{
  "schema": "truthbeam-open-questions/0.5",
  "status": "research invitations, not established claims",
  "participation_boundary": {
    "independent_work": "Verification of the published data, bounded analysis, and comparable measurements from a compatible independently operated rig are welcome.",
    "collaboration_route": "For cross-witnessing, protocol development, applications, licensing, or wider joint work, collaborate with Cathal Ryan Hynes and the project through CONTRIBUTING.md.",
    "no_implied_rights": "The invitation does not itself grant copyright, patent, trademark, commercial, or identifiable-capture redistribution rights."
  },
  "routes": {
    "human_agenda": "OPEN_QUESTIONS.md",
    "downloads": "START_WITH_DATA.md",
    "contributing": "CONTRIBUTING.md",
    "permission": "RESEARCH_PERMISSION.md",
    "results": "DOWNLOADS.md#results-2026-09"
  },
  "questions": [
    {
      "id": "TB-Q1",
      "tier": "scores-2mb",
      "topic": "finite-sample uncertainty and calibration",
      "question": "Which block-aware uncertainty and threshold-transfer summaries are justified by two fixed same-rig session strata?",
      "scope": "D2 and V10 are fixed strata, not a population sample of rigs, people, or scenes.",
      "requires_new_capture": false,
      "evidence_anchor": "REPRODUCE.md",
      "decisive_test": "nominal intervals and thresholds that stay stable across reasonable block-aware resamples and hold on a held-out session."
    },
    {
      "id": "TB-Q2",
      "tier": "sample-then-full",
      "topic": "train-free optical coupling statistics",
      "question": "Can a preregistered analytic statistic distinguish named wrong-emission or substitution families after capture-only, emission-only, and block-aware permutation controls?",
      "scope": "Analytic same-rig diagnostic, not a universal verifier.",
      "requires_new_capture": false,
      "evidence_anchor": "how-it-works.md",
      "decisive_test": "above-chance performance under proper held-out controls, with a threshold that transfers between sessions.",
      "progress": {
        "date": "2026-09-06",
        "status": "partly answered",
        "result": "No Training Required: a grid-correlation statistic with no fitted parameters separates matched from mismatched frames on the 975 held-out tail frames of both sessions, pooled AUROC 0.683 to 0.756 at each of the five grid sizes tested (4, 8, 16, 32 and 64), under five recorded within-session shuffles and four registered hard-negative families; Sixteen Cells, One Threshold proves the 4 by 4 case in exact arithmetic for two public examples.",
        "open": "a threshold that transfers between sessions; any claim about another rig",
        "urls": [
          "https://github.com/poliebotics/dark-lantern/tree/main/results/train_free_coupling_20260906",
          "https://github.com/poliebotics/dark-lantern/tree/main/proofs/train_free_grid_correlation_20260906"
        ]
      }
    },
    {
      "id": "TB-Q3",
      "tier": "sample-then-full",
      "topic": "evidence-preserving data reduction",
      "question": "Which reduced representations preserve predeclared check families within numerical tolerances while reducing storage?",
      "scope": "Only the checks and tolerances declared by the experiment.",
      "requires_new_capture": false,
      "evidence_anchor": "CID_MANIFEST.json",
      "decisive_test": "at a predeclared compression ratio, the reduced representation preserves the named checks within their tolerances across both sessions and reproduces from its manifest."
    },
    {
      "id": "TB-Q4",
      "tier": "full-corpus-and-new-captures",
      "topic": "causal location of measured coupling",
      "question": "Beyond the released off-body segmentation result, what fractions are causally attributable to the scene, projector, optics, sensor, timing, and apparatus-specific nuisance features?",
      "scope": "The released 106-frame Phase G/F-A v1 segmentation result is the starting boundary.",
      "requires_new_capture": true,
      "evidence_anchor": "results/redteam_segmentation_evals/",
      "decisive_test": "predeclared causal interventions yield stable, separately reported attributable fractions for the scene, projector, optics, sensor, timing, body, and nuisance partitions, and identify which effects persist under each declared controlled change."
    },
    {
      "id": "TB-Q5",
      "tier": "full-corpus",
      "topic": "verification-side diagnostic lag",
      "question": "Does a predeclared lagged or multi-frame diagnostic improve held-out discrimination and transfer between D2 and V10?",
      "scope": "The recording convention remains C_t paired with the emission derived from S_t at offset zero.",
      "requires_new_capture": false,
      "evidence_anchor": "paper/sections/02_system_threat.tex",
      "decisive_test": "an optimum that holds across sessions and challenge families, with gains that persist on held-out blocks."
    },
    {
      "id": "TB-Q6",
      "tier": "full-corpus-plus-new-rig",
      "topic": "self-supervised cross-rig adaptation",
      "question": "Can paired pretraining on D2 and V10 reduce labelled calibration requirements for an independently operated rig?",
      "scope": "One released rig plus a separately consented calibration and evaluation record.",
      "requires_new_capture": true,
      "evidence_anchor": "START_WITH_DATA.md",
      "decisive_test": "a held-out new-rig gain from pretraining attributable to optical coupling, with session and performer identity controlled."
    },
    {
      "id": "TB-Q7",
      "tier": "full-corpus-models-compute",
      "topic": "Can AI test a preregistered surrogate-transfer robustness bound?",
      "question": "Under a predeclared no-target-query protocol, does the public Phase G evaluation-only checkpoint retain its declared acceptance boundary against automated search or a forger trained on fresh surrogate verifiers within a fixed budget?",
      "scope": "Surrogate-adaptive transfer test; target-aware robustness requires a new independent holdout mechanism.",
      "requires_new_capture": false,
      "evidence_anchor": "results/eval/",
      "decisive_test": "no valid candidate reaches the predeclared target region within the committed search budget after leakage and model-selection channels are removed, with the complete search record replayable under the declared constraints."
    },
    {
      "id": "TB-Q9",
      "tier": "new-rig",
      "topic": "controlled cross-rig, cross-scene, material, and operator transfer",
      "question": "Which fixed-budget calibration or adaptation procedure transfers when controlled physical factors change?",
      "scope": "Both transfer directions and each changed factor are reported separately.",
      "requires_new_capture": true,
      "evidence_anchor": "CONTRIBUTING.md",
      "decisive_test": "performance that holds without full retraining and exceeds a capture-only baseline."
    },
    {
      "id": "TB-Q10",
      "tier": "new-rig",
      "topic": "informative optical challenge design",
      "question": "Can a precommitted challenge family improve one primary verification endpoint at fixed visible energy, bandwidth, and exposure?",
      "scope": "Safety, comfort, recoverability, and robustness are declared secondary endpoints unless one is chosen as primary.",
      "requires_new_capture": true,
      "evidence_anchor": "how-it-works.md",
      "decisive_test": "gains that survive equalised visible energy and bandwidth and persist across sessions."
    },
    {
      "id": "TB-Q11",
      "tier": "new-liveness-sessions",
      "topic": "reproducible seeded human-action matching",
      "question": "Can a predeclared action ontology and matcher score committed instruction-response correspondence with calibrated error and useful inter-rater agreement?",
      "scope": "Action correspondence is separate from participant identity, consent, and authority.",
      "requires_new_capture": true,
      "evidence_anchor": "LLM_LIVENESS.md",
      "decisive_test": "material agreement between independent raters and implementations, meeting the predeclared calibration and error thresholds with no uncommitted semantic judgement in the matcher."
    }
  ]
}