{ "title": "Ropedia Xperience-10M Glossary", "status": "published", "purpose": "Define reader-facing terms that can be confused across the repo, website, Hugging Face mirrors, result matrices, and model-package surfaces.", "categories": [ { "id": "dataset_scope", "label": "Dataset and scope", "description": "Public data boundaries, evidence lines, and how each result family should be read." }, { "id": "files_features", "label": "Files and features", "description": "How raw sample files, windows, feature manifests, and public-safe derivatives relate." }, { "id": "tasks_metrics", "label": "Tasks and metrics", "description": "Task contracts, scored records, direct scores, compact proxies, and audits." }, { "id": "models_runs", "label": "Models and runs", "description": "Baseline families, Qwen3-Omni, Cosmos3, LoRA adapters, and full-parameter gates." }, { "id": "public_surfaces", "label": "Public surfaces", "description": "GitHub, website, Hugging Face repos, parity checks, and package validation." } ], "entries": [ { "term": "Xperience-10M", "category": "dataset_scope", "plain_meaning": "The upstream embodied human-interaction dataset.", "project_usage": "Source dataset behind the public sample, selected-128 features, task suite, and model diagnostics.", "do_not_confuse_with": "This repo, which only redistributes public-safe derived artifacts.", "primary_files": ["XPERIENCE10M_DATASET_CARD_ALIGNMENT.md", "docs/data/xperience10m_dataset_card_alignment.json"] }, { "term": "Public sample episode", "category": "dataset_scope", "plain_meaning": "One officially available sample episode.", "project_usage": "The fully inspectable Line 1 unit used for raw-file browsing, 20-frame windows, task construction, and single-episode baselines.", "do_not_confuse_with": "The selected-128 comparison rows.", "primary_files": ["docs/data/raw_sample_files.json", "docs/single_episode_explorer.html"] }, { "term": "Selected 128 episodes", "category": "dataset_scope", "plain_meaning": "A public-safe selected subset of official gated episode paths.", "project_usage": "Line 2 uses derived windows/features and keeps links back to official episode ids and gated source paths.", "do_not_confuse_with": "Redistributed raw MP4/HDF5/RRD data.", "primary_files": ["XPERIENCE10M_128_EPISODE_FEATURE_INDEX.md", "docs/data/xperience10m_128_episode_feature_index.json"] }, { "term": "Evidence line", "category": "dataset_scope", "plain_meaning": "A reading lane for a group of results.", "project_usage": "Line 1 is one public sample episode; Line 2 is selected-128 held-out comparison.", "do_not_confuse_with": "Qwen run versions v1-v6, which are model-run lineage.", "primary_files": ["TWO_EVIDENCE_LINES.md", "docs/data/two_evidence_lines.json"] }, { "term": "Official gated data", "category": "dataset_scope", "plain_meaning": "Upstream files that require official dataset access.", "project_usage": "Raw Xperience-10M MP4/HDF5/RRD files and full source directories remain outside the public repo.", "do_not_confuse_with": "Public-safe metrics, derived features, figures, and manifests.", "primary_files": ["DATA_NOTICE.md", "REPRODUCIBILITY.md"] }, { "term": "Public-safe artifact", "category": "public_surfaces", "plain_meaning": "A file that can be mirrored publicly without raw gated content.", "project_usage": "Metrics, JSON summaries, model cards, figures, derived manifests, and approved lightweight weights/adapters.", "do_not_confuse_with": "Raw dataset redistribution.", "primary_files": ["ARTIFACT_GUIDE.md", "docs/data/artifact_index.json"] }, { "term": "Episode", "category": "files_features", "plain_meaning": "One recorded interaction sequence.", "project_usage": "The basic source unit behind windows, labels, and train/val/test splits.", "do_not_confuse_with": "A 20-frame window.", "primary_files": ["docs/data/raw_sample_files.json", "docs/data/xperience10m_128_episode_feature_index.json"] }, { "term": "20-frame window", "category": "files_features", "plain_meaning": "A fixed short clip slice.", "project_usage": "The sample episode is converted into aligned 20-frame units for features, labels, and many task heads.", "do_not_confuse_with": "A full episode or arbitrary video segment.", "primary_files": ["results/episode_task_suite/windows.csv", "EVALUATION_PROTOCOL.md"] }, { "term": "Feature manifest", "category": "files_features", "plain_meaning": "A map from model-input columns to source modalities.", "project_usage": "Explains feature groups and dimensions for the sample task suite.", "do_not_confuse_with": "The raw annotation file.", "primary_files": ["results/episode_task_suite/feature_manifest.json"] }, { "term": "annotation.hdf5", "category": "files_features", "plain_meaning": "Upstream annotation container for the sample.", "project_usage": "Contains original labels/metadata; some public derived files expose processed features instead of every raw text field.", "do_not_confuse_with": "Task result summaries.", "primary_files": ["docs/data/raw_sample_files.json"] }, { "term": "Interaction text", "category": "files_features", "plain_meaning": "Natural-language interaction/caption content.", "project_usage": "Used by task 15 and some derived text features; public matrices record direct or compact-proxy status.", "do_not_confuse_with": "Numeric action ids or subtask ids.", "primary_files": ["TASK_SUITE_20.md", "docs/data/task_method_20_result_matrix.json"] }, { "term": "Modality", "category": "files_features", "plain_meaning": "A type of signal.", "project_usage": "Video, audio, depth, pose/SLAM, motion capture, inertial, calibration, and language-derived signals.", "do_not_confuse_with": "A task target.", "primary_files": ["docs/data/modality_atlas.json", "results/episode_task_suite/feature_manifest.json"] }, { "term": "Task contract", "category": "tasks_metrics", "plain_meaning": "The definition of one benchmark task.", "project_usage": "Includes input, target/output, metric, split, source artifact, and limitation.", "do_not_confuse_with": "A model architecture.", "primary_files": ["TASK_SUITE_20.md", "docs/data/task_suite_20.json"] }, { "term": "Unified 20-task suite", "category": "tasks_metrics", "plain_meaning": "The current task surface.", "project_usage": "All 20 task contracts are presented together and scored across methods where real artifacts exist.", "do_not_confuse_with": "Historical tier2_task_suite filenames, which are provenance paths rather than a second suite.", "primary_files": ["TASK_SUITE_20.md", "docs/data/task_suite_20.json"] }, { "term": "Task-method record", "category": "tasks_metrics", "plain_meaning": "One method evaluated on one task.", "project_usage": "9 methods x 20 tasks gives 180 public result records.", "do_not_confuse_with": "A single prediction row.", "primary_files": ["TASK_METHOD_20_RESULT_MATRIX.md", "docs/data/task_method_20_result_matrix.json"] }, { "term": "Direct score", "category": "tasks_metrics", "plain_meaning": "A metric computed against the task target directly.", "project_usage": "The preferred score type in the 20-task matrix.", "do_not_confuse_with": "Compact-proxy score.", "primary_files": ["TASK_METHOD_20_GAP_AUDIT.md", "docs/data/task_method_20_gap_audit.json"] }, { "term": "Compact-proxy score", "category": "tasks_metrics", "plain_meaning": "A bounded proxy metric when a direct raw target is not publicly available.", "project_usage": "Kept explicit in the matrix and gap audit so readers do not over-read it.", "do_not_confuse_with": "A direct target measurement.", "primary_files": ["TASK_METHOD_20_GAP_AUDIT.md", "docs/data/task_method_20_gap_audit.json"] }, { "term": "Raw metric value", "category": "tasks_metrics", "plain_meaning": "The original metric value emitted by the runner or verified result package.", "project_usage": "This is the value to cite from the 180-result table.", "do_not_confuse_with": "The normalized radar value.", "primary_files": ["TASK_METHOD_20_RESULT_MATRIX.md", "docs/data/task_method_20_result_matrix.json"] }, { "term": "Normalized radar value", "category": "tasks_metrics", "plain_meaning": "A 0-1 plotting value used only to draw comparable radar polygons.", "project_usage": "Helps visualize metrics with different scales and directions.", "do_not_confuse_with": "The raw metric value to cite.", "primary_files": ["docs/data/unified_task_model_radar.json", "docs/assets/charts/unified_task_model_radar.svg"] }, { "term": "Gap audit", "category": "tasks_metrics", "plain_meaning": "A coverage and source-status audit.", "project_usage": "Explains scored, proxy, and unsupported cells.", "do_not_confuse_with": "A performance leaderboard.", "primary_files": ["TASK_METHOD_20_GAP_AUDIT.md", "docs/data/task_method_20_gap_audit.json"] }, { "term": "Leakage control", "category": "tasks_metrics", "plain_meaning": "A split or feature rule that prevents using target information unfairly.", "project_usage": "Chronological splits, held-out splits, and source audits protect task interpretation.", "do_not_confuse_with": "Lower training accuracy.", "primary_files": ["EVALUATION_PROTOCOL.md", "docs/data/evaluation_protocol.json"] }, { "term": "Minimal baseline", "category": "models_runs", "plain_meaning": "A simple non-neural task head; the \"minimum\" reference row in casual wording.", "project_usage": "Provides a reproducible lower-complexity comparison for task feasibility.", "do_not_confuse_with": "Metadata-only selected-128 baseline family.", "primary_files": ["RESEARCH_TAKEAWAYS.md", "docs/data/task_method_20_result_matrix.json"] }, { "term": "Simple baseline", "category": "models_runs", "plain_meaning": "A non-neural baseline family for the selected-128 rows.", "project_usage": "Used for metadata/text and raw-feature 128-episode comparisons before NN/foundation-model rows.", "do_not_confuse_with": "The single-episode Minimal baseline.", "primary_files": ["RESEARCH_TAKEAWAYS.md", "docs/data/task_method_20_result_matrix.json"] }, { "term": "Neural MLP", "category": "models_runs", "plain_meaning": "A compact neural task head.", "project_usage": "Used for single-episode and selected-128 baseline comparisons.", "do_not_confuse_with": "Foundation-model fine-tuning.", "primary_files": ["results/episode_task_suite/neural_mlp/", "docs/data/task_method_20_result_matrix.json"] }, { "term": "Qwen3-Omni", "category": "models_runs", "plain_meaning": "The multimodal foundation-model family used for the Qwen branch.", "project_usage": "The current public 20-task Qwen row is Qwen3-Omni v6 LoRA plus task-specific probes.", "do_not_confuse_with": "Cosmos3 or single-episode task-head baselines.", "primary_files": ["QWEN3_OMNI_RUN_LINEAGE.md", "docs/data/qwen3_omni_run_lineage.json"] }, { "term": "Qwen v1-v6", "category": "models_runs", "plain_meaning": "The Qwen3-Omni run lineage.", "project_usage": "v1-v4 are earlier pipeline/ablation evidence, v5 is the prior pinned release, and v6 is the current public 20-task row.", "do_not_confuse_with": "Six different evidence lines.", "primary_files": ["QWEN3_OMNI_RUN_LINEAGE.md", "docs/data/qwen3_omni_run_lineage.json"] }, { "term": "Cosmos3-Super", "category": "models_runs", "plain_meaning": "The larger Cosmos3-style branch tracked in this project.", "project_usage": "Published as Reasoner diagnostics and a separate forward-dynamics LoRA adapter/result branch when verified.", "do_not_confuse_with": "Cosmos3-Nano.", "primary_files": ["docs/data/omni_model_comparison.json"] }, { "term": "Cosmos3-Nano", "category": "models_runs", "plain_meaning": "A smaller Cosmos3 compatibility/future-window branch.", "project_usage": "Used for the Nano Future Window row and related diagnostics.", "do_not_confuse_with": "Cosmos3-Super fine-tuned adapter.", "primary_files": ["docs/data/omni_model_comparison.json"] }, { "term": "LoRA adapter", "category": "models_runs", "plain_meaning": "A lightweight set of trainable adapter weights.", "project_usage": "Published only when the package is verified and public-safe.", "do_not_confuse_with": "Full base-model weights.", "primary_files": ["OMNI_MODEL_EXTENSION_CONTRACT.md", "docs/data/omni_model_comparison.json"] }, { "term": "Full-parameter fine-tuning", "category": "models_runs", "plain_meaning": "Updating the whole model rather than only adapters.", "project_usage": "This project records feasibility gates and short pilots, but does not publish full checkpoints.", "do_not_confuse_with": "LoRA adapter publication.", "primary_files": ["docs/data/qwen3_full_parameter_gates.json"] }, { "term": "Foundation pipeline", "category": "models_runs", "plain_meaning": "A high-level training direction.", "project_usage": "Spatial intelligence, human-video world modeling, and vision-language-action are documented as trainable directions with task mappings.", "do_not_confuse_with": "A completed public result row.", "primary_files": ["THREE_FOUNDATION_PIPELINES.md", "docs/data/three_foundation_pipelines.json"] }, { "term": "Spatial intelligence", "category": "models_runs", "plain_meaning": "Learning geometry and spatial reasoning from egocentric data.", "project_usage": "Uses video, depth, camera pose, and language tasks to target 3D/space reasoning.", "do_not_confuse_with": "World-model future prediction.", "primary_files": ["THREE_FOUNDATION_PIPELINES.md", "docs/data/three_foundation_pipelines.json"] }, { "term": "Human-video world model", "category": "models_runs", "plain_meaning": "Learning future frames, actions, and interaction dynamics from human video.", "project_usage": "Uses temporal prediction, next-action, transition, and object-forecast tasks.", "do_not_confuse_with": "Robot policy execution.", "primary_files": ["THREE_FOUNDATION_PIPELINES.md", "docs/data/three_foundation_pipelines.json"] }, { "term": "Vision-language-action", "category": "models_runs", "plain_meaning": "Mapping perception and language to action chunks.", "project_usage": "A future policy/VLA direction that needs action-target conversion and stronger policy packaging.", "do_not_confuse_with": "Qwen3-Omni diagnostic scoring.", "primary_files": ["THREE_FOUNDATION_PIPELINES.md", "docs/data/three_foundation_pipelines.json"] }, { "term": "HF Space", "category": "public_surfaces", "plain_meaning": "Hugging Face-hosted app/site surface.", "project_usage": "Mirrors the dashboard and static website assets.", "do_not_confuse_with": "HF artifact dataset or model repo.", "primary_files": ["PUBLIC_READER_MAP.md", "docs/data/public_reader_map.json"] }, { "term": "HF artifact dataset", "category": "public_surfaces", "plain_meaning": "Hugging Face dataset repo for derived evidence.", "project_usage": "Stores public-safe reports, metrics, website JSON, and sanitized result packages.", "do_not_confuse_with": "Original Xperience-10M dataset.", "primary_files": ["ARTIFACT_GUIDE.md", "docs/data/artifact_index.json"] }, { "term": "HF baseline model repo", "category": "public_surfaces", "plain_meaning": "Hugging Face model repo for lightweight baseline artifacts.", "project_usage": "Mirrors baseline weights, figures, metrics, and task artifacts.", "do_not_confuse_with": "Qwen/Cosmos adapter-specific repos.", "primary_files": ["PUBLIC_READER_MAP.md", "docs/data/public_reader_map.json"] }, { "term": "Mirror parity", "category": "public_surfaces", "plain_meaning": "A check that public copies match the source files.", "project_usage": "Records whether GitHub, website, and HF mirrors agree.", "do_not_confuse_with": "A model-quality metric.", "primary_files": ["docs/data/mirror_parity.json"] }, { "term": "Verified package", "category": "public_surfaces", "plain_meaning": "A result or artifact bundle that passed local/public validators.", "project_usage": "Only verified packages are promoted to README, website, and HF surfaces as public evidence.", "do_not_confuse_with": "A running or exploratory experiment.", "primary_files": ["docs/data/publication_audit.json", "PUBLIC_SURFACE_QA.md"] } ], "file_entry_points": [ { "need": "Reader navigation", "files": ["PUBLIC_READER_MAP.md", "docs/data/public_reader_map.json"] }, { "need": "Task definitions", "files": ["TASK_SUITE_20.md", "docs/data/task_suite_20.json"] }, { "need": "Result matrix", "files": ["TASK_METHOD_20_RESULT_MATRIX.md", "docs/data/task_method_20_result_matrix.json"] }, { "need": "Direct/proxy status", "files": ["TASK_METHOD_20_GAP_AUDIT.md", "docs/data/task_method_20_gap_audit.json"] }, { "need": "Qwen lineage", "files": ["QWEN3_OMNI_RUN_LINEAGE.md", "docs/data/qwen3_omni_run_lineage.json"] }, { "need": "128-episode source/features", "files": ["XPERIENCE10M_128_EPISODE_FEATURE_INDEX.md", "docs/data/xperience10m_128_episode_feature_index.json"] }, { "need": "Public mirrors", "files": ["PUBLIC_SURFACE_QA.md", "docs/data/mirror_parity.json", "docs/data/live_publication_status.json"] } ] }