Expand glossary with technical field terms

GLOSSARY.md

@@ -1,69 +1,179 @@
 # Glossary
 
-This glossary defines project terms that can be easy to confuse across the
-GitHub repo, website, Hugging Face Space, artifact dataset, model repos, and
-result matrices. Use it with `PUBLIC_READER_MAP.md` when choosing what to read
-first, and with `docs/data/glossary.json` when a tool needs the same terms in
-machine-readable form.
+This glossary defines project-specific terms and adjacent technical field terms that can be easy to confuse across the GitHub repo, website, Hugging Face Space, artifact dataset, model repos, result matrices, and embodied-AI training discussions. Use it with `PUBLIC_READER_MAP.md` when choosing what to read first, and with `docs/data/glossary.json` when a tool needs the same terms in machine-readable form.
 
 ## How To Read The Terms
 
 | Category | What it clarifies |
 | --- | --- |
-| Dataset and scope | Which data is public, which data is gated upstream, and how each evidence line should be read. |
-| Files and features | How raw sample files, derived windows, feature manifests, and public-safe artifacts relate to each other. |
-| Tasks and metrics | What a scored task row means, when a score is direct, and when a compact proxy is being used. |
-| Models and runs | How simple/NN baselines, Qwen3-Omni, Cosmos3, LoRA adapters, and full-parameter gates differ. |
-| Public surfaces | Which repo or Hub surface owns which part of the public package. |
+| Dataset and scope | Public data boundaries, evidence lines, and how each result family should be read. |
+| Files and features | Raw sample files, windows, feature manifests, and public-safe derivatives. |
+| Multimodal sensing | Video, audio, depth, IMU, motion capture, calibration, and synchronization terms. |
+| Spatial geometry | Camera pose, SLAM, coordinate frames, point clouds, 3D reconstruction, and spatial grounding. |
+| Temporal and world models | Future prediction, rollouts, forward dynamics, long-horizon forecasting, and temporal leakage. |
+| Robotics and VLA | Vision-language-action, policies, action chunks, imitation learning, contact, and dexterity. |
+| Tasks and metrics | Task contracts, scored records, direct scores, compact proxies, and audits. |
+| Training and evaluation | Splits, held-out evaluation, metric types, prompt/schema checks, adapters, and distributed training. |
+| Models and runs | Baseline families, Qwen3-Omni, Cosmos3, LoRA adapters, and full-parameter gates. |
+| Public surfaces | GitHub, website, Hugging Face repos, parity checks, and package validation. |
 
-## Core Terms
+## Core And Field Terms
+
+### Dataset and scope
 
 | Term | Plain meaning | In this project | Do not confuse with |
 | --- | --- | --- | --- |
-| Xperience-10M | The upstream embodied human-interaction dataset. | The source dataset behind the public sample, selected-128 features, task suite, and model diagnostics. | This repo itself; the repo only redistributes public-safe derived artifacts. |
+| Evidence line | A reading lane for a group of results. | Line 1 is one public sample episode; Line 2 is selected-128 held-out comparison. | Qwen run versions v1-v6, which are model-run lineage. |
+| Official gated data | Upstream files that require official dataset access. | Raw Xperience-10M MP4/HDF5/RRD files and full source directories remain outside the public repo. | Public-safe metrics, derived features, figures, and manifests. |
 | Public sample episode | One officially available sample episode. | The fully inspectable Line 1 unit used for raw-file browsing, 20-frame windows, task construction, and single-episode baselines. | The selected-128 comparison rows. |
 | Selected 128 episodes | A public-safe selected subset of official gated episode paths. | Line 2 uses derived windows/features and keeps links back to official episode ids and gated source paths. | Redistributed raw MP4/HDF5/RRD data. |
-| Evidence line | A reading lane for a group of results. | Line 1 is one public sample episode; Line 2 is selected-128 held-out comparison. | Qwen run versions v1-v6, which are model-run lineage, not evidence lines. |
-| Official gated data | Upstream files that require official dataset access. | Raw Xperience-10M MP4/HDF5/RRD files and full source directories remain outside the public repo. | Public-safe metrics, derived features, figures, and manifests. |
-| Public-safe artifact | A file that can be mirrored publicly without raw gated content. | Metrics, JSON summaries, model cards, figures, derived manifests, and approved lightweight weights/adapters. | Raw dataset redistribution. |
-| Episode | One recorded interaction sequence. | The basic source unit behind windows, labels, and train/val/test splits. | A 20-frame window, which is a smaller model input slice. |
-| 20-frame window | A fixed short clip slice. | The sample episode is converted into aligned 20-frame units for features, labels, and many task heads. | A full episode or an arbitrary video segment. |
-| Window stride | The frame step between neighboring windows. | Used to create overlapping examples while preserving chronological order and leakage controls. | Video frame rate. |
-| Feature manifest | A map from model-input columns to source modalities. | `results/episode_task_suite/feature_manifest.json` explains the feature groups and dimensions. | The raw annotation file. |
-| Raw sample file map | A human-readable inventory of the sample episode files. | `docs/data/raw_sample_files.json` explains videos, annotations, calibration, motion, and derived previews. | A training manifest. |
-| annotation.hdf5 | Upstream annotation container for the sample. | Contains original labels/metadata; some public derived files expose hashed or processed features rather than every raw text field. | `summary_report.json` or task result JSON. |
-| visualization.rrd | Rerun viewer recording for visual inspection. | Lets readers inspect the sample episode in Rerun 0.29.0 when they download the official file. It is not used for the published training or metric rows. | MP4 video streams or model inputs. |
-| Interaction text | Natural-language interaction/caption content. | Used by task 15 and some derived text features; public matrices record when text targets are direct or compact-proxy. | Numeric action ids or subtask ids. |
+| Xperience-10M | The upstream embodied human-interaction dataset. | Source dataset behind the public sample, selected-128 features, task suite, and model diagnostics. | This repo, which only redistributes public-safe derived artifacts. |
+
+### Files and features
+
+| Term | Plain meaning | In this project | Do not confuse with |
+| --- | --- | --- | --- |
+| 20-frame window | A fixed short clip slice. | The sample episode is converted into aligned 20-frame units for features, labels, and many task heads. | A full episode or arbitrary video segment. |
+| annotation.hdf5 | Upstream annotation container for the sample. | Contains original labels/metadata; some public derived files expose processed features instead of every raw text field. | Task result summaries. |
+| Episode | One recorded interaction sequence. | The basic source unit behind windows, labels, and train/val/test splits. | A 20-frame window. |
+| Feature manifest | A map from model-input columns to source modalities. | Explains feature groups and dimensions for the sample task suite. | The raw annotation file. |
+| Interaction text | Natural-language interaction/caption content. | Used by task 15 and some derived text features; public matrices record direct or compact-proxy status. | Numeric action ids or subtask ids. |
 | Modality | A type of signal. | Video, audio, depth, pose/SLAM, motion capture, inertial, calibration, and language-derived signals. | A task target. |
-| Task contract | The definition of one benchmark task. | Includes input, target/output, metric, split, source artifact, and limitation. | A model architecture. |
-| Unified 20-task suite | The current task surface. | All 20 task contracts are presented together and scored across methods where real artifacts exist. | Historical `tier2_task_suite` filenames; those are provenance paths, not a second suite. |
-| Task-method record | One method evaluated on one task. | 9 methods x 20 tasks gives 180 public result records. | A single prediction row. |
-| Direct score | A metric computed against the task target directly. | The preferred score type in the 20-task matrix. | Compact-proxy score. |
+| Raw sample file map | A human-readable inventory of the sample episode files. | Explains videos, annotations, calibration, motion, and derived previews. | A training manifest. |
+| visualization.rrd | Rerun viewer recording for visual inspection. | Can be downloaded from the official sample dataset and opened in Rerun 0.29.0 to inspect the sample episode. It is not used for published training or metric rows. | MP4 video streams or model inputs. |
+| Window stride | The frame step between neighboring windows. | Creates overlapping examples while preserving chronological order and leakage controls. | Video frame rate. |
+
+### Multimodal sensing
+
+| Term | Plain meaning | In this project | Do not confuse with |
+| --- | --- | --- | --- |
+| Audio waveform | A time-series pressure signal from sound. | The audio ablation measures whether embedded audio helps selected task contracts. | Language captions or text labels. |
+| Calibration | Parameters that relate sensors to each other and to physical space. | Needed to interpret camera streams, depth, pose, and synchronized multimodal features together. | A model training hyperparameter. |
+| Camera extrinsics | A camera position and orientation relative to another coordinate frame. | Connects different camera streams and world coordinates. | Camera intrinsics. |
+| Camera intrinsics | Internal camera parameters such as focal length and distortion. | Explain how image pixels project to rays for geometry tasks. | Camera extrinsics. |
+| Depth map | A per-pixel estimate of distance from the camera. | Depth-derived signals support spatial and geometry-oriented tasks. | RGB brightness or semantic segmentation. |
+| Egocentric video | Video captured from a first-person or body-mounted viewpoint. | The sample streams are egocentric views of human interaction and are the visual basis for many tasks. | Third-person robot-camera footage. |
+| Fisheye camera | A wide-angle camera with strong lens distortion. | Multiple fisheye MP4 streams give broad room coverage but need calibration-aware interpretation. | A rectilinear pinhole camera image. |
+| IMU | An inertial measurement unit with accelerometer and gyroscope signals. | Supports motion, temporal, and sensor-bridging tasks. | Motion capture skeleton data. |
+| Metric depth | Depth expressed in physical units rather than arbitrary relative scale. | Useful for distance-sensitive spatial reasoning and reconstruction targets. | Relative monocular depth. |
+| Motion capture | A system that records body or hand motion over time. | Provides hand/body motion evidence when exposed through public-safe derived features. | Video-only pose estimation. |
+| RGB frame | A color image frame from a video stream. | Used for visual statistics, previews, and many model inputs. | Depth values or point-cloud coordinates. |
+| Sensor alignment | Putting different sensor streams into a shared temporal or spatial reference. | Used to make video, audio, pose, depth, IMU, and mocap usable in the same task input. | Model ensembling. |
+| Stereo camera | A paired-camera setup that supports depth or geometry estimation. | The sample browser exposes stereo streams as part of the visual modality set. | Single-view RGB video. |
+| Timestamp synchronization | Aligning sensor samples by time. | The task suite assumes aligned windows across modalities so labels and features refer to the same moment. | Randomly joining files with similar names. |
+
+### Spatial geometry
+
+| Term | Plain meaning | In this project | Do not confuse with |
+| --- | --- | --- | --- |
+| 3D reconstruction | Recovering 3D scene structure from sensor data. | One core spatial-intelligence direction for Xperience-style data. | Next-action classification. |
+| Affordance | An action possibility offered by an object or scene. | Relevant when moving from observed human interaction to robot-action or VLA tasks. | A detected object category alone. |
+| Camera pose | The camera position and orientation at a time step. | Supports spatial-intelligence tasks, view synchronization, and geometry diagnostics. | The human body pose. |
+| Coordinate frame | A reference system for positions and orientations. | Needed when comparing camera, body, object, and world measurements. | A video frame. |
+| Object-centric representation | A representation organized around objects and their relations. | Useful for object relevance, object-set forecast, and action-object relation tasks. | A flat feature vector without object identity. |
+| Odometry | Motion estimated from sensor changes over time. | A relevant spatial term for ego-motion and camera-pose reasoning. | Ground-truth motion capture. |
+| Point cloud | A set of 3D points representing scene structure. | A likely target or intermediate representation for spatial-intelligence extensions. | A 2D image grid. |
+| SLAM | Simultaneous localization and mapping. | A field term for estimating camera motion and scene structure from sensor observations. | A task label or action class. |
+| Spatial grounding | Linking language or labels to locations, objects, or geometry. | Connects language grounding tasks with 3D/spatial reasoning. | General text classification. |
+| Trajectory | A sequence of positions over time. | Used for hand motion, camera motion, and future-path tasks. | A single coordinate or label. |
+
+### Temporal and world models
+
+| Term | Plain meaning | In this project | Do not confuse with |
+| --- | --- | --- | --- |
+| Action forecasting | Predicting a future action before it happens. | Covered by next-action and long-horizon task contracts. | Recognizing the current action only. |
+| Autoregressive prediction | Generating each future token, state, or frame conditioned on prior outputs. | Relevant for model branches that produce structured JSON or temporal predictions. | A one-shot classifier. |
+| Forward dynamics | Predicting the next state from the current state and action/context. | The Cosmos3-Super LoRA branch uses a forward-dynamics-style diagnostic contract. | Reverse inference from result back to cause. |
+| Latent state | A hidden representation that summarizes observed context. | Useful for future foundation-model and world-model training plans. | A visible annotation column. |
+| Long-horizon prediction | Predicting outcomes several seconds or steps ahead. | Tasks 13 and 14 test longer temporal context beyond immediate recognition. | Single-frame classification. |
+| Next-frame prediction | Predicting future visual frames from past frames. | A field-level world-model objective related to the human-video world-model direction. | Next-action prediction. |
+| Object persistence | Tracking that an object remains present over time even when view or interaction changes. | Relevant for object-set forecast and long-video reasoning. | A single-frame object detection. |
+| Rollout | Repeatedly predicting future steps from a model state. | Important for judging world models beyond one-step prediction. | A held-out static test row. |
+| Subtask forecasting | Predicting the next higher-level step in an activity. | Used in the future-task probe line for Qwen3-Omni. | Frame-level action classification. |
+| Teacher forcing | Training a sequence model using ground-truth previous outputs. | A likely training option for future sequence/world-model baselines. | Free-running rollout evaluation. |
+| Temporal leakage | Using future information that would not be available at prediction time. | Avoided by chronological splits and target-side feature controls. | A low model score. |
+| Transition timing | Estimating when the next state or action transition happens. | Task 20 turns temporal change into a regression target. | Classifying the transition type only. |
+
+### Robotics and VLA
+
+| Term | Plain meaning | In this project | Do not confuse with |
+| --- | --- | --- | --- |
+| Action chunk | A short sequence of low-level actions predicted together. | The VLA figure and plan use action chunks as the policy-output concept. | A natural-language action label. |
+| Behavior cloning | A supervised imitation-learning method for predicting demonstrated actions. | A plausible baseline once action targets are converted. | Generative video modeling. |
+| Contact event | A moment when a hand, body, or tool touches an object or surface. | Used in contact-related tasks and action-quality interpretation. | Visual co-occurrence without touch. |
+| Dexterity | Fine-grained physical manipulation ability. | Relevant to hand-object interaction, contact, and VLA/policy directions. | High text-generation accuracy. |
+| End effector | The robot part that acts on the world, such as a gripper or hand. | A key target frame for future manipulation-policy conversion. | A camera or global scene coordinate. |
+| Hand-object interaction | A physical interaction between hands and objects. | A central signal family behind action, contact, object relevance, and interaction-text tasks. | Object detection without action. |
+| Imitation learning | Training a policy to imitate demonstrated behavior. | Relevant when converting human video/motion into action supervision. | Reinforcement learning from online robot trials. |
+| Language grounding | Connecting text to observed objects, actions, or spatial context. | Task 8 and VLA directions use language as grounded supervision rather than standalone text. | Caption fluency alone. |
+| Policy | A mapping from observations to actions. | A future target for robot-compatible Xperience-derived action data. | A benchmark metric. |
+| Robot-compatible action target | An action representation a robot policy can execute or imitate. | Needed before OpenVLA/openpi/GR00T-style policy training is meaningful here. | Human-only caption text. |
+| Vision-language-action model | A model that maps visual context and language into actions. | The VLA direction is a future path after action targets are converted into robot-compatible chunks. | A vision-language model that only answers text. |
+
+### Tasks and metrics
+
+| Term | Plain meaning | In this project | Do not confuse with |
+| --- | --- | --- | --- |
 | Compact-proxy score | A bounded proxy metric when a direct raw target is not publicly available. | Kept explicit in the matrix and gap audit so readers do not over-read it. | A direct target measurement. |
-| Raw metric value | The original metric value emitted by the runner or verified result package. | This is the value to cite from the 180-result table. | The normalized radar value. |
+| Direct score | A metric computed against the task target directly. | The preferred score type in the 20-task matrix. | Compact-proxy score. |
+| Gap audit | A coverage and source-status audit. | Explains scored, proxy, and unsupported cells. | A performance leaderboard. |
+| Leakage control | A split or feature rule that prevents using target information unfairly. | Chronological splits, held-out splits, and source audits protect task interpretation. | Lower training accuracy. |
 | Normalized radar value | A 0-1 plotting value used only to draw comparable radar polygons. | Helps visualize metrics with different scales and directions. | The raw metric value to cite. |
-| Gap audit | A coverage and source-status audit. | `docs/data/task_method_20_gap_audit.json` explains scored, proxy, and unsupported cells. | A performance leaderboard. |
-| Leakage control | A split or feature rule that prevents using future/target information unfairly. | Chronological splits, held-out splits, and source audits protect task interpretation. | Lower training accuracy. |
-| Minimal baseline | A simple non-neural task head; the "minimum" reference row in casual wording. | Provides a reproducible lower-complexity comparison for task feasibility. | The metadata-only baseline family in the selected-128 matrix. |
-| Simple baseline | A non-neural baseline family for the selected-128 rows. | Used for metadata/text and raw-feature 128-episode comparisons before NN/foundation-model rows. | The single-episode Minimal baseline. |
-| Neural MLP | A compact neural task head. | Used for single-episode and selected-128 baseline comparisons. | Foundation-model fine-tuning. |
-| Metadata baseline | A selected-128 baseline using metadata/text-derived public-safe features. | Helps compare simple and neural heads on the held-out split. | Raw video/depth/audio feature baselines. |
-| Raw-feature baseline | A selected-128 baseline using exported public-safe raw-feature groups. | Tracks what non-foundation heads can do with richer processed inputs. | Raw gated media redistribution. |
-| Qwen3-Omni | The multimodal foundation-model family used for the Qwen branch. | The current public 20-task Qwen row is Qwen3-Omni v6 LoRA plus task-specific probes. | Cosmos3 or the single-episode task-head baselines. |
-| Qwen v1-v6 | The Qwen3-Omni run lineage. | v1-v4 are earlier pipeline/ablation evidence, v5 is the prior pinned release, and v6 is the current public 20-task row. | Six different evidence lines. |
-| Cosmos3-Super | The larger Cosmos3-style branch tracked in this project. | Published as Reasoner diagnostics and a separate forward-dynamics LoRA adapter/result branch when verified. | Cosmos3-Nano. |
+| Raw metric value | The original metric value emitted by the runner or verified result package. | This is the value to cite from the 180-result table. | The normalized radar value. |
+| Task contract | The definition of one benchmark task. | Includes input, target/output, metric, split, source artifact, and limitation. | A model architecture. |
+| Task-method record | One method evaluated on one task. | 9 methods x 20 tasks gives 180 public result records. | A single prediction row. |
+| Unified 20-task suite | The current task surface. | All 20 task contracts are presented together and scored across methods where real artifacts exist. | Historical tier2_task_suite filenames, which are provenance paths rather than a second suite. |
+
+### Training and evaluation
+
+| Term | Plain meaning | In this project | Do not confuse with |
+| --- | --- | --- | --- |
+| Adapter checkpoint | Saved adapter weights from a fine-tuning run. | The public model branches publish adapters when validated and public-safe. | Full base-model checkpoint. |
+| Balanced accuracy | Accuracy averaged across classes to reduce majority-class dominance. | Useful for imbalanced task labels. | Overall accuracy. |
+| Chronological split | A split ordered by time. | Used for the single-episode baselines to reduce future-window leakage. | A random row split. |
+| Confusion matrix | A table of predicted classes versus true classes. | Helps inspect which task labels a method confuses. | A scalar leaderboard score. |
+| FSDP | Fully Sharded Data Parallel, a distributed training strategy. | Appears in full-parameter feasibility and multi-GPU training notes. | A model architecture. |
+| Held-out evaluation | Testing on examples not used for training. | Required before promoting Qwen/Cosmos results to public evidence. | Training-set loss. |
+| JSON validity | Whether model output parses as the required JSON schema. | A key diagnostic for Qwen3-Omni structured-output runs. | Task correctness after parsing. |
+| Macro F1 | The average F1 score across classes, usually treating classes equally. | Used when class imbalance matters in classification tasks. | Accuracy dominated by frequent classes. |
+| Mean absolute error | The average absolute difference between predicted and true numeric values. | Used for regression-style task rows such as timing or trajectory targets. | A classification F1 score. |
+| Overfit check | A small training test that verifies a model can learn a tiny subset. | Useful for catching data/model wiring bugs before full training. | Evidence of generalization. |
+| Parameter-efficient fine-tuning | Updating a small number of added or selected parameters. | LoRA is the current parameter-efficient path for Qwen/Cosmos branches. | Full-parameter fine-tuning. |
+| Schema compliance | Whether an output follows the expected field names and value types. | Needed for structured task probes and public package validation. | High semantic accuracy. |
+| Smoke run | A short run that checks whether a pipeline can start and execute key steps. | Used for feasibility gates before expensive full runs. | A complete benchmark result. |
+| Top-k accuracy | A score that counts a prediction correct if the target is among the k highest-ranked outputs. | Useful for large-label or retrieval-style tasks. | Top-1 exact accuracy. |
+| Train/validation/test split | A partition that separates model fitting, tuning, and final evaluation examples. | The selected-128 setup uses a held-out split discipline for model branches. | A random shuffle without temporal or episode boundaries. |
+
+### Models and runs
+
+| Term | Plain meaning | In this project | Do not confuse with |
+| --- | --- | --- | --- |
 | Cosmos3-Nano | A smaller Cosmos3 compatibility/future-window branch. | Used for the Nano Future Window row and related diagnostics. | Cosmos3-Super fine-tuned adapter. |
-| LoRA adapter | A lightweight set of trainable adapter weights. | Published only when the package is verified and public-safe. | Full base-model weights. |
-| Full-parameter fine-tuning | Updating the whole model rather than only adapters. | This project records feasibility gates and short pilots, but does not publish full checkpoints. | LoRA adapter publication. |
+| Cosmos3-Super | The larger Cosmos3-style branch tracked in this project. | Published as Reasoner diagnostics and a separate forward-dynamics LoRA adapter/result branch when verified. | Cosmos3-Nano. |
 | Foundation pipeline | A high-level training direction. | Spatial intelligence, human-video world modeling, and vision-language-action are documented as trainable directions with task mappings. | A completed public result row. |
-| Spatial intelligence | Learning geometry and spatial reasoning from egocentric data. | Uses video, depth, camera pose, and language tasks to target 3D/space reasoning. | World-model future prediction. |
+| Full-parameter fine-tuning | Updating the whole model rather than only adapters. | This project records feasibility gates and short pilots, but does not publish full checkpoints. | LoRA adapter publication. |
 | Human-video world model | Learning future frames, actions, and interaction dynamics from human video. | Uses temporal prediction, next-action, transition, and object-forecast tasks. | Robot policy execution. |
+| LoRA adapter | A lightweight set of trainable adapter weights. | Published only when the package is verified and public-safe. | Full base-model weights. |
+| Metadata baseline | A selected-128 baseline using metadata or text-derived public-safe features. | Compares simple and neural heads on the held-out split. | Raw video, depth, or audio feature baselines. |
+| Minimal baseline | A simple non-neural task head; the "minimum" reference row in casual wording. | Provides a reproducible lower-complexity comparison for task feasibility. | Metadata-only selected-128 baseline family. |
+| Neural MLP | A compact neural task head. | Used for single-episode and selected-128 baseline comparisons. | Foundation-model fine-tuning. |
+| Qwen v1-v6 | The Qwen3-Omni run lineage. | v1-v4 are earlier pipeline/ablation evidence, v5 is the prior pinned release, and v6 is the current public 20-task row. | Six different evidence lines. |
+| Qwen3-Omni | The multimodal foundation-model family used for the Qwen branch. | The current public 20-task Qwen row is Qwen3-Omni v6 LoRA plus task-specific probes. | Cosmos3 or single-episode task-head baselines. |
+| Raw-feature baseline | A selected-128 baseline using exported public-safe raw-feature groups. | Tracks what non-foundation heads can do with richer processed inputs. | Raw gated media redistribution. |
+| Simple baseline | A non-neural baseline family for the selected-128 rows. | Used for metadata/text and raw-feature 128-episode comparisons before NN/foundation-model rows. | The single-episode Minimal baseline. |
+| Spatial intelligence | Learning geometry and spatial reasoning from egocentric data. | Uses video, depth, camera pose, and language tasks to target 3D/space reasoning. | World-model future prediction. |
 | Vision-language-action | Mapping perception and language to action chunks. | A future policy/VLA direction that needs action-target conversion and stronger policy packaging. | Qwen3-Omni diagnostic scoring. |
-| HF Space | Hugging Face-hosted app/site surface. | Mirrors the dashboard and static website assets. | HF artifact dataset or model repo. |
+
+### Public surfaces
+
+| Term | Plain meaning | In this project | Do not confuse with |
+| --- | --- | --- | --- |
 | HF artifact dataset | Hugging Face dataset repo for derived evidence. | Stores public-safe reports, metrics, website JSON, and sanitized result packages. | Original Xperience-10M dataset. |
 | HF baseline model repo | Hugging Face model repo for lightweight baseline artifacts. | Mirrors baseline weights, figures, metrics, and task artifacts. | Qwen/Cosmos adapter-specific repos. |
-| HF weights/results repo | Consolidated public-safe model-result bundle. | Groups baseline weights, verified Qwen/Cosmos artifacts, analysis files, and manifests. | The upstream raw dataset. |
-| Mirror parity | A check that public copies match the source files. | `docs/data/mirror_parity.json` records whether GitHub, website, and HF mirrors agree. | A model-quality metric. |
+| HF Space | Hugging Face-hosted app/site surface. | Mirrors the dashboard and static website assets. | HF artifact dataset or model repo. |
+| HF weights/results repo | A consolidated public-safe model-result bundle. | Groups baseline weights, verified model artifacts, analysis files, and manifests. | The upstream raw dataset. |
+| Mirror parity | A check that public copies match the source files. | Records whether GitHub, website, and HF mirrors agree. | A model-quality metric. |
+| Public-safe artifact | A file that can be mirrored publicly without raw gated content. | Metrics, JSON summaries, model cards, figures, derived manifests, and approved lightweight weights/adapters. | Raw dataset redistribution. |
 | Publication audit | A public-package validation report. | Confirms required files exist and forbidden raw/private assets are not included. | Scientific peer review. |
 | Verified package | A result or artifact bundle that passed local/public validators. | Only verified packages are promoted to README, website, and HF surfaces as public evidence. | A running or exploratory experiment. |
 

data/artifact_index.json

@@ -1,6 +1,6 @@
 {
   "title": "Ropedia Xperience-10M Task Suite Artifact Index",
-  "generated_at_utc": "2026-06-22T14:40:39+00:00",
+  "generated_at_utc": "2026-06-22T15:09:12+00:00",
   "status": "pass",
   "artifact_count": 228,
   "missing": [],
@@ -92,8 +92,8 @@
       "surface": "repo_hf",
       "shows": "Defines terminology that can be confused across data scope, task metrics, model branches, and public mirrors.",
       "exists": true,
-      "bytes": 11122,
-      "sha256": "fe781a4eb5dd56454b5e0cb3383c88a2106c7bbf269888a0a7613b1618c8d196"
+      "bytes": 24120,
+      "sha256": "32812ee8aa021e7c9f7937ba1d436651290d6f3f61e290b17aee5ffc06746465"
     },
     {
       "id": "glossary_json",
@@ -103,8 +103,8 @@
       "surface": "website_hf",
       "shows": "Machine-readable terminology layer for the website, artifact dataset, model mirror, and public QA checks.",
       "exists": true,
-      "bytes": 19260,
-      "sha256": "525de375608793cd34ab386819eac5291177b53ca5839d54a9046707206e844a"
+      "bytes": 49025,
+      "sha256": "2939ed7aefd6c1f330785fae28492f8af49cc6d9ef10620e0dd79e9e3ae8c2cb"
     },
     {
       "id": "research_roadmap",
@@ -632,7 +632,7 @@
       "shows": "Machine-readable source-alignment pass/fail check for repo, website, and HF surfaces.",
       "exists": true,
       "bytes": 4432,
-      "sha256": "77199b03fc4d589a648033e359183a252d43dd6900e19ca609349bd972939e84"
+      "sha256": "c00a89e8694e08a6bb844924da00ba78bcf6c5da96690d548d670bf0baa4fa9f"
     },
     {
       "id": "source_alignment_validator",
@@ -1182,7 +1182,7 @@
       "shows": "Machine-readable release-check summary for validators, mirrors, and public project surfaces.",
       "exists": true,
       "bytes": 8640,
-      "sha256": "745b9c41da785ab84af8aba58babe3d8decd3c0bb07e3187b64f82ae42d91ec5"
+      "sha256": "e3f97614b47251d02f560db20a4bc918088b9aadc45c1d1397c2756b4d4867bf"
     },
     {
       "id": "public_surface_qa",

data/glossary.json

@@ -1,7 +1,7 @@
 {
   "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.",
+  "purpose": "Define reader-facing project terms and adjacent technical field terms for embodied AI, egocentric multimodal data, spatial intelligence, world models, VLA/policy learning, evaluation, and public artifact reading.",
   "categories": [
     {
       "id": "dataset_scope",
@@ -11,13 +11,38 @@
     {
       "id": "files_features",
       "label": "Files and features",
-      "description": "How raw sample files, windows, feature manifests, and public-safe derivatives relate."
+      "description": "Raw sample files, windows, feature manifests, and public-safe derivatives."
+    },
+    {
+      "id": "multimodal_sensing",
+      "label": "Multimodal sensing",
+      "description": "Video, audio, depth, IMU, motion capture, calibration, and synchronization terms."
+    },
+    {
+      "id": "spatial_geometry",
+      "label": "Spatial geometry",
+      "description": "Camera pose, SLAM, coordinate frames, point clouds, 3D reconstruction, and spatial grounding."
+    },
+    {
+      "id": "temporal_world_models",
+      "label": "Temporal and world models",
+      "description": "Future prediction, rollouts, forward dynamics, long-horizon forecasting, and temporal leakage."
+    },
+    {
+      "id": "robotics_vla",
+      "label": "Robotics and VLA",
+      "description": "Vision-language-action, policies, action chunks, imitation learning, contact, and dexterity."
     },
     {
       "id": "tasks_metrics",
       "label": "Tasks and metrics",
       "description": "Task contracts, scored records, direct scores, compact proxies, and audits."
     },
+    {
+      "id": "training_eval",
+      "label": "Training and evaluation",
+      "description": "Splits, held-out evaluation, metric types, prompt/schema checks, adapters, and distributed training."
+    },
     {
       "id": "models_runs",
       "label": "Models and runs",
@@ -31,12 +56,26 @@
   ],
   "entries": [
     {
-      "term": "Xperience-10M",
+      "term": "Evidence line",
       "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"]
+      "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 sample episode",
@@ -44,7 +83,10 @@
       "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"]
+      "primary_files": [
+        "docs/data/raw_sample_files.json",
+        "docs/single_episode_explorer.html"
+      ]
     },
     {
       "term": "Selected 128 episodes",
@@ -52,31 +94,42 @@
       "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"]
+      "primary_files": [
+        "XPERIENCE10M_128_EPISODE_FEATURE_INDEX.md",
+        "docs/data/xperience10m_128_episode_feature_index.json"
+      ]
     },
     {
-      "term": "Evidence line",
+      "term": "Xperience-10M",
       "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"]
+      "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": "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": "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": "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": "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": "Episode",
@@ -84,15 +137,10 @@
       "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"]
+      "primary_files": [
+        "docs/data/raw_sample_files.json",
+        "docs/data/xperience10m_128_episode_feature_index.json"
+      ]
     },
     {
       "term": "Feature manifest",
@@ -100,15 +148,41 @@
       "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"]
+      "primary_files": [
+        "results/episode_task_suite/feature_manifest.json"
+      ]
     },
     {
-      "term": "annotation.hdf5",
+      "term": "Interaction text",
       "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"]
+      "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": "Raw sample file map",
+      "category": "files_features",
+      "plain_meaning": "A human-readable inventory of the sample episode files.",
+      "project_usage": "Explains videos, annotations, calibration, motion, and derived previews.",
+      "do_not_confuse_with": "A training manifest.",
+      "primary_files": [
+        "docs/data/raw_sample_files.json"
+      ]
     },
     {
       "term": "visualization.rrd",
@@ -116,47 +190,501 @@
       "plain_meaning": "Rerun viewer recording for visual inspection.",
       "project_usage": "Can be downloaded from the official sample dataset and opened in Rerun 0.29.0 to inspect the sample episode. It is not used for published training or metric rows.",
       "do_not_confuse_with": "MP4 video streams or model inputs.",
-      "primary_files": ["docs/data/raw_sample_files.json", "REPRODUCIBILITY.md"]
+      "primary_files": [
+        "docs/data/raw_sample_files.json",
+        "REPRODUCIBILITY.md"
+      ]
     },
     {
-      "term": "Interaction text",
+      "term": "Window stride",
       "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"]
+      "plain_meaning": "The frame step between neighboring windows.",
+      "project_usage": "Creates overlapping examples while preserving chronological order and leakage controls.",
+      "do_not_confuse_with": "Video frame rate.",
+      "primary_files": [
+        "EVALUATION_PROTOCOL.md"
+      ]
     },
     {
-      "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": "Audio waveform",
+      "category": "multimodal_sensing",
+      "plain_meaning": "A time-series pressure signal from sound.",
+      "project_usage": "The audio ablation measures whether embedded audio helps selected task contracts.",
+      "do_not_confuse_with": "Language captions or text labels.",
+      "primary_files": [
+        "docs/data/audio_ablation_summary.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": "Calibration",
+      "category": "multimodal_sensing",
+      "plain_meaning": "Parameters that relate sensors to each other and to physical space.",
+      "project_usage": "Needed to interpret camera streams, depth, pose, and synchronized multimodal features together.",
+      "do_not_confuse_with": "A model training hyperparameter.",
+      "primary_files": [
+        "docs/data/raw_sample_files.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": "Camera extrinsics",
+      "category": "multimodal_sensing",
+      "plain_meaning": "A camera position and orientation relative to another coordinate frame.",
+      "project_usage": "Connects different camera streams and world coordinates.",
+      "do_not_confuse_with": "Camera intrinsics.",
+      "primary_files": [
+        "docs/data/raw_sample_files.json"
+      ]
     },
     {
-      "term": "Task-method record",
+      "term": "Camera intrinsics",
+      "category": "multimodal_sensing",
+      "plain_meaning": "Internal camera parameters such as focal length and distortion.",
+      "project_usage": "Explain how image pixels project to rays for geometry tasks.",
+      "do_not_confuse_with": "Camera extrinsics.",
+      "primary_files": [
+        "docs/data/raw_sample_files.json"
+      ]
+    },
+    {
+      "term": "Depth map",
+      "category": "multimodal_sensing",
+      "plain_meaning": "A per-pixel estimate of distance from the camera.",
+      "project_usage": "Depth-derived signals support spatial and geometry-oriented tasks.",
+      "do_not_confuse_with": "RGB brightness or semantic segmentation.",
+      "primary_files": [
+        "docs/data/modality_atlas.json"
+      ]
+    },
+    {
+      "term": "Egocentric video",
+      "category": "multimodal_sensing",
+      "plain_meaning": "Video captured from a first-person or body-mounted viewpoint.",
+      "project_usage": "The sample streams are egocentric views of human interaction and are the visual basis for many tasks.",
+      "do_not_confuse_with": "Third-person robot-camera footage.",
+      "primary_files": [
+        "docs/data/raw_sample_files.json"
+      ]
+    },
+    {
+      "term": "Fisheye camera",
+      "category": "multimodal_sensing",
+      "plain_meaning": "A wide-angle camera with strong lens distortion.",
+      "project_usage": "Multiple fisheye MP4 streams give broad room coverage but need calibration-aware interpretation.",
+      "do_not_confuse_with": "A rectilinear pinhole camera image.",
+      "primary_files": [
+        "docs/data/raw_sample_files.json"
+      ]
+    },
+    {
+      "term": "IMU",
+      "category": "multimodal_sensing",
+      "plain_meaning": "An inertial measurement unit with accelerometer and gyroscope signals.",
+      "project_usage": "Supports motion, temporal, and sensor-bridging tasks.",
+      "do_not_confuse_with": "Motion capture skeleton data.",
+      "primary_files": [
+        "docs/data/modality_atlas.json"
+      ]
+    },
+    {
+      "term": "Metric depth",
+      "category": "multimodal_sensing",
+      "plain_meaning": "Depth expressed in physical units rather than arbitrary relative scale.",
+      "project_usage": "Useful for distance-sensitive spatial reasoning and reconstruction targets.",
+      "do_not_confuse_with": "Relative monocular depth.",
+      "primary_files": [
+        "docs/data/modality_atlas.json"
+      ]
+    },
+    {
+      "term": "Motion capture",
+      "category": "multimodal_sensing",
+      "plain_meaning": "A system that records body or hand motion over time.",
+      "project_usage": "Provides hand/body motion evidence when exposed through public-safe derived features.",
+      "do_not_confuse_with": "Video-only pose estimation.",
+      "primary_files": [
+        "docs/data/modality_atlas.json"
+      ]
+    },
+    {
+      "term": "RGB frame",
+      "category": "multimodal_sensing",
+      "plain_meaning": "A color image frame from a video stream.",
+      "project_usage": "Used for visual statistics, previews, and many model inputs.",
+      "do_not_confuse_with": "Depth values or point-cloud coordinates.",
+      "primary_files": [
+        "results/episode_task_suite/feature_manifest.json"
+      ]
+    },
+    {
+      "term": "Sensor alignment",
+      "category": "multimodal_sensing",
+      "plain_meaning": "Putting different sensor streams into a shared temporal or spatial reference.",
+      "project_usage": "Used to make video, audio, pose, depth, IMU, and mocap usable in the same task input.",
+      "do_not_confuse_with": "Model ensembling.",
+      "primary_files": [
+        "docs/data/modality_atlas.json"
+      ]
+    },
+    {
+      "term": "Stereo camera",
+      "category": "multimodal_sensing",
+      "plain_meaning": "A paired-camera setup that supports depth or geometry estimation.",
+      "project_usage": "The sample browser exposes stereo streams as part of the visual modality set.",
+      "do_not_confuse_with": "Single-view RGB video.",
+      "primary_files": [
+        "docs/data/raw_sample_files.json"
+      ]
+    },
+    {
+      "term": "Timestamp synchronization",
+      "category": "multimodal_sensing",
+      "plain_meaning": "Aligning sensor samples by time.",
+      "project_usage": "The task suite assumes aligned windows across modalities so labels and features refer to the same moment.",
+      "do_not_confuse_with": "Randomly joining files with similar names.",
+      "primary_files": [
+        "EVALUATION_PROTOCOL.md"
+      ]
+    },
+    {
+      "term": "3D reconstruction",
+      "category": "spatial_geometry",
+      "plain_meaning": "Recovering 3D scene structure from sensor data.",
+      "project_usage": "One core spatial-intelligence direction for Xperience-style data.",
+      "do_not_confuse_with": "Next-action classification.",
+      "primary_files": [
+        "docs/data/three_foundation_pipelines.json"
+      ]
+    },
+    {
+      "term": "Affordance",
+      "category": "spatial_geometry",
+      "plain_meaning": "An action possibility offered by an object or scene.",
+      "project_usage": "Relevant when moving from observed human interaction to robot-action or VLA tasks.",
+      "do_not_confuse_with": "A detected object category alone.",
+      "primary_files": [
+        "docs/data/three_foundation_pipelines.json"
+      ]
+    },
+    {
+      "term": "Camera pose",
+      "category": "spatial_geometry",
+      "plain_meaning": "The camera position and orientation at a time step.",
+      "project_usage": "Supports spatial-intelligence tasks, view synchronization, and geometry diagnostics.",
+      "do_not_confuse_with": "The human body pose.",
+      "primary_files": [
+        "docs/data/modality_atlas.json"
+      ]
+    },
+    {
+      "term": "Coordinate frame",
+      "category": "spatial_geometry",
+      "plain_meaning": "A reference system for positions and orientations.",
+      "project_usage": "Needed when comparing camera, body, object, and world measurements.",
+      "do_not_confuse_with": "A video frame.",
+      "primary_files": [
+        "EVALUATION_PROTOCOL.md"
+      ]
+    },
+    {
+      "term": "Object-centric representation",
+      "category": "spatial_geometry",
+      "plain_meaning": "A representation organized around objects and their relations.",
+      "project_usage": "Useful for object relevance, object-set forecast, and action-object relation tasks.",
+      "do_not_confuse_with": "A flat feature vector without object identity.",
+      "primary_files": [
+        "docs/data/task_suite_20.json"
+      ]
+    },
+    {
+      "term": "Odometry",
+      "category": "spatial_geometry",
+      "plain_meaning": "Motion estimated from sensor changes over time.",
+      "project_usage": "A relevant spatial term for ego-motion and camera-pose reasoning.",
+      "do_not_confuse_with": "Ground-truth motion capture.",
+      "primary_files": [
+        "docs/data/modality_atlas.json"
+      ]
+    },
+    {
+      "term": "Point cloud",
+      "category": "spatial_geometry",
+      "plain_meaning": "A set of 3D points representing scene structure.",
+      "project_usage": "A likely target or intermediate representation for spatial-intelligence extensions.",
+      "do_not_confuse_with": "A 2D image grid.",
+      "primary_files": [
+        "docs/data/three_foundation_pipelines.json"
+      ]
+    },
+    {
+      "term": "SLAM",
+      "category": "spatial_geometry",
+      "plain_meaning": "Simultaneous localization and mapping.",
+      "project_usage": "A field term for estimating camera motion and scene structure from sensor observations.",
+      "do_not_confuse_with": "A task label or action class.",
+      "primary_files": [
+        "docs/data/modality_atlas.json"
+      ]
+    },
+    {
+      "term": "Spatial grounding",
+      "category": "spatial_geometry",
+      "plain_meaning": "Linking language or labels to locations, objects, or geometry.",
+      "project_usage": "Connects language grounding tasks with 3D/spatial reasoning.",
+      "do_not_confuse_with": "General text classification.",
+      "primary_files": [
+        "docs/data/research_directions.json"
+      ]
+    },
+    {
+      "term": "Trajectory",
+      "category": "spatial_geometry",
+      "plain_meaning": "A sequence of positions over time.",
+      "project_usage": "Used for hand motion, camera motion, and future-path tasks.",
+      "do_not_confuse_with": "A single coordinate or label.",
+      "primary_files": [
+        "TASK_SUITE_20.md"
+      ]
+    },
+    {
+      "term": "Action forecasting",
+      "category": "temporal_world_models",
+      "plain_meaning": "Predicting a future action before it happens.",
+      "project_usage": "Covered by next-action and long-horizon task contracts.",
+      "do_not_confuse_with": "Recognizing the current action only.",
+      "primary_files": [
+        "docs/data/task_suite_20.json"
+      ]
+    },
+    {
+      "term": "Autoregressive prediction",
+      "category": "temporal_world_models",
+      "plain_meaning": "Generating each future token, state, or frame conditioned on prior outputs.",
+      "project_usage": "Relevant for model branches that produce structured JSON or temporal predictions.",
+      "do_not_confuse_with": "A one-shot classifier.",
+      "primary_files": [
+        "docs/data/foundation_model_plan.json"
+      ]
+    },
+    {
+      "term": "Forward dynamics",
+      "category": "temporal_world_models",
+      "plain_meaning": "Predicting the next state from the current state and action/context.",
+      "project_usage": "The Cosmos3-Super LoRA branch uses a forward-dynamics-style diagnostic contract.",
+      "do_not_confuse_with": "Reverse inference from result back to cause.",
+      "primary_files": [
+        "docs/data/omni_model_comparison.json"
+      ]
+    },
+    {
+      "term": "Latent state",
+      "category": "temporal_world_models",
+      "plain_meaning": "A hidden representation that summarizes observed context.",
+      "project_usage": "Useful for future foundation-model and world-model training plans.",
+      "do_not_confuse_with": "A visible annotation column.",
+      "primary_files": [
+        "docs/data/foundation_model_plan.json"
+      ]
+    },
+    {
+      "term": "Long-horizon prediction",
+      "category": "temporal_world_models",
+      "plain_meaning": "Predicting outcomes several seconds or steps ahead.",
+      "project_usage": "Tasks 13 and 14 test longer temporal context beyond immediate recognition.",
+      "do_not_confuse_with": "Single-frame classification.",
+      "primary_files": [
+        "docs/data/task_suite_20.json"
+      ]
+    },
+    {
+      "term": "Next-frame prediction",
+      "category": "temporal_world_models",
+      "plain_meaning": "Predicting future visual frames from past frames.",
+      "project_usage": "A field-level world-model objective related to the human-video world-model direction.",
+      "do_not_confuse_with": "Next-action prediction.",
+      "primary_files": [
+        "docs/data/three_foundation_pipelines.json"
+      ]
+    },
+    {
+      "term": "Object persistence",
+      "category": "temporal_world_models",
+      "plain_meaning": "Tracking that an object remains present over time even when view or interaction changes.",
+      "project_usage": "Relevant for object-set forecast and long-video reasoning.",
+      "do_not_confuse_with": "A single-frame object detection.",
+      "primary_files": [
+        "docs/data/task_suite_20.json"
+      ]
+    },
+    {
+      "term": "Rollout",
+      "category": "temporal_world_models",
+      "plain_meaning": "Repeatedly predicting future steps from a model state.",
+      "project_usage": "Important for judging world models beyond one-step prediction.",
+      "do_not_confuse_with": "A held-out static test row.",
+      "primary_files": [
+        "docs/data/three_foundation_pipelines.json"
+      ]
+    },
+    {
+      "term": "Subtask forecasting",
+      "category": "temporal_world_models",
+      "plain_meaning": "Predicting the next higher-level step in an activity.",
+      "project_usage": "Used in the future-task probe line for Qwen3-Omni.",
+      "do_not_confuse_with": "Frame-level action classification.",
+      "primary_files": [
+        "docs/data/task_method_20_result_matrix.json"
+      ]
+    },
+    {
+      "term": "Teacher forcing",
+      "category": "temporal_world_models",
+      "plain_meaning": "Training a sequence model using ground-truth previous outputs.",
+      "project_usage": "A likely training option for future sequence/world-model baselines.",
+      "do_not_confuse_with": "Free-running rollout evaluation.",
+      "primary_files": [
+        "docs/data/foundation_model_plan.json"
+      ]
+    },
+    {
+      "term": "Temporal leakage",
+      "category": "temporal_world_models",
+      "plain_meaning": "Using future information that would not be available at prediction time.",
+      "project_usage": "Avoided by chronological splits and target-side feature controls.",
+      "do_not_confuse_with": "A low model score.",
+      "primary_files": [
+        "EVALUATION_PROTOCOL.md"
+      ]
+    },
+    {
+      "term": "Transition timing",
+      "category": "temporal_world_models",
+      "plain_meaning": "Estimating when the next state or action transition happens.",
+      "project_usage": "Task 20 turns temporal change into a regression target.",
+      "do_not_confuse_with": "Classifying the transition type only.",
+      "primary_files": [
+        "docs/data/task_suite_20.json"
+      ]
+    },
+    {
+      "term": "Action chunk",
+      "category": "robotics_vla",
+      "plain_meaning": "A short sequence of low-level actions predicted together.",
+      "project_usage": "The VLA figure and plan use action chunks as the policy-output concept.",
+      "do_not_confuse_with": "A natural-language action label.",
+      "primary_files": [
+        "docs/data/three_foundation_pipelines.json"
+      ]
+    },
+    {
+      "term": "Behavior cloning",
+      "category": "robotics_vla",
+      "plain_meaning": "A supervised imitation-learning method for predicting demonstrated actions.",
+      "project_usage": "A plausible baseline once action targets are converted.",
+      "do_not_confuse_with": "Generative video modeling.",
+      "primary_files": [
+        "docs/data/foundation_model_plan.json"
+      ]
+    },
+    {
+      "term": "Contact event",
+      "category": "robotics_vla",
+      "plain_meaning": "A moment when a hand, body, or tool touches an object or surface.",
+      "project_usage": "Used in contact-related tasks and action-quality interpretation.",
+      "do_not_confuse_with": "Visual co-occurrence without touch.",
+      "primary_files": [
+        "docs/data/task_suite_20.json"
+      ]
+    },
+    {
+      "term": "Dexterity",
+      "category": "robotics_vla",
+      "plain_meaning": "Fine-grained physical manipulation ability.",
+      "project_usage": "Relevant to hand-object interaction, contact, and VLA/policy directions.",
+      "do_not_confuse_with": "High text-generation accuracy.",
+      "primary_files": [
+        "docs/data/research_directions.json"
+      ]
+    },
+    {
+      "term": "End effector",
+      "category": "robotics_vla",
+      "plain_meaning": "The robot part that acts on the world, such as a gripper or hand.",
+      "project_usage": "A key target frame for future manipulation-policy conversion.",
+      "do_not_confuse_with": "A camera or global scene coordinate.",
+      "primary_files": [
+        "docs/data/three_foundation_pipelines.json"
+      ]
+    },
+    {
+      "term": "Hand-object interaction",
+      "category": "robotics_vla",
+      "plain_meaning": "A physical interaction between hands and objects.",
+      "project_usage": "A central signal family behind action, contact, object relevance, and interaction-text tasks.",
+      "do_not_confuse_with": "Object detection without action.",
+      "primary_files": [
+        "docs/data/task_suite_20.json"
+      ]
+    },
+    {
+      "term": "Imitation learning",
+      "category": "robotics_vla",
+      "plain_meaning": "Training a policy to imitate demonstrated behavior.",
+      "project_usage": "Relevant when converting human video/motion into action supervision.",
+      "do_not_confuse_with": "Reinforcement learning from online robot trials.",
+      "primary_files": [
+        "docs/data/foundation_model_plan.json"
+      ]
+    },
+    {
+      "term": "Language grounding",
+      "category": "robotics_vla",
+      "plain_meaning": "Connecting text to observed objects, actions, or spatial context.",
+      "project_usage": "Task 8 and VLA directions use language as grounded supervision rather than standalone text.",
+      "do_not_confuse_with": "Caption fluency alone.",
+      "primary_files": [
+        "docs/data/task_suite_20.json"
+      ]
+    },
+    {
+      "term": "Policy",
+      "category": "robotics_vla",
+      "plain_meaning": "A mapping from observations to actions.",
+      "project_usage": "A future target for robot-compatible Xperience-derived action data.",
+      "do_not_confuse_with": "A benchmark metric.",
+      "primary_files": [
+        "docs/data/foundation_model_plan.json"
+      ]
+    },
+    {
+      "term": "Robot-compatible action target",
+      "category": "robotics_vla",
+      "plain_meaning": "An action representation a robot policy can execute or imitate.",
+      "project_usage": "Needed before OpenVLA/openpi/GR00T-style policy training is meaningful here.",
+      "do_not_confuse_with": "Human-only caption text.",
+      "primary_files": [
+        "docs/data/foundation_model_plan.json"
+      ]
+    },
+    {
+      "term": "Vision-language-action model",
+      "category": "robotics_vla",
+      "plain_meaning": "A model that maps visual context and language into actions.",
+      "project_usage": "The VLA direction is a future path after action targets are converted into robot-compatible chunks.",
+      "do_not_confuse_with": "A vision-language model that only answers text.",
+      "primary_files": [
+        "docs/data/three_foundation_pipelines.json"
+      ]
+    },
+    {
+      "term": "Compact-proxy score",
       "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"]
+      "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": "Direct score",
@@ -164,23 +692,32 @@
       "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"]
+      "primary_files": [
+        "TASK_METHOD_20_GAP_AUDIT.md",
+        "docs/data/task_method_20_gap_audit.json"
+      ]
     },
     {
-      "term": "Compact-proxy score",
+      "term": "Gap audit",
       "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"]
+      "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": "Raw metric value",
+      "term": "Leakage control",
       "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"]
+      "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": "Normalized radar value",
@@ -188,63 +725,214 @@
       "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"]
+      "primary_files": [
+        "docs/data/unified_task_model_radar.json",
+        "docs/assets/charts/unified_task_model_radar.svg"
+      ]
     },
     {
-      "term": "Gap audit",
+      "term": "Raw metric value",
       "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"]
+      "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": "Leakage control",
+      "term": "Task contract",
       "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"]
+      "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": "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": "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": "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": "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": "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": "Adapter checkpoint",
+      "category": "training_eval",
+      "plain_meaning": "Saved adapter weights from a fine-tuning run.",
+      "project_usage": "The public model branches publish adapters when validated and public-safe.",
+      "do_not_confuse_with": "Full base-model checkpoint.",
+      "primary_files": [
+        "docs/data/omni_model_comparison.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": "Balanced accuracy",
+      "category": "training_eval",
+      "plain_meaning": "Accuracy averaged across classes to reduce majority-class dominance.",
+      "project_usage": "Useful for imbalanced task labels.",
+      "do_not_confuse_with": "Overall accuracy.",
+      "primary_files": [
+        "docs/data/task_method_20_result_matrix.json"
+      ]
     },
     {
-      "term": "Qwen v1-v6",
+      "term": "Chronological split",
+      "category": "training_eval",
+      "plain_meaning": "A split ordered by time.",
+      "project_usage": "Used for the single-episode baselines to reduce future-window leakage.",
+      "do_not_confuse_with": "A random row split.",
+      "primary_files": [
+        "EVALUATION_PROTOCOL.md"
+      ]
+    },
+    {
+      "term": "Confusion matrix",
+      "category": "training_eval",
+      "plain_meaning": "A table of predicted classes versus true classes.",
+      "project_usage": "Helps inspect which task labels a method confuses.",
+      "do_not_confuse_with": "A scalar leaderboard score.",
+      "primary_files": [
+        "results/episode_task_suite/neural_mlp"
+      ]
+    },
+    {
+      "term": "FSDP",
+      "category": "training_eval",
+      "plain_meaning": "Fully Sharded Data Parallel, a distributed training strategy.",
+      "project_usage": "Appears in full-parameter feasibility and multi-GPU training notes.",
+      "do_not_confuse_with": "A model architecture.",
+      "primary_files": [
+        "docs/data/qwen3_full_parameter_gates.json"
+      ]
+    },
+    {
+      "term": "Held-out evaluation",
+      "category": "training_eval",
+      "plain_meaning": "Testing on examples not used for training.",
+      "project_usage": "Required before promoting Qwen/Cosmos results to public evidence.",
+      "do_not_confuse_with": "Training-set loss.",
+      "primary_files": [
+        "docs/data/omni_model_comparison.json"
+      ]
+    },
+    {
+      "term": "JSON validity",
+      "category": "training_eval",
+      "plain_meaning": "Whether model output parses as the required JSON schema.",
+      "project_usage": "A key diagnostic for Qwen3-Omni structured-output runs.",
+      "do_not_confuse_with": "Task correctness after parsing.",
+      "primary_files": [
+        "docs/data/omni_model_comparison.json"
+      ]
+    },
+    {
+      "term": "Macro F1",
+      "category": "training_eval",
+      "plain_meaning": "The average F1 score across classes, usually treating classes equally.",
+      "project_usage": "Used when class imbalance matters in classification tasks.",
+      "do_not_confuse_with": "Accuracy dominated by frequent classes.",
+      "primary_files": [
+        "docs/data/task_method_20_result_matrix.json"
+      ]
+    },
+    {
+      "term": "Mean absolute error",
+      "category": "training_eval",
+      "plain_meaning": "The average absolute difference between predicted and true numeric values.",
+      "project_usage": "Used for regression-style task rows such as timing or trajectory targets.",
+      "do_not_confuse_with": "A classification F1 score.",
+      "primary_files": [
+        "docs/data/task_method_20_result_matrix.json"
+      ]
+    },
+    {
+      "term": "Overfit check",
+      "category": "training_eval",
+      "plain_meaning": "A small training test that verifies a model can learn a tiny subset.",
+      "project_usage": "Useful for catching data/model wiring bugs before full training.",
+      "do_not_confuse_with": "Evidence of generalization.",
+      "primary_files": [
+        "docs/data/foundation_model_plan.json"
+      ]
+    },
+    {
+      "term": "Parameter-efficient fine-tuning",
+      "category": "training_eval",
+      "plain_meaning": "Updating a small number of added or selected parameters.",
+      "project_usage": "LoRA is the current parameter-efficient path for Qwen/Cosmos branches.",
+      "do_not_confuse_with": "Full-parameter fine-tuning.",
+      "primary_files": [
+        "docs/data/foundation_model_plan.json"
+      ]
+    },
+    {
+      "term": "Schema compliance",
+      "category": "training_eval",
+      "plain_meaning": "Whether an output follows the expected field names and value types.",
+      "project_usage": "Needed for structured task probes and public package validation.",
+      "do_not_confuse_with": "High semantic accuracy.",
+      "primary_files": [
+        "docs/data/omni_model_comparison.json"
+      ]
+    },
+    {
+      "term": "Smoke run",
+      "category": "training_eval",
+      "plain_meaning": "A short run that checks whether a pipeline can start and execute key steps.",
+      "project_usage": "Used for feasibility gates before expensive full runs.",
+      "do_not_confuse_with": "A complete benchmark result.",
+      "primary_files": [
+        "docs/data/qwen3_full_parameter_gates.json"
+      ]
+    },
+    {
+      "term": "Top-k accuracy",
+      "category": "training_eval",
+      "plain_meaning": "A score that counts a prediction correct if the target is among the k highest-ranked outputs.",
+      "project_usage": "Useful for large-label or retrieval-style tasks.",
+      "do_not_confuse_with": "Top-1 exact accuracy.",
+      "primary_files": [
+        "docs/data/task_method_20_result_matrix.json"
+      ]
+    },
+    {
+      "term": "Train/validation/test split",
+      "category": "training_eval",
+      "plain_meaning": "A partition that separates model fitting, tuning, and final evaluation examples.",
+      "project_usage": "The selected-128 setup uses a held-out split discipline for model branches.",
+      "do_not_confuse_with": "A random shuffle without temporal or episode boundaries.",
+      "primary_files": [
+        "EVALUATION_PROTOCOL.md"
+      ]
+    },
+    {
+      "term": "Cosmos3-Nano",
       "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"]
+      "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": "Cosmos3-Super",
@@ -252,15 +940,41 @@
       "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"]
+      "primary_files": [
+        "docs/data/omni_model_comparison.json"
+      ]
     },
     {
-      "term": "Cosmos3-Nano",
+      "term": "Foundation pipeline",
       "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"]
+      "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": "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": "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": "LoRA adapter",
@@ -268,23 +982,85 @@
       "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"]
+      "primary_files": [
+        "OMNI_MODEL_EXTENSION_CONTRACT.md",
+        "docs/data/omni_model_comparison.json"
+      ]
     },
     {
-      "term": "Full-parameter fine-tuning",
+      "term": "Metadata baseline",
       "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"]
+      "plain_meaning": "A selected-128 baseline using metadata or text-derived public-safe features.",
+      "project_usage": "Compares simple and neural heads on the held-out split.",
+      "do_not_confuse_with": "Raw video, depth, or audio feature baselines.",
+      "primary_files": [
+        "docs/data/task_method_20_result_matrix.json"
+      ]
     },
     {
-      "term": "Foundation pipeline",
+      "term": "Minimal baseline",
       "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"]
+      "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": "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": "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": "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": "Raw-feature baseline",
+      "category": "models_runs",
+      "plain_meaning": "A selected-128 baseline using exported public-safe raw-feature groups.",
+      "project_usage": "Tracks what non-foundation heads can do with richer processed inputs.",
+      "do_not_confuse_with": "Raw gated media redistribution.",
+      "primary_files": [
+        "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": "Spatial intelligence",
@@ -292,15 +1068,10 @@
       "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"]
+      "primary_files": [
+        "THREE_FOUNDATION_PIPELINES.md",
+        "docs/data/three_foundation_pipelines.json"
+      ]
     },
     {
       "term": "Vision-language-action",
@@ -308,15 +1079,10 @@
       "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"]
+      "primary_files": [
+        "THREE_FOUNDATION_PIPELINES.md",
+        "docs/data/three_foundation_pipelines.json"
+      ]
     },
     {
       "term": "HF artifact dataset",
@@ -324,7 +1090,10 @@
       "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"]
+      "primary_files": [
+        "ARTIFACT_GUIDE.md",
+        "docs/data/artifact_index.json"
+      ]
     },
     {
       "term": "HF baseline model repo",
@@ -332,7 +1101,31 @@
       "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"]
+      "primary_files": [
+        "PUBLIC_READER_MAP.md",
+        "docs/data/public_reader_map.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 weights/results repo",
+      "category": "public_surfaces",
+      "plain_meaning": "A consolidated public-safe model-result bundle.",
+      "project_usage": "Groups baseline weights, verified model artifacts, analysis files, and manifests.",
+      "do_not_confuse_with": "The upstream raw dataset.",
+      "primary_files": [
+        "PUBLIC_READER_MAP.md"
+      ]
     },
     {
       "term": "Mirror parity",
@@ -340,7 +1133,30 @@
       "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"]
+      "primary_files": [
+        "docs/data/mirror_parity.json"
+      ]
+    },
+    {
+      "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": "Publication audit",
+      "category": "public_surfaces",
+      "plain_meaning": "A public-package validation report.",
+      "project_usage": "Confirms required files exist and forbidden raw/private assets are not included.",
+      "do_not_confuse_with": "Scientific peer review.",
+      "primary_files": [
+        "docs/data/publication_audit.json"
+      ]
     },
     {
       "term": "Verified package",
@@ -348,37 +1164,62 @@
       "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"]
+      "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"]
+      "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"]
+      "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"]
+      "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"]
+      "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"]
+      "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"]
+      "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"]
+      "files": [
+        "PUBLIC_SURFACE_QA.md",
+        "docs/data/mirror_parity.json",
+        "docs/data/live_publication_status.json"
+      ]
     }
   ]
 }

data/mirror_parity.json

@@ -1,6 +1,6 @@
 {
   "status": "pass",
-  "generated_at_utc": "2026-06-22T14:36:09+00:00",
+  "generated_at_utc": "2026-06-22T14:44:16+00:00",
   "hf_root": "hf_publish",
   "summary": {
     "group_count": 1306,
@@ -139,44 +139,44 @@
         "path": "repo:docs/data/artifact_index.json",
         "exists": true,
         "bytes": 124477,
-        "sha256": "e987021a89d254d0e3caafe7281872511a85357cc475dd33494f00edfed1e098"
+        "sha256": "246b4e257b7ac1abe2967fe08925b5c7d2ea3454edc55d0c0589eeba8dd1518e"
       },
       "mirrors": {
         "hf_space": {
           "path": "hf_space:data/artifact_index.json",
           "exists": true,
           "bytes": 124477,
-          "sha256": "e987021a89d254d0e3caafe7281872511a85357cc475dd33494f00edfed1e098"
+          "sha256": "246b4e257b7ac1abe2967fe08925b5c7d2ea3454edc55d0c0589eeba8dd1518e"
         },
         "hf_artifacts_data": {
           "path": "hf_artifacts:data/artifact_index.json",
           "exists": true,
           "bytes": 124477,
-          "sha256": "e987021a89d254d0e3caafe7281872511a85357cc475dd33494f00edfed1e098"
+          "sha256": "246b4e257b7ac1abe2967fe08925b5c7d2ea3454edc55d0c0589eeba8dd1518e"
         },
         "hf_artifacts": {
           "path": "hf_artifacts:docs/data/artifact_index.json",
           "exists": true,
           "bytes": 124477,
-          "sha256": "e987021a89d254d0e3caafe7281872511a85357cc475dd33494f00edfed1e098"
+          "sha256": "246b4e257b7ac1abe2967fe08925b5c7d2ea3454edc55d0c0589eeba8dd1518e"
         },
         "hf_model_data": {
           "path": "hf_model:data/artifact_index.json",
           "exists": true,
           "bytes": 124477,
-          "sha256": "e987021a89d254d0e3caafe7281872511a85357cc475dd33494f00edfed1e098"
+          "sha256": "246b4e257b7ac1abe2967fe08925b5c7d2ea3454edc55d0c0589eeba8dd1518e"
         },
         "hf_model_docs_data": {
           "path": "hf_model:docs/data/artifact_index.json",
           "exists": true,
           "bytes": 124477,
-          "sha256": "e987021a89d254d0e3caafe7281872511a85357cc475dd33494f00edfed1e098"
+          "sha256": "246b4e257b7ac1abe2967fe08925b5c7d2ea3454edc55d0c0589eeba8dd1518e"
         },
         "hf_model": {
           "path": "hf_model:metrics/artifact_index.json",
           "exists": true,
           "bytes": 124477,
-          "sha256": "e987021a89d254d0e3caafe7281872511a85357cc475dd33494f00edfed1e098"
+          "sha256": "246b4e257b7ac1abe2967fe08925b5c7d2ea3454edc55d0c0589eeba8dd1518e"
         }
       },
       "failures": []
@@ -972,44 +972,44 @@
         "path": "repo:docs/data/publication_audit.json",
         "exists": true,
         "bytes": 10940,
-        "sha256": "5b02a1990d8e14421a9cb9803f059b70c0c816440cc0dcaeee939a0efc4341bc"
+        "sha256": "2b988f2440692a77049129e6f1f6495ae541e371395fa2b7527ec68d6bd9fb8d"
       },
       "mirrors": {
         "hf_space": {
           "path": "hf_space:data/publication_audit.json",
           "exists": true,
           "bytes": 10940,
-          "sha256": "5b02a1990d8e14421a9cb9803f059b70c0c816440cc0dcaeee939a0efc4341bc"
+          "sha256": "2b988f2440692a77049129e6f1f6495ae541e371395fa2b7527ec68d6bd9fb8d"
         },
         "hf_artifacts_data": {
           "path": "hf_artifacts:data/publication_audit.json",
           "exists": true,
           "bytes": 10940,
-          "sha256": "5b02a1990d8e14421a9cb9803f059b70c0c816440cc0dcaeee939a0efc4341bc"
+          "sha256": "2b988f2440692a77049129e6f1f6495ae541e371395fa2b7527ec68d6bd9fb8d"
         },
         "hf_artifacts": {
           "path": "hf_artifacts:docs/data/publication_audit.json",
           "exists": true,
           "bytes": 10940,
-          "sha256": "5b02a1990d8e14421a9cb9803f059b70c0c816440cc0dcaeee939a0efc4341bc"
+          "sha256": "2b988f2440692a77049129e6f1f6495ae541e371395fa2b7527ec68d6bd9fb8d"
         },
         "hf_model_data": {
           "path": "hf_model:data/publication_audit.json",
           "exists": true,
           "bytes": 10940,
-          "sha256": "5b02a1990d8e14421a9cb9803f059b70c0c816440cc0dcaeee939a0efc4341bc"
+          "sha256": "2b988f2440692a77049129e6f1f6495ae541e371395fa2b7527ec68d6bd9fb8d"
         },
         "hf_model_docs_data": {
           "path": "hf_model:docs/data/publication_audit.json",
           "exists": true,
           "bytes": 10940,
-          "sha256": "5b02a1990d8e14421a9cb9803f059b70c0c816440cc0dcaeee939a0efc4341bc"
+          "sha256": "2b988f2440692a77049129e6f1f6495ae541e371395fa2b7527ec68d6bd9fb8d"
         },
         "hf_model": {
           "path": "hf_model:metrics/publication_audit.json",
           "exists": true,
           "bytes": 10940,
-          "sha256": "5b02a1990d8e14421a9cb9803f059b70c0c816440cc0dcaeee939a0efc4341bc"
+          "sha256": "2b988f2440692a77049129e6f1f6495ae541e371395fa2b7527ec68d6bd9fb8d"
         }
       },
       "failures": []
@@ -1021,44 +1021,44 @@
         "path": "repo:docs/data/public_surface_qa.json",
         "exists": true,
         "bytes": 7690,
-        "sha256": "4c171fb154654cba54cc0697dcd0c53a9dbd805d22bfa3d2941b4fdc155bf146"
+        "sha256": "28467929e81e827a7f3de5508a981f20ab7f271ddac2b11188ffe655363ef50e"
       },
       "mirrors": {
         "hf_space": {
           "path": "hf_space:data/public_surface_qa.json",
           "exists": true,
           "bytes": 7690,
-          "sha256": "4c171fb154654cba54cc0697dcd0c53a9dbd805d22bfa3d2941b4fdc155bf146"
+          "sha256": "28467929e81e827a7f3de5508a981f20ab7f271ddac2b11188ffe655363ef50e"
         },
         "hf_artifacts_data": {
           "path": "hf_artifacts:data/public_surface_qa.json",
           "exists": true,
           "bytes": 7690,
-          "sha256": "4c171fb154654cba54cc0697dcd0c53a9dbd805d22bfa3d2941b4fdc155bf146"
+          "sha256": "28467929e81e827a7f3de5508a981f20ab7f271ddac2b11188ffe655363ef50e"
         },
         "hf_artifacts": {
           "path": "hf_artifacts:docs/data/public_surface_qa.json",
           "exists": true,
           "bytes": 7690,
-          "sha256": "4c171fb154654cba54cc0697dcd0c53a9dbd805d22bfa3d2941b4fdc155bf146"
+          "sha256": "28467929e81e827a7f3de5508a981f20ab7f271ddac2b11188ffe655363ef50e"
         },
         "hf_model_data": {
           "path": "hf_model:data/public_surface_qa.json",
           "exists": true,
           "bytes": 7690,
-          "sha256": "4c171fb154654cba54cc0697dcd0c53a9dbd805d22bfa3d2941b4fdc155bf146"
+          "sha256": "28467929e81e827a7f3de5508a981f20ab7f271ddac2b11188ffe655363ef50e"
         },
         "hf_model_docs_data": {
           "path": "hf_model:docs/data/public_surface_qa.json",
           "exists": true,
           "bytes": 7690,
-          "sha256": "4c171fb154654cba54cc0697dcd0c53a9dbd805d22bfa3d2941b4fdc155bf146"
+          "sha256": "28467929e81e827a7f3de5508a981f20ab7f271ddac2b11188ffe655363ef50e"
         },
         "hf_model": {
           "path": "hf_model:metrics/public_surface_qa.json",
           "exists": true,
           "bytes": 7690,
-          "sha256": "4c171fb154654cba54cc0697dcd0c53a9dbd805d22bfa3d2941b4fdc155bf146"
+          "sha256": "28467929e81e827a7f3de5508a981f20ab7f271ddac2b11188ffe655363ef50e"
         }
       },
       "failures": []
@@ -1217,44 +1217,44 @@
         "path": "repo:docs/data/quality_gates.json",
         "exists": true,
         "bytes": 8640,
-        "sha256": "745b9c41da785ab84af8aba58babe3d8decd3c0bb07e3187b64f82ae42d91ec5"
+        "sha256": "e3f97614b47251d02f560db20a4bc918088b9aadc45c1d1397c2756b4d4867bf"
       },
       "mirrors": {
         "hf_space": {
           "path": "hf_space:data/quality_gates.json",
           "exists": true,
           "bytes": 8640,
-          "sha256": "745b9c41da785ab84af8aba58babe3d8decd3c0bb07e3187b64f82ae42d91ec5"
+          "sha256": "e3f97614b47251d02f560db20a4bc918088b9aadc45c1d1397c2756b4d4867bf"
         },
         "hf_artifacts_data": {
           "path": "hf_artifacts:data/quality_gates.json",
           "exists": true,
           "bytes": 8640,
-          "sha256": "745b9c41da785ab84af8aba58babe3d8decd3c0bb07e3187b64f82ae42d91ec5"
+          "sha256": "e3f97614b47251d02f560db20a4bc918088b9aadc45c1d1397c2756b4d4867bf"
         },
         "hf_artifacts": {
           "path": "hf_artifacts:docs/data/quality_gates.json",
           "exists": true,
           "bytes": 8640,
-          "sha256": "745b9c41da785ab84af8aba58babe3d8decd3c0bb07e3187b64f82ae42d91ec5"
+          "sha256": "e3f97614b47251d02f560db20a4bc918088b9aadc45c1d1397c2756b4d4867bf"
         },
         "hf_model_data": {
           "path": "hf_model:data/quality_gates.json",
           "exists": true,
           "bytes": 8640,
-          "sha256": "745b9c41da785ab84af8aba58babe3d8decd3c0bb07e3187b64f82ae42d91ec5"
+          "sha256": "e3f97614b47251d02f560db20a4bc918088b9aadc45c1d1397c2756b4d4867bf"
         },
         "hf_model_docs_data": {
           "path": "hf_model:docs/data/quality_gates.json",
           "exists": true,
           "bytes": 8640,
-          "sha256": "745b9c41da785ab84af8aba58babe3d8decd3c0bb07e3187b64f82ae42d91ec5"
+          "sha256": "e3f97614b47251d02f560db20a4bc918088b9aadc45c1d1397c2756b4d4867bf"
         },
         "hf_model": {
           "path": "hf_model:metrics/quality_gates.json",
           "exists": true,
           "bytes": 8640,
-          "sha256": "745b9c41da785ab84af8aba58babe3d8decd3c0bb07e3187b64f82ae42d91ec5"
+          "sha256": "e3f97614b47251d02f560db20a4bc918088b9aadc45c1d1397c2756b4d4867bf"
         }
       },
       "failures": []
@@ -1756,44 +1756,44 @@
         "path": "repo:docs/data/source_alignment_audit.json",
         "exists": true,
         "bytes": 4432,
-        "sha256": "c3fafe7f03e0558e2acfd702d222bba9a53e391cfd711ee7a2b65299579fcf28"
+        "sha256": "77199b03fc4d589a648033e359183a252d43dd6900e19ca609349bd972939e84"
       },
       "mirrors": {
         "hf_space": {
           "path": "hf_space:data/source_alignment_audit.json",
           "exists": true,
           "bytes": 4432,
-          "sha256": "c3fafe7f03e0558e2acfd702d222bba9a53e391cfd711ee7a2b65299579fcf28"
+          "sha256": "77199b03fc4d589a648033e359183a252d43dd6900e19ca609349bd972939e84"
         },
         "hf_artifacts_data": {
           "path": "hf_artifacts:data/source_alignment_audit.json",
           "exists": true,
           "bytes": 4432,
-          "sha256": "c3fafe7f03e0558e2acfd702d222bba9a53e391cfd711ee7a2b65299579fcf28"
+          "sha256": "77199b03fc4d589a648033e359183a252d43dd6900e19ca609349bd972939e84"
         },
         "hf_artifacts": {
           "path": "hf_artifacts:docs/data/source_alignment_audit.json",
           "exists": true,
           "bytes": 4432,
-          "sha256": "c3fafe7f03e0558e2acfd702d222bba9a53e391cfd711ee7a2b65299579fcf28"
+          "sha256": "77199b03fc4d589a648033e359183a252d43dd6900e19ca609349bd972939e84"
         },
         "hf_model_data": {
           "path": "hf_model:data/source_alignment_audit.json",
           "exists": true,
           "bytes": 4432,
-          "sha256": "c3fafe7f03e0558e2acfd702d222bba9a53e391cfd711ee7a2b65299579fcf28"
+          "sha256": "77199b03fc4d589a648033e359183a252d43dd6900e19ca609349bd972939e84"
         },
         "hf_model_docs_data": {
           "path": "hf_model:docs/data/source_alignment_audit.json",
           "exists": true,
           "bytes": 4432,
-          "sha256": "c3fafe7f03e0558e2acfd702d222bba9a53e391cfd711ee7a2b65299579fcf28"
+          "sha256": "77199b03fc4d589a648033e359183a252d43dd6900e19ca609349bd972939e84"
         },
         "hf_model": {
           "path": "hf_model:metrics/source_alignment_audit.json",
           "exists": true,
           "bytes": 4432,
-          "sha256": "c3fafe7f03e0558e2acfd702d222bba9a53e391cfd711ee7a2b65299579fcf28"
+          "sha256": "77199b03fc4d589a648033e359183a252d43dd6900e19ca609349bd972939e84"
         }
       },
       "failures": []
@@ -2197,44 +2197,44 @@
         "path": "repo:docs/data/task_surface_integrity.json",
         "exists": true,
         "bytes": 46246,
-        "sha256": "fa0d28c31c066a6f53af8ced3c67d0f1fe9573d92b03bb9f5d0c3d5b239fc6cc"
+        "sha256": "525d6b4f05bb6ffa6f7524d573d5f51d6cd8fe55a4d1a4853282e25247f90a12"
       },
       "mirrors": {
         "hf_space": {
           "path": "hf_space:data/task_surface_integrity.json",
           "exists": true,
           "bytes": 46246,
-          "sha256": "fa0d28c31c066a6f53af8ced3c67d0f1fe9573d92b03bb9f5d0c3d5b239fc6cc"
+          "sha256": "525d6b4f05bb6ffa6f7524d573d5f51d6cd8fe55a4d1a4853282e25247f90a12"
         },
         "hf_artifacts_data": {
           "path": "hf_artifacts:data/task_surface_integrity.json",
           "exists": true,
           "bytes": 46246,
-          "sha256": "fa0d28c31c066a6f53af8ced3c67d0f1fe9573d92b03bb9f5d0c3d5b239fc6cc"
+          "sha256": "525d6b4f05bb6ffa6f7524d573d5f51d6cd8fe55a4d1a4853282e25247f90a12"
         },
         "hf_artifacts": {
           "path": "hf_artifacts:docs/data/task_surface_integrity.json",
           "exists": true,
           "bytes": 46246,
-          "sha256": "fa0d28c31c066a6f53af8ced3c67d0f1fe9573d92b03bb9f5d0c3d5b239fc6cc"
+          "sha256": "525d6b4f05bb6ffa6f7524d573d5f51d6cd8fe55a4d1a4853282e25247f90a12"
         },
         "hf_model_data": {
           "path": "hf_model:data/task_surface_integrity.json",
           "exists": true,
           "bytes": 46246,
-          "sha256": "fa0d28c31c066a6f53af8ced3c67d0f1fe9573d92b03bb9f5d0c3d5b239fc6cc"
+          "sha256": "525d6b4f05bb6ffa6f7524d573d5f51d6cd8fe55a4d1a4853282e25247f90a12"
         },
         "hf_model_docs_data": {
           "path": "hf_model:docs/data/task_surface_integrity.json",
           "exists": true,
           "bytes": 46246,
-          "sha256": "fa0d28c31c066a6f53af8ced3c67d0f1fe9573d92b03bb9f5d0c3d5b239fc6cc"
+          "sha256": "525d6b4f05bb6ffa6f7524d573d5f51d6cd8fe55a4d1a4853282e25247f90a12"
         },
         "hf_model": {
           "path": "hf_model:metrics/task_surface_integrity.json",
           "exists": true,
           "bytes": 46246,
-          "sha256": "fa0d28c31c066a6f53af8ced3c67d0f1fe9573d92b03bb9f5d0c3d5b239fc6cc"
+          "sha256": "525d6b4f05bb6ffa6f7524d573d5f51d6cd8fe55a4d1a4853282e25247f90a12"
         }
       },
       "failures": []
@@ -2589,44 +2589,44 @@
         "path": "repo:docs/data/website_integrity.json",
         "exists": true,
         "bytes": 24948,
-        "sha256": "1ff9661e301b73102f61efec5e6a2003d53569ab6cdf6f016fadbcaba608d7ca"
+        "sha256": "6657a1f46aea0d58ea5b1ce5b0dbeb12096ffc1bc5ec498c94b9bd5ba4ed1bf2"
       },
       "mirrors": {
         "hf_space": {
           "path": "hf_space:data/website_integrity.json",
           "exists": true,
           "bytes": 24948,
-          "sha256": "1ff9661e301b73102f61efec5e6a2003d53569ab6cdf6f016fadbcaba608d7ca"
+          "sha256": "6657a1f46aea0d58ea5b1ce5b0dbeb12096ffc1bc5ec498c94b9bd5ba4ed1bf2"
         },
         "hf_artifacts_data": {
           "path": "hf_artifacts:data/website_integrity.json",
           "exists": true,
           "bytes": 24948,
-          "sha256": "1ff9661e301b73102f61efec5e6a2003d53569ab6cdf6f016fadbcaba608d7ca"
+          "sha256": "6657a1f46aea0d58ea5b1ce5b0dbeb12096ffc1bc5ec498c94b9bd5ba4ed1bf2"
         },
         "hf_artifacts": {
           "path": "hf_artifacts:docs/data/website_integrity.json",
           "exists": true,
           "bytes": 24948,
-          "sha256": "1ff9661e301b73102f61efec5e6a2003d53569ab6cdf6f016fadbcaba608d7ca"
+          "sha256": "6657a1f46aea0d58ea5b1ce5b0dbeb12096ffc1bc5ec498c94b9bd5ba4ed1bf2"
         },
         "hf_model_data": {
           "path": "hf_model:data/website_integrity.json",
           "exists": true,
           "bytes": 24948,
-          "sha256": "1ff9661e301b73102f61efec5e6a2003d53569ab6cdf6f016fadbcaba608d7ca"
+          "sha256": "6657a1f46aea0d58ea5b1ce5b0dbeb12096ffc1bc5ec498c94b9bd5ba4ed1bf2"
         },
         "hf_model_docs_data": {
           "path": "hf_model:docs/data/website_integrity.json",
           "exists": true,
           "bytes": 24948,
-          "sha256": "1ff9661e301b73102f61efec5e6a2003d53569ab6cdf6f016fadbcaba608d7ca"
+          "sha256": "6657a1f46aea0d58ea5b1ce5b0dbeb12096ffc1bc5ec498c94b9bd5ba4ed1bf2"
         },
         "hf_model": {
           "path": "hf_model:metrics/website_integrity.json",
           "exists": true,
           "bytes": 24948,
-          "sha256": "1ff9661e301b73102f61efec5e6a2003d53569ab6cdf6f016fadbcaba608d7ca"
+          "sha256": "6657a1f46aea0d58ea5b1ce5b0dbeb12096ffc1bc5ec498c94b9bd5ba4ed1bf2"
         }
       },
       "failures": []
@@ -7429,39 +7429,39 @@
       "local": {
         "path": "repo:docs/index.html",
         "exists": true,
-        "bytes": 367548,
-        "sha256": "bafcdcb8aaa1647eea5fdbae52c318991b64a5a0cfa8727edc624e5c7b8bcf01"
+        "bytes": 367575,
+        "sha256": "2f2325aedf5f02d17e82a76310b7d66a676985bc8b343e5ac3658be2a7c802da"
       },
       "mirrors": {
         "hf_space": {
           "path": "hf_space:index.html",
           "exists": true,
-          "bytes": 367548,
-          "sha256": "bafcdcb8aaa1647eea5fdbae52c318991b64a5a0cfa8727edc624e5c7b8bcf01"
+          "bytes": 367575,
+          "sha256": "2f2325aedf5f02d17e82a76310b7d66a676985bc8b343e5ac3658be2a7c802da"
         },
         "hf_artifacts_root": {
           "path": "hf_artifacts:index.html",
           "exists": true,
-          "bytes": 367548,
-          "sha256": "bafcdcb8aaa1647eea5fdbae52c318991b64a5a0cfa8727edc624e5c7b8bcf01"
+          "bytes": 367575,
+          "sha256": "2f2325aedf5f02d17e82a76310b7d66a676985bc8b343e5ac3658be2a7c802da"
         },
         "hf_artifacts_docs": {
           "path": "hf_artifacts:docs/index.html",
           "exists": true,
-          "bytes": 367548,
-          "sha256": "bafcdcb8aaa1647eea5fdbae52c318991b64a5a0cfa8727edc624e5c7b8bcf01"
+          "bytes": 367575,
+          "sha256": "2f2325aedf5f02d17e82a76310b7d66a676985bc8b343e5ac3658be2a7c802da"
         },
         "hf_model": {
           "path": "hf_model:index.html",
           "exists": true,
-          "bytes": 367548,
-          "sha256": "bafcdcb8aaa1647eea5fdbae52c318991b64a5a0cfa8727edc624e5c7b8bcf01"
+          "bytes": 367575,
+          "sha256": "2f2325aedf5f02d17e82a76310b7d66a676985bc8b343e5ac3658be2a7c802da"
         },
         "hf_model_docs": {
           "path": "hf_model:docs/index.html",
           "exists": true,
-          "bytes": 367548,
-          "sha256": "bafcdcb8aaa1647eea5fdbae52c318991b64a5a0cfa8727edc624e5c7b8bcf01"
+          "bytes": 367575,
+          "sha256": "2f2325aedf5f02d17e82a76310b7d66a676985bc8b343e5ac3658be2a7c802da"
         }
       },
       "failures": []

data/public_surface_qa.json

@@ -1,7 +1,7 @@
 {
   "title": "Ropedia Xperience-10M Public Project Surface",
   "status": "pass",
-  "generated_at_utc": "2026-06-22T14:40:34+00:00",
+  "generated_at_utc": "2026-06-22T15:09:11+00:00",
   "scope": "Repo README, GitHub Pages HTML, Hugging Face Space card, artifact dataset card, and model card.",
   "checks": [
     {
@@ -18,7 +18,7 @@
         "website_integrity": {
           "exists": true,
           "status": "pass",
-          "generated_at_utc": "2026-06-22T14:39:11+00:00"
+          "generated_at_utc": "2026-06-22T15:07:54+00:00"
         },
         "rendered_site_check": {
           "exists": true,
@@ -28,12 +28,12 @@
         "task_surface_integrity": {
           "exists": true,
           "status": "pass",
-          "generated_at_utc": "2026-06-22T14:31:36+00:00"
+          "generated_at_utc": "2026-06-22T15:07:48+00:00"
         },
         "source_alignment": {
           "exists": true,
           "status": "pass",
-          "generated_at_utc": "2026-06-22T14:31:36+00:00"
+          "generated_at_utc": "2026-06-22T15:07:48+00:00"
         },
         "scale_up_status": {
           "exists": true,
@@ -43,12 +43,12 @@
         "publication_package": {
           "exists": true,
           "status": "pass",
-          "generated_at_utc": "2026-06-22T14:37:20+00:00"
+          "generated_at_utc": "2026-06-22T15:08:27+00:00"
         },
         "mirror_parity": {
           "exists": true,
           "status": "pass",
-          "generated_at_utc": "2026-06-22T14:36:09+00:00"
+          "generated_at_utc": "2026-06-22T14:44:16+00:00"
         }
       },
       "failures": {}
@@ -96,9 +96,9 @@
       "reason": "Public copy should consistently present the project as Ropedia Xperience-10M, with the Qwen3-Omni scale-up status.",
       "marker_counts": {
         "Ropedia Xperience-10M Task Suite": 22,
-        "Xperience-10M": 171,
+        "Xperience-10M": 170,
         "20-task": 117,
-        "Qwen3-Omni": 236,
+        "Qwen3-Omni": 233,
         "128-episode pilot": 1
       }
     },

data/publication_audit.json

@@ -1,6 +1,6 @@
 {
   "status": "pass",
-  "generated_at_utc": "2026-06-22T14:41:37+00:00",
+  "generated_at_utc": "2026-06-22T15:08:27+00:00",
   "checks": [
     {
       "name": "required_publication_assets_present",

data/quality_gates.json

@@ -1,7 +1,7 @@
 {
   "title": "Ropedia Xperience-10M Release Checks",
   "status": "pass",
-  "generated_at_utc": "2026-06-22T14:42:50+00:00",
+  "generated_at_utc": "2026-06-22T15:09:12+00:00",
   "rule": "A release is current when the automated reports pass and the live GitHub/Hugging Face mirrors are verified after publishing.",
   "automated_gates": [
     {

data/source_alignment_audit.json

@@ -1,7 +1,7 @@
 {
   "title": "Ropedia Xperience-10M Source Alignment Note",
   "status": "pass",
-  "generated_at_utc": "2026-06-22T14:40:33+00:00",
+  "generated_at_utc": "2026-06-22T15:07:48+00:00",
   "alignment_json": "docs/data/xperience10m_dataset_card_alignment.json",
   "alignment_summary": {
     "full_dataset_repo": "ropedia-ai/xperience-10m",

data/task_surface_integrity.json

@@ -1,6 +1,6 @@
 {
   "status": "pass",
-  "generated_at_utc": "2026-06-22T14:40:33+00:00",
+  "generated_at_utc": "2026-06-22T15:07:48+00:00",
   "summary": {
     "original_walkthrough_task_count": 12,
     "expected_original_walkthrough_task_count": 12,

data/website_integrity.json

@@ -1,6 +1,6 @@
 {
   "status": "pass",
-  "generated_at_utc": "2026-06-22T14:39:11+00:00",
+  "generated_at_utc": "2026-06-22T15:07:54+00:00",
   "docs_root": "docs",
   "site_base": "/ropedia-xperience-10m-task-suite/",
   "summary": {
@@ -80,8 +80,8 @@
       "name": "project_overview_precedes_progress_ledger",
       "status": "pass",
       "reason": "The project overview should appear before the deeper progress ledger.",
-      "overview_index": 151226,
-      "evidence_index": 199040
+      "overview_index": 151515,
+      "evidence_index": 199329
     },
     {
       "name": "project_status_links_json",
@@ -159,9 +159,9 @@
       "name": "evaluation_protocol_between_overview_and_progress",
       "status": "pass",
       "reason": "The evaluation protocol should appear before the deeper evidence ledger.",
-      "overview_index": 151226,
-      "protocol_index": 195245,
-      "evidence_index": 199040
+      "overview_index": 151515,
+      "protocol_index": 195534,
+      "evidence_index": 199329
     },
     {
       "name": "evaluation_protocol_links_json",
@@ -290,7 +290,7 @@
     },
     {
       "path": "index.html",
-      "id_count": 101,
+      "id_count": 102,
       "reference_count": 252,
       "image_count": 54
     },
@@ -355,7 +355,7 @@
     },
     {
       "path": "data/glossary.json",
-      "bytes": 19260,
+      "bytes": 49025,
       "top_level_type": "dict"
     },
     {

docs/data/artifact_index.json

@@ -1,6 +1,6 @@
 {
   "title": "Ropedia Xperience-10M Task Suite Artifact Index",
-  "generated_at_utc": "2026-06-22T14:40:39+00:00",
+  "generated_at_utc": "2026-06-22T15:09:12+00:00",
   "status": "pass",
   "artifact_count": 228,
   "missing": [],
@@ -92,8 +92,8 @@
       "surface": "repo_hf",
       "shows": "Defines terminology that can be confused across data scope, task metrics, model branches, and public mirrors.",
       "exists": true,
-      "bytes": 11122,
-      "sha256": "fe781a4eb5dd56454b5e0cb3383c88a2106c7bbf269888a0a7613b1618c8d196"
+      "bytes": 24120,
+      "sha256": "32812ee8aa021e7c9f7937ba1d436651290d6f3f61e290b17aee5ffc06746465"
     },
     {
       "id": "glossary_json",
@@ -103,8 +103,8 @@
       "surface": "website_hf",
       "shows": "Machine-readable terminology layer for the website, artifact dataset, model mirror, and public QA checks.",
       "exists": true,
-      "bytes": 19260,
-      "sha256": "525de375608793cd34ab386819eac5291177b53ca5839d54a9046707206e844a"
+      "bytes": 49025,
+      "sha256": "2939ed7aefd6c1f330785fae28492f8af49cc6d9ef10620e0dd79e9e3ae8c2cb"
     },
     {
       "id": "research_roadmap",
@@ -632,7 +632,7 @@
       "shows": "Machine-readable source-alignment pass/fail check for repo, website, and HF surfaces.",
       "exists": true,
       "bytes": 4432,
-      "sha256": "77199b03fc4d589a648033e359183a252d43dd6900e19ca609349bd972939e84"
+      "sha256": "c00a89e8694e08a6bb844924da00ba78bcf6c5da96690d548d670bf0baa4fa9f"
     },
     {
       "id": "source_alignment_validator",
@@ -1182,7 +1182,7 @@
       "shows": "Machine-readable release-check summary for validators, mirrors, and public project surfaces.",
       "exists": true,
       "bytes": 8640,
-      "sha256": "745b9c41da785ab84af8aba58babe3d8decd3c0bb07e3187b64f82ae42d91ec5"
+      "sha256": "e3f97614b47251d02f560db20a4bc918088b9aadc45c1d1397c2756b4d4867bf"
     },
     {
       "id": "public_surface_qa",

docs/data/glossary.json

@@ -1,7 +1,7 @@
 {
   "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.",
+  "purpose": "Define reader-facing project terms and adjacent technical field terms for embodied AI, egocentric multimodal data, spatial intelligence, world models, VLA/policy learning, evaluation, and public artifact reading.",
   "categories": [
     {
       "id": "dataset_scope",
@@ -11,13 +11,38 @@
     {
       "id": "files_features",
       "label": "Files and features",
-      "description": "How raw sample files, windows, feature manifests, and public-safe derivatives relate."
+      "description": "Raw sample files, windows, feature manifests, and public-safe derivatives."
+    },
+    {
+      "id": "multimodal_sensing",
+      "label": "Multimodal sensing",
+      "description": "Video, audio, depth, IMU, motion capture, calibration, and synchronization terms."
+    },
+    {
+      "id": "spatial_geometry",
+      "label": "Spatial geometry",
+      "description": "Camera pose, SLAM, coordinate frames, point clouds, 3D reconstruction, and spatial grounding."
+    },
+    {
+      "id": "temporal_world_models",
+      "label": "Temporal and world models",
+      "description": "Future prediction, rollouts, forward dynamics, long-horizon forecasting, and temporal leakage."
+    },
+    {
+      "id": "robotics_vla",
+      "label": "Robotics and VLA",
+      "description": "Vision-language-action, policies, action chunks, imitation learning, contact, and dexterity."
     },
     {
       "id": "tasks_metrics",
       "label": "Tasks and metrics",
       "description": "Task contracts, scored records, direct scores, compact proxies, and audits."
     },
+    {
+      "id": "training_eval",
+      "label": "Training and evaluation",
+      "description": "Splits, held-out evaluation, metric types, prompt/schema checks, adapters, and distributed training."
+    },
     {
       "id": "models_runs",
       "label": "Models and runs",
@@ -31,12 +56,26 @@
   ],
   "entries": [
     {
-      "term": "Xperience-10M",
+      "term": "Evidence line",
       "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"]
+      "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 sample episode",
@@ -44,7 +83,10 @@
       "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"]
+      "primary_files": [
+        "docs/data/raw_sample_files.json",
+        "docs/single_episode_explorer.html"
+      ]
     },
     {
       "term": "Selected 128 episodes",
@@ -52,31 +94,42 @@
       "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"]
+      "primary_files": [
+        "XPERIENCE10M_128_EPISODE_FEATURE_INDEX.md",
+        "docs/data/xperience10m_128_episode_feature_index.json"
+      ]
     },
     {
-      "term": "Evidence line",
+      "term": "Xperience-10M",
       "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"]
+      "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": "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": "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": "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": "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": "Episode",
@@ -84,15 +137,10 @@
       "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"]
+      "primary_files": [
+        "docs/data/raw_sample_files.json",
+        "docs/data/xperience10m_128_episode_feature_index.json"
+      ]
     },
     {
       "term": "Feature manifest",
@@ -100,15 +148,41 @@
       "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"]
+      "primary_files": [
+        "results/episode_task_suite/feature_manifest.json"
+      ]
     },
     {
-      "term": "annotation.hdf5",
+      "term": "Interaction text",
       "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"]
+      "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": "Raw sample file map",
+      "category": "files_features",
+      "plain_meaning": "A human-readable inventory of the sample episode files.",
+      "project_usage": "Explains videos, annotations, calibration, motion, and derived previews.",
+      "do_not_confuse_with": "A training manifest.",
+      "primary_files": [
+        "docs/data/raw_sample_files.json"
+      ]
     },
     {
       "term": "visualization.rrd",
@@ -116,47 +190,501 @@
       "plain_meaning": "Rerun viewer recording for visual inspection.",
       "project_usage": "Can be downloaded from the official sample dataset and opened in Rerun 0.29.0 to inspect the sample episode. It is not used for published training or metric rows.",
       "do_not_confuse_with": "MP4 video streams or model inputs.",
-      "primary_files": ["docs/data/raw_sample_files.json", "REPRODUCIBILITY.md"]
+      "primary_files": [
+        "docs/data/raw_sample_files.json",
+        "REPRODUCIBILITY.md"
+      ]
     },
     {
-      "term": "Interaction text",
+      "term": "Window stride",
       "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"]
+      "plain_meaning": "The frame step between neighboring windows.",
+      "project_usage": "Creates overlapping examples while preserving chronological order and leakage controls.",
+      "do_not_confuse_with": "Video frame rate.",
+      "primary_files": [
+        "EVALUATION_PROTOCOL.md"
+      ]
     },
     {
-      "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": "Audio waveform",
+      "category": "multimodal_sensing",
+      "plain_meaning": "A time-series pressure signal from sound.",
+      "project_usage": "The audio ablation measures whether embedded audio helps selected task contracts.",
+      "do_not_confuse_with": "Language captions or text labels.",
+      "primary_files": [
+        "docs/data/audio_ablation_summary.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": "Calibration",
+      "category": "multimodal_sensing",
+      "plain_meaning": "Parameters that relate sensors to each other and to physical space.",
+      "project_usage": "Needed to interpret camera streams, depth, pose, and synchronized multimodal features together.",
+      "do_not_confuse_with": "A model training hyperparameter.",
+      "primary_files": [
+        "docs/data/raw_sample_files.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": "Camera extrinsics",
+      "category": "multimodal_sensing",
+      "plain_meaning": "A camera position and orientation relative to another coordinate frame.",
+      "project_usage": "Connects different camera streams and world coordinates.",
+      "do_not_confuse_with": "Camera intrinsics.",
+      "primary_files": [
+        "docs/data/raw_sample_files.json"
+      ]
     },
     {
-      "term": "Task-method record",
+      "term": "Camera intrinsics",
+      "category": "multimodal_sensing",
+      "plain_meaning": "Internal camera parameters such as focal length and distortion.",
+      "project_usage": "Explain how image pixels project to rays for geometry tasks.",
+      "do_not_confuse_with": "Camera extrinsics.",
+      "primary_files": [
+        "docs/data/raw_sample_files.json"
+      ]
+    },
+    {
+      "term": "Depth map",
+      "category": "multimodal_sensing",
+      "plain_meaning": "A per-pixel estimate of distance from the camera.",
+      "project_usage": "Depth-derived signals support spatial and geometry-oriented tasks.",
+      "do_not_confuse_with": "RGB brightness or semantic segmentation.",
+      "primary_files": [
+        "docs/data/modality_atlas.json"
+      ]
+    },
+    {
+      "term": "Egocentric video",
+      "category": "multimodal_sensing",
+      "plain_meaning": "Video captured from a first-person or body-mounted viewpoint.",
+      "project_usage": "The sample streams are egocentric views of human interaction and are the visual basis for many tasks.",
+      "do_not_confuse_with": "Third-person robot-camera footage.",
+      "primary_files": [
+        "docs/data/raw_sample_files.json"
+      ]
+    },
+    {
+      "term": "Fisheye camera",
+      "category": "multimodal_sensing",
+      "plain_meaning": "A wide-angle camera with strong lens distortion.",
+      "project_usage": "Multiple fisheye MP4 streams give broad room coverage but need calibration-aware interpretation.",
+      "do_not_confuse_with": "A rectilinear pinhole camera image.",
+      "primary_files": [
+        "docs/data/raw_sample_files.json"
+      ]
+    },
+    {
+      "term": "IMU",
+      "category": "multimodal_sensing",
+      "plain_meaning": "An inertial measurement unit with accelerometer and gyroscope signals.",
+      "project_usage": "Supports motion, temporal, and sensor-bridging tasks.",
+      "do_not_confuse_with": "Motion capture skeleton data.",
+      "primary_files": [
+        "docs/data/modality_atlas.json"
+      ]
+    },
+    {
+      "term": "Metric depth",
+      "category": "multimodal_sensing",
+      "plain_meaning": "Depth expressed in physical units rather than arbitrary relative scale.",
+      "project_usage": "Useful for distance-sensitive spatial reasoning and reconstruction targets.",
+      "do_not_confuse_with": "Relative monocular depth.",
+      "primary_files": [
+        "docs/data/modality_atlas.json"
+      ]
+    },
+    {
+      "term": "Motion capture",
+      "category": "multimodal_sensing",
+      "plain_meaning": "A system that records body or hand motion over time.",
+      "project_usage": "Provides hand/body motion evidence when exposed through public-safe derived features.",
+      "do_not_confuse_with": "Video-only pose estimation.",
+      "primary_files": [
+        "docs/data/modality_atlas.json"
+      ]
+    },
+    {
+      "term": "RGB frame",
+      "category": "multimodal_sensing",
+      "plain_meaning": "A color image frame from a video stream.",
+      "project_usage": "Used for visual statistics, previews, and many model inputs.",
+      "do_not_confuse_with": "Depth values or point-cloud coordinates.",
+      "primary_files": [
+        "results/episode_task_suite/feature_manifest.json"
+      ]
+    },
+    {
+      "term": "Sensor alignment",
+      "category": "multimodal_sensing",
+      "plain_meaning": "Putting different sensor streams into a shared temporal or spatial reference.",
+      "project_usage": "Used to make video, audio, pose, depth, IMU, and mocap usable in the same task input.",
+      "do_not_confuse_with": "Model ensembling.",
+      "primary_files": [
+        "docs/data/modality_atlas.json"
+      ]
+    },
+    {
+      "term": "Stereo camera",
+      "category": "multimodal_sensing",
+      "plain_meaning": "A paired-camera setup that supports depth or geometry estimation.",
+      "project_usage": "The sample browser exposes stereo streams as part of the visual modality set.",
+      "do_not_confuse_with": "Single-view RGB video.",
+      "primary_files": [
+        "docs/data/raw_sample_files.json"
+      ]
+    },
+    {
+      "term": "Timestamp synchronization",
+      "category": "multimodal_sensing",
+      "plain_meaning": "Aligning sensor samples by time.",
+      "project_usage": "The task suite assumes aligned windows across modalities so labels and features refer to the same moment.",
+      "do_not_confuse_with": "Randomly joining files with similar names.",
+      "primary_files": [
+        "EVALUATION_PROTOCOL.md"
+      ]
+    },
+    {
+      "term": "3D reconstruction",
+      "category": "spatial_geometry",
+      "plain_meaning": "Recovering 3D scene structure from sensor data.",
+      "project_usage": "One core spatial-intelligence direction for Xperience-style data.",
+      "do_not_confuse_with": "Next-action classification.",
+      "primary_files": [
+        "docs/data/three_foundation_pipelines.json"
+      ]
+    },
+    {
+      "term": "Affordance",
+      "category": "spatial_geometry",
+      "plain_meaning": "An action possibility offered by an object or scene.",
+      "project_usage": "Relevant when moving from observed human interaction to robot-action or VLA tasks.",
+      "do_not_confuse_with": "A detected object category alone.",
+      "primary_files": [
+        "docs/data/three_foundation_pipelines.json"
+      ]
+    },
+    {
+      "term": "Camera pose",
+      "category": "spatial_geometry",
+      "plain_meaning": "The camera position and orientation at a time step.",
+      "project_usage": "Supports spatial-intelligence tasks, view synchronization, and geometry diagnostics.",
+      "do_not_confuse_with": "The human body pose.",
+      "primary_files": [
+        "docs/data/modality_atlas.json"
+      ]
+    },
+    {
+      "term": "Coordinate frame",
+      "category": "spatial_geometry",
+      "plain_meaning": "A reference system for positions and orientations.",
+      "project_usage": "Needed when comparing camera, body, object, and world measurements.",
+      "do_not_confuse_with": "A video frame.",
+      "primary_files": [
+        "EVALUATION_PROTOCOL.md"
+      ]
+    },
+    {
+      "term": "Object-centric representation",
+      "category": "spatial_geometry",
+      "plain_meaning": "A representation organized around objects and their relations.",
+      "project_usage": "Useful for object relevance, object-set forecast, and action-object relation tasks.",
+      "do_not_confuse_with": "A flat feature vector without object identity.",
+      "primary_files": [
+        "docs/data/task_suite_20.json"
+      ]
+    },
+    {
+      "term": "Odometry",
+      "category": "spatial_geometry",
+      "plain_meaning": "Motion estimated from sensor changes over time.",
+      "project_usage": "A relevant spatial term for ego-motion and camera-pose reasoning.",
+      "do_not_confuse_with": "Ground-truth motion capture.",
+      "primary_files": [
+        "docs/data/modality_atlas.json"
+      ]
+    },
+    {
+      "term": "Point cloud",
+      "category": "spatial_geometry",
+      "plain_meaning": "A set of 3D points representing scene structure.",
+      "project_usage": "A likely target or intermediate representation for spatial-intelligence extensions.",
+      "do_not_confuse_with": "A 2D image grid.",
+      "primary_files": [
+        "docs/data/three_foundation_pipelines.json"
+      ]
+    },
+    {
+      "term": "SLAM",
+      "category": "spatial_geometry",
+      "plain_meaning": "Simultaneous localization and mapping.",
+      "project_usage": "A field term for estimating camera motion and scene structure from sensor observations.",
+      "do_not_confuse_with": "A task label or action class.",
+      "primary_files": [
+        "docs/data/modality_atlas.json"
+      ]
+    },
+    {
+      "term": "Spatial grounding",
+      "category": "spatial_geometry",
+      "plain_meaning": "Linking language or labels to locations, objects, or geometry.",
+      "project_usage": "Connects language grounding tasks with 3D/spatial reasoning.",
+      "do_not_confuse_with": "General text classification.",
+      "primary_files": [
+        "docs/data/research_directions.json"
+      ]
+    },
+    {
+      "term": "Trajectory",
+      "category": "spatial_geometry",
+      "plain_meaning": "A sequence of positions over time.",
+      "project_usage": "Used for hand motion, camera motion, and future-path tasks.",
+      "do_not_confuse_with": "A single coordinate or label.",
+      "primary_files": [
+        "TASK_SUITE_20.md"
+      ]
+    },
+    {
+      "term": "Action forecasting",
+      "category": "temporal_world_models",
+      "plain_meaning": "Predicting a future action before it happens.",
+      "project_usage": "Covered by next-action and long-horizon task contracts.",
+      "do_not_confuse_with": "Recognizing the current action only.",
+      "primary_files": [
+        "docs/data/task_suite_20.json"
+      ]
+    },
+    {
+      "term": "Autoregressive prediction",
+      "category": "temporal_world_models",
+      "plain_meaning": "Generating each future token, state, or frame conditioned on prior outputs.",
+      "project_usage": "Relevant for model branches that produce structured JSON or temporal predictions.",
+      "do_not_confuse_with": "A one-shot classifier.",
+      "primary_files": [
+        "docs/data/foundation_model_plan.json"
+      ]
+    },
+    {
+      "term": "Forward dynamics",
+      "category": "temporal_world_models",
+      "plain_meaning": "Predicting the next state from the current state and action/context.",
+      "project_usage": "The Cosmos3-Super LoRA branch uses a forward-dynamics-style diagnostic contract.",
+      "do_not_confuse_with": "Reverse inference from result back to cause.",
+      "primary_files": [
+        "docs/data/omni_model_comparison.json"
+      ]
+    },
+    {
+      "term": "Latent state",
+      "category": "temporal_world_models",
+      "plain_meaning": "A hidden representation that summarizes observed context.",
+      "project_usage": "Useful for future foundation-model and world-model training plans.",
+      "do_not_confuse_with": "A visible annotation column.",
+      "primary_files": [
+        "docs/data/foundation_model_plan.json"
+      ]
+    },
+    {
+      "term": "Long-horizon prediction",
+      "category": "temporal_world_models",
+      "plain_meaning": "Predicting outcomes several seconds or steps ahead.",
+      "project_usage": "Tasks 13 and 14 test longer temporal context beyond immediate recognition.",
+      "do_not_confuse_with": "Single-frame classification.",
+      "primary_files": [
+        "docs/data/task_suite_20.json"
+      ]
+    },
+    {
+      "term": "Next-frame prediction",
+      "category": "temporal_world_models",
+      "plain_meaning": "Predicting future visual frames from past frames.",
+      "project_usage": "A field-level world-model objective related to the human-video world-model direction.",
+      "do_not_confuse_with": "Next-action prediction.",
+      "primary_files": [
+        "docs/data/three_foundation_pipelines.json"
+      ]
+    },
+    {
+      "term": "Object persistence",
+      "category": "temporal_world_models",
+      "plain_meaning": "Tracking that an object remains present over time even when view or interaction changes.",
+      "project_usage": "Relevant for object-set forecast and long-video reasoning.",
+      "do_not_confuse_with": "A single-frame object detection.",
+      "primary_files": [
+        "docs/data/task_suite_20.json"
+      ]
+    },
+    {
+      "term": "Rollout",
+      "category": "temporal_world_models",
+      "plain_meaning": "Repeatedly predicting future steps from a model state.",
+      "project_usage": "Important for judging world models beyond one-step prediction.",
+      "do_not_confuse_with": "A held-out static test row.",
+      "primary_files": [
+        "docs/data/three_foundation_pipelines.json"
+      ]
+    },
+    {
+      "term": "Subtask forecasting",
+      "category": "temporal_world_models",
+      "plain_meaning": "Predicting the next higher-level step in an activity.",
+      "project_usage": "Used in the future-task probe line for Qwen3-Omni.",
+      "do_not_confuse_with": "Frame-level action classification.",
+      "primary_files": [
+        "docs/data/task_method_20_result_matrix.json"
+      ]
+    },
+    {
+      "term": "Teacher forcing",
+      "category": "temporal_world_models",
+      "plain_meaning": "Training a sequence model using ground-truth previous outputs.",
+      "project_usage": "A likely training option for future sequence/world-model baselines.",
+      "do_not_confuse_with": "Free-running rollout evaluation.",
+      "primary_files": [
+        "docs/data/foundation_model_plan.json"
+      ]
+    },
+    {
+      "term": "Temporal leakage",
+      "category": "temporal_world_models",
+      "plain_meaning": "Using future information that would not be available at prediction time.",
+      "project_usage": "Avoided by chronological splits and target-side feature controls.",
+      "do_not_confuse_with": "A low model score.",
+      "primary_files": [
+        "EVALUATION_PROTOCOL.md"
+      ]
+    },
+    {
+      "term": "Transition timing",
+      "category": "temporal_world_models",
+      "plain_meaning": "Estimating when the next state or action transition happens.",
+      "project_usage": "Task 20 turns temporal change into a regression target.",
+      "do_not_confuse_with": "Classifying the transition type only.",
+      "primary_files": [
+        "docs/data/task_suite_20.json"
+      ]
+    },
+    {
+      "term": "Action chunk",
+      "category": "robotics_vla",
+      "plain_meaning": "A short sequence of low-level actions predicted together.",
+      "project_usage": "The VLA figure and plan use action chunks as the policy-output concept.",
+      "do_not_confuse_with": "A natural-language action label.",
+      "primary_files": [
+        "docs/data/three_foundation_pipelines.json"
+      ]
+    },
+    {
+      "term": "Behavior cloning",
+      "category": "robotics_vla",
+      "plain_meaning": "A supervised imitation-learning method for predicting demonstrated actions.",
+      "project_usage": "A plausible baseline once action targets are converted.",
+      "do_not_confuse_with": "Generative video modeling.",
+      "primary_files": [
+        "docs/data/foundation_model_plan.json"
+      ]
+    },
+    {
+      "term": "Contact event",
+      "category": "robotics_vla",
+      "plain_meaning": "A moment when a hand, body, or tool touches an object or surface.",
+      "project_usage": "Used in contact-related tasks and action-quality interpretation.",
+      "do_not_confuse_with": "Visual co-occurrence without touch.",
+      "primary_files": [
+        "docs/data/task_suite_20.json"
+      ]
+    },
+    {
+      "term": "Dexterity",
+      "category": "robotics_vla",
+      "plain_meaning": "Fine-grained physical manipulation ability.",
+      "project_usage": "Relevant to hand-object interaction, contact, and VLA/policy directions.",
+      "do_not_confuse_with": "High text-generation accuracy.",
+      "primary_files": [
+        "docs/data/research_directions.json"
+      ]
+    },
+    {
+      "term": "End effector",
+      "category": "robotics_vla",
+      "plain_meaning": "The robot part that acts on the world, such as a gripper or hand.",
+      "project_usage": "A key target frame for future manipulation-policy conversion.",
+      "do_not_confuse_with": "A camera or global scene coordinate.",
+      "primary_files": [
+        "docs/data/three_foundation_pipelines.json"
+      ]
+    },
+    {
+      "term": "Hand-object interaction",
+      "category": "robotics_vla",
+      "plain_meaning": "A physical interaction between hands and objects.",
+      "project_usage": "A central signal family behind action, contact, object relevance, and interaction-text tasks.",
+      "do_not_confuse_with": "Object detection without action.",
+      "primary_files": [
+        "docs/data/task_suite_20.json"
+      ]
+    },
+    {
+      "term": "Imitation learning",
+      "category": "robotics_vla",
+      "plain_meaning": "Training a policy to imitate demonstrated behavior.",
+      "project_usage": "Relevant when converting human video/motion into action supervision.",
+      "do_not_confuse_with": "Reinforcement learning from online robot trials.",
+      "primary_files": [
+        "docs/data/foundation_model_plan.json"
+      ]
+    },
+    {
+      "term": "Language grounding",
+      "category": "robotics_vla",
+      "plain_meaning": "Connecting text to observed objects, actions, or spatial context.",
+      "project_usage": "Task 8 and VLA directions use language as grounded supervision rather than standalone text.",
+      "do_not_confuse_with": "Caption fluency alone.",
+      "primary_files": [
+        "docs/data/task_suite_20.json"
+      ]
+    },
+    {
+      "term": "Policy",
+      "category": "robotics_vla",
+      "plain_meaning": "A mapping from observations to actions.",
+      "project_usage": "A future target for robot-compatible Xperience-derived action data.",
+      "do_not_confuse_with": "A benchmark metric.",
+      "primary_files": [
+        "docs/data/foundation_model_plan.json"
+      ]
+    },
+    {
+      "term": "Robot-compatible action target",
+      "category": "robotics_vla",
+      "plain_meaning": "An action representation a robot policy can execute or imitate.",
+      "project_usage": "Needed before OpenVLA/openpi/GR00T-style policy training is meaningful here.",
+      "do_not_confuse_with": "Human-only caption text.",
+      "primary_files": [
+        "docs/data/foundation_model_plan.json"
+      ]
+    },
+    {
+      "term": "Vision-language-action model",
+      "category": "robotics_vla",
+      "plain_meaning": "A model that maps visual context and language into actions.",
+      "project_usage": "The VLA direction is a future path after action targets are converted into robot-compatible chunks.",
+      "do_not_confuse_with": "A vision-language model that only answers text.",
+      "primary_files": [
+        "docs/data/three_foundation_pipelines.json"
+      ]
+    },
+    {
+      "term": "Compact-proxy score",
       "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"]
+      "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": "Direct score",
@@ -164,23 +692,32 @@
       "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"]
+      "primary_files": [
+        "TASK_METHOD_20_GAP_AUDIT.md",
+        "docs/data/task_method_20_gap_audit.json"
+      ]
     },
     {
-      "term": "Compact-proxy score",
+      "term": "Gap audit",
       "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"]
+      "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": "Raw metric value",
+      "term": "Leakage control",
       "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"]
+      "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": "Normalized radar value",
@@ -188,63 +725,214 @@
       "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"]
+      "primary_files": [
+        "docs/data/unified_task_model_radar.json",
+        "docs/assets/charts/unified_task_model_radar.svg"
+      ]
     },
     {
-      "term": "Gap audit",
+      "term": "Raw metric value",
       "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"]
+      "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": "Leakage control",
+      "term": "Task contract",
       "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"]
+      "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": "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": "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": "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": "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": "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": "Adapter checkpoint",
+      "category": "training_eval",
+      "plain_meaning": "Saved adapter weights from a fine-tuning run.",
+      "project_usage": "The public model branches publish adapters when validated and public-safe.",
+      "do_not_confuse_with": "Full base-model checkpoint.",
+      "primary_files": [
+        "docs/data/omni_model_comparison.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": "Balanced accuracy",
+      "category": "training_eval",
+      "plain_meaning": "Accuracy averaged across classes to reduce majority-class dominance.",
+      "project_usage": "Useful for imbalanced task labels.",
+      "do_not_confuse_with": "Overall accuracy.",
+      "primary_files": [
+        "docs/data/task_method_20_result_matrix.json"
+      ]
     },
     {
-      "term": "Qwen v1-v6",
+      "term": "Chronological split",
+      "category": "training_eval",
+      "plain_meaning": "A split ordered by time.",
+      "project_usage": "Used for the single-episode baselines to reduce future-window leakage.",
+      "do_not_confuse_with": "A random row split.",
+      "primary_files": [
+        "EVALUATION_PROTOCOL.md"
+      ]
+    },
+    {
+      "term": "Confusion matrix",
+      "category": "training_eval",
+      "plain_meaning": "A table of predicted classes versus true classes.",
+      "project_usage": "Helps inspect which task labels a method confuses.",
+      "do_not_confuse_with": "A scalar leaderboard score.",
+      "primary_files": [
+        "results/episode_task_suite/neural_mlp"
+      ]
+    },
+    {
+      "term": "FSDP",
+      "category": "training_eval",
+      "plain_meaning": "Fully Sharded Data Parallel, a distributed training strategy.",
+      "project_usage": "Appears in full-parameter feasibility and multi-GPU training notes.",
+      "do_not_confuse_with": "A model architecture.",
+      "primary_files": [
+        "docs/data/qwen3_full_parameter_gates.json"
+      ]
+    },
+    {
+      "term": "Held-out evaluation",
+      "category": "training_eval",
+      "plain_meaning": "Testing on examples not used for training.",
+      "project_usage": "Required before promoting Qwen/Cosmos results to public evidence.",
+      "do_not_confuse_with": "Training-set loss.",
+      "primary_files": [
+        "docs/data/omni_model_comparison.json"
+      ]
+    },
+    {
+      "term": "JSON validity",
+      "category": "training_eval",
+      "plain_meaning": "Whether model output parses as the required JSON schema.",
+      "project_usage": "A key diagnostic for Qwen3-Omni structured-output runs.",
+      "do_not_confuse_with": "Task correctness after parsing.",
+      "primary_files": [
+        "docs/data/omni_model_comparison.json"
+      ]
+    },
+    {
+      "term": "Macro F1",
+      "category": "training_eval",
+      "plain_meaning": "The average F1 score across classes, usually treating classes equally.",
+      "project_usage": "Used when class imbalance matters in classification tasks.",
+      "do_not_confuse_with": "Accuracy dominated by frequent classes.",
+      "primary_files": [
+        "docs/data/task_method_20_result_matrix.json"
+      ]
+    },
+    {
+      "term": "Mean absolute error",
+      "category": "training_eval",
+      "plain_meaning": "The average absolute difference between predicted and true numeric values.",
+      "project_usage": "Used for regression-style task rows such as timing or trajectory targets.",
+      "do_not_confuse_with": "A classification F1 score.",
+      "primary_files": [
+        "docs/data/task_method_20_result_matrix.json"
+      ]
+    },
+    {
+      "term": "Overfit check",
+      "category": "training_eval",
+      "plain_meaning": "A small training test that verifies a model can learn a tiny subset.",
+      "project_usage": "Useful for catching data/model wiring bugs before full training.",
+      "do_not_confuse_with": "Evidence of generalization.",
+      "primary_files": [
+        "docs/data/foundation_model_plan.json"
+      ]
+    },
+    {
+      "term": "Parameter-efficient fine-tuning",
+      "category": "training_eval",
+      "plain_meaning": "Updating a small number of added or selected parameters.",
+      "project_usage": "LoRA is the current parameter-efficient path for Qwen/Cosmos branches.",
+      "do_not_confuse_with": "Full-parameter fine-tuning.",
+      "primary_files": [
+        "docs/data/foundation_model_plan.json"
+      ]
+    },
+    {
+      "term": "Schema compliance",
+      "category": "training_eval",
+      "plain_meaning": "Whether an output follows the expected field names and value types.",
+      "project_usage": "Needed for structured task probes and public package validation.",
+      "do_not_confuse_with": "High semantic accuracy.",
+      "primary_files": [
+        "docs/data/omni_model_comparison.json"
+      ]
+    },
+    {
+      "term": "Smoke run",
+      "category": "training_eval",
+      "plain_meaning": "A short run that checks whether a pipeline can start and execute key steps.",
+      "project_usage": "Used for feasibility gates before expensive full runs.",
+      "do_not_confuse_with": "A complete benchmark result.",
+      "primary_files": [
+        "docs/data/qwen3_full_parameter_gates.json"
+      ]
+    },
+    {
+      "term": "Top-k accuracy",
+      "category": "training_eval",
+      "plain_meaning": "A score that counts a prediction correct if the target is among the k highest-ranked outputs.",
+      "project_usage": "Useful for large-label or retrieval-style tasks.",
+      "do_not_confuse_with": "Top-1 exact accuracy.",
+      "primary_files": [
+        "docs/data/task_method_20_result_matrix.json"
+      ]
+    },
+    {
+      "term": "Train/validation/test split",
+      "category": "training_eval",
+      "plain_meaning": "A partition that separates model fitting, tuning, and final evaluation examples.",
+      "project_usage": "The selected-128 setup uses a held-out split discipline for model branches.",
+      "do_not_confuse_with": "A random shuffle without temporal or episode boundaries.",
+      "primary_files": [
+        "EVALUATION_PROTOCOL.md"
+      ]
+    },
+    {
+      "term": "Cosmos3-Nano",
       "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"]
+      "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": "Cosmos3-Super",
@@ -252,15 +940,41 @@
       "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"]
+      "primary_files": [
+        "docs/data/omni_model_comparison.json"
+      ]
     },
     {
-      "term": "Cosmos3-Nano",
+      "term": "Foundation pipeline",
       "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"]
+      "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": "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": "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": "LoRA adapter",
@@ -268,23 +982,85 @@
       "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"]
+      "primary_files": [
+        "OMNI_MODEL_EXTENSION_CONTRACT.md",
+        "docs/data/omni_model_comparison.json"
+      ]
     },
     {
-      "term": "Full-parameter fine-tuning",
+      "term": "Metadata baseline",
       "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"]
+      "plain_meaning": "A selected-128 baseline using metadata or text-derived public-safe features.",
+      "project_usage": "Compares simple and neural heads on the held-out split.",
+      "do_not_confuse_with": "Raw video, depth, or audio feature baselines.",
+      "primary_files": [
+        "docs/data/task_method_20_result_matrix.json"
+      ]
     },
     {
-      "term": "Foundation pipeline",
+      "term": "Minimal baseline",
       "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"]
+      "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": "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": "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": "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": "Raw-feature baseline",
+      "category": "models_runs",
+      "plain_meaning": "A selected-128 baseline using exported public-safe raw-feature groups.",
+      "project_usage": "Tracks what non-foundation heads can do with richer processed inputs.",
+      "do_not_confuse_with": "Raw gated media redistribution.",
+      "primary_files": [
+        "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": "Spatial intelligence",
@@ -292,15 +1068,10 @@
       "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"]
+      "primary_files": [
+        "THREE_FOUNDATION_PIPELINES.md",
+        "docs/data/three_foundation_pipelines.json"
+      ]
     },
     {
       "term": "Vision-language-action",
@@ -308,15 +1079,10 @@
       "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"]
+      "primary_files": [
+        "THREE_FOUNDATION_PIPELINES.md",
+        "docs/data/three_foundation_pipelines.json"
+      ]
     },
     {
       "term": "HF artifact dataset",
@@ -324,7 +1090,10 @@
       "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"]
+      "primary_files": [
+        "ARTIFACT_GUIDE.md",
+        "docs/data/artifact_index.json"
+      ]
     },
     {
       "term": "HF baseline model repo",
@@ -332,7 +1101,31 @@
       "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"]
+      "primary_files": [
+        "PUBLIC_READER_MAP.md",
+        "docs/data/public_reader_map.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 weights/results repo",
+      "category": "public_surfaces",
+      "plain_meaning": "A consolidated public-safe model-result bundle.",
+      "project_usage": "Groups baseline weights, verified model artifacts, analysis files, and manifests.",
+      "do_not_confuse_with": "The upstream raw dataset.",
+      "primary_files": [
+        "PUBLIC_READER_MAP.md"
+      ]
     },
     {
       "term": "Mirror parity",
@@ -340,7 +1133,30 @@
       "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"]
+      "primary_files": [
+        "docs/data/mirror_parity.json"
+      ]
+    },
+    {
+      "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": "Publication audit",
+      "category": "public_surfaces",
+      "plain_meaning": "A public-package validation report.",
+      "project_usage": "Confirms required files exist and forbidden raw/private assets are not included.",
+      "do_not_confuse_with": "Scientific peer review.",
+      "primary_files": [
+        "docs/data/publication_audit.json"
+      ]
     },
     {
       "term": "Verified package",
@@ -348,37 +1164,62 @@
       "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"]
+      "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"]
+      "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"]
+      "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"]
+      "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"]
+      "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"]
+      "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"]
+      "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"]
+      "files": [
+        "PUBLIC_SURFACE_QA.md",
+        "docs/data/mirror_parity.json",
+        "docs/data/live_publication_status.json"
+      ]
     }
   ]
 }

docs/data/mirror_parity.json

@@ -1,6 +1,6 @@
 {
   "status": "pass",
-  "generated_at_utc": "2026-06-22T14:36:09+00:00",
+  "generated_at_utc": "2026-06-22T14:44:16+00:00",
   "hf_root": "hf_publish",
   "summary": {
     "group_count": 1306,
@@ -139,44 +139,44 @@
         "path": "repo:docs/data/artifact_index.json",
         "exists": true,
         "bytes": 124477,
-        "sha256": "e987021a89d254d0e3caafe7281872511a85357cc475dd33494f00edfed1e098"
+        "sha256": "246b4e257b7ac1abe2967fe08925b5c7d2ea3454edc55d0c0589eeba8dd1518e"
       },
       "mirrors": {
         "hf_space": {
           "path": "hf_space:data/artifact_index.json",
           "exists": true,
           "bytes": 124477,
-          "sha256": "e987021a89d254d0e3caafe7281872511a85357cc475dd33494f00edfed1e098"
+          "sha256": "246b4e257b7ac1abe2967fe08925b5c7d2ea3454edc55d0c0589eeba8dd1518e"
         },
         "hf_artifacts_data": {
           "path": "hf_artifacts:data/artifact_index.json",
           "exists": true,
           "bytes": 124477,
-          "sha256": "e987021a89d254d0e3caafe7281872511a85357cc475dd33494f00edfed1e098"
+          "sha256": "246b4e257b7ac1abe2967fe08925b5c7d2ea3454edc55d0c0589eeba8dd1518e"
         },
         "hf_artifacts": {
           "path": "hf_artifacts:docs/data/artifact_index.json",
           "exists": true,
           "bytes": 124477,
-          "sha256": "e987021a89d254d0e3caafe7281872511a85357cc475dd33494f00edfed1e098"
+          "sha256": "246b4e257b7ac1abe2967fe08925b5c7d2ea3454edc55d0c0589eeba8dd1518e"
         },
         "hf_model_data": {
           "path": "hf_model:data/artifact_index.json",
           "exists": true,
           "bytes": 124477,
-          "sha256": "e987021a89d254d0e3caafe7281872511a85357cc475dd33494f00edfed1e098"
+          "sha256": "246b4e257b7ac1abe2967fe08925b5c7d2ea3454edc55d0c0589eeba8dd1518e"
         },
         "hf_model_docs_data": {
           "path": "hf_model:docs/data/artifact_index.json",
           "exists": true,
           "bytes": 124477,
-          "sha256": "e987021a89d254d0e3caafe7281872511a85357cc475dd33494f00edfed1e098"
+          "sha256": "246b4e257b7ac1abe2967fe08925b5c7d2ea3454edc55d0c0589eeba8dd1518e"
         },
         "hf_model": {
           "path": "hf_model:metrics/artifact_index.json",
           "exists": true,
           "bytes": 124477,
-          "sha256": "e987021a89d254d0e3caafe7281872511a85357cc475dd33494f00edfed1e098"
+          "sha256": "246b4e257b7ac1abe2967fe08925b5c7d2ea3454edc55d0c0589eeba8dd1518e"
         }
       },
       "failures": []
@@ -972,44 +972,44 @@
         "path": "repo:docs/data/publication_audit.json",
         "exists": true,
         "bytes": 10940,
-        "sha256": "5b02a1990d8e14421a9cb9803f059b70c0c816440cc0dcaeee939a0efc4341bc"
+        "sha256": "2b988f2440692a77049129e6f1f6495ae541e371395fa2b7527ec68d6bd9fb8d"
       },
       "mirrors": {
         "hf_space": {
           "path": "hf_space:data/publication_audit.json",
           "exists": true,
           "bytes": 10940,
-          "sha256": "5b02a1990d8e14421a9cb9803f059b70c0c816440cc0dcaeee939a0efc4341bc"
+          "sha256": "2b988f2440692a77049129e6f1f6495ae541e371395fa2b7527ec68d6bd9fb8d"
         },
         "hf_artifacts_data": {
           "path": "hf_artifacts:data/publication_audit.json",
           "exists": true,
           "bytes": 10940,
-          "sha256": "5b02a1990d8e14421a9cb9803f059b70c0c816440cc0dcaeee939a0efc4341bc"
+          "sha256": "2b988f2440692a77049129e6f1f6495ae541e371395fa2b7527ec68d6bd9fb8d"
         },
         "hf_artifacts": {
           "path": "hf_artifacts:docs/data/publication_audit.json",
           "exists": true,
           "bytes": 10940,
-          "sha256": "5b02a1990d8e14421a9cb9803f059b70c0c816440cc0dcaeee939a0efc4341bc"
+          "sha256": "2b988f2440692a77049129e6f1f6495ae541e371395fa2b7527ec68d6bd9fb8d"
         },
         "hf_model_data": {
           "path": "hf_model:data/publication_audit.json",
           "exists": true,
           "bytes": 10940,
-          "sha256": "5b02a1990d8e14421a9cb9803f059b70c0c816440cc0dcaeee939a0efc4341bc"
+          "sha256": "2b988f2440692a77049129e6f1f6495ae541e371395fa2b7527ec68d6bd9fb8d"
         },
         "hf_model_docs_data": {
           "path": "hf_model:docs/data/publication_audit.json",
           "exists": true,
           "bytes": 10940,
-          "sha256": "5b02a1990d8e14421a9cb9803f059b70c0c816440cc0dcaeee939a0efc4341bc"
+          "sha256": "2b988f2440692a77049129e6f1f6495ae541e371395fa2b7527ec68d6bd9fb8d"
         },
         "hf_model": {
           "path": "hf_model:metrics/publication_audit.json",
           "exists": true,
           "bytes": 10940,
-          "sha256": "5b02a1990d8e14421a9cb9803f059b70c0c816440cc0dcaeee939a0efc4341bc"
+          "sha256": "2b988f2440692a77049129e6f1f6495ae541e371395fa2b7527ec68d6bd9fb8d"
         }
       },
       "failures": []
@@ -1021,44 +1021,44 @@
         "path": "repo:docs/data/public_surface_qa.json",
         "exists": true,
         "bytes": 7690,
-        "sha256": "4c171fb154654cba54cc0697dcd0c53a9dbd805d22bfa3d2941b4fdc155bf146"
+        "sha256": "28467929e81e827a7f3de5508a981f20ab7f271ddac2b11188ffe655363ef50e"
       },
       "mirrors": {
         "hf_space": {
           "path": "hf_space:data/public_surface_qa.json",
           "exists": true,
           "bytes": 7690,
-          "sha256": "4c171fb154654cba54cc0697dcd0c53a9dbd805d22bfa3d2941b4fdc155bf146"
+          "sha256": "28467929e81e827a7f3de5508a981f20ab7f271ddac2b11188ffe655363ef50e"
         },
         "hf_artifacts_data": {
           "path": "hf_artifacts:data/public_surface_qa.json",
           "exists": true,
           "bytes": 7690,
-          "sha256": "4c171fb154654cba54cc0697dcd0c53a9dbd805d22bfa3d2941b4fdc155bf146"
+          "sha256": "28467929e81e827a7f3de5508a981f20ab7f271ddac2b11188ffe655363ef50e"
         },
         "hf_artifacts": {
           "path": "hf_artifacts:docs/data/public_surface_qa.json",
           "exists": true,
           "bytes": 7690,
-          "sha256": "4c171fb154654cba54cc0697dcd0c53a9dbd805d22bfa3d2941b4fdc155bf146"
+          "sha256": "28467929e81e827a7f3de5508a981f20ab7f271ddac2b11188ffe655363ef50e"
         },
         "hf_model_data": {
           "path": "hf_model:data/public_surface_qa.json",
           "exists": true,
           "bytes": 7690,
-          "sha256": "4c171fb154654cba54cc0697dcd0c53a9dbd805d22bfa3d2941b4fdc155bf146"
+          "sha256": "28467929e81e827a7f3de5508a981f20ab7f271ddac2b11188ffe655363ef50e"
         },
         "hf_model_docs_data": {
           "path": "hf_model:docs/data/public_surface_qa.json",
           "exists": true,
           "bytes": 7690,
-          "sha256": "4c171fb154654cba54cc0697dcd0c53a9dbd805d22bfa3d2941b4fdc155bf146"
+          "sha256": "28467929e81e827a7f3de5508a981f20ab7f271ddac2b11188ffe655363ef50e"
         },
         "hf_model": {
           "path": "hf_model:metrics/public_surface_qa.json",
           "exists": true,
           "bytes": 7690,
-          "sha256": "4c171fb154654cba54cc0697dcd0c53a9dbd805d22bfa3d2941b4fdc155bf146"
+          "sha256": "28467929e81e827a7f3de5508a981f20ab7f271ddac2b11188ffe655363ef50e"
         }
       },
       "failures": []
@@ -1217,44 +1217,44 @@
         "path": "repo:docs/data/quality_gates.json",
         "exists": true,
         "bytes": 8640,
-        "sha256": "745b9c41da785ab84af8aba58babe3d8decd3c0bb07e3187b64f82ae42d91ec5"
+        "sha256": "e3f97614b47251d02f560db20a4bc918088b9aadc45c1d1397c2756b4d4867bf"
       },
       "mirrors": {
         "hf_space": {
           "path": "hf_space:data/quality_gates.json",
           "exists": true,
           "bytes": 8640,
-          "sha256": "745b9c41da785ab84af8aba58babe3d8decd3c0bb07e3187b64f82ae42d91ec5"
+          "sha256": "e3f97614b47251d02f560db20a4bc918088b9aadc45c1d1397c2756b4d4867bf"
         },
         "hf_artifacts_data": {
           "path": "hf_artifacts:data/quality_gates.json",
           "exists": true,
           "bytes": 8640,
-          "sha256": "745b9c41da785ab84af8aba58babe3d8decd3c0bb07e3187b64f82ae42d91ec5"
+          "sha256": "e3f97614b47251d02f560db20a4bc918088b9aadc45c1d1397c2756b4d4867bf"
         },
         "hf_artifacts": {
           "path": "hf_artifacts:docs/data/quality_gates.json",
           "exists": true,
           "bytes": 8640,
-          "sha256": "745b9c41da785ab84af8aba58babe3d8decd3c0bb07e3187b64f82ae42d91ec5"
+          "sha256": "e3f97614b47251d02f560db20a4bc918088b9aadc45c1d1397c2756b4d4867bf"
         },
         "hf_model_data": {
           "path": "hf_model:data/quality_gates.json",
           "exists": true,
           "bytes": 8640,
-          "sha256": "745b9c41da785ab84af8aba58babe3d8decd3c0bb07e3187b64f82ae42d91ec5"
+          "sha256": "e3f97614b47251d02f560db20a4bc918088b9aadc45c1d1397c2756b4d4867bf"
         },
         "hf_model_docs_data": {
           "path": "hf_model:docs/data/quality_gates.json",
           "exists": true,
           "bytes": 8640,
-          "sha256": "745b9c41da785ab84af8aba58babe3d8decd3c0bb07e3187b64f82ae42d91ec5"
+          "sha256": "e3f97614b47251d02f560db20a4bc918088b9aadc45c1d1397c2756b4d4867bf"
         },
         "hf_model": {
           "path": "hf_model:metrics/quality_gates.json",
           "exists": true,
           "bytes": 8640,
-          "sha256": "745b9c41da785ab84af8aba58babe3d8decd3c0bb07e3187b64f82ae42d91ec5"
+          "sha256": "e3f97614b47251d02f560db20a4bc918088b9aadc45c1d1397c2756b4d4867bf"
         }
       },
       "failures": []
@@ -1756,44 +1756,44 @@
         "path": "repo:docs/data/source_alignment_audit.json",
         "exists": true,
         "bytes": 4432,
-        "sha256": "c3fafe7f03e0558e2acfd702d222bba9a53e391cfd711ee7a2b65299579fcf28"
+        "sha256": "77199b03fc4d589a648033e359183a252d43dd6900e19ca609349bd972939e84"
       },
       "mirrors": {
         "hf_space": {
           "path": "hf_space:data/source_alignment_audit.json",
           "exists": true,
           "bytes": 4432,
-          "sha256": "c3fafe7f03e0558e2acfd702d222bba9a53e391cfd711ee7a2b65299579fcf28"
+          "sha256": "77199b03fc4d589a648033e359183a252d43dd6900e19ca609349bd972939e84"
         },
         "hf_artifacts_data": {
           "path": "hf_artifacts:data/source_alignment_audit.json",
           "exists": true,
           "bytes": 4432,
-          "sha256": "c3fafe7f03e0558e2acfd702d222bba9a53e391cfd711ee7a2b65299579fcf28"
+          "sha256": "77199b03fc4d589a648033e359183a252d43dd6900e19ca609349bd972939e84"
         },
         "hf_artifacts": {
           "path": "hf_artifacts:docs/data/source_alignment_audit.json",
           "exists": true,
           "bytes": 4432,
-          "sha256": "c3fafe7f03e0558e2acfd702d222bba9a53e391cfd711ee7a2b65299579fcf28"
+          "sha256": "77199b03fc4d589a648033e359183a252d43dd6900e19ca609349bd972939e84"
         },
         "hf_model_data": {
           "path": "hf_model:data/source_alignment_audit.json",
           "exists": true,
           "bytes": 4432,
-          "sha256": "c3fafe7f03e0558e2acfd702d222bba9a53e391cfd711ee7a2b65299579fcf28"
+          "sha256": "77199b03fc4d589a648033e359183a252d43dd6900e19ca609349bd972939e84"
         },
         "hf_model_docs_data": {
           "path": "hf_model:docs/data/source_alignment_audit.json",
           "exists": true,
           "bytes": 4432,
-          "sha256": "c3fafe7f03e0558e2acfd702d222bba9a53e391cfd711ee7a2b65299579fcf28"
+          "sha256": "77199b03fc4d589a648033e359183a252d43dd6900e19ca609349bd972939e84"
         },
         "hf_model": {
           "path": "hf_model:metrics/source_alignment_audit.json",
           "exists": true,
           "bytes": 4432,
-          "sha256": "c3fafe7f03e0558e2acfd702d222bba9a53e391cfd711ee7a2b65299579fcf28"
+          "sha256": "77199b03fc4d589a648033e359183a252d43dd6900e19ca609349bd972939e84"
         }
       },
       "failures": []
@@ -2197,44 +2197,44 @@
         "path": "repo:docs/data/task_surface_integrity.json",
         "exists": true,
         "bytes": 46246,
-        "sha256": "fa0d28c31c066a6f53af8ced3c67d0f1fe9573d92b03bb9f5d0c3d5b239fc6cc"
+        "sha256": "525d6b4f05bb6ffa6f7524d573d5f51d6cd8fe55a4d1a4853282e25247f90a12"
       },
       "mirrors": {
         "hf_space": {
           "path": "hf_space:data/task_surface_integrity.json",
           "exists": true,
           "bytes": 46246,
-          "sha256": "fa0d28c31c066a6f53af8ced3c67d0f1fe9573d92b03bb9f5d0c3d5b239fc6cc"
+          "sha256": "525d6b4f05bb6ffa6f7524d573d5f51d6cd8fe55a4d1a4853282e25247f90a12"
         },
         "hf_artifacts_data": {
           "path": "hf_artifacts:data/task_surface_integrity.json",
           "exists": true,
           "bytes": 46246,
-          "sha256": "fa0d28c31c066a6f53af8ced3c67d0f1fe9573d92b03bb9f5d0c3d5b239fc6cc"
+          "sha256": "525d6b4f05bb6ffa6f7524d573d5f51d6cd8fe55a4d1a4853282e25247f90a12"
         },
         "hf_artifacts": {
           "path": "hf_artifacts:docs/data/task_surface_integrity.json",
           "exists": true,
           "bytes": 46246,
-          "sha256": "fa0d28c31c066a6f53af8ced3c67d0f1fe9573d92b03bb9f5d0c3d5b239fc6cc"
+          "sha256": "525d6b4f05bb6ffa6f7524d573d5f51d6cd8fe55a4d1a4853282e25247f90a12"
         },
         "hf_model_data": {
           "path": "hf_model:data/task_surface_integrity.json",
           "exists": true,
           "bytes": 46246,
-          "sha256": "fa0d28c31c066a6f53af8ced3c67d0f1fe9573d92b03bb9f5d0c3d5b239fc6cc"
+          "sha256": "525d6b4f05bb6ffa6f7524d573d5f51d6cd8fe55a4d1a4853282e25247f90a12"
         },
         "hf_model_docs_data": {
           "path": "hf_model:docs/data/task_surface_integrity.json",
           "exists": true,
           "bytes": 46246,
-          "sha256": "fa0d28c31c066a6f53af8ced3c67d0f1fe9573d92b03bb9f5d0c3d5b239fc6cc"
+          "sha256": "525d6b4f05bb6ffa6f7524d573d5f51d6cd8fe55a4d1a4853282e25247f90a12"
         },
         "hf_model": {
           "path": "hf_model:metrics/task_surface_integrity.json",
           "exists": true,
           "bytes": 46246,
-          "sha256": "fa0d28c31c066a6f53af8ced3c67d0f1fe9573d92b03bb9f5d0c3d5b239fc6cc"
+          "sha256": "525d6b4f05bb6ffa6f7524d573d5f51d6cd8fe55a4d1a4853282e25247f90a12"
         }
       },
       "failures": []
@@ -2589,44 +2589,44 @@
         "path": "repo:docs/data/website_integrity.json",
         "exists": true,
         "bytes": 24948,
-        "sha256": "1ff9661e301b73102f61efec5e6a2003d53569ab6cdf6f016fadbcaba608d7ca"
+        "sha256": "6657a1f46aea0d58ea5b1ce5b0dbeb12096ffc1bc5ec498c94b9bd5ba4ed1bf2"
       },
       "mirrors": {
         "hf_space": {
           "path": "hf_space:data/website_integrity.json",
           "exists": true,
           "bytes": 24948,
-          "sha256": "1ff9661e301b73102f61efec5e6a2003d53569ab6cdf6f016fadbcaba608d7ca"
+          "sha256": "6657a1f46aea0d58ea5b1ce5b0dbeb12096ffc1bc5ec498c94b9bd5ba4ed1bf2"
         },
         "hf_artifacts_data": {
           "path": "hf_artifacts:data/website_integrity.json",
           "exists": true,
           "bytes": 24948,
-          "sha256": "1ff9661e301b73102f61efec5e6a2003d53569ab6cdf6f016fadbcaba608d7ca"
+          "sha256": "6657a1f46aea0d58ea5b1ce5b0dbeb12096ffc1bc5ec498c94b9bd5ba4ed1bf2"
         },
         "hf_artifacts": {
           "path": "hf_artifacts:docs/data/website_integrity.json",
           "exists": true,
           "bytes": 24948,
-          "sha256": "1ff9661e301b73102f61efec5e6a2003d53569ab6cdf6f016fadbcaba608d7ca"
+          "sha256": "6657a1f46aea0d58ea5b1ce5b0dbeb12096ffc1bc5ec498c94b9bd5ba4ed1bf2"
         },
         "hf_model_data": {
           "path": "hf_model:data/website_integrity.json",
           "exists": true,
           "bytes": 24948,
-          "sha256": "1ff9661e301b73102f61efec5e6a2003d53569ab6cdf6f016fadbcaba608d7ca"
+          "sha256": "6657a1f46aea0d58ea5b1ce5b0dbeb12096ffc1bc5ec498c94b9bd5ba4ed1bf2"
         },
         "hf_model_docs_data": {
           "path": "hf_model:docs/data/website_integrity.json",
           "exists": true,
           "bytes": 24948,
-          "sha256": "1ff9661e301b73102f61efec5e6a2003d53569ab6cdf6f016fadbcaba608d7ca"
+          "sha256": "6657a1f46aea0d58ea5b1ce5b0dbeb12096ffc1bc5ec498c94b9bd5ba4ed1bf2"
         },
         "hf_model": {
           "path": "hf_model:metrics/website_integrity.json",
           "exists": true,
           "bytes": 24948,
-          "sha256": "1ff9661e301b73102f61efec5e6a2003d53569ab6cdf6f016fadbcaba608d7ca"
+          "sha256": "6657a1f46aea0d58ea5b1ce5b0dbeb12096ffc1bc5ec498c94b9bd5ba4ed1bf2"
         }
       },
       "failures": []
@@ -7429,39 +7429,39 @@
       "local": {
         "path": "repo:docs/index.html",
         "exists": true,
-        "bytes": 367548,
-        "sha256": "bafcdcb8aaa1647eea5fdbae52c318991b64a5a0cfa8727edc624e5c7b8bcf01"
+        "bytes": 367575,
+        "sha256": "2f2325aedf5f02d17e82a76310b7d66a676985bc8b343e5ac3658be2a7c802da"
       },
       "mirrors": {
         "hf_space": {
           "path": "hf_space:index.html",
           "exists": true,
-          "bytes": 367548,
-          "sha256": "bafcdcb8aaa1647eea5fdbae52c318991b64a5a0cfa8727edc624e5c7b8bcf01"
+          "bytes": 367575,
+          "sha256": "2f2325aedf5f02d17e82a76310b7d66a676985bc8b343e5ac3658be2a7c802da"
         },
         "hf_artifacts_root": {
           "path": "hf_artifacts:index.html",
           "exists": true,
-          "bytes": 367548,
-          "sha256": "bafcdcb8aaa1647eea5fdbae52c318991b64a5a0cfa8727edc624e5c7b8bcf01"
+          "bytes": 367575,
+          "sha256": "2f2325aedf5f02d17e82a76310b7d66a676985bc8b343e5ac3658be2a7c802da"
         },
         "hf_artifacts_docs": {
           "path": "hf_artifacts:docs/index.html",
           "exists": true,
-          "bytes": 367548,
-          "sha256": "bafcdcb8aaa1647eea5fdbae52c318991b64a5a0cfa8727edc624e5c7b8bcf01"
+          "bytes": 367575,
+          "sha256": "2f2325aedf5f02d17e82a76310b7d66a676985bc8b343e5ac3658be2a7c802da"
         },
         "hf_model": {
           "path": "hf_model:index.html",
           "exists": true,
-          "bytes": 367548,
-          "sha256": "bafcdcb8aaa1647eea5fdbae52c318991b64a5a0cfa8727edc624e5c7b8bcf01"
+          "bytes": 367575,
+          "sha256": "2f2325aedf5f02d17e82a76310b7d66a676985bc8b343e5ac3658be2a7c802da"
         },
         "hf_model_docs": {
           "path": "hf_model:docs/index.html",
           "exists": true,
-          "bytes": 367548,
-          "sha256": "bafcdcb8aaa1647eea5fdbae52c318991b64a5a0cfa8727edc624e5c7b8bcf01"
+          "bytes": 367575,
+          "sha256": "2f2325aedf5f02d17e82a76310b7d66a676985bc8b343e5ac3658be2a7c802da"
         }
       },
       "failures": []

docs/data/public_surface_qa.json

@@ -1,7 +1,7 @@
 {
   "title": "Ropedia Xperience-10M Public Project Surface",
   "status": "pass",
-  "generated_at_utc": "2026-06-22T14:40:34+00:00",
+  "generated_at_utc": "2026-06-22T15:09:11+00:00",
   "scope": "Repo README, GitHub Pages HTML, Hugging Face Space card, artifact dataset card, and model card.",
   "checks": [
     {
@@ -18,7 +18,7 @@
         "website_integrity": {
           "exists": true,
           "status": "pass",
-          "generated_at_utc": "2026-06-22T14:39:11+00:00"
+          "generated_at_utc": "2026-06-22T15:07:54+00:00"
         },
         "rendered_site_check": {
           "exists": true,
@@ -28,12 +28,12 @@
         "task_surface_integrity": {
           "exists": true,
           "status": "pass",
-          "generated_at_utc": "2026-06-22T14:31:36+00:00"
+          "generated_at_utc": "2026-06-22T15:07:48+00:00"
         },
         "source_alignment": {
           "exists": true,
           "status": "pass",
-          "generated_at_utc": "2026-06-22T14:31:36+00:00"
+          "generated_at_utc": "2026-06-22T15:07:48+00:00"
         },
         "scale_up_status": {
           "exists": true,
@@ -43,12 +43,12 @@
         "publication_package": {
           "exists": true,
           "status": "pass",
-          "generated_at_utc": "2026-06-22T14:37:20+00:00"
+          "generated_at_utc": "2026-06-22T15:08:27+00:00"
         },
         "mirror_parity": {
           "exists": true,
           "status": "pass",
-          "generated_at_utc": "2026-06-22T14:36:09+00:00"
+          "generated_at_utc": "2026-06-22T14:44:16+00:00"
         }
       },
       "failures": {}
@@ -96,9 +96,9 @@
       "reason": "Public copy should consistently present the project as Ropedia Xperience-10M, with the Qwen3-Omni scale-up status.",
       "marker_counts": {
         "Ropedia Xperience-10M Task Suite": 22,
-        "Xperience-10M": 171,
+        "Xperience-10M": 170,
         "20-task": 117,
-        "Qwen3-Omni": 236,
+        "Qwen3-Omni": 233,
         "128-episode pilot": 1
       }
     },

docs/data/publication_audit.json

@@ -1,6 +1,6 @@
 {
   "status": "pass",
-  "generated_at_utc": "2026-06-22T14:41:37+00:00",
+  "generated_at_utc": "2026-06-22T15:08:27+00:00",
   "checks": [
     {
       "name": "required_publication_assets_present",

docs/data/quality_gates.json

@@ -1,7 +1,7 @@
 {
   "title": "Ropedia Xperience-10M Release Checks",
   "status": "pass",
-  "generated_at_utc": "2026-06-22T14:42:50+00:00",
+  "generated_at_utc": "2026-06-22T15:09:12+00:00",
   "rule": "A release is current when the automated reports pass and the live GitHub/Hugging Face mirrors are verified after publishing.",
   "automated_gates": [
     {

docs/data/source_alignment_audit.json

@@ -1,7 +1,7 @@
 {
   "title": "Ropedia Xperience-10M Source Alignment Note",
   "status": "pass",
-  "generated_at_utc": "2026-06-22T14:40:33+00:00",
+  "generated_at_utc": "2026-06-22T15:07:48+00:00",
   "alignment_json": "docs/data/xperience10m_dataset_card_alignment.json",
   "alignment_summary": {
     "full_dataset_repo": "ropedia-ai/xperience-10m",

docs/data/task_surface_integrity.json

@@ -1,6 +1,6 @@
 {
   "status": "pass",
-  "generated_at_utc": "2026-06-22T14:40:33+00:00",
+  "generated_at_utc": "2026-06-22T15:07:48+00:00",
   "summary": {
     "original_walkthrough_task_count": 12,
     "expected_original_walkthrough_task_count": 12,

docs/data/website_integrity.json

@@ -1,6 +1,6 @@
 {
   "status": "pass",
-  "generated_at_utc": "2026-06-22T14:39:11+00:00",
+  "generated_at_utc": "2026-06-22T15:07:54+00:00",
   "docs_root": "docs",
   "site_base": "/ropedia-xperience-10m-task-suite/",
   "summary": {
@@ -80,8 +80,8 @@
       "name": "project_overview_precedes_progress_ledger",
       "status": "pass",
       "reason": "The project overview should appear before the deeper progress ledger.",
-      "overview_index": 151226,
-      "evidence_index": 199040
+      "overview_index": 151515,
+      "evidence_index": 199329
     },
     {
       "name": "project_status_links_json",
@@ -159,9 +159,9 @@
       "name": "evaluation_protocol_between_overview_and_progress",
       "status": "pass",
       "reason": "The evaluation protocol should appear before the deeper evidence ledger.",
-      "overview_index": 151226,
-      "protocol_index": 195245,
-      "evidence_index": 199040
+      "overview_index": 151515,
+      "protocol_index": 195534,
+      "evidence_index": 199329
     },
     {
       "name": "evaluation_protocol_links_json",
@@ -290,7 +290,7 @@
     },
     {
       "path": "index.html",
-      "id_count": 101,
+      "id_count": 102,
       "reference_count": 252,
       "image_count": 54
     },
@@ -355,7 +355,7 @@
     },
     {
       "path": "data/glossary.json",
-      "bytes": 19260,
+      "bytes": 49025,
       "top_level_type": "dict"
     },
     {

docs/index.html

@@ -2465,6 +2465,17 @@
       font-size: 15px;
       font-weight: 750;
     }
+    .glossary-table td:first-child small {
+      display: block;
+      margin-top: 6px;
+      color: var(--green);
+      font-family: var(--font-mono);
+      font-size: 10px;
+      font-weight: 800;
+      letter-spacing: 0.05em;
+      text-transform: uppercase;
+      opacity: 0.82;
+    }
     .glossary-table td:nth-child(2) {
       width: 32%;
       color: #d8e2d4;
@@ -6624,25 +6635,30 @@
     <section id="glossary" data-project-tab="resources" role="tabpanel" aria-labelledby="tab-resources" tabindex="-1">
       <div class="wrap">
         <div class="section-head">
-          <h2>Glossary for overloaded terms.</h2>
-          <p>These are the terms readers most often confuse when moving between the repo, website, Hugging Face mirrors, result matrices, and model-package cards. The full glossary is mirrored as a reader note and structured data.</p>
+          <h2>Glossary for project and field terms.</h2>
+          <p>This glossary covers the overloaded project terms plus adjacent technical terms from embodied AI, egocentric multimodal data, spatial geometry, world models, VLA/policy learning, training, evaluation, and public artifact reading.</p>
         </div>
         <div class="glossary-panel">
           <div class="glossary-summary" aria-label="Glossary categories">
             <article>
-              <small>scope</small>
-              <strong>Separate data, result lanes, and mirrors</strong>
-              <p>Evidence lines, public-safe artifacts, and gated upstream data are different objects. The glossary keeps those boundaries visible.</p>
+              <small>data</small>
+              <strong>Read multimodal sample terms</strong>
+              <p>Episode, window, modality, fisheye, depth, IMU, calibration, and synchronization terms are grouped with the project data boundary terms.</p>
+            </article>
+            <article>
+              <small>space + time</small>
+              <strong>Decode geometry and world-model language</strong>
+              <p>Camera pose, SLAM, point clouds, rollouts, forward dynamics, long-horizon forecasting, and temporal leakage are defined next to the relevant tasks.</p>
             </article>
             <article>
-              <small>results</small>
-              <strong>Read scores by source type</strong>
-              <p>Direct scores, compact-proxy scores, gap audits, and task-method records should not be interpreted as the same kind of evidence.</p>
+              <small>robotics</small>
+              <strong>Connect VLA and policy terms</strong>
+              <p>Action chunks, policies, imitation learning, behavior cloning, end effectors, dexterity, contact, and language grounding are included for extension readers.</p>
             </article>
             <article>
-              <small>models</small>
-              <strong>Keep branches distinct</strong>
-              <p>Minimal/NN heads, metadata/raw baselines, Qwen3-Omni v1-v6, Cosmos3-Super, Cosmos3-Nano, LoRA adapters, and full-parameter gates each mean something specific.</p>
+              <small>evidence</small>
+              <strong>Keep scores and public surfaces distinct</strong>
+              <p>Direct/proxy scores, raw metrics, radar values, held-out evaluation, Qwen/Cosmos branches, adapters, and HF mirrors remain explicitly separated.</p>
             </article>
           </div>
           <div>
@@ -6656,29 +6672,19 @@
                     <th>Do not confuse with</th>
                   </tr>
                 </thead>
-                <tbody>
-                  <tr><td>Evidence line</td><td>A reading lane for a group of results.</td><td>Line 1 is the public sample episode; Line 2 is selected-128 held-out comparison.</td><td>Qwen v1-v6 run versions.</td></tr>
-                  <tr><td>Public sample episode</td><td>The one fully inspectable official sample episode.</td><td>Raw-file browsing, task construction, single-episode baselines.</td><td>The selected-128 comparison rows.</td></tr>
-                  <tr><td>Selected 128 episodes</td><td>Public-safe derived features linked to official gated episode paths.</td><td>Same-split Line 2 baseline/model comparisons.</td><td>Redistributed raw MP4/HDF5/RRD files.</td></tr>
-                  <tr><td>20-frame window</td><td>A fixed short clip slice used as a model input unit.</td><td>Feature rows, labels, tasks, and many baseline heads.</td><td>A full episode.</td></tr>
-                  <tr><td>Task-method record</td><td>One method evaluated on one task.</td><td>The 9 x 20 public matrix, now 180 scored records.</td><td>A single prediction row.</td></tr>
-                  <tr><td>Direct score</td><td>A metric computed against the task target directly.</td><td>Primary interpretation in the result matrix.</td><td>Compact-proxy score.</td></tr>
-                  <tr><td>Compact-proxy score</td><td>A bounded proxy when the direct raw target is not public.</td><td>Explicitly marked cells in the gap audit and matrix.</td><td>A direct target measurement.</td></tr>
-                  <tr><td>Raw metric value</td><td>The original value emitted by the runner or verified package.</td><td>The value to cite from the 180-result table.</td><td>Normalized radar value.</td></tr>
-                  <tr><td>Normalized radar value</td><td>A 0-1 plotting value used only for comparable radar polygons.</td><td>Visual comparison across metrics with different scales.</td><td>The raw metric value to cite.</td></tr>
-                  <tr><td>Minimal baseline</td><td>A simple non-neural task head; the "minimum" reference row in casual wording.</td><td>Single-episode lower-complexity comparison.</td><td>Selected-128 Simple baseline rows.</td></tr>
-                  <tr><td>Simple baseline</td><td>A non-neural selected-128 baseline family.</td><td>Metadata/text and raw-feature 128-episode comparisons before NN/foundation rows.</td><td>The single-episode Minimal baseline.</td></tr>
-                  <tr><td>Qwen3-Omni v6</td><td>The current public Qwen 20-task row.</td><td>Qwen3-Omni LoRA plus task-specific probes.</td><td>All Qwen v1-v6 experiments.</td></tr>
-                  <tr><td>Cosmos3-Super</td><td>The larger Cosmos-style branch.</td><td>Reasoner diagnostics and a verified forward-dynamics LoRA branch.</td><td>Cosmos3-Nano Future Window.</td></tr>
-                  <tr><td>LoRA adapter</td><td>Lightweight trainable adapter weights.</td><td>Public model-branch artifacts when verified.</td><td>Full base-model weights.</td></tr>
-                  <tr><td>HF artifact dataset</td><td>Hugging Face dataset repo for derived evidence.</td><td>Reports, metrics, website data, sanitized result packages.</td><td>The upstream Xperience-10M dataset.</td></tr>
-                  <tr><td>Mirror parity</td><td>A check that public copies match source files.</td><td>Verifying GitHub, website, and HF mirrors.</td><td>A model-quality metric.</td></tr>
+                <tbody id="glossaryRows">
+                  <tr><td>Egocentric video<small>multimodal sensing</small></td><td>Video captured from a first-person or body-mounted viewpoint.</td><td>The sample streams are egocentric views of human interaction and are the visual basis for many tasks.</td><td>Third-person robot-camera footage.</td></tr>
+                  <tr><td>Camera pose<small>spatial geometry</small></td><td>The camera position and orientation at a time step.</td><td>Supports spatial-intelligence tasks, view synchronization, and geometry diagnostics.</td><td>The human body pose.</td></tr>
+                  <tr><td>Forward dynamics<small>temporal and world models</small></td><td>Predicting the next state from the current state and action/context.</td><td>The Cosmos3-Super LoRA branch uses a forward-dynamics-style diagnostic contract.</td><td>Reverse inference from result back to cause.</td></tr>
+                  <tr><td>Vision-language-action model<small>robotics and VLA</small></td><td>A model that maps visual context and language into actions.</td><td>The VLA direction is a future path after action targets are converted into robot-compatible chunks.</td><td>A vision-language model that only answers text.</td></tr>
+                  <tr><td>Direct score<small>tasks and metrics</small></td><td>A metric computed against the task target directly.</td><td>The preferred score type in the 20-task matrix.</td><td>Compact-proxy score.</td></tr>
+                  <tr><td>Held-out evaluation<small>training and evaluation</small></td><td>Testing on examples not used for training.</td><td>Required before promoting Qwen/Cosmos results to public evidence.</td><td>Training-set loss.</td></tr>
                 </tbody>
               </table>
             </div>
             <div class="glossary-links">
-              <a href="https://github.com/ChaoYue0307/ropedia-xperience-10m-task-suite/blob/main/GLOSSARY.md">Open full glossary</a>
-              <a href="data/glossary.json">Open glossary data</a>
+              <a href="https://github.com/ChaoYue0307/ropedia-xperience-10m-task-suite/blob/main/GLOSSARY.md">Open complete glossary note</a>
+              <a href="data/glossary.json">Open structured glossary data</a>
               <a href="https://github.com/ChaoYue0307/ropedia-xperience-10m-task-suite/blob/main/PUBLIC_READER_MAP.md">Reader map</a>
               <a href="data/task_method_20_gap_audit.json">Score/proxy audit</a>
             </div>
@@ -6764,7 +6770,7 @@
             </div>
             <div class="artifact-grid">
               <article class="artifact primary-artifact"><div><h3>Public reader map</h3><p>Single navigation view for GitHub, GitHub Pages, HF Space, artifact dataset, baseline model repo, Qwen3-Omni/Cosmos3 repos, and result-reading lanes.</p></div><a href="data/public_reader_map.json">reader map</a></article>
-              <article class="artifact primary-artifact"><div><h3>Glossary</h3><p>Definitions for evidence lines, windows, direct/proxy scores, Qwen v1-v6, Cosmos3 branches, adapters, and public mirror terms.</p></div><a href="data/glossary.json">glossary data</a><a href="https://github.com/ChaoYue0307/ropedia-xperience-10m-task-suite/blob/main/GLOSSARY.md">full glossary note</a></article>
+              <article class="artifact primary-artifact"><div><h3>Glossary</h3><p>Definitions for project-specific terms plus broader embodied-AI, egocentric multimodal data, spatial geometry, world-model, VLA, training, evaluation, adapter, and public mirror terms.</p></div><a href="data/glossary.json">glossary data</a><a href="https://github.com/ChaoYue0307/ropedia-xperience-10m-task-suite/blob/main/GLOSSARY.md">full glossary note</a></article>
               <article class="artifact primary-artifact"><div><h3>Artifact guide</h3><p>Human-readable map from project scope to data contract, task evidence, platform mirrors, and scale-up status.</p></div><a href="https://github.com/ChaoYue0307/ropedia-xperience-10m-task-suite/blob/main/ARTIFACT_GUIDE.md">artifact guide</a></article>
               <article class="artifact"><h3>Reproduction scripts</h3><p>Training, visualization, taxonomy, walkthrough, validator, and omni-readiness scripts.</p><a href="https://github.com/ChaoYue0307/ropedia-xperience-10m-task-suite/tree/main/scripts">scripts/</a></article>
               <article class="artifact"><h3>Hugging Face Space</h3><p>The dashboard packaged as a public static Space.</p><a href="https://huggingface.co/spaces/cy0307/ropedia-xperience-10m-task-suite">HF Space</a></article>
@@ -7733,6 +7739,47 @@ python scripts/validate_publication_package.py</code></pre>
       });
     }
 
+    function renderGlossaryTable(payload) {
+      const tableBody = document.getElementById("glossaryRows");
+      if (!tableBody || !Array.isArray(payload?.entries)) return;
+      const categoryLabels = new Map((payload.categories || []).map((category) => [category.id, category.label]));
+      tableBody.innerHTML = payload.entries.map((entry) => {
+        const category = categoryLabels.get(entry.category) || entry.category || "Glossary";
+        return `
+          <tr>
+            <td>${escapeHtml(entry.term)}<small>${escapeHtml(category)}</small></td>
+            <td>${escapeHtml(entry.plain_meaning || "")}</td>
+            <td>${escapeHtml(entry.project_usage || "")}</td>
+            <td>${escapeHtml(entry.do_not_confuse_with || "")}</td>
+          </tr>
+        `;
+      }).join("");
+    }
+
+    let glossaryDataPromise = null;
+
+    async function loadGlossaryData() {
+      if (!glossaryDataPromise) {
+        glossaryDataPromise = fetch("data/glossary.json", { cache: "no-cache" })
+          .then((response) => {
+            if (!response.ok) throw new Error(`glossary ${response.status}`);
+            return response.json();
+          });
+      }
+      const payload = await glossaryDataPromise;
+      registerGlossaryTerms(payload);
+      renderGlossaryTable(payload);
+      return payload;
+    }
+
+    async function initGlossaryData() {
+      try {
+        await loadGlossaryData();
+      } catch (error) {
+        // Keep the static fallback rows if the structured glossary is unavailable.
+      }
+    }
+
     function termDefinition(term) {
       const key = normalizeTermKey(term);
       const entry = glossaryTermMap.get(key);
@@ -7902,12 +7949,11 @@ python scripts/validate_publication_package.py</code></pre>
       const table = document.getElementById("resultScoreTable");
       if (!table) return;
       try {
-        const [response, glossaryResponse] = await Promise.all([
+        const [response] = await Promise.all([
           fetch("data/task_method_20_result_matrix.json", { cache: "no-cache" }),
-          fetch("data/glossary.json", { cache: "no-cache" }).catch(() => null)
+          loadGlossaryData().catch(() => null)
         ]);
         if (!response.ok) throw new Error(`matrix ${response.status}`);
-        if (glossaryResponse?.ok) registerGlossaryTerms(await glossaryResponse.json());
         renderResultScoreTable(await response.json());
       } catch (error) {
         table.querySelector("tbody").innerHTML = '<tr><td colspan="2">Result matrix could not be loaded. Open the source-data link above.</td></tr>';
@@ -8249,6 +8295,7 @@ python scripts/validate_publication_package.py</code></pre>
       });
     });
     labelResponsiveTables();
+    initGlossaryData();
     initResultMatrixTable();
     initTaskSurface();
 

index.html

@@ -2465,6 +2465,17 @@
       font-size: 15px;
       font-weight: 750;
     }
+    .glossary-table td:first-child small {
+      display: block;
+      margin-top: 6px;
+      color: var(--green);
+      font-family: var(--font-mono);
+      font-size: 10px;
+      font-weight: 800;
+      letter-spacing: 0.05em;
+      text-transform: uppercase;
+      opacity: 0.82;
+    }
     .glossary-table td:nth-child(2) {
       width: 32%;
       color: #d8e2d4;
@@ -6624,25 +6635,30 @@
     <section id="glossary" data-project-tab="resources" role="tabpanel" aria-labelledby="tab-resources" tabindex="-1">
       <div class="wrap">
         <div class="section-head">
-          <h2>Glossary for overloaded terms.</h2>
-          <p>These are the terms readers most often confuse when moving between the repo, website, Hugging Face mirrors, result matrices, and model-package cards. The full glossary is mirrored as a reader note and structured data.</p>
+          <h2>Glossary for project and field terms.</h2>
+          <p>This glossary covers the overloaded project terms plus adjacent technical terms from embodied AI, egocentric multimodal data, spatial geometry, world models, VLA/policy learning, training, evaluation, and public artifact reading.</p>
         </div>
         <div class="glossary-panel">
           <div class="glossary-summary" aria-label="Glossary categories">
             <article>
-              <small>scope</small>
-              <strong>Separate data, result lanes, and mirrors</strong>
-              <p>Evidence lines, public-safe artifacts, and gated upstream data are different objects. The glossary keeps those boundaries visible.</p>
+              <small>data</small>
+              <strong>Read multimodal sample terms</strong>
+              <p>Episode, window, modality, fisheye, depth, IMU, calibration, and synchronization terms are grouped with the project data boundary terms.</p>
+            </article>
+            <article>
+              <small>space + time</small>
+              <strong>Decode geometry and world-model language</strong>
+              <p>Camera pose, SLAM, point clouds, rollouts, forward dynamics, long-horizon forecasting, and temporal leakage are defined next to the relevant tasks.</p>
             </article>
             <article>
-              <small>results</small>
-              <strong>Read scores by source type</strong>
-              <p>Direct scores, compact-proxy scores, gap audits, and task-method records should not be interpreted as the same kind of evidence.</p>
+              <small>robotics</small>
+              <strong>Connect VLA and policy terms</strong>
+              <p>Action chunks, policies, imitation learning, behavior cloning, end effectors, dexterity, contact, and language grounding are included for extension readers.</p>
             </article>
             <article>
-              <small>models</small>
-              <strong>Keep branches distinct</strong>
-              <p>Minimal/NN heads, metadata/raw baselines, Qwen3-Omni v1-v6, Cosmos3-Super, Cosmos3-Nano, LoRA adapters, and full-parameter gates each mean something specific.</p>
+              <small>evidence</small>
+              <strong>Keep scores and public surfaces distinct</strong>
+              <p>Direct/proxy scores, raw metrics, radar values, held-out evaluation, Qwen/Cosmos branches, adapters, and HF mirrors remain explicitly separated.</p>
             </article>
           </div>
           <div>
@@ -6656,29 +6672,19 @@
                     <th>Do not confuse with</th>
                   </tr>
                 </thead>
-                <tbody>
-                  <tr><td>Evidence line</td><td>A reading lane for a group of results.</td><td>Line 1 is the public sample episode; Line 2 is selected-128 held-out comparison.</td><td>Qwen v1-v6 run versions.</td></tr>
-                  <tr><td>Public sample episode</td><td>The one fully inspectable official sample episode.</td><td>Raw-file browsing, task construction, single-episode baselines.</td><td>The selected-128 comparison rows.</td></tr>
-                  <tr><td>Selected 128 episodes</td><td>Public-safe derived features linked to official gated episode paths.</td><td>Same-split Line 2 baseline/model comparisons.</td><td>Redistributed raw MP4/HDF5/RRD files.</td></tr>
-                  <tr><td>20-frame window</td><td>A fixed short clip slice used as a model input unit.</td><td>Feature rows, labels, tasks, and many baseline heads.</td><td>A full episode.</td></tr>
-                  <tr><td>Task-method record</td><td>One method evaluated on one task.</td><td>The 9 x 20 public matrix, now 180 scored records.</td><td>A single prediction row.</td></tr>
-                  <tr><td>Direct score</td><td>A metric computed against the task target directly.</td><td>Primary interpretation in the result matrix.</td><td>Compact-proxy score.</td></tr>
-                  <tr><td>Compact-proxy score</td><td>A bounded proxy when the direct raw target is not public.</td><td>Explicitly marked cells in the gap audit and matrix.</td><td>A direct target measurement.</td></tr>
-                  <tr><td>Raw metric value</td><td>The original value emitted by the runner or verified package.</td><td>The value to cite from the 180-result table.</td><td>Normalized radar value.</td></tr>
-                  <tr><td>Normalized radar value</td><td>A 0-1 plotting value used only for comparable radar polygons.</td><td>Visual comparison across metrics with different scales.</td><td>The raw metric value to cite.</td></tr>
-                  <tr><td>Minimal baseline</td><td>A simple non-neural task head; the "minimum" reference row in casual wording.</td><td>Single-episode lower-complexity comparison.</td><td>Selected-128 Simple baseline rows.</td></tr>
-                  <tr><td>Simple baseline</td><td>A non-neural selected-128 baseline family.</td><td>Metadata/text and raw-feature 128-episode comparisons before NN/foundation rows.</td><td>The single-episode Minimal baseline.</td></tr>
-                  <tr><td>Qwen3-Omni v6</td><td>The current public Qwen 20-task row.</td><td>Qwen3-Omni LoRA plus task-specific probes.</td><td>All Qwen v1-v6 experiments.</td></tr>
-                  <tr><td>Cosmos3-Super</td><td>The larger Cosmos-style branch.</td><td>Reasoner diagnostics and a verified forward-dynamics LoRA branch.</td><td>Cosmos3-Nano Future Window.</td></tr>
-                  <tr><td>LoRA adapter</td><td>Lightweight trainable adapter weights.</td><td>Public model-branch artifacts when verified.</td><td>Full base-model weights.</td></tr>
-                  <tr><td>HF artifact dataset</td><td>Hugging Face dataset repo for derived evidence.</td><td>Reports, metrics, website data, sanitized result packages.</td><td>The upstream Xperience-10M dataset.</td></tr>
-                  <tr><td>Mirror parity</td><td>A check that public copies match source files.</td><td>Verifying GitHub, website, and HF mirrors.</td><td>A model-quality metric.</td></tr>
+                <tbody id="glossaryRows">
+                  <tr><td>Egocentric video<small>multimodal sensing</small></td><td>Video captured from a first-person or body-mounted viewpoint.</td><td>The sample streams are egocentric views of human interaction and are the visual basis for many tasks.</td><td>Third-person robot-camera footage.</td></tr>
+                  <tr><td>Camera pose<small>spatial geometry</small></td><td>The camera position and orientation at a time step.</td><td>Supports spatial-intelligence tasks, view synchronization, and geometry diagnostics.</td><td>The human body pose.</td></tr>
+                  <tr><td>Forward dynamics<small>temporal and world models</small></td><td>Predicting the next state from the current state and action/context.</td><td>The Cosmos3-Super LoRA branch uses a forward-dynamics-style diagnostic contract.</td><td>Reverse inference from result back to cause.</td></tr>
+                  <tr><td>Vision-language-action model<small>robotics and VLA</small></td><td>A model that maps visual context and language into actions.</td><td>The VLA direction is a future path after action targets are converted into robot-compatible chunks.</td><td>A vision-language model that only answers text.</td></tr>
+                  <tr><td>Direct score<small>tasks and metrics</small></td><td>A metric computed against the task target directly.</td><td>The preferred score type in the 20-task matrix.</td><td>Compact-proxy score.</td></tr>
+                  <tr><td>Held-out evaluation<small>training and evaluation</small></td><td>Testing on examples not used for training.</td><td>Required before promoting Qwen/Cosmos results to public evidence.</td><td>Training-set loss.</td></tr>
                 </tbody>
               </table>
             </div>
             <div class="glossary-links">
-              <a href="https://github.com/ChaoYue0307/ropedia-xperience-10m-task-suite/blob/main/GLOSSARY.md">Open full glossary</a>
-              <a href="data/glossary.json">Open glossary data</a>
+              <a href="https://github.com/ChaoYue0307/ropedia-xperience-10m-task-suite/blob/main/GLOSSARY.md">Open complete glossary note</a>
+              <a href="data/glossary.json">Open structured glossary data</a>
               <a href="https://github.com/ChaoYue0307/ropedia-xperience-10m-task-suite/blob/main/PUBLIC_READER_MAP.md">Reader map</a>
               <a href="data/task_method_20_gap_audit.json">Score/proxy audit</a>
             </div>
@@ -6764,7 +6770,7 @@
             </div>
             <div class="artifact-grid">
               <article class="artifact primary-artifact"><div><h3>Public reader map</h3><p>Single navigation view for GitHub, GitHub Pages, HF Space, artifact dataset, baseline model repo, Qwen3-Omni/Cosmos3 repos, and result-reading lanes.</p></div><a href="data/public_reader_map.json">reader map</a></article>
-              <article class="artifact primary-artifact"><div><h3>Glossary</h3><p>Definitions for evidence lines, windows, direct/proxy scores, Qwen v1-v6, Cosmos3 branches, adapters, and public mirror terms.</p></div><a href="data/glossary.json">glossary data</a><a href="https://github.com/ChaoYue0307/ropedia-xperience-10m-task-suite/blob/main/GLOSSARY.md">full glossary note</a></article>
+              <article class="artifact primary-artifact"><div><h3>Glossary</h3><p>Definitions for project-specific terms plus broader embodied-AI, egocentric multimodal data, spatial geometry, world-model, VLA, training, evaluation, adapter, and public mirror terms.</p></div><a href="data/glossary.json">glossary data</a><a href="https://github.com/ChaoYue0307/ropedia-xperience-10m-task-suite/blob/main/GLOSSARY.md">full glossary note</a></article>
               <article class="artifact primary-artifact"><div><h3>Artifact guide</h3><p>Human-readable map from project scope to data contract, task evidence, platform mirrors, and scale-up status.</p></div><a href="https://github.com/ChaoYue0307/ropedia-xperience-10m-task-suite/blob/main/ARTIFACT_GUIDE.md">artifact guide</a></article>
               <article class="artifact"><h3>Reproduction scripts</h3><p>Training, visualization, taxonomy, walkthrough, validator, and omni-readiness scripts.</p><a href="https://github.com/ChaoYue0307/ropedia-xperience-10m-task-suite/tree/main/scripts">scripts/</a></article>
               <article class="artifact"><h3>Hugging Face Space</h3><p>The dashboard packaged as a public static Space.</p><a href="https://huggingface.co/spaces/cy0307/ropedia-xperience-10m-task-suite">HF Space</a></article>
@@ -7733,6 +7739,47 @@ python scripts/validate_publication_package.py</code></pre>
       });
     }
 
+    function renderGlossaryTable(payload) {
+      const tableBody = document.getElementById("glossaryRows");
+      if (!tableBody || !Array.isArray(payload?.entries)) return;
+      const categoryLabels = new Map((payload.categories || []).map((category) => [category.id, category.label]));
+      tableBody.innerHTML = payload.entries.map((entry) => {
+        const category = categoryLabels.get(entry.category) || entry.category || "Glossary";
+        return `
+          <tr>
+            <td>${escapeHtml(entry.term)}<small>${escapeHtml(category)}</small></td>
+            <td>${escapeHtml(entry.plain_meaning || "")}</td>
+            <td>${escapeHtml(entry.project_usage || "")}</td>
+            <td>${escapeHtml(entry.do_not_confuse_with || "")}</td>
+          </tr>
+        `;
+      }).join("");
+    }
+
+    let glossaryDataPromise = null;
+
+    async function loadGlossaryData() {
+      if (!glossaryDataPromise) {
+        glossaryDataPromise = fetch("data/glossary.json", { cache: "no-cache" })
+          .then((response) => {
+            if (!response.ok) throw new Error(`glossary ${response.status}`);
+            return response.json();
+          });
+      }
+      const payload = await glossaryDataPromise;
+      registerGlossaryTerms(payload);
+      renderGlossaryTable(payload);
+      return payload;
+    }
+
+    async function initGlossaryData() {
+      try {
+        await loadGlossaryData();
+      } catch (error) {
+        // Keep the static fallback rows if the structured glossary is unavailable.
+      }
+    }
+
     function termDefinition(term) {
       const key = normalizeTermKey(term);
       const entry = glossaryTermMap.get(key);
@@ -7902,12 +7949,11 @@ python scripts/validate_publication_package.py</code></pre>
       const table = document.getElementById("resultScoreTable");
       if (!table) return;
       try {
-        const [response, glossaryResponse] = await Promise.all([
+        const [response] = await Promise.all([
           fetch("data/task_method_20_result_matrix.json", { cache: "no-cache" }),
-          fetch("data/glossary.json", { cache: "no-cache" }).catch(() => null)
+          loadGlossaryData().catch(() => null)
         ]);
         if (!response.ok) throw new Error(`matrix ${response.status}`);
-        if (glossaryResponse?.ok) registerGlossaryTerms(await glossaryResponse.json());
         renderResultScoreTable(await response.json());
       } catch (error) {
         table.querySelector("tbody").innerHTML = '<tr><td colspan="2">Result matrix could not be loaded. Open the source-data link above.</td></tr>';
@@ -8249,6 +8295,7 @@ python scripts/validate_publication_package.py</code></pre>
       });
     });
     labelResponsiveTables();
+    initGlossaryData();
     initResultMatrixTable();
     initTaskSurface();