{
  "schema": "ULTRACON_AI_EPISTEMIC_UPDATE_EXPERIMENTS_V1",
  "date": "2026-09-20",
  "experiments": [
    {
      "id": "UCF-48",
      "name": "Latent psychological-construct steering validity gate",
      "axis": [
        "META",
        "SPIRAL"
      ],
      "access": "white-box activation access preferred; matched behavioral and human-proxy validation",
      "design": "For a declared psychological construct, derive candidate activation directions from contrastive examples, then evaluate four separable claims: latent predictivity, causal steering, construct specificity, and human-proxy validity. Use held-out prompts, negative-control constructs, direction-shuffling, multiple layers/coefficients, cross-model replication, independent human or validated instrument scoring, and prompt-only baselines. Test whether the direction predicts and causally changes only the declared construct rather than generic valence, compliance, style or verbosity.",
      "measures": [
        "held-out construct AUC/correlation",
        "causal effect size under activation steering",
        "negative-control spillover",
        "cross-model replication",
        "prompt-vs-activation stability",
        "judge-dependence sensitivity",
        "human-rating agreement",
        "OOD construct specificity"
      ],
      "supports": "Whether a latent direction carries construct-specific predictive information and participates causally in output-level expression under the tested model and operationalization.",
      "does_not_support": "Literal possession of a human clinical schema, stable personality identity, endogenous self-model, introspective access, subjective affect, psychiatric diagnosis or phenomenal consciousness.",
      "source_anchors": [
        "SRC-ZHANG-MECH-SIMPART-NPJAI-2026",
        "SRC-RIVA-LATENT-PERSONA-NPJAI-2026",
        "SRC-KARETNIKOV-HUMAN-PROXIES-2026"
      ]
    }
  ]
}