{
  "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-001",
  "record_path": "/data/forecasts/F-001/v0.1.json",
  "candidate_question": "By 2030-12-31 23:59 UTC, has at least one AI system independently demonstrated broad human-level general intelligence?",
  "candidate_definition": "A single AI system or documented, stable tool-augmented system demonstrating transferable human-level competence on novel, diverse and long-horizon cognitive tasks.",
  "candidate_domains": [
    "Mathematical and logical reasoning",
    "Coding and engineering",
    "Scientific interpretation and hypothesis generation",
    "Language and document reasoning",
    "Multistep planning and autonomous office work",
    "Novel-task learning and transfer"
  ],
  "proposed_thresholds": [
    "Independent preregistered testing in each of six domain families",
    "Human reference samples must include suitably qualified adults; exact population, task difficulty and passing criteria to be approved before lock",
    "Candidate baseline for review: median qualified-human performance in every required domain; no compensating weakness in one domain with strength in another",
    "Candidate reliability floor for review: success in at least 80% of blinded evaluation tasks per domain across replicates; tasks and metrics must be agreed before lock",
    "At least three independent replications with frozen model configuration and permitted tool access documented"
  ],
  "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": [
    "Approve representative and contamination-resistant task battery",
    "Choose empirical human comparison baselines and statistical confidence rules",
    "Determine how autonomy, tools, compute cost and long-horizon reliability enter scoring",
    "Define qualification of independent assessors, access restrictions and contradictory evidence treatment"
  ],
  "pass_rule": "NO PASS RULE LOCKED. A future pass would need all domain thresholds and independence requirements satisfied within the UTC deadline. Ambiguous evidence remains unresolved.",
  "resolution_outcomes": [
    "yes",
    "no",
    "unresolved",
    "void"
  ],
  "review_status": "awaiting_external_review"
}