docs/data/evidence_contract.json

{
  "project": "Ropedia Xperience-10M Task Suite",
  "scope": "single public Xperience-10M sample episode",
  "claims": [
    {
      "id": "project_status",
      "claim": "A first-pass reader has a compact current-state summary.",
      "status": "verified",
      "evidence": [
        "PROJECT_STATUS.md",
        "docs/data/project_status.json"
      ],
      "boundary": "summarizes existing evidence and current limitations"
    },
    {
      "id": "research_roadmap",
      "claim": "The research roadmap is explicit.",
      "status": "current",
      "evidence": [
        "RESEARCH_ROADMAP.md",
        "docs/data/research_roadmap.json"
      ],
      "boundary": "connects public-sample task development to multi-episode data preparation, Qwen3-Omni LoRA, robustness runs, and larger omni-model extensions"
    },
        {
            "id": "official_dataset_card_alignment",
            "claim": "The public dataset description is aligned with the official gated Xperience-10M dataset card and public sample card.",
      "status": "verified",
      "evidence": [
        "XPERIENCE10M_DATASET_CARD_ALIGNMENT.md",
        "docs/data/xperience10m_dataset_card_alignment.json",
        "https://huggingface.co/datasets/ropedia-ai/xperience-10m"
      ],
            "boundary": "summarizes upstream public metadata, API listing facts, sample license/tooling, and dataset-card facts; does not grant access or mirror raw data"
        },
        {
            "id": "source_alignment",
            "claim": "Source facts, sample details, API-listing notes, and project coverage are validated across repo, website, and HF cards.",
            "status": "verified",
            "evidence": [
                "SOURCE_ALIGNMENT_AUDIT.md",
                "docs/data/source_alignment_audit.json",
                "scripts/validate_source_alignment.py"
            ],
            "boundary": "offline committed-fact check; does not fetch private gated data"
        },
        {
            "id": "aligned_windows",
      "claim": "The public Xperience-10M sample has been converted into aligned model windows.",
      "status": "verified",
      "evidence": [
        "results/episode_task_suite/windows.csv",
        "results/episode_task_suite/shared_windows.npz",
        "results/episode_task_suite/summary_report.json"
      ],
      "boundary": "5,821 frames, 1,161 windows, one public sample episode"
    },
        {
            "id": "feature_contract",
            "claim": "The current feature contract is explicit and inspectable.",
            "status": "verified",
            "evidence": [
                "results/episode_task_suite/feature_manifest.json",
                "results/episode_task_suite/available_modalities.json"
            ],
            "boundary": "8,546-dimensional aligned multimodal window representation"
        },
        {
            "id": "evaluation_protocol",
            "claim": "The task evaluation protocol is explicit and generated from committed metrics.",
            "status": "verified",
            "evidence": [
                "EVALUATION_PROTOCOL.md",
                "docs/data/evaluation_protocol.json",
                "scripts/build_evaluation_protocol.py"
            ],
            "boundary": "defines windows, split, per-task metrics, leakage controls, and current limitations"
        },
    {
      "id": "modality_atlas",
      "claim": "The public sample modalities are inspectable without raw data redistribution.",
      "status": "verified",
      "evidence": [
        "docs/data/modality_atlas.json",
        "docs/assets/modalities/",
        "docs/index.html"
      ],
      "boundary": "derived thumbnails for presentation; raw data remains excluded"
    },
    {
      "id": "task_surface_integrity",
      "claim": "Public task cards stay readable for non-expert readers.",
      "status": "verified",
      "evidence": [
        "docs/data/task_surface_integrity.json",
        "scripts/validate_task_surface.py",
        "docs/index.html"
      ],
      "boundary": "presentation integrity for the public task surface"
    },
    {
      "id": "figure_index",
      "claim": "Public figures, charts, and modality thumbnails are indexed as project evidence.",
      "status": "verified",
      "evidence": [
        "FIGURE_INDEX.md",
        "docs/data/figure_index.json",
        "scripts/build_figure_index.py"
      ],
      "boundary": "records derived visual assets, dimensions, hashes, roles, and source scripts; raw Xperience-10M data remains excluded"
    },
    {
      "id": "brand_assets",
      "claim": "A project logo is consistently applied across public surfaces.",
      "status": "verified",
      "evidence": [
        "docs/assets/brand/",
        "docs/data/brand_assets.json",
        "scripts/build_brand_assets.py"
      ],
      "boundary": "generated logo and deterministic derivatives only; no raw dataset data or model weights"
    },
    {
      "id": "twelve_tasks",
      "claim": "The 12 task heads are implemented as scripts with saved metrics and predictions.",
      "status": "verified",
      "evidence": [
        "scripts/episode_task_suite.py",
        "results/episode_task_suite/*/metrics.json",
        "results/episode_task_suite/*/predictions.*"
      ],
      "boundary": "chronological single-episode split, not cross-episode generalization"
    },
    {
      "id": "minimal_vs_neural",
      "claim": "Minimal and neural heads use the same task contracts.",
      "status": "verified",
      "evidence": [
        "scripts/neural_task_models.py",
        "results/episode_task_suite/neural_mlp/",
        "docs/assets/task_architectures.png"
      ],
      "boundary": "small heads only; not a foundation model"
    },
    {
      "id": "research_directions",
      "claim": "Four Ropedia research directions are mapped honestly as direct, proxy, or diagnostic evidence.",
      "status": "verified",
      "evidence": [
        "results/episode_task_suite/research_directions/research_direction_taxonomy.json",
        "docs/data/research_directions.json"
      ],
      "boundary": "some directions remain proxy-only"
    },
    {
      "id": "direction_extensions",
      "claim": "Four extra direction probes are coded and evaluated.",
      "status": "verified",
      "evidence": [
        "results/episode_task_suite/research_direction_extensions/research_direction_extension_results.json",
        "docs/data/research_direction_extensions.json"
      ],
      "boundary": "single-episode probes, not full research-direction solutions"
    },
    {
      "id": "qwen3_omni_setup",
      "claim": "Qwen3-Omni infrastructure has passed technical setup checks.",
      "status": "setup_stage",
      "evidence": [
        "results/omni_finetune/RUN_REPORT.md",
        "results/omni_finetune/dataset_manifest.json",
        "results/omni_finetune/metrics_eval.json"
      ],
      "boundary": "one episode and 128 train windows; full metrics require completed multi-episode data preparation and held-out evaluation"
    },
    {
      "id": "multi_episode_staging",
      "claim": "The Qwen3-Omni LoRA pilot is in selected multi-episode preparation.",
      "status": "data_preparation",
      "evidence": [
        "results/omni_finetune/DATA_ACCESS_STATUS.md",
        "results/omni_finetune/MULTI_EPISODE_ACCESS_STATUS.md",
        "results/omni_finetune/source_discovery.json"
      ],
      "boundary": "the gated Xperience-10M dataset is available for the selected pilot; held-out metrics come after manifest construction, training, and test evaluation"
    },
    {
      "id": "scale_up_status_check",
      "claim": "Older pilot path strings are tracked as setup-file provenance.",
      "status": "verified",
      "evidence": [
        "scripts/validate_scope_claims.py",
        "docs/data/scope_claims_audit.json"
      ],
      "boundary": "run/path identifiers stay separate from completed held-out-episode results"
    },
    {
      "id": "mirror_parity",
      "claim": "Prepared GitHub and Hugging Face mirrors carry matching critical data, visual, HTML, and validator files.",
      "status": "verified",
      "evidence": [
        "scripts/validate_mirror_parity.py",
        "docs/data/mirror_parity.json"
      ],
      "boundary": "compares prepared local mirror bundles before upload; live URLs are checked after publishing"
    },
    {
      "id": "publication_package",
      "claim": "The public GitHub and Hugging Face bundles contain the intended release files.",
      "status": "verified",
      "evidence": [
        "scripts/validate_publication_package.py",
        "docs/data/publication_audit.json"
      ],
      "boundary": "checks public files, HF bundles, and current public-card assets; temporary local outputs are excluded"
    },
    {
      "id": "website_integrity",
      "claim": "The public website has checked local references.",
      "status": "verified",
      "evidence": [
        "scripts/validate_website_integrity.py",
        "docs/data/website_integrity.json"
      ],
      "boundary": "checks local links, anchors, JSON data, and referenced images; external URLs are not fetched"
    },
    {
      "id": "rendered_site_check",
      "claim": "The rendered website walkthrough has a browser-level interaction check.",
      "status": "verified",
      "evidence": [
        "RENDERED_SITE_CHECK.md",
        "scripts/build_rendered_site_check.py",
        "docs/data/rendered_site_check.json"
      ],
      "boundary": "checks local page load, tab switch, walkthrough deep link, player controls, and console health"
    },
    {
      "id": "quality_gates",
      "claim": "The release gate is explicit.",
      "status": "verified",
      "evidence": [
        "QUALITY_GATES.md",
        "scripts/build_quality_gates.py",
        "docs/data/quality_gates.json"
      ],
      "boundary": "summarizes packaging and live-mirror checks; cross-episode model quality is measured by later held-out reports"
    },
    {
      "id": "live_publication_status",
      "claim": "The live public mirrors are checked after upload.",
      "status": "verified",
      "evidence": [
        "scripts/verify_live_publication.py",
        "docs/data/live_publication_status.json"
      ],
      "boundary": "fetches public GitHub/HF URLs; it does not validate private training state"
    },
    {
      "id": "citation_metadata",
      "claim": "The project is externally citable and machine-readable.",
      "status": "verified",
      "evidence": [
        "CITATION.cff",
        "codemeta.json",
        "docs/data/project_manifest.json",
        "LICENSE"
      ],
      "boundary": "code license does not override original Xperience-10M dataset terms"
    },
    {
      "id": "project_path",
      "claim": "A first-time reader has an explicit project path.",
      "status": "verified",
      "evidence": [
        "docs/data/project_packet.json",
        "ARTIFACT_GUIDE.md",
        "docs/data/artifact_index.json",
        "README.md",
        "docs/index.html"
      ],
      "boundary": "guides inspection across data, tasks, results, and scale-up status"
    },
    {
      "id": "artifact_index",
      "claim": "The core project artifacts are grouped for human reading and indexed with existence, size, and hash metadata where stable.",
      "status": "verified",
      "evidence": [
        "ARTIFACT_GUIDE.md",
        "scripts/build_artifact_index.py",
        "docs/data/artifact_index.json"
      ],
      "boundary": "selective source-of-truth catalog, not a complete inventory of every output file"
    },
    {
      "id": "reproducibility_contract",
      "claim": "The public reproduction path is documented with commands, expected outputs, and exact-match reproduction evidence.",
      "status": "verified",
      "evidence": [
        "REPRODUCIBILITY.md",
        "docs/data/reproducibility_matrix.json",
        "notes/reproducibility_audit.md"
      ],
      "boundary": "publicly reproduces the single-episode pipeline; multi-episode Qwen3-Omni metrics are added only after staging and held-out evaluation"
    }
  ]
}