access and causal contribution",
    "At least three independent replication runs and meaningful uncertainty quantification"
  ],
  "shared_evaluation": {
    "protocol": "Before any lock, external reviewers must approve a preregistered task set, hidden holdouts, disjoint data, scoring rubrics and a recorded human comparison baseline. Tests must include at least three independent replications. Report access, tool use, error bars, sampling uncertainty and contamination checks. No selected demo or company marketing alone qualifies.",
    "scope": "The same named AI system or reproducible controlled configuration, not selectively combining the best outputs of unrelated systems.",
    "evidence": "Reproducible public technical findings, access to evaluation artifacts, independent assessor reports and challenge period. Restricted materials may be separately reviewed but findings must be auditable."
  },
  "review_decisions_required": [
    "Validate a defensible measure of substantial advantage over top expert teams",
    "Approve aggregation metrics across heterogeneous tasks and domain weighting",
    "Assess speed and resource parity to humans and avoid cherry-picked outputs",
    "Handle multi-agent composition, unobservable frontier labs, economic constraints and inconsistent evidence"
  ],
  "pass_rule": "NO PASS RULE LOCKED. Candidate scoring margins are proposals for critical independent review, not scientifically established thresholds. Outcome must be based on predeclared rules and evidence.",
  "resolution_outcomes": [
    "yes",
    "no",
    "unresolved",
    "void"
  ],
  "review_status": "awaiting_external_review"
}