Add Xperience embodied foundation pretraining goal

ARTIFACT_GUIDE.md

@@ -3,15 +3,17 @@
 This guide is the human-readable map for the public Ropedia Xperience-10M task
 suite artifacts. It is organized around what a reader usually wants to do:
 understand the project, inspect the sample episode, compare baselines, read the
-task results, and follow the Qwen3-Omni scale-up path.
+task results, follow the Qwen3-Omni scale-up path, and understand the longer
+Xperience-native pretraining goal.
 
 ## Start Here
 
 | Artifact | Why to open it first |
 | --- | --- |
 | [`PROJECT_STATUS.md`](PROJECT_STATUS.md) | Gives the fastest current-state table: implemented, in staging, and outside current scope. |
-| [`RESEARCH_ROADMAP.md`](RESEARCH_ROADMAP.md) | Shows the roadmap from public-sample task development to multi-episode data preparation, Qwen3-Omni LoRA, robustness runs, and larger omni-model extensions. |
+| [`RESEARCH_ROADMAP.md`](RESEARCH_ROADMAP.md) | Shows the roadmap from public-sample task development to multi-episode data preparation, Qwen3-Omni LoRA, robustness runs, model branches, and the future native-pretraining goal. |
 | [`FOUNDATION_MODEL_PLAN.md`](FOUNDATION_MODEL_PLAN.md) | Explains which foundation backbones fit which Xperience-10M objective: Qwen3-Omni first, Cosmos 3 for world modeling, and VLA/policy models after action-target conversion. |
+| [`XPERIENCE_EMBODIED_FOUNDATION_MODEL_PRETRAINING.md`](XPERIENCE_EMBODIED_FOUNDATION_MODEL_PRETRAINING.md) | Describes the future full-corpus Xperience Embodied Foundation Model goal, including modules, objectives, staged scale-up, hardware ranges, and evaluation. |
 | [`EVALUATION_PROTOCOL.md`](EVALUATION_PROTOCOL.md) | Defines the task unit, chronological split, metrics, leakage controls, and current limitations. |
 | [`REPRODUCIBILITY.md`](REPRODUCIBILITY.md) | Defines public reproduction commands, expected outputs, and unreproducible boundaries. |
 | [`results/audio_ablation/AUDIO_ABLATION_SUMMARY.md`](results/audio_ablation/AUDIO_ABLATION_SUMMARY.md) | Shows measured current-audio and raw log-mel replacement deltas across the 12 task contracts. |
@@ -107,6 +109,7 @@ research project.
 | [`scripts/omni/train_qwen3_omni_lora.py`](scripts/omni/train_qwen3_omni_lora.py) | Training entrypoint for the Qwen3-Omni LoRA pilot after the data gate passes. |
 | [`FOUNDATION_MODEL_PLAN.md`](FOUNDATION_MODEL_PLAN.md) | Adds the post-data-gate backbone selection plan: Qwen3-Omni first, Cosmos 3 for world modeling, and OpenVLA/openpi/GR00T for policy/action branches. |
 | [`docs/data/foundation_model_plan.json`](docs/data/foundation_model_plan.json) | Machine-readable model-family registry with source links, entry conditions, and evaluation additions. |
+| [`XPERIENCE_EMBODIED_FOUNDATION_MODEL_PRETRAINING.md`](XPERIENCE_EMBODIED_FOUNDATION_MODEL_PRETRAINING.md) | Future full-corpus Xperience-native pretraining plan; not a current model result. |
 
 ## What Is Not Included
 

FOUNDATION_MODEL_PLAN.md

@@ -20,6 +20,7 @@ run a held-out multi-episode foundation-model evaluation.
 | 5 | openpi pi0/pi0.5 | Open robot policy and action expert baseline | Useful for action chunking, policy fine-tuning, and embodiment transfer experiments | Candidate for policy branch once action labels are retargeted |
 | 6 | Gemini Robotics | Closed/API embodied reasoning reference | Strong candidate for qualitative reasoning and task interpretation, but not a local fine-tune target | Use only as an external comparison or annotation assistant |
 | 7 | Octo / SmolVLA-style lightweight policies | Smaller reproducible robot-policy baselines | Good for cheaper action-policy experiments, but less directly omni-modal | Optional baseline branch after selected-episode data preparation |
+| Future | Xperience Embodied Foundation Model | Xperience-native domain model pretrained from scratch on full-corpus embodied experience | Would learn a shared temporal representation across video, audio, depth, pose, mocap, IMU, and language | Long-term goal after smaller pilots prove value and full-corpus storage/compute are available |
 
 ## Why Qwen3-Omni Still Goes First
 
@@ -38,6 +39,46 @@ prepare video/audio/language prompts and adapter inputs. It is also suitable for
 the 12 current task contracts, which mostly produce labels, structured JSON, or
 short task answers.
 
+The executable Qwen branch and future branch contracts are now represented as
+config files under `configs/omni_backbones/`. Validate them with:
+
+```bash
+python scripts/omni/backbone_registry.py --validate --json
+```
+
+The shared extension rules are in
+[`OMNI_MODEL_EXTENSION_CONTRACT.md`](OMNI_MODEL_EXTENSION_CONTRACT.md). A new
+foundation branch should add a config first, then implement the exporter,
+trainer, evaluator, and launcher required by that config.
+
+## Long-Term Native Pretraining Goal
+
+Qwen3-Omni, Cosmos 3, GR00T, OpenVLA, and openpi are backbone choices for the
+next experiments. The longer-term goal is different: train an
+**Xperience Embodied Foundation Model** that is native to the Xperience-10M
+modality structure.
+
+That model would not start as a general internet-scale omni model. It would be
+a domain model over synchronized embodied experience: multi-view egocentric
+video, audio, depth, pose/SLAM, hand and body mocap, IMU, calibration, and
+language annotations. Its pretraining should combine masked multimodal
+modeling, cross-modal contrastive alignment, future-state prediction,
+ego-motion and hand-motion forecasting, action/procedure prediction, language
+grounding, contact/affordance prediction, and optional policy-style targets
+after action conversion.
+
+This is not a current result in the repo. It becomes appropriate only after:
+
+- the selected multi-episode pipeline trains and evaluates cleanly,
+- scaling from 128 episodes to thousands of episodes shows measurable value,
+- raw-corpus storage and derived-shard capacity are available,
+- distributed training and checkpoint/restart infrastructure are reliable,
+- evaluation covers held-out episodes, sessions, activities, objects, and
+  missing-modality robustness.
+
+The full plan is documented in
+[`XPERIENCE_EMBODIED_FOUNDATION_MODEL_PRETRAINING.md`](XPERIENCE_EMBODIED_FOUNDATION_MODEL_PRETRAINING.md).
+
 ## Why Cosmos 3 Should Be Added Next
 
 Cosmos 3 should not replace the Qwen3-Omni pilot. It should become the first
@@ -105,6 +146,9 @@ The foundation-model stage should add metrics beyond the current 12-task suite:
    retargeting artifacts are traceable.
 6. Update public cards only when a branch has real manifests, predictions,
    metrics, and qualitative examples.
+7. Start Xperience-native pretraining only after smaller scaling stages,
+   full-corpus storage, multi-node compute, and held-out evaluation protocols
+   are in place.
 
 ## Source Links
 
@@ -116,3 +160,5 @@ The foundation-model stage should add metrics beyond the current 12-task suite:
 - Gemini Robotics: https://deepmind.google/discover/blog/gemini-robotics-brings-ai-into-the-physical-world/
 - Octo: https://octo-models.github.io/
 - LeRobot / SmolVLA: https://github.com/huggingface/lerobot
+- Xperience Embodied Foundation Model pretraining plan:
+  `XPERIENCE_EMBODIED_FOUNDATION_MODEL_PRETRAINING.md`

PROJECT_README.md

@@ -42,7 +42,7 @@ embodied-AI research infrastructure:
 | Multimodal data understanding | Parses the public sample into synchronized windows across video, audio, depth, pose/SLAM, mocap, IMU, calibration, and language-derived signals |
 | Task design | Defines 12 human-readable tasks plus four direction-extension probes with inputs, outputs, process modules, metrics, and case-study walkthroughs |
 | Model and evaluation discipline | Runs minimal and compact neural baselines, records predictions/metrics, keeps chronological split boundaries explicit, and separates sample evidence from held-out claims |
-| Scale-up planning | Connects the public-sample pipeline to 32/128-episode held-out pilots, Qwen3-Omni LoRA, Cosmos-style world-model branches, and later policy-model branches |
+| Scale-up planning | Connects the public-sample pipeline to 32/128-episode held-out pilots, Qwen3-Omni LoRA, Cosmos-style world-model branches, policy-model branches, and the future Xperience-native foundation-model pretraining goal |
 
 ## Start Here
 
@@ -59,6 +59,7 @@ before the multi-episode omni-model stage becomes a real held-out evaluation.
 | Navigate the 12 tasks, four tracks, and scale-up plan | [Interactive research roadmap](https://chaoyue0307.github.io/ropedia-xperience-10m-task-suite/research_roadmap.html), [`docs/data/research_roadmap_interactive.json`](docs/data/research_roadmap_interactive.json) |
 | Compare current task metrics | [`RESEARCH_TAKEAWAYS.md`](RESEARCH_TAKEAWAYS.md), [`docs/data/summary_metrics.json`](docs/data/summary_metrics.json) |
 | Compare possible foundation backbones | [`FOUNDATION_MODEL_PLAN.md`](FOUNDATION_MODEL_PLAN.md), [`docs/data/foundation_model_plan.json`](docs/data/foundation_model_plan.json) |
+| Understand the future native pretraining goal | [`XPERIENCE_EMBODIED_FOUNDATION_MODEL_PRETRAINING.md`](XPERIENCE_EMBODIED_FOUNDATION_MODEL_PRETRAINING.md) |
 | Understand one model input | [`results/episode_task_suite/feature_manifest.json`](results/episode_task_suite/feature_manifest.json), [`results/episode_task_suite/windows.csv`](results/episode_task_suite/windows.csv) |
 | Check multi-episode data status | [`results/omni_finetune/DATA_ACCESS_STATUS.md`](results/omni_finetune/DATA_ACCESS_STATUS.md) |
 
@@ -71,7 +72,7 @@ before the multi-episode omni-model stage becomes a real held-out evaluation.
 | Task suite | 12 human-readable embodied-AI task contracts with input, process, output, metrics, predictions, and case-study walkthroughs |
 | Baselines | Minimal linear/ridge/logistic heads plus compact PyTorch MLP task heads over the same chronological split |
 | Research directions | Task mapping and extension probes for human modeling, 3D/4D reconstruction, egocentric interaction, and world modeling |
-| Scale-up path | The gated Xperience-10M dataset is available for a selected 128-episode pilot before Qwen3-Omni LoRA, followed by Cosmos 3/world-model and VLA/policy branches |
+| Scale-up path | The gated Xperience-10M dataset is available for a selected 128-episode pilot before Qwen3-Omni LoRA, followed by Cosmos 3/world-model and VLA/policy branches; the long-term goal is an Xperience-native embodied foundation model if full-corpus data, storage, and compute are available |
 | Public surfaces | GitHub repo, GitHub Pages dashboard, HF Space, HF artifact dataset, HF baseline-model repo, and HF collection |
 
 For the fastest interpretation of the current metrics, start with
@@ -93,100 +94,27 @@ Current contributions:
 - human-readable research task cards and an interactive scrub/play walkthrough storyboard for every task,
 - an interactive research roadmap connecting 12 tasks, four research tracks, current sample evidence, the Qwen3-Omni scale-up path, and foundation-model branch selection,
 - a next-milestone track for Qwen3-Omni fine-tuning, Cosmos 3 world modeling, and sensor-bridge evaluation,
+- a future pretraining plan for an Xperience Embodied Foundation Model over the full corpus after smaller multi-episode stages prove value,
 - metrics, predictions, model weights, manifests, charts, and a two-level
   tabbed static research website,
 - a clear explanation of what is implemented now and what moves to the multi-episode stage.
 
 ## Current Research Scope
 
-This repo separates implemented single-episode research artifacts from future
-multi-episode held-out model metrics:
+This project is best read as a staged embodied-AI research study:
 
-| Project layer | Evidence | Current scope |
+| Layer | Current scope | Where to start |
 | --- | --- | --- |
-| Official Xperience-10M description | `XPERIENCE10M_DATASET_CARD_ALIGNMENT.md`, `docs/data/xperience10m_dataset_card_alignment.json` | aligns public wording with the official gated dataset card, public sample card, and HF API metadata; does not mirror raw data |
-| Source alignment | `SOURCE_ALIGNMENT_AUDIT.md`, `docs/data/source_alignment_audit.json`, `scripts/validate_source_alignment.py` | records the same official dataset facts, public sample details, API-listing notes, and project coverage across repo, website, and HF cards |
-| Figure index | `FIGURE_INDEX.md`, `docs/data/figure_index.json`, `scripts/build_figure_index.py` | catalogs public figures, charts, modality thumbnails, dimensions, hashes, roles, and source scripts |
-| Brand assets | `docs/assets/brand/`, `docs/favicon.png`, `docs/apple-touch-icon.png`, `scripts/build_brand_assets.py` | applies the generated project logo system across the website, README, HF cards, favicon, and social previews |
-| Data windows | `results/episode_task_suite/windows.csv`, `shared_windows.npz`, `summary_report.json` | one public sample episode |
-| Feature contract | `results/episode_task_suite/feature_manifest.json`, `available_modalities.json` | documents the 8,546-dimensional multimodal representation and source coverage |
-| Evaluation protocol | `EVALUATION_PROTOCOL.md`, `docs/data/evaluation_protocol.json`, `scripts/build_evaluation_protocol.py` | defines windowing, chronological split, leakage controls, per-task metrics, and current limitations |
-| Research Takeaways | `RESEARCH_TAKEAWAYS.md`, `docs/data/research_takeaways.json`, `scripts/build_research_takeaways.py` | summarizes result interpretation from committed metrics and identifies which experiments need held-out episodes |
-| Audio ablation | `scripts/audio_ablation_and_raw_upgrade.py`, `results/audio_ablation/`, `docs/data/audio_ablation_summary.json` | measures whether audio helps each of the 12 task contracts |
-| Research roadmap | `RESEARCH_ROADMAP.md`, `docs/research_roadmap.html`, `docs/data/research_roadmap.json`, `docs/data/research_roadmap_interactive.json` | stages and visualizes the path from public-sample task development to multi-episode held-out evaluation, foundation-model selection, and larger omni/world-model extensions |
-| Foundation-model plan | `FOUNDATION_MODEL_PLAN.md`, `docs/data/foundation_model_plan.json` | keeps Qwen3-Omni as the first trainable pilot, adds Cosmos 3 as the first world-model branch, and tracks OpenVLA/openpi/GR00T policy candidates |
-| 12-task suite | `scripts/episode_task_suite.py`, per-task `metrics.json`, predictions | chronological single-episode split |
-| Single-episode diagnostics | `scripts/single_episode_diagnostics.py`, `results/single_episode_diagnostics/`, `docs/single_episode_explorer.html` | modality ablations, timeline overlay, object-label export, alignment stress tests, and interactive window inspection from one sample episode |
-| Neural heads | `scripts/neural_task_models.py`, `results/episode_task_suite/neural_mlp/` | compact MLP heads, not a foundation model |
-| Research directions | `research_direction_taxonomy.json`, extension probe results | direct/proxy/diagnostic evidence, not full solutions |
-| Task surface integrity | `docs/data/task_surface_integrity.json`, `scripts/validate_task_surface.py` | public task cards stay human-readable, thumbnail-backed, and wired to the scrub/play walkthrough storyboard |
-| Rendered website check | `RENDERED_SITE_CHECK.md`, `docs/data/rendered_site_check.json`, `scripts/build_rendered_site_check.py` | records a browser-level load, tab, walkthrough deep-link, control-click, and console-health check |
-| Public project surface | `PUBLIC_SURFACE_QA.md`, `docs/data/public_surface_qa.json`, `scripts/build_public_surface_qa.py` | presents the repo, website, and Hugging Face cards as one research project surface |
-| Qwen3-Omni | `results/omni_finetune/DATA_ACCESS_STATUS.md`, `MULTI_EPISODE_ACCESS_STATUS.md` | the gated full dataset is available for a selected 128-episode pilot before held-out evaluation |
-| Multi-episode pilot status | `scripts/validate_scope_claims.py`, `docs/data/scope_claims_audit.json` | separates setup artifacts, selected-episode preparation, and completed held-out-episode metrics |
-| Mirror parity | `scripts/validate_mirror_parity.py`, `docs/data/mirror_parity.json` | prepared GitHub/HF mirrors carry matching data, figure, website HTML, and validator files |
-| Public bundle contents | `scripts/validate_publication_package.py`, `docs/data/publication_audit.json` | summarizes the public repo and HF bundles, including raw-data exclusion and temporary local-file exclusion |
-| Release checks | `QUALITY_GATES.md`, `docs/data/quality_gates.json`, `metrics/quality_gates.json`, `scripts/build_quality_gates.py` | one map for automated checks and live post-publish verification; the `metrics/` path is the Hugging Face model-repo mirror |
-| Artifact index | `scripts/build_artifact_index.py`, `docs/data/artifact_index.json` | selective source-of-truth catalog with existence, size, and stable-file hashes |
-| Project status | `PROJECT_STATUS.md`, `docs/data/project_status.json` | compact current-state table for first-pass readers |
-| Citation and metadata | `CITATION.cff`, `codemeta.json`, `docs/data/project_manifest.json`, `LICENSE` | code is MIT-scoped; raw-data use follows Xperience-10M terms |
-| Project path | `docs/data/project_packet.json`, website project path section | navigation guide across data, tasks, results, and scale-up status |
-
-Read the full scope note in [`EVIDENCE_CONTRACT.md`](EVIDENCE_CONTRACT.md), or
-consume the machine-readable copy at
-[`docs/data/evidence_contract.json`](docs/data/evidence_contract.json).
-The current release package report is at
-[`docs/data/publication_audit.json`](docs/data/publication_audit.json).
-The release-check summary is at
-[`QUALITY_GATES.md`](QUALITY_GATES.md) and
-[`docs/data/quality_gates.json`](docs/data/quality_gates.json).
-The last live-publication verification report is at
-[`docs/data/live_publication_status.json`](docs/data/live_publication_status.json).
-The current prepared-mirror parity report is at
-[`docs/data/mirror_parity.json`](docs/data/mirror_parity.json).
-The current multi-episode pilot status note is at
-[`docs/data/scope_claims_audit.json`](docs/data/scope_claims_audit.json).
-The task-card and walkthrough-storyboard integrity report is at
-[`docs/data/task_surface_integrity.json`](docs/data/task_surface_integrity.json).
-The public presentation report is at
-[`PUBLIC_SURFACE_QA.md`](PUBLIC_SURFACE_QA.md) and
-[`docs/data/public_surface_qa.json`](docs/data/public_surface_qa.json).
-The generated evaluation protocol is at
-[`EVALUATION_PROTOCOL.md`](EVALUATION_PROTOCOL.md) and
-[`docs/data/evaluation_protocol.json`](docs/data/evaluation_protocol.json).
-The generated research takeaways are at
-[`RESEARCH_TAKEAWAYS.md`](RESEARCH_TAKEAWAYS.md) and
-[`docs/data/research_takeaways.json`](docs/data/research_takeaways.json).
-The research roadmap is at
-[`RESEARCH_ROADMAP.md`](RESEARCH_ROADMAP.md) and
-[`docs/data/research_roadmap.json`](docs/data/research_roadmap.json).
-The foundation-model selection plan is at
-[`FOUNDATION_MODEL_PLAN.md`](FOUNDATION_MODEL_PLAN.md) and
-[`docs/data/foundation_model_plan.json`](docs/data/foundation_model_plan.json).
-The source-of-truth artifact index is at
-[`docs/data/artifact_index.json`](docs/data/artifact_index.json).
-For a human-readable artifact map, use
-[`ARTIFACT_GUIDE.md`](ARTIFACT_GUIDE.md).
-For reproduction commands and expected outputs, use
-[`REPRODUCIBILITY.md`](REPRODUCIBILITY.md) and
-[`docs/data/reproducibility_matrix.json`](docs/data/reproducibility_matrix.json).
-Project citation and machine-readable metadata live in
-[`CITATION.cff`](CITATION.cff), [`codemeta.json`](codemeta.json), and
-[`docs/data/project_manifest.json`](docs/data/project_manifest.json).
-The upstream dataset-card alignment note is
-[`XPERIENCE10M_DATASET_CARD_ALIGNMENT.md`](XPERIENCE10M_DATASET_CARD_ALIGNMENT.md),
-with a machine-readable copy at
-[`docs/data/xperience10m_dataset_card_alignment.json`](docs/data/xperience10m_dataset_card_alignment.json).
-The generated source-alignment note is at
-[`SOURCE_ALIGNMENT_AUDIT.md`](SOURCE_ALIGNMENT_AUDIT.md) and
-[`docs/data/source_alignment_audit.json`](docs/data/source_alignment_audit.json).
-The generated figure index is at
-[`FIGURE_INDEX.md`](FIGURE_INDEX.md) and
-[`docs/data/figure_index.json`](docs/data/figure_index.json).
-The project logo system is packaged by
-[`scripts/build_brand_assets.py`](scripts/build_brand_assets.py), stored under
-[`docs/assets/brand/`](docs/assets/brand/), and indexed in
-[`docs/data/brand_assets.json`](docs/data/brand_assets.json).
+| Data understanding | One public Xperience-10M sample episode is converted into 5,821 frames, 1,161 aligned windows, and an 8,546-dimensional multimodal representation. | [`PROJECT_BRIEF.md`](PROJECT_BRIEF.md), [`PROJECT_STATUS.md`](PROJECT_STATUS.md) |
+| Task suite | Twelve human-readable tasks cover action, procedure, contact, object, language, retrieval, reconstruction, order, and synchronization questions. | [`RESEARCH_TAKEAWAYS.md`](RESEARCH_TAKEAWAYS.md), [`results/episode_task_suite/summary_report.json`](results/episode_task_suite/summary_report.json) |
+| Baselines | Minimal heads and compact PyTorch MLP heads provide a first controlled comparison on the same chronological split. | [`results/episode_task_suite/neural_mlp/`](results/episode_task_suite/neural_mlp/) |
+| Diagnostics | Audio contribution, modality ablations, timeline overlays, object labels, and alignment stress tests show which signals are useful and which tasks remain hard. | [`results/audio_ablation/AUDIO_ABLATION_SUMMARY.md`](results/audio_ablation/AUDIO_ABLATION_SUMMARY.md), [`docs/single_episode_explorer.html`](docs/single_episode_explorer.html) |
+| Scale-up | A selected 128-episode Qwen3-Omni LoRA pilot is being prepared from the gated dataset; held-out model metrics will be added only after training and evaluation finish. The long-term native-pretraining plan is documented separately as a future research goal. | [`RESEARCH_ROADMAP.md`](RESEARCH_ROADMAP.md), [`FOUNDATION_MODEL_PLAN.md`](FOUNDATION_MODEL_PLAN.md), [`XPERIENCE_EMBODIED_FOUNDATION_MODEL_PRETRAINING.md`](XPERIENCE_EMBODIED_FOUNDATION_MODEL_PRETRAINING.md), [`results/omni_finetune/DATA_ACCESS_STATUS.md`](results/omni_finetune/DATA_ACCESS_STATUS.md) |
+
+Detailed dataset notes, reproduction checks, and generated JSON reports are
+included for readers who want to inspect the implementation, but they are
+supporting materials rather than the main reading path. Use
+[`ARTIFACT_GUIDE.md`](ARTIFACT_GUIDE.md) when you want the full file map.
 
 ## Project Status
 
@@ -200,10 +128,9 @@ They give the current research state in one compact table:
 | Public-sample pipeline | Verified on one public sample episode: 5,821 frames, 1,161 windows, 8,546 dimensions |
 | 12-task suite | Verified minimal baselines with committed metrics, predictions, and manifests |
 | Neural heads | Verified compact PyTorch MLP heads over the same task contracts and chronological splits |
-| Official dataset wording | Verified against the public `ropedia-ai/xperience-10m` dataset card/API metadata |
-| Source alignment | Source facts, sample details, API-listing notes, and project coverage are consistent across repo, website, and HF cards |
+| Dataset context | Official Xperience-10M links, sample-vs-gated-data boundary, modality coverage, and redistribution policy are documented |
 | Evaluation protocol | Verified generated protocol for windowing, split policy, leakage controls, and per-task metrics |
-| Website and HF mirrors | Verified by website reference reports, public presentation reports, mirror parity, and live-publication checks; the public dashboard uses six top-level tabs, including an explicit Directions tab, plus subsection tabs for dataset, task-suite, method, result, direction, and resource views |
+| Website and Hub pages | Public dashboard, Hugging Face Space, artifact dataset, baseline model repo, and collection use the same project framing and links |
 | Qwen3-Omni multi-episode pilot | The gated Xperience-10M dataset is available for selected 128-episode preparation, with full metrics pending completed preprocessing, training, and held-out evaluation |
 | Raw Xperience-10M data / full Qwen weights | Not redistributed |
 
@@ -213,33 +140,31 @@ If you are reading the project cold, open these in order:
 
 | Step | Question | Primary artifacts | What should be true |
 | --- | --- | --- | --- |
-| 1 | What has been implemented? | [`PROJECT_BRIEF.md`](PROJECT_BRIEF.md), [`PROJECT_STATUS.md`](PROJECT_STATUS.md), [`docs/data/project_status.json`](docs/data/project_status.json), [`ARTIFACT_GUIDE.md`](ARTIFACT_GUIDE.md), [`docs/data/artifact_index.json`](docs/data/artifact_index.json), [`docs/data/figure_index.json`](docs/data/figure_index.json) | Single-episode task engineering, visual assets, mirrors, and scale-up status are summarized for first-pass reading. |
-| 2 | What is the official upstream dataset? | [`XPERIENCE10M_DATASET_CARD_ALIGNMENT.md`](XPERIENCE10M_DATASET_CARD_ALIGNMENT.md), [`docs/data/xperience10m_dataset_card_alignment.json`](docs/data/xperience10m_dataset_card_alignment.json), [official HF dataset](https://huggingface.co/datasets/ropedia-ai/xperience-10m) | The full dataset is described as a gated large-scale 4D multimodal egocentric source; this repo validates only one public sample episode. |
-| 3 | Are source facts consistently presented? | [`SOURCE_ALIGNMENT_AUDIT.md`](SOURCE_ALIGNMENT_AUDIT.md), [`docs/data/source_alignment_audit.json`](docs/data/source_alignment_audit.json), [`scripts/validate_source_alignment.py`](scripts/validate_source_alignment.py) | Repo, website, and HF cards use the same full-dataset facts, sample-card facts, API-listing notes, and project coverage. |
-| 4 | How exactly are tasks evaluated? | [`EVALUATION_PROTOCOL.md`](EVALUATION_PROTOCOL.md), [`docs/data/evaluation_protocol.json`](docs/data/evaluation_protocol.json), [`scripts/build_evaluation_protocol.py`](scripts/build_evaluation_protocol.py) | The window unit, chronological split, leakage controls, task metrics, and current limitations are explicit. |
-| 5 | What do the current results mean? | [`RESEARCH_TAKEAWAYS.md`](RESEARCH_TAKEAWAYS.md), [`docs/data/research_takeaways.json`](docs/data/research_takeaways.json), [`docs/data/summary_metrics.json`](docs/data/summary_metrics.json) | The takeaways are generated from committed metrics and identify which signals are ready for larger held-out experiments. |
-| 6 | What is the research roadmap? | [`RESEARCH_ROADMAP.md`](RESEARCH_ROADMAP.md), [`docs/data/research_roadmap.json`](docs/data/research_roadmap.json), [`DATA_ACCESS_STATUS.md`](results/omni_finetune/DATA_ACCESS_STATUS.md) | The roadmap connects public-sample task development to multi-episode data preparation, Qwen3-Omni LoRA, foundation-model selection, robustness runs, and larger omni/world-model extensions. |
-| 7 | Which foundation model comes next? | [`FOUNDATION_MODEL_PLAN.md`](FOUNDATION_MODEL_PLAN.md), [`docs/data/foundation_model_plan.json`](docs/data/foundation_model_plan.json) | Qwen3-Omni remains the first held-out LoRA baseline; Cosmos 3 is the first world-model branch; OpenVLA/openpi/GR00T wait for explicit action targets. |
-| 8 | How do I reproduce it? | [`REPRODUCIBILITY.md`](REPRODUCIBILITY.md), [`docs/data/reproducibility_matrix.json`](docs/data/reproducibility_matrix.json), [`notes/reproducibility_audit.md`](notes/reproducibility_audit.md) | Public commands, expected outputs, and the latest exact-match reproduction record are explicit. |
-| 9 | What is one model input? | [`windows.csv`](results/episode_task_suite/windows.csv), [`feature_manifest.json`](results/episode_task_suite/feature_manifest.json), [`available_modalities.json`](results/episode_task_suite/available_modalities.json) | The input is an aligned 8,546-dimensional multimodal window with synchronized video, audio, sensor, and language signals. |
-| 10 | Are the task results backed by files? | [`summary_report.json`](results/episode_task_suite/summary_report.json), [`neural_mlp/`](results/episode_task_suite/neural_mlp/), [`docs/data/summary_metrics.json`](docs/data/summary_metrics.json) | Each task has minimal and neural-head evidence over the same window contracts. |
-| 11 | Is the website self-consistent? | [`docs/data/website_integrity.json`](docs/data/website_integrity.json), [`scripts/validate_website_integrity.py`](scripts/validate_website_integrity.py) | Local links, anchors, tab routing, JSON data, and referenced images are checked before publishing. |
-| 12 | What is still pending? | [`DATA_ACCESS_STATUS.md`](results/omni_finetune/DATA_ACCESS_STATUS.md), [`MULTI_EPISODE_ACCESS_STATUS.md`](results/omni_finetune/MULTI_EPISODE_ACCESS_STATUS.md), [`scripts/omni/discover_xperience10m_sources.py`](scripts/omni/discover_xperience10m_sources.py) | The multi-episode Qwen3-Omni run is prepared at the episode-selection level; final model metrics require completed preprocessing, training, and held-out evaluation. |
-
-The machine-readable project packet is
+| 1 | What is this project? | [`PROJECT_BRIEF.md`](PROJECT_BRIEF.md), [`PROJECT_STATUS.md`](PROJECT_STATUS.md), [dashboard](https://chaoyue0307.github.io/ropedia-xperience-10m-task-suite/) | A public-sample Xperience-10M research project with 12 tasks, baselines, and a scale-up plan. |
+| 2 | What data is used? | [`XPERIENCE10M_DATASET_CARD_ALIGNMENT.md`](XPERIENCE10M_DATASET_CARD_ALIGNMENT.md), [official HF dataset](https://huggingface.co/datasets/ropedia-ai/xperience-10m), [sample HF dataset](https://huggingface.co/datasets/ropedia-ai/xperience-10m-sample) | The implemented suite uses one public sample episode; the gated dataset is reserved for selected multi-episode training. |
+| 3 | What does one model input contain? | [`windows.csv`](results/episode_task_suite/windows.csv), [`feature_manifest.json`](results/episode_task_suite/feature_manifest.json), [`available_modalities.json`](results/episode_task_suite/available_modalities.json) | Each window is an aligned multimodal unit with video, audio, depth, pose/SLAM, mocap, IMU, calibration, and language-derived signals. |
+| 4 | What are the 12 tasks? | [`results/episode_task_suite/task_walkthroughs/`](results/episode_task_suite/task_walkthroughs/), [`docs/data/task_walkthroughs.json`](docs/data/task_walkthroughs.json) | Every task has a human-readable name, case study, input, process modules, output, metric, and limitation. |
+| 5 | How are tasks evaluated? | [`EVALUATION_PROTOCOL.md`](EVALUATION_PROTOCOL.md), [`docs/data/evaluation_protocol.json`](docs/data/evaluation_protocol.json) | The window unit, chronological split, leakage controls, task metrics, and current limitations are explicit. |
+| 6 | What do the current results mean? | [`RESEARCH_TAKEAWAYS.md`](RESEARCH_TAKEAWAYS.md), [`docs/data/research_takeaways.json`](docs/data/research_takeaways.json), [`docs/data/summary_metrics.json`](docs/data/summary_metrics.json) | Current metrics describe sample-level task behavior and identify which signals need larger held-out experiments. |
+| 7 | Which models are implemented? | [`results/episode_task_suite/summary_report.json`](results/episode_task_suite/summary_report.json), [`results/episode_task_suite/neural_mlp/`](results/episode_task_suite/neural_mlp/), [HF baseline repo](https://huggingface.co/cy0307/ropedia-xperience-10m-task-baselines) | Each task has minimal and neural-head evidence over the same feature windows. |
+| 8 | What research directions does this support? | [`RESEARCH_ROADMAP.md`](RESEARCH_ROADMAP.md), [`docs/data/research_directions.json`](docs/data/research_directions.json), [`docs/data/research_direction_extensions.json`](docs/data/research_direction_extensions.json) | The tasks are mapped to human modeling, 3D/4D reconstruction, egocentric interaction, and world modeling. |
+| 9 | Which foundation model comes next? | [`FOUNDATION_MODEL_PLAN.md`](FOUNDATION_MODEL_PLAN.md), [`docs/data/foundation_model_plan.json`](docs/data/foundation_model_plan.json), [`XPERIENCE_EMBODIED_FOUNDATION_MODEL_PRETRAINING.md`](XPERIENCE_EMBODIED_FOUNDATION_MODEL_PRETRAINING.md) | Qwen3-Omni is the first held-out LoRA baseline; Cosmos 3 is the first world-model branch; policy models wait for explicit action targets; Xperience-native pretraining is the full-corpus future goal. |
+| 10 | How do I reproduce it? | [`REPRODUCIBILITY.md`](REPRODUCIBILITY.md), [`notes/reproducibility_audit.md`](notes/reproducibility_audit.md) | Public commands and expected outputs are documented for the sample-episode task suite. |
+| 11 | What is still pending? | [`DATA_ACCESS_STATUS.md`](results/omni_finetune/DATA_ACCESS_STATUS.md), [`MULTI_EPISODE_ACCESS_STATUS.md`](results/omni_finetune/MULTI_EPISODE_ACCESS_STATUS.md) | Multi-episode Qwen3-Omni model quality will be reported after preprocessing, training, and held-out evaluation complete. |
+
+A compact reader-path summary is available at
 [`docs/data/project_packet.json`](docs/data/project_packet.json).
 
-## Artifact Index
+## Supporting Files
 
-[`docs/data/artifact_index.json`](docs/data/artifact_index.json) is the compact
-project artifact map for the repo. It lists the core supporting artifacts, whether each exists,
-its size, and a SHA-256 hash for stable files. Volatile generated files, such as
-the publication package report with a run timestamp, are marked so readers know they
-are checked for presence and size rather than treated as fixed hashes.
+[`ARTIFACT_GUIDE.md`](ARTIFACT_GUIDE.md) is the human-readable map for readers
+who want to inspect the project files after the first pass. It groups the main
+briefs, task outputs, baseline results, visual assets, data notes, and
+scale-up documents.
 
-[`ARTIFACT_GUIDE.md`](ARTIFACT_GUIDE.md) is the human-readable companion. It
-groups the same project evidence into start-here files, data-contract files,
-task-evidence files, platform mirrors, and scale-up status artifacts.
+[`docs/data/artifact_index.json`](docs/data/artifact_index.json) is the compact
+machine-readable companion used by the website and Hugging Face artifact
+dataset.
 
 ## Evaluation Protocol
 
@@ -256,41 +181,20 @@ generated from committed metric artifacts. They define:
   audio-visual learning, pixel-depth reconstruction, and real held-out
   multi-episode Qwen3-Omni quality.
 
-## Official Dataset Alignment
+## Dataset Context
 
 The official [`ropedia-ai/xperience-10m`](https://huggingface.co/datasets/ropedia-ai/xperience-10m)
-card describes Xperience-10M as a large-scale gated egocentric multimodal
-dataset for embodied AI, robotics, world models, and spatial intelligence. Its
-public metadata lists video classification, image-to-text, depth estimation,
-and robotics task categories; 3D, audio, and video modalities; English
-language; `other` license; and manually reviewed non-commercial access.
-
-At full scale, the official card describes about 10 million experience units,
-about 10,000 hours, six RGB streams per episode, audio, stereo depth, camera
-pose/SLAM, hand and full-body mocap, IMU, captions, metadata, and calibration.
-The card also reports headline counts such as billions of RGB/depth/IMU records
-and large caption/object annotations. The live HF page/API separately shows a
-31.9 TB currently hosted file-size display; this is kept separate from the
-card's about-1PB full-scale storage statement. This repo records those upstream facts in
-[`XPERIENCE10M_DATASET_CARD_ALIGNMENT.md`](XPERIENCE10M_DATASET_CARD_ALIGNMENT.md)
-and [`docs/data/xperience10m_dataset_card_alignment.json`](docs/data/xperience10m_dataset_card_alignment.json).
+dataset is a gated large-scale egocentric multimodal dataset for embodied AI,
+robotics, spatial intelligence, and world modeling. The public
+[`ropedia-ai/xperience-10m-sample`](https://huggingface.co/datasets/ropedia-ai/xperience-10m-sample)
+repo provides the sample episode used for the implemented task suite here.
 
-The current HF API snapshot for the gated dataset reports commit
-`ce943cf271a758b60240084892d05cf6dc12dd90`, last modified
-`2026-04-21T05:03:45.000Z`, manual gating, and a metadata file listing with
-803 session folders and 12,103 episode folders carrying `annotation.hdf5`.
-Those counts are upstream listing metadata only; they are not local downloads,
-not redistributed files, and not evidence of model quality in this repo.
+This project keeps those layers separate: the public sample supports the
+current 12-task study, while the gated full dataset is used only for the
+selected multi-episode Qwen3-Omni pilot. Raw Xperience-10M MP4/HDF5/RRD files
+are not redistributed in this repo or in the Hugging Face mirrors.
 
-The public sample repo,
-[`ropedia-ai/xperience-10m-sample`](https://huggingface.co/datasets/ropedia-ai/xperience-10m-sample),
-is separately documented as `Xperience-10M-Sample` with sample metadata,
-`cc-by-nc-4.0` license, HOMIE Toolkit usage, and Rerun 0.29.0 `.rrd`
-visualization. This project preserves that distinction: the sample powers the
-current 5,821-frame task suite, while the full gated dataset is the source for
-the selected 128-episode held-out multi-episode pilot now in preparation.
-
-This repo's current verified subset is much smaller and intentionally explicit:
+The current verified public-sample subset is:
 
 - one public sample episode, 5,821 frames, and 1,161 aligned windows,
 - raw sample files with six MP4 video streams and audio streams,
@@ -299,15 +203,11 @@ This repo's current verified subset is much smaller and intentionally explicit:
 - an 8,546-dimensional baseline representation using video, audio, depth,
   pose/SLAM, mocap, IMU, calibration, and language-derived signals.
 
-The same alignment note also records what is outside the current implemented subset: real
-audio-visual learning, caption generation, pixel-depth estimation, SLAM
-estimation, neural rendering, policy learning, cross-episode generalization,
-and real held-out multi-episode Qwen3-Omni model quality.
-It also preserves the official responsible-use scope: the open-source
-dataset is limited in diversity and showcase/production quality, and it should
-not be used for identity recognition, re-identification, biometric profiling,
-surveillance, sensitive attribute inference, or safety-critical deployment
-without appropriate safeguards.
+Detailed dataset notes are available in
+[`XPERIENCE10M_DATASET_CARD_ALIGNMENT.md`](XPERIENCE10M_DATASET_CARD_ALIGNMENT.md)
+for readers who need the full upstream-card and access-term context. The
+practical boundary is simple: current results come from the public sample, and
+multi-episode model quality is pending the selected held-out pilot.
 
 Start with the visual dashboard:
 
@@ -323,22 +223,15 @@ Hugging Face Space app:
 | --- | --- | --- |
 | Project status | `PROJECT_STATUS.md`, `docs/data/project_status.json` | Gives a one-table current project summary before reading the full artifact trail |
 | Data contract | `windows.csv`, `feature_manifest.json`, modality manifests | Confirms what each sample window contains before modeling |
-| Official dataset alignment | `XPERIENCE10M_DATASET_CARD_ALIGNMENT.md`, `docs/data/xperience10m_dataset_card_alignment.json` | Keeps public descriptions aligned with the official gated dataset card |
-| Source alignment | `SOURCE_ALIGNMENT_AUDIT.md`, `docs/data/source_alignment_audit.json` | Summarizes official dataset facts, sample-card facts, API-listing notes, and project coverage across repo, website, and HF cards |
-| Figure index | `FIGURE_INDEX.md`, `docs/data/figure_index.json` | Indexes public figures, charts, modality thumbnails, dimensions, hashes, and source scripts |
-| Brand assets | `docs/data/brand_assets.json`, `docs/assets/brand/` | Indexes the generated logo, favicon, README/HF card image, app icon, and social preview |
+| Dataset context | `XPERIENCE10M_DATASET_CARD_ALIGNMENT.md`, official dataset links | Explains the official dataset, public sample, modalities, access boundary, and what this repo uses |
+| Visual assets | `FIGURE_INDEX.md`, `docs/assets/` | Shows the task-suite graphic, modality thumbnails, pipeline diagrams, charts, and logo assets |
 | Evaluation protocol | `EVALUATION_PROTOCOL.md`, `docs/data/evaluation_protocol.json` | Defines the task unit, split, metrics, leakage controls, and current limitations |
-| Task surface integrity | `docs/data/task_surface_integrity.json` | Checks the public task cards, readable task names, representative modality thumbnails, and interactive walkthrough storyboard |
-| Rendered website check | `RENDERED_SITE_CHECK.md`, `docs/data/rendered_site_check.json` | Records the browser-level page load, tab navigation, walkthrough deep link, player interaction, and console-health result |
-| Research roadmap | `RESEARCH_ROADMAP.md`, `docs/data/research_roadmap.json` | Shows the path from sample-level task development to multi-episode and larger omni-model work |
+| Research roadmap | `RESEARCH_ROADMAP.md`, `docs/data/research_roadmap.json` | Shows the path from sample-level task development to multi-episode work, larger model branches, and the future native-pretraining goal |
+| Xperience Embodied Foundation Model plan | `XPERIENCE_EMBODIED_FOUNDATION_MODEL_PRETRAINING.md` | Describes the long-term full-corpus pretraining goal, target modules, objectives, staged scale-up, hardware ranges, and evaluation protocol |
 | Minimal heads | softmax, ridge projection/regression, multi-label logistic heads | Keeps every input/output contract visible and inspectable |
 | Neural heads | PyTorch MLP classifiers/regressors under `neural_mlp/` | Checks whether nonlinear heads improve each task without changing features |
 | Evidence | metrics, predictions, confusion matrices, diagrams, dashboard | Makes the single-episode task development inspectable without rerunning first |
-| Release checks | `QUALITY_GATES.md`, `docs/data/quality_gates.json` | Shows the automated and post-publish checks used to keep the public release current |
-| Live publication status | `docs/data/live_publication_status.json` | Records the last live GitHub Pages, GitHub raw, and Hugging Face mirror verification |
-| Public bundle contents | `docs/data/publication_audit.json` | Summarizes public bundle contents, raw Xperience-10M data exclusion, cache exclusion, archive exclusion, credential-text checks, and public-card figure references |
-| Artifact index | `docs/data/artifact_index.json` | Gives readers a compact source-of-truth catalog with stable hashes |
-| Artifact guide | `ARTIFACT_GUIDE.md` | Groups the public evidence into research-project layers |
+| Artifact guide | `ARTIFACT_GUIDE.md` | Groups the public evidence into research-project layers after the first-pass overview |
 | Reproducibility contract | `REPRODUCIBILITY.md`, `docs/data/reproducibility_matrix.json` | States public commands, expected outputs, exact-match reproduction evidence, and non-reproducible boundaries |
 | Citation metadata | `CITATION.cff`, `codemeta.json`, `LICENSE` | Makes the repo easier to cite, index, and reuse without confusing code license and dataset terms |
 
@@ -421,12 +314,12 @@ scripts/
   export_modality_atlas_assets.py   # exports responsive modality-card assets
   render_overview_figures.py        # renders polished pipeline/architecture PNGs
   build_brand_assets.py             # derives logo sizes, favicon, social card
-  build_artifact_index.py           # builds the source-of-truth artifact index
+  build_artifact_index.py           # builds the compact artifact guide data
   build_quality_gates.py            # builds release checks
   validate_mirror_parity.py         # checks prepared GitHub/HF mirror file parity
-  validate_scope_claims.py          # keeps Qwen3-Omni setup and result states separate
+  validate_scope_claims.py          # separates setup artifacts from completed model metrics
   validate_task_surface.py          # checks readable task cards and interactive storyboard wiring
-  validate_website_integrity.py     # checks local site links, anchors, JSON, images
+  validate_website_integrity.py     # checks local site links, anchors, and images
   validate_publication_package.py   # checks public repo + HF bundle contents
   publish_hf_bundles.py             # uploads prepared HF Space/artifact/model bundles
   omni/
@@ -454,11 +347,9 @@ docs/
   data/artifact_index.json          # compact project-artifact catalog
   data/live_publication_status.json # live GitHub/HF publication verification
   data/quality_gates.json           # machine-readable release checks
-  data/publication_audit.json       # machine-readable public bundle report
   data/task_surface_integrity.json  # machine-readable task-card/storyboard integrity check
-  data/website_integrity.json       # machine-readable website integrity check
   data/project_manifest.json        # machine-readable public-surface metadata
-  data/project_packet.json          # machine-readable project path and scope summary
+  data/project_packet.json          # compact project path and scope summary
   data/research_roadmap.json        # multi-episode and omni-model roadmap
   data/research_directions.json     # four-track website data bundle
   data/research_direction_extensions.json # four extra probe data bundle
@@ -671,6 +562,59 @@ uses the same split guard, exports episodes in parallel CPU shards, skips and
 reports episodes that contain no labeled windows under the configured label
 rule, then launches Qwen3-Omni LoRA with `NUM_PROCESSES=8`.
 
+### Full 128-Episode Held-Out Pilot
+
+Once all selected episodes are complete, use the fixed selected-episode split:
+
+- 96 train episodes,
+- 16 validation episodes,
+- 16 held-out test episodes.
+
+The clean full-run launcher validates the selected split, exports all splits in
+parallel, trains Qwen3-Omni LoRA on train/val only, then evaluates on the held-
+out test split:
+
+```bash
+RUN_ID=xperience10m_qwen3_omni_128ep_fullsplit_fast8gpu \
+DATA_ROOT=/path/to/xperience10m_128 \
+SELECTION_JSON=results/omni_finetune/xperience10m_128_episode_selection.json \
+MODEL_DIR=/path/to/Qwen__Qwen3-Omni-30B-A3B-Instruct \
+NUM_PROCESSES=8 \
+scripts/omni/run_128_fullsplit_parallel_export_8gpu.sh
+```
+
+Monitor the run with:
+
+```bash
+python scripts/omni/monitor_omni_progress.py \
+  --run-id xperience10m_qwen3_omni_128ep_fullsplit_fast8gpu
+```
+
+Validate the run artifacts stage by stage:
+
+```bash
+python scripts/omni/validate_omni_finetune_run.py \
+  --run-id xperience10m_qwen3_omni_128ep_fullsplit_fast8gpu \
+  --require-stage manifest
+
+python scripts/omni/validate_omni_finetune_run.py \
+  --run-id xperience10m_qwen3_omni_128ep_fullsplit_fast8gpu \
+  --require-stage eval \
+  --min-json-validity 0.98
+```
+
+After dataset export, a model-neutral window index can be created for future
+backbones:
+
+```bash
+python scripts/omni/export_model_neutral_window_index.py \
+  --dataset-jsonl results/omni_finetune/xperience10m_qwen3_omni_128ep_fullsplit_fast8gpu_dataset/dataset.jsonl
+```
+
+This produces `window_index.jsonl` and `window_index_manifest.json` so Cosmos-
+style world models and VLA/policy branches can reuse the same split-checked
+windows without depending on Qwen chat-message records.
+
 ### Uploading the pilot Qwen3-Omni LoRA
 
 A prepared upload package is available at `results/omni_finetune/hf_upload`.
@@ -697,11 +641,23 @@ assuming one backbone solves every Xperience-10M objective.
 | GR00T | Humanoid/action-policy branch | Use after mocap/contact retargeting creates well-defined humanoid action targets. |
 | OpenVLA / openpi | Open VLA/policy baselines | Use after the project defines robot-compatible or action-token targets. |
 | Gemini Robotics | External reasoning reference | Use only for qualitative comparison or annotation support unless local trainable access exists. |
+| Xperience Embodied Foundation Model | Future Xperience-native pretraining goal | Use only after multi-episode pilots, full-corpus storage, distributed training infrastructure, and scaling evidence justify a from-scratch domain model. |
 
 See [`FOUNDATION_MODEL_PLAN.md`](FOUNDATION_MODEL_PLAN.md) and
 [`docs/data/foundation_model_plan.json`](docs/data/foundation_model_plan.json)
 for the full selection matrix, source links, and model-specific evaluation
-additions.
+additions. See
+[`XPERIENCE_EMBODIED_FOUNDATION_MODEL_PRETRAINING.md`](XPERIENCE_EMBODIED_FOUNDATION_MODEL_PRETRAINING.md)
+for the long-term full-corpus pretraining plan.
+
+Backbone-specific contracts now live in [`configs/omni_backbones`](configs/omni_backbones).
+The extension contract is documented in
+[`OMNI_MODEL_EXTENSION_CONTRACT.md`](OMNI_MODEL_EXTENSION_CONTRACT.md), and the
+registry can be checked with:
+
+```bash
+python scripts/omni/backbone_registry.py --validate --json
+```
 
 ## Four Research Directions
 

PROJECT_STATUS.md

@@ -21,8 +21,9 @@ scale-up readiness; it is not presented as final full-dataset model quality.
 | Neural heads | Verified | `scripts/neural_task_models.py`, `results/episode_task_suite/neural_mlp/` | Each task also has a compact PyTorch MLP run over the same feature tensor and chronological split. |
 | Audio contribution study | Verified | `scripts/audio_ablation_and_raw_upgrade.py`, `results/audio_ablation/`, `docs/data/audio_ablation_summary.json` | Audio variants are compared across all 12 task contracts; audio improves the primary metric on 6 of 12 tasks, and a 588-d audio-window representation improves over the baseline audio variant on 6 of 12 tasks. |
 | Research takeaways | Verified | `RESEARCH_TAKEAWAYS.md`, `docs/data/research_takeaways.json`, `scripts/build_research_takeaways.py` | The main result interpretation is generated from committed metrics: chronological class shift, neural gains on dynamics/order/alignment, open retrieval/reconstruction problems, and the need for held-out episodes. |
-| Research roadmap | Current | `RESEARCH_ROADMAP.md`, `docs/data/research_roadmap.json` | The roadmap connects public-sample task development to 128-episode data preparation, Qwen3-Omni LoRA, foundation-model selection, robustness runs, and larger omni/world-model extensions. |
+| Research roadmap | Current | `RESEARCH_ROADMAP.md`, `docs/data/research_roadmap.json` | The roadmap connects public-sample task development to 128-episode data preparation, Qwen3-Omni LoRA, foundation-model selection, robustness runs, world/policy branches, and the future Xperience-native pretraining goal. |
 | Foundation-model plan | Current | `FOUNDATION_MODEL_PLAN.md`, `docs/data/foundation_model_plan.json` | Qwen3-Omni remains the first trainable held-out LoRA baseline; Cosmos 3 is added as the first world-model/action-generation branch; OpenVLA/openpi/GR00T are policy candidates after action targets are explicit. |
+| Xperience Embodied Foundation Model | Future goal | `XPERIENCE_EMBODIED_FOUNDATION_MODEL_PRETRAINING.md` | A future full-corpus pretraining plan describes target modules, objectives, staged scale-up, hardware ranges, and evaluation for a domain-specific embodied foundation model. |
 | Evaluation protocol | Verified | `EVALUATION_PROTOCOL.md`, `docs/data/evaluation_protocol.json`, `scripts/build_evaluation_protocol.py` | Windowing, chronological split, per-task metrics, leakage controls, and current limitations are generated from committed metric artifacts. |
 | Dataset context | Verified | `XPERIENCE10M_DATASET_CARD_ALIGNMENT.md`, official Xperience-10M and sample cards | The README and dashboard distinguish the public sample used here from the gated full dataset used for the selected multi-episode pilot. |
 | Public dashboard and Hub pages | Verified | GitHub Pages, HF Space, artifact dataset, baseline model repo, Qwen3-Omni LoRA repo | Readers can move between the website, code, derived artifacts, baseline weights, and Qwen3-Omni pilot status without needing internal setup details. |
@@ -42,15 +43,17 @@ scale-up readiness; it is not presented as final full-dataset model quality.
    the path from public-sample task work to multi-episode modeling.
 5. Inspect `FOUNDATION_MODEL_PLAN.md` and
    `docs/data/foundation_model_plan.json` before choosing a backbone branch.
-6. Inspect `docs/data/summary_metrics.json` and
+6. Inspect `XPERIENCE_EMBODIED_FOUNDATION_MODEL_PRETRAINING.md` for the
+   long-term full-corpus pretraining goal.
+7. Inspect `docs/data/summary_metrics.json` and
    `results/episode_task_suite/neural_mlp/` to check the 12-task outputs.
-7. Inspect `results/audio_ablation/AUDIO_ABLATION_SUMMARY.md` before judging
+8. Inspect `results/audio_ablation/AUDIO_ABLATION_SUMMARY.md` before judging
    whether audio helps the current task suite.
-8. Inspect `EVALUATION_PROTOCOL.md` before judging task metrics or leakage
+9. Inspect `EVALUATION_PROTOCOL.md` before judging task metrics or leakage
    controls.
-9. Inspect `XPERIENCE10M_DATASET_CARD_ALIGNMENT.md` only if you need the
+10. Inspect `XPERIENCE10M_DATASET_CARD_ALIGNMENT.md` only if you need the
    detailed upstream dataset-card context.
-10. Inspect `results/omni_finetune/DATA_ACCESS_STATUS.md` before judging
+11. Inspect `results/omni_finetune/DATA_ACCESS_STATUS.md` before judging
    Qwen3-Omni scale-up status.
 
 ## Current Reading Notes
@@ -67,3 +70,5 @@ scale-up readiness; it is not presented as final full-dataset model quality.
 - Foundation-model selection is now explicit: Qwen3-Omni is the immediate
   trainable pilot, Cosmos 3 is the first world-model branch, and policy models
   such as OpenVLA/openpi/GR00T wait for action-target conversion.
+- The Xperience Embodied Foundation Model is a future native-pretraining goal,
+  not a completed model or current benchmark.

README.md

@@ -64,7 +64,7 @@ embodied-AI research infrastructure:
 | Multimodal data understanding | Parses the public sample into synchronized windows across video, audio, depth, pose/SLAM, mocap, IMU, calibration, and language-derived signals |
 | Task design | Defines 12 human-readable tasks plus four direction-extension probes with inputs, outputs, process modules, metrics, and case-study walkthroughs |
 | Model and evaluation discipline | Runs minimal and compact neural baselines, records predictions/metrics, keeps chronological split boundaries explicit, and separates sample evidence from held-out claims |
-| Scale-up planning | Connects the public-sample pipeline to 32/128-episode held-out pilots, Qwen3-Omni LoRA, Cosmos-style world-model branches, and later policy-model branches |
+| Scale-up planning | Connects the public-sample pipeline to 32/128-episode held-out pilots, Qwen3-Omni LoRA, Cosmos-style world-model branches, policy-model branches, and the future Xperience-native foundation-model pretraining goal |
 
 ## Start Here
 
@@ -81,6 +81,7 @@ before the multi-episode omni-model stage becomes a real held-out evaluation.
 | Navigate the 12 tasks, four tracks, and scale-up plan | [Interactive research roadmap](https://chaoyue0307.github.io/ropedia-xperience-10m-task-suite/research_roadmap.html), [`docs/data/research_roadmap_interactive.json`](docs/data/research_roadmap_interactive.json) |
 | Compare current task metrics | [`RESEARCH_TAKEAWAYS.md`](RESEARCH_TAKEAWAYS.md), [`docs/data/summary_metrics.json`](docs/data/summary_metrics.json) |
 | Compare possible foundation backbones | [`FOUNDATION_MODEL_PLAN.md`](FOUNDATION_MODEL_PLAN.md), [`docs/data/foundation_model_plan.json`](docs/data/foundation_model_plan.json) |
+| Understand the future native pretraining goal | [`XPERIENCE_EMBODIED_FOUNDATION_MODEL_PRETRAINING.md`](XPERIENCE_EMBODIED_FOUNDATION_MODEL_PRETRAINING.md) |
 | Understand one model input | [`results/episode_task_suite/feature_manifest.json`](results/episode_task_suite/feature_manifest.json), [`results/episode_task_suite/windows.csv`](results/episode_task_suite/windows.csv) |
 | Check multi-episode data status | [`results/omni_finetune/DATA_ACCESS_STATUS.md`](results/omni_finetune/DATA_ACCESS_STATUS.md) |
 
@@ -93,7 +94,7 @@ before the multi-episode omni-model stage becomes a real held-out evaluation.
 | Task suite | 12 human-readable embodied-AI task contracts with input, process, output, metrics, predictions, and case-study walkthroughs |
 | Baselines | Minimal linear/ridge/logistic heads plus compact PyTorch MLP task heads over the same chronological split |
 | Research directions | Task mapping and extension probes for human modeling, 3D/4D reconstruction, egocentric interaction, and world modeling |
-| Scale-up path | The gated Xperience-10M dataset is available for a selected 128-episode pilot before Qwen3-Omni LoRA, followed by Cosmos 3/world-model and VLA/policy branches |
+| Scale-up path | The gated Xperience-10M dataset is available for a selected 128-episode pilot before Qwen3-Omni LoRA, followed by Cosmos 3/world-model and VLA/policy branches; the long-term goal is an Xperience-native embodied foundation model if full-corpus data, storage, and compute are available |
 | Public surfaces | GitHub repo, GitHub Pages dashboard, HF Space, HF artifact dataset, HF baseline-model repo, and HF collection |
 
 For the fastest interpretation of the current metrics, start with
@@ -115,6 +116,7 @@ Current contributions:
 - human-readable research task cards and an interactive scrub/play walkthrough storyboard for every task,
 - an interactive research roadmap connecting 12 tasks, four research tracks, current sample evidence, the Qwen3-Omni scale-up path, and foundation-model branch selection,
 - a next-milestone track for Qwen3-Omni fine-tuning, Cosmos 3 world modeling, and sensor-bridge evaluation,
+- a future pretraining plan for an Xperience Embodied Foundation Model over the full corpus after smaller multi-episode stages prove value,
 - metrics, predictions, model weights, manifests, charts, and a two-level
   tabbed static research website,
 - a clear explanation of what is implemented now and what moves to the multi-episode stage.
@@ -129,7 +131,7 @@ This project is best read as a staged embodied-AI research study:
 | Task suite | Twelve human-readable tasks cover action, procedure, contact, object, language, retrieval, reconstruction, order, and synchronization questions. | [`RESEARCH_TAKEAWAYS.md`](RESEARCH_TAKEAWAYS.md), [`results/episode_task_suite/summary_report.json`](results/episode_task_suite/summary_report.json) |
 | Baselines | Minimal heads and compact PyTorch MLP heads provide a first controlled comparison on the same chronological split. | [`results/episode_task_suite/neural_mlp/`](results/episode_task_suite/neural_mlp/) |
 | Diagnostics | Audio contribution, modality ablations, timeline overlays, object labels, and alignment stress tests show which signals are useful and which tasks remain hard. | [`results/audio_ablation/AUDIO_ABLATION_SUMMARY.md`](results/audio_ablation/AUDIO_ABLATION_SUMMARY.md), [`docs/single_episode_explorer.html`](docs/single_episode_explorer.html) |
-| Scale-up | A selected 128-episode Qwen3-Omni LoRA pilot is being prepared from the gated dataset; held-out model metrics will be added only after training and evaluation finish. | [`RESEARCH_ROADMAP.md`](RESEARCH_ROADMAP.md), [`FOUNDATION_MODEL_PLAN.md`](FOUNDATION_MODEL_PLAN.md), [`results/omni_finetune/DATA_ACCESS_STATUS.md`](results/omni_finetune/DATA_ACCESS_STATUS.md) |
+| Scale-up | A selected 128-episode Qwen3-Omni LoRA pilot is being prepared from the gated dataset; held-out model metrics will be added only after training and evaluation finish. The long-term native-pretraining plan is documented separately as a future research goal. | [`RESEARCH_ROADMAP.md`](RESEARCH_ROADMAP.md), [`FOUNDATION_MODEL_PLAN.md`](FOUNDATION_MODEL_PLAN.md), [`XPERIENCE_EMBODIED_FOUNDATION_MODEL_PRETRAINING.md`](XPERIENCE_EMBODIED_FOUNDATION_MODEL_PRETRAINING.md), [`results/omni_finetune/DATA_ACCESS_STATUS.md`](results/omni_finetune/DATA_ACCESS_STATUS.md) |
 
 Detailed dataset notes, reproduction checks, and generated JSON reports are
 included for readers who want to inspect the implementation, but they are
@@ -168,7 +170,7 @@ If you are reading the project cold, open these in order:
 | 6 | What do the current results mean? | [`RESEARCH_TAKEAWAYS.md`](RESEARCH_TAKEAWAYS.md), [`docs/data/research_takeaways.json`](docs/data/research_takeaways.json), [`docs/data/summary_metrics.json`](docs/data/summary_metrics.json) | Current metrics describe sample-level task behavior and identify which signals need larger held-out experiments. |
 | 7 | Which models are implemented? | [`results/episode_task_suite/summary_report.json`](results/episode_task_suite/summary_report.json), [`results/episode_task_suite/neural_mlp/`](results/episode_task_suite/neural_mlp/), [HF baseline repo](https://huggingface.co/cy0307/ropedia-xperience-10m-task-baselines) | Each task has minimal and neural-head evidence over the same feature windows. |
 | 8 | What research directions does this support? | [`RESEARCH_ROADMAP.md`](RESEARCH_ROADMAP.md), [`docs/data/research_directions.json`](docs/data/research_directions.json), [`docs/data/research_direction_extensions.json`](docs/data/research_direction_extensions.json) | The tasks are mapped to human modeling, 3D/4D reconstruction, egocentric interaction, and world modeling. |
-| 9 | Which foundation model comes next? | [`FOUNDATION_MODEL_PLAN.md`](FOUNDATION_MODEL_PLAN.md), [`docs/data/foundation_model_plan.json`](docs/data/foundation_model_plan.json) | Qwen3-Omni is the first held-out LoRA baseline; Cosmos 3 is the first world-model branch; policy models wait for explicit action targets. |
+| 9 | Which foundation model comes next? | [`FOUNDATION_MODEL_PLAN.md`](FOUNDATION_MODEL_PLAN.md), [`docs/data/foundation_model_plan.json`](docs/data/foundation_model_plan.json), [`XPERIENCE_EMBODIED_FOUNDATION_MODEL_PRETRAINING.md`](XPERIENCE_EMBODIED_FOUNDATION_MODEL_PRETRAINING.md) | Qwen3-Omni is the first held-out LoRA baseline; Cosmos 3 is the first world-model branch; policy models wait for explicit action targets; Xperience-native pretraining is the full-corpus future goal. |
 | 10 | How do I reproduce it? | [`REPRODUCIBILITY.md`](REPRODUCIBILITY.md), [`notes/reproducibility_audit.md`](notes/reproducibility_audit.md) | Public commands and expected outputs are documented for the sample-episode task suite. |
 | 11 | What is still pending? | [`DATA_ACCESS_STATUS.md`](results/omni_finetune/DATA_ACCESS_STATUS.md), [`MULTI_EPISODE_ACCESS_STATUS.md`](results/omni_finetune/MULTI_EPISODE_ACCESS_STATUS.md) | Multi-episode Qwen3-Omni model quality will be reported after preprocessing, training, and held-out evaluation complete. |
 
@@ -246,7 +248,8 @@ Hugging Face Space app:
 | Dataset context | `XPERIENCE10M_DATASET_CARD_ALIGNMENT.md`, official dataset links | Explains the official dataset, public sample, modalities, access boundary, and what this repo uses |
 | Visual assets | `FIGURE_INDEX.md`, `docs/assets/` | Shows the task-suite graphic, modality thumbnails, pipeline diagrams, charts, and logo assets |
 | Evaluation protocol | `EVALUATION_PROTOCOL.md`, `docs/data/evaluation_protocol.json` | Defines the task unit, split, metrics, leakage controls, and current limitations |
-| Research roadmap | `RESEARCH_ROADMAP.md`, `docs/data/research_roadmap.json` | Shows the path from sample-level task development to multi-episode and larger omni-model work |
+| Research roadmap | `RESEARCH_ROADMAP.md`, `docs/data/research_roadmap.json` | Shows the path from sample-level task development to multi-episode work, larger model branches, and the future native-pretraining goal |
+| Xperience Embodied Foundation Model plan | `XPERIENCE_EMBODIED_FOUNDATION_MODEL_PRETRAINING.md` | Describes the long-term full-corpus pretraining goal, target modules, objectives, staged scale-up, hardware ranges, and evaluation protocol |
 | Minimal heads | softmax, ridge projection/regression, multi-label logistic heads | Keeps every input/output contract visible and inspectable |
 | Neural heads | PyTorch MLP classifiers/regressors under `neural_mlp/` | Checks whether nonlinear heads improve each task without changing features |
 | Evidence | metrics, predictions, confusion matrices, diagrams, dashboard | Makes the single-episode task development inspectable without rerunning first |
@@ -607,11 +610,14 @@ assuming one backbone solves every Xperience-10M objective.
 | GR00T | Humanoid/action-policy branch | Use after mocap/contact retargeting creates well-defined humanoid action targets. |
 | OpenVLA / openpi | Open VLA/policy baselines | Use after the project defines robot-compatible or action-token targets. |
 | Gemini Robotics | External reasoning reference | Use only for qualitative comparison or annotation support unless local trainable access exists. |
+| Xperience Embodied Foundation Model | Future Xperience-native pretraining goal | Use only after multi-episode pilots, full-corpus storage, distributed training infrastructure, and scaling evidence justify a from-scratch domain model. |
 
 See [`FOUNDATION_MODEL_PLAN.md`](FOUNDATION_MODEL_PLAN.md) and
 [`docs/data/foundation_model_plan.json`](docs/data/foundation_model_plan.json)
 for the full selection matrix, source links, and model-specific evaluation
-additions.
+additions. See
+[`XPERIENCE_EMBODIED_FOUNDATION_MODEL_PRETRAINING.md`](XPERIENCE_EMBODIED_FOUNDATION_MODEL_PRETRAINING.md)
+for the long-term full-corpus pretraining plan.
 
 ## Four Research Directions
 

RESEARCH_ROADMAP.md

@@ -15,6 +15,7 @@ should exist before the stage is treated as complete.
 | Foundation-Model Selection Matrix | Next | The selected pilot episodes are prepared, or a 3-8 episode dry run is available for preprocessing checks. | Backbone registry, Cosmos 3 world-model branch plan, Qwen3-Omni baseline plan, OpenVLA/openpi/GR00T policy candidates, and model-specific evaluation additions. | `FOUNDATION_MODEL_PLAN.md`, `docs/data/foundation_model_plan.json`, `research_roadmap_interactive.json` |
 | 64-128 Episode Robustness Run | Planned | The selected-episode pilot trains and evaluates cleanly. | Split-by-session metrics, modality ablations, calibration/object/language error analysis, and sensitivity to missing views. | Held-out metrics by session, task, and modality; ablation tables; qualitative error analysis. |
 | Cosmos 3 and Policy-Model Extensions | Planned | Enough multi-episode data, compute budget, and model-specific action/world-state targets. | Cosmos 3 future-window or action-conditioned world-model probes, OpenVLA/openpi/GR00T action-policy baselines, modality-conditioning checks, affordance tasks, and synthetic-data usefulness tests. | Task-specific held-out evaluations, qualitative inspection, and updated model cards. |
+| Xperience Embodied Foundation Model Pretraining | Future | Full-corpus access, PB-scale storage path, multi-node compute, and positive scaling evidence from smaller runs. | Xperience-native temporal multimodal model, full-corpus manifests, pretraining shards, scaling curves, held-out evaluations, and model card. | Pretraining metadata, checkpoint inventory, held-out metrics, scaling report, and data-boundary report. |
 
 ## Current Decision Point
 
@@ -24,9 +25,11 @@ episodes to run the held-out Qwen3-Omni pilot, then choose larger model branches
 by task fit. Qwen3-Omni remains the first trainable multimodal LoRA target.
 Cosmos 3 becomes the first world-model/action-generation branch. OpenVLA,
 openpi, GR00T, Octo, and SmolVLA-style models become policy/action branches only
-after the action target is explicit. The public sample is already enough for
-task design, feature contracts, walkthroughs, and baseline comparisons. It is
-not enough to measure general embodied-AI model quality.
+after the action target is explicit. A from-scratch Xperience Embodied
+Foundation Model is the long-term native-pretraining goal, not the immediate
+experiment. The public sample is already enough for task design, feature
+contracts, walkthroughs, and baseline comparisons. It is not enough to measure
+general embodied-AI model quality.
 
 ## Stage Details
 
@@ -109,6 +112,27 @@ objectives: audio-visible alignment, future-window prediction,
 action-conditioned world modeling, synthetic-data usefulness tests, policy-style
 next action, contact, object relevance, and affordance reasoning.
 
+### 7. Xperience Embodied Foundation Model Pretraining
+
+This stage is the long-term full-corpus goal. Instead of adapting an existing
+backbone, it would pretrain a domain model directly on the synchronized
+Xperience-10M modality structure: video, audio, depth, pose/SLAM, hand/body
+mocap, IMU, calibration, and language annotations.
+
+The first realistic target is a 3B-7B Xperience-native domain model after
+smaller 0.3B-1B and 1B-3B pilots prove that the objectives and data loaders
+scale. The training objective should combine masked multimodal modeling,
+cross-modal alignment, future-state prediction, ego-motion and hand-motion
+forecasting, action/procedure prediction, language grounding, contact and
+affordance prediction, and optional policy-style targets after action
+conversion.
+
+This stage needs full-corpus access, PB-scale storage planning, high-throughput
+media decoding, distributed training, reliable checkpoints, and held-out
+evaluation across episodes, sessions, activities, objects, and missing
+modalities. The plan is reader-facing in
+`XPERIENCE_EMBODIED_FOUNDATION_MODEL_PRETRAINING.md`.
+
 ## Public Artifacts That Should Move Together
 
 When a roadmap stage advances, update these public surfaces together:
@@ -118,6 +142,7 @@ When a roadmap stage advances, update these public surfaces together:
 - `RESEARCH_TAKEAWAYS.md`
 - `EVALUATION_PROTOCOL.md`
 - `ARTIFACT_GUIDE.md`
+- `XPERIENCE_EMBODIED_FOUNDATION_MODEL_PRETRAINING.md`
 - `docs/index.html`
 - `docs/data/research_roadmap.json`
 - Hugging Face Space, artifact dataset, and model cards

XPERIENCE_EMBODIED_FOUNDATION_MODEL_PRETRAINING.md

@@ -0,0 +1,178 @@
+# Xperience Embodied Foundation Model Pretraining Goal
+
+This document describes a future research direction for the project: a
+domain-specific embodied foundation model pretrained on the full Xperience-10M
+corpus, if full-episode access, storage, and compute become available.
+
+Current status: this is a planning artifact. The public project currently
+contains a public-sample task suite, lightweight baselines, Qwen3-Omni LoRA
+preparation, and a smoke LoRA artifact. It does not currently contain a
+from-scratch Xperience foundation model or full-corpus pretraining run.
+
+## Why This Is A Natural Long-Term Goal
+
+Xperience-10M is designed for physical-AI pretraining rather than only
+single-task supervised learning. The official dataset card describes 10 million
+experiences, 10,000 hours of synchronized first-person recordings, six video
+streams, audio, stereo depth, camera pose, hand and full-body mocap, IMU, and
+hierarchical language annotations. It also reports 2.88B RGB frames, 720M depth
+frames, 576M pose/mocap frames, 7.2B IMU frames, and about 1 PB of total data.
+
+That scale and alignment make a specific Xperience-native model plausible:
+not a general web-scale omni model, but an embodied model specialized for
+egocentric perception, human-object interaction, temporal dynamics, physical
+state, and task intent.
+
+## Target Model
+
+The proposed model name is **Xperience Embodied Foundation Model**.
+
+The model should learn a shared temporal representation of embodied experience:
+what the wearer sees and hears, how the camera moves, how the body and hands
+move, what objects are involved, what geometry is present, and what task is
+being performed.
+
+Expected modules:
+
+| Module | Input | Role |
+| --- | --- | --- |
+| Multi-view video encoder | fisheye/stereo/RGB streams | visual state, egocentric context, object interaction |
+| Audio encoder | synchronized MP4 audio | event cues, contact-like sound, temporal grounding |
+| Depth and geometry encoder | depth, confidence, calibration | spatial structure and 3D/4D scene cues |
+| Pose/SLAM encoder | camera trajectory and orientation | ego-motion, viewpoint, scene traversal |
+| Mocap encoder | hand/body joints | human motion, hand-object interaction, affordance cues |
+| IMU encoder | accelerometer/gyroscope streams | inertial dynamics and wearable motion |
+| Language encoder/decoder | task/subtask/action/object annotations | semantic grounding and structured generation |
+| Temporal fusion transformer | aligned per-window modality tokens | shared embodied representation across time |
+| Task heads / decoders | fused representation | action, caption, future motion, retrieval, reconstruction, and world-state outputs |
+
+## Pretraining Objectives
+
+The model should not rely on one loss. It should combine complementary
+objectives so that every modality contributes to the shared representation.
+
+| Objective | What the model learns | Example output |
+| --- | --- | --- |
+| Masked multimodal modeling | recover hidden video/depth/sensor tokens from context | reconstructed latent patches or sensor features |
+| Cross-modal contrastive alignment | align video, motion, audio, geometry, and language from the same time window | matching score or retrieval embedding |
+| Future-state prediction | predict what changes after the current window | future visual/depth/motion latent |
+| Ego-motion and hand-motion forecasting | model wearer/body dynamics | future camera delta or hand trajectory |
+| Action and procedure prediction | connect physical state to task semantics | action, subtask, transition, next action |
+| Language grounding and captioning | connect temporal windows to natural language | caption, object/action grounding, structured JSON |
+| Contact and affordance prediction | learn interaction state from human-object motion | contact state, relevant object set |
+| Optional policy-style targets | learn action-like outputs after target conversion | action token, motion chunk, retargeted policy target |
+
+## Staged Pretraining Plan
+
+### Stage 0: Data Contract And Quality Gate
+
+Use the existing public-sample task suite to define the data contract. Before
+pretraining, every episode must pass a strict manifest check:
+
+- `annotation.hdf5` exists and is readable,
+- video streams are present or missing views are explicitly recorded,
+- audio can be extracted or marked unavailable,
+- depth, pose, mocap, IMU, calibration, and language fields are indexed,
+- windows are aligned by timestamp or frame index,
+- train/val/test splits are episode-level, not window-level leakage splits,
+- raw data remains outside public repos and Hugging Face artifact mirrors.
+
+### Stage 1: 128-1,000 Episode Representation Pilot
+
+Start with a smaller model and a selected subset. The goal is to test whether
+the multimodal objectives train stably and improve held-out task performance.
+
+Recommended scale:
+
+- 128 to 1,000 episodes,
+- frozen or lightly trainable video/audio encoders at first,
+- 0.3B-1B temporal fusion model,
+- all available sensor modalities represented as tokens,
+- evaluation on the existing 12-task suite plus future-state/retrieval probes.
+
+### Stage 2: 10K Episode Domain Model
+
+Scale after the pilot proves value. This stage should train a stronger
+Xperience-specific representation model rather than only fine-tuning a general
+omni model.
+
+Recommended scale:
+
+- thousands to 10K episodes,
+- 1B-3B parameter multimodal temporal model,
+- mixed supervised, contrastive, and predictive objectives,
+- held-out sessions and held-out activities,
+- robustness to missing camera views and sensor dropout.
+
+### Stage 3: Full-Corpus Xperience Embodied Foundation Model
+
+Use this stage only if storage, data throughput, and multi-node compute are
+available. The goal is a domain foundation model over embodied human experience,
+not a general internet-scale language model.
+
+Recommended scale:
+
+- all available Xperience-10M episodes,
+- 3B-7B domain model as a realistic first full-corpus target,
+- larger models only after scaling curves justify the cost,
+- mixture of reconstruction, retrieval, forecasting, language, and world-model
+  objectives,
+- downstream evaluation on held-out episodes, held-out sessions, unseen
+  objects, unseen activities, and downstream robotics/world-model tasks.
+
+## Hardware Requirements
+
+These are planning ranges, not completed run measurements from this repo.
+
+| Training goal | Typical compute | Storage and data path | Practical use |
+| --- | --- | --- | --- |
+| 0.3B-1B pilot | 8-32 modern 80GB-class data-center GPUs | tens of TB plus fast local cache | prove objectives and data loaders |
+| 1B-3B domain model | 32-128 GPUs | 100TB-scale cache, high-throughput decoding | serious research-scale pretraining |
+| 3B-7B full-corpus domain model | 128-512 GPUs | PB-scale storage plus 100-400Gbps networking | first full Xperience-native foundation model |
+| 30B-class omni model from scratch | 512-2,000+ GPUs | PB-scale storage, multi-node orchestration, large checkpoint budget | lab-scale project, not the first target |
+| frontier general omni model | thousands of GPUs | data beyond Xperience-10M plus large infrastructure | out of scope for this project |
+
+For full-corpus work, storage is as important as GPU count:
+
+- raw corpus storage around the official dataset scale,
+- 1.5-3x extra capacity for derived shards, caches, checkpoints, and metadata,
+- fast NVMe cache for active shards,
+- parallel media decoding and feature extraction workers,
+- distributed training with reliable checkpoint/restart,
+- per-episode provenance and split manifests.
+
+## Evaluation Protocol
+
+The model should not be judged only by training loss. Evaluation should include:
+
+- JSON validity and structured task metrics from the current task suite,
+- action/subtask/contact/object metrics on held-out episodes,
+- text-to-window and window-to-text retrieval,
+- future ego-motion and hand-motion forecasting,
+- cross-modal reconstruction and missing-modality robustness,
+- held-out object/activity/session generalization,
+- qualitative inspection of retrieved or generated future states,
+- downstream transfer to Qwen3-Omni, Cosmos-style world modeling, and
+  policy/action branches.
+
+## Relationship To Existing Public Work
+
+The current public project is the harness for this future model:
+
+- the 12-task suite defines concrete input/output contracts,
+- minimal and neural baselines provide initial supervised targets,
+- audio/modality diagnostics show which signals contribute,
+- Qwen3-Omni LoRA provides the first trainable multi-episode adapter path,
+- Cosmos and policy branches define downstream model families,
+- the pretraining goal unifies these into a long-term representation-learning
+  direction.
+
+The next practical step is still selected multi-episode preparation and
+held-out Qwen3-Omni LoRA evaluation. Full-corpus pretraining should come after
+the smaller scaling stages show measurable value.
+
+## Source Links
+
+- Official Xperience-10M dataset: https://huggingface.co/datasets/ropedia-ai/xperience-10m
+- Ropedia Xperience-10M release page: https://ropedia.com/blog/20260316_xperience_10m
+- Ropedia physical-AI data infrastructure page: https://ropedia-dev.com/

data/artifact_index.json

@@ -1,11 +1,11 @@
 {
   "title": "Ropedia Xperience-10M Task Suite Artifact Index",
-  "generated_at_utc": "2026-06-04T16:42:13+00:00",
+  "generated_at_utc": "2026-06-04T20:40:52+00:00",
   "status": "pass",
-  "artifact_count": 72,
+  "artifact_count": 73,
   "missing": [],
   "by_kind": {
-    "project_path": 11,
+    "project_path": 12,
     "project_scope": 1,
     "source_alignment": 5,
     "publication_workflow": 1,
@@ -62,8 +62,8 @@
       "surface": "repo_hf",
       "shows": "Gives a compact current-state table for first-pass readers.",
       "exists": true,
-      "bytes": 7138,
-      "sha256": "67d85a198ee90082e47d790bd0f4d9dafbc97625cd39b17cc94b9785ec25104a"
+      "bytes": 7207,
+      "sha256": "7baaba976ccc254da1a03ee2653057d1e08f3fb0c0cad035886c362442828720"
     },
     {
       "id": "project_status_json",
@@ -73,8 +73,8 @@
       "surface": "website_hf",
       "shows": "Machine-readable copy of the current project status for website and HF mirrors.",
       "exists": true,
-      "bytes": 9169,
-      "sha256": "50d3c87b774c8375dcb897bd363d25e392e5fd6571571c41d56e623df15063f8"
+      "bytes": 9874,
+      "sha256": "600c95726eae3404127a8b2110f35468ff2ba02943cae0fbcd3ea43c66109d3e"
     },
     {
       "id": "research_roadmap",
@@ -84,8 +84,8 @@
       "surface": "repo_hf",
       "shows": "Defines the path from public-sample task development to multi-episode held-out evaluation and larger omni-model extensions.",
       "exists": true,
-      "bytes": 6677,
-      "sha256": "58491bfb68ad3e6b7569bdb1a3cac3de7682a49beb9de368a114d58ebf0b118b"
+      "bytes": 8388,
+      "sha256": "0b3e3356076998ad94dc39f708cc783a4ebeab76c9da661cdd37ea12a3bb3665"
     },
     {
       "id": "research_roadmap_json",
@@ -95,8 +95,8 @@
       "surface": "website_hf",
       "shows": "Machine-readable research roadmap for the website and Hugging Face mirrors.",
       "exists": true,
-      "bytes": 5758,
-      "sha256": "54657eb8824416d2128d6e5710543bdaf9e41d7c2fa46dd14ad6b58fede3b5db"
+      "bytes": 7161,
+      "sha256": "cc96118c2c05108c831616151bc027441f7545495adeeb6a4a6a6bffe8da7801"
     },
     {
       "id": "foundation_model_plan",
@@ -106,8 +106,8 @@
       "surface": "repo_hf",
       "shows": "Defines the post-data-gate backbone choices: Qwen3-Omni first, Cosmos 3 for world modeling, and VLA/policy models after action-target conversion.",
       "exists": true,
-      "bytes": 6559,
-      "sha256": "955be6559b554f1c6c4141dd6ca2818127d89585df3940c2bd9b975ad9047926"
+      "bytes": 9075,
+      "sha256": "444d13ab556d2e16a199a7fca191b87c85ab8685d167aab357bc6341839299a2"
     },
     {
       "id": "foundation_model_plan_json",
@@ -117,8 +117,19 @@
       "surface": "website_hf",
       "shows": "Machine-readable foundation-model selection matrix with source links, entry conditions, and evaluation additions.",
       "exists": true,
-      "bytes": 8889,
-      "sha256": "e9b11114fa290253000b921575586780ccc3ba17665235259d4326c524f6ce97"
+      "bytes": 12981,
+      "sha256": "9cce52025a2e2f8afb4660e2af3353aea6ad0a1af380849218dd74c0acc271bb"
+    },
+    {
+      "id": "xperience_embodied_foundation_pretraining",
+      "title": "Xperience Embodied Foundation Model pretraining goal",
+      "path": "XPERIENCE_EMBODIED_FOUNDATION_MODEL_PRETRAINING.md",
+      "kind": "project_path",
+      "surface": "repo_hf",
+      "shows": "Describes the future full-corpus Xperience-native pretraining goal, target modules, objectives, staged scale-up, hardware ranges, and evaluation protocol.",
+      "exists": true,
+      "bytes": 9182,
+      "sha256": "b5a6ddc58647cd895a4772b110ecc9f4d685427fb37b81b22c6c02d2b9b323f1"
     },
     {
       "id": "evidence_contract",
@@ -150,8 +161,8 @@
       "surface": "repo_hf",
       "shows": "Gives the human-readable map from project scope to data, tasks, platform mirrors, and scale-up status.",
       "exists": true,
-      "bytes": 16890,
-      "sha256": "8bce9a773daf36214e377a7154b72a4493efd0f7d1a1941d5e0fc9bf784a29e5"
+      "bytes": 11440,
+      "sha256": "9b8821a9b14fe1744f2e6b5c419b2c5daaf70b57f1944caf1105c36c0c66c119"
     },
     {
       "id": "official_dataset_card_alignment",
@@ -195,7 +206,7 @@
       "shows": "Machine-readable source-alignment pass/fail check for repo, website, and HF surfaces.",
       "exists": true,
       "bytes": 4432,
-      "sha256": "96c7adc61c869fab71ef34ec2f6ec4f5f88af844509bd3d51d3818732d1f84b6"
+      "sha256": "06c6e2d111c72df01ed127fd288e6675b63e35a21ae12a2523931a072bd0bc49"
     },
     {
       "id": "source_alignment_validator",
@@ -573,8 +584,8 @@
       "surface": "repo_hf",
       "shows": "Generates the selective artifact catalog from local files.",
       "exists": true,
-      "bytes": 26568,
-      "sha256": "a611b399e858560f6afb41e121f033724753c5167d04e0d7bf243e569de88f04"
+      "bytes": 27020,
+      "sha256": "0ca7ed96f24caecbab31687cffa99f0eba8471258986412a294614e688c5aff5"
     },
     {
       "id": "publication_audit",
@@ -585,7 +596,7 @@
       "volatile": true,
       "shows": "Confirms public bundles exclude raw data, caches, heavy archives, and credential text.",
       "exists": true,
-      "bytes": 7289,
+      "bytes": 11811,
       "hash_policy": "existence_and_size_only"
     },
     {
@@ -597,7 +608,7 @@
       "volatile": true,
       "shows": "Separates setup paths from completed held-out-episode results.",
       "exists": true,
-      "bytes": 19505,
+      "bytes": 18981,
       "hash_policy": "existence_and_size_only"
     },
     {
@@ -609,7 +620,7 @@
       "volatile": true,
       "shows": "Confirms prepared GitHub/HF Space/artifact/model mirrors share the same critical data, figure, website HTML, and validator files.",
       "exists": true,
-      "bytes": 108617,
+      "bytes": 108621,
       "hash_policy": "existence_and_size_only"
     },
     {
@@ -621,7 +632,7 @@
       "volatile": true,
       "shows": "Confirms local website links, anchors, JSON data files, and referenced images resolve.",
       "exists": true,
-      "bytes": 14923,
+      "bytes": 14891,
       "hash_policy": "existence_and_size_only"
     },
     {

data/foundation_model_plan.json

@@ -2,6 +2,16 @@
   "title": "Xperience-10M Foundation Model Plan",
   "status": "planning_artifact",
   "current_boundary": "No held-out multi-episode foundation-model result has been completed in this repo. The current foundation-model artifacts are setup-stage until enough valid episodes are prepared and evaluated.",
+  "backbone_registry": {
+    "config_dir": "configs/omni_backbones",
+    "validator": "scripts/omni/backbone_registry.py --validate --json",
+    "extension_contract": "OMNI_MODEL_EXTENSION_CONTRACT.md",
+    "implemented_backbone": "qwen3_omni_lora",
+    "planned_backbones": [
+      "cosmos_world_model",
+      "policy_vla_branch"
+    ]
+  },
   "decision": {
     "immediate_trainable_backbone": "Qwen3-Omni",
     "first_world_model_branch": "Cosmos 3",
@@ -10,7 +20,65 @@
       "openpi pi0/pi0.5",
       "NVIDIA GR00T"
     ],
-    "external_reasoning_reference": "Gemini Robotics"
+    "external_reasoning_reference": "Gemini Robotics",
+    "long_term_native_pretraining_goal": "Xperience Embodied Foundation Model"
+  },
+  "future_pretraining_goal": {
+    "name": "Xperience Embodied Foundation Model",
+    "status": "future_planning_goal",
+    "role": "Domain-specific embodied foundation model pretrained on full Xperience-10M if full-corpus data, storage, and compute become available.",
+    "not_current_result": true,
+    "document": "XPERIENCE_EMBODIED_FOUNDATION_MODEL_PRETRAINING.md",
+    "entry_conditions": [
+      "Selected multi-episode Qwen3-Omni pilot trains and evaluates cleanly.",
+      "Scaling from 128 episodes to thousands of episodes shows measurable value.",
+      "Full-corpus storage, derived-shard storage, and fast active-cache capacity are available.",
+      "Distributed training, checkpoint/restart, and provenance tracking are reliable.",
+      "Evaluation covers held-out episodes, sessions, activities, objects, and missing-modality robustness."
+    ],
+    "target_modules": [
+      "multi-view video encoder",
+      "audio encoder",
+      "depth and geometry encoder",
+      "pose/SLAM encoder",
+      "hand/body mocap encoder",
+      "IMU encoder",
+      "language encoder/decoder",
+      "temporal fusion transformer",
+      "task heads and decoders"
+    ],
+    "pretraining_objectives": [
+      "masked multimodal modeling",
+      "cross-modal contrastive alignment",
+      "future-state prediction",
+      "ego-motion and hand-motion forecasting",
+      "action and procedure prediction",
+      "language grounding and captioning",
+      "contact and affordance prediction",
+      "optional policy-style targets after action conversion"
+    ],
+    "hardware_ranges": [
+      {
+        "goal": "0.3B-1B pilot",
+        "compute": "8-32 modern 80GB-class data-center GPUs",
+        "use": "prove objectives and data loaders"
+      },
+      {
+        "goal": "1B-3B domain model",
+        "compute": "32-128 GPUs",
+        "use": "research-scale Xperience representation learning"
+      },
+      {
+        "goal": "3B-7B full-corpus domain model",
+        "compute": "128-512 GPUs",
+        "use": "first realistic full Xperience-native foundation model"
+      },
+      {
+        "goal": "30B-class omni model from scratch",
+        "compute": "512-2000+ GPUs",
+        "use": "lab-scale project after scaling curves justify cost"
+      }
+    ]
   },
   "model_families": [
     {
@@ -112,6 +180,21 @@
       "current_decision": "optional_baseline_after_data_staging",
       "entry_condition": "Action labels and baseline protocol exist.",
       "public_source": "https://github.com/huggingface/lerobot"
+    },
+    {
+      "priority": 8,
+      "family": "Xperience Embodied Foundation Model",
+      "category": "xperience_native_pretraining_goal",
+      "openness": "future project-specific model if full-corpus access and compute exist",
+      "best_role": "Domain model over synchronized embodied experience.",
+      "xperience10m_fit": [
+        "Uses the full aligned modality stack rather than treating sensors as auxiliary metadata.",
+        "Targets temporal embodied representation learning across perception, motion, geometry, audio, and language.",
+        "Can become the shared pretraining backbone for Qwen-style instruction tasks, Cosmos-style world modeling, and policy/action branches."
+      ],
+      "current_decision": "future_goal_after_scaling_evidence",
+      "entry_condition": "Full-corpus data path, PB-scale storage, multi-node compute, and positive smaller-run scaling evidence.",
+      "public_source": "XPERIENCE_EMBODIED_FOUNDATION_MODEL_PRETRAINING.md"
     }
   ],
   "execution_order": [
@@ -144,6 +227,11 @@
       "step": 6,
       "name": "Publishing threshold",
       "action": "Publish branch results only with real manifests, predictions, metrics, and qualitative examples."
+    },
+    {
+      "step": 7,
+      "name": "Xperience-native pretraining",
+      "action": "Start a from-scratch Xperience Embodied Foundation Model only after smaller scaling stages, full-corpus storage, multi-node compute, and held-out evaluation protocols are in place."
     }
   ],
   "evaluation_additions": [
@@ -230,6 +318,10 @@
     {
       "label": "LeRobot / SmolVLA",
       "url": "https://github.com/huggingface/lerobot"
+    },
+    {
+      "label": "Xperience Embodied Foundation Model pretraining plan",
+      "url": "XPERIENCE_EMBODIED_FOUNDATION_MODEL_PRETRAINING.md"
     }
   ]
 }

data/mirror_parity.json

@@ -1,6 +1,6 @@
 {
   "status": "pass",
-  "generated_at_utc": "2026-06-04T18:33:44+00:00",
+  "generated_at_utc": "2026-06-04T20:45:22+00:00",
   "hf_root": "hf_publish",
   "summary": {
     "group_count": 101,
@@ -71,27 +71,27 @@
       "local": {
         "path": "repo:docs/data/artifact_index.json",
         "exists": true,
-        "bytes": 32296,
-        "sha256": "5494e5ee1e40bc50d44a9cd6f77c8de694175939bda4f174fb5b1554e53ec508"
+        "bytes": 32864,
+        "sha256": "ec7d17898c42fd76109567c201f9638059b6a9a11a48817b32677a0eb2662178"
       },
       "mirrors": {
         "hf_space": {
           "path": "hf_space:data/artifact_index.json",
           "exists": true,
-          "bytes": 32296,
-          "sha256": "5494e5ee1e40bc50d44a9cd6f77c8de694175939bda4f174fb5b1554e53ec508"
+          "bytes": 32864,
+          "sha256": "ec7d17898c42fd76109567c201f9638059b6a9a11a48817b32677a0eb2662178"
         },
         "hf_artifacts": {
           "path": "hf_artifacts:docs/data/artifact_index.json",
           "exists": true,
-          "bytes": 32296,
-          "sha256": "5494e5ee1e40bc50d44a9cd6f77c8de694175939bda4f174fb5b1554e53ec508"
+          "bytes": 32864,
+          "sha256": "ec7d17898c42fd76109567c201f9638059b6a9a11a48817b32677a0eb2662178"
         },
         "hf_model": {
           "path": "hf_model:metrics/artifact_index.json",
           "exists": true,
-          "bytes": 32296,
-          "sha256": "5494e5ee1e40bc50d44a9cd6f77c8de694175939bda4f174fb5b1554e53ec508"
+          "bytes": 32864,
+          "sha256": "ec7d17898c42fd76109567c201f9638059b6a9a11a48817b32677a0eb2662178"
         }
       },
       "failures": []
@@ -226,27 +226,27 @@
       "local": {
         "path": "repo:docs/data/foundation_model_plan.json",
         "exists": true,
-        "bytes": 8889,
-        "sha256": "e9b11114fa290253000b921575586780ccc3ba17665235259d4326c524f6ce97"
+        "bytes": 12981,
+        "sha256": "9cce52025a2e2f8afb4660e2af3353aea6ad0a1af380849218dd74c0acc271bb"
       },
       "mirrors": {
         "hf_space": {
           "path": "hf_space:data/foundation_model_plan.json",
           "exists": true,
-          "bytes": 8889,
-          "sha256": "e9b11114fa290253000b921575586780ccc3ba17665235259d4326c524f6ce97"
+          "bytes": 12981,
+          "sha256": "9cce52025a2e2f8afb4660e2af3353aea6ad0a1af380849218dd74c0acc271bb"
         },
         "hf_artifacts": {
           "path": "hf_artifacts:docs/data/foundation_model_plan.json",
           "exists": true,
-          "bytes": 8889,
-          "sha256": "e9b11114fa290253000b921575586780ccc3ba17665235259d4326c524f6ce97"
+          "bytes": 12981,
+          "sha256": "9cce52025a2e2f8afb4660e2af3353aea6ad0a1af380849218dd74c0acc271bb"
         },
         "hf_model": {
           "path": "hf_model:metrics/foundation_model_plan.json",
           "exists": true,
-          "bytes": 8889,
-          "sha256": "e9b11114fa290253000b921575586780ccc3ba17665235259d4326c524f6ce97"
+          "bytes": 12981,
+          "sha256": "9cce52025a2e2f8afb4660e2af3353aea6ad0a1af380849218dd74c0acc271bb"
         }
       },
       "failures": []
@@ -412,27 +412,27 @@
       "local": {
         "path": "repo:docs/data/project_status.json",
         "exists": true,
-        "bytes": 9169,
-        "sha256": "50d3c87b774c8375dcb897bd363d25e392e5fd6571571c41d56e623df15063f8"
+        "bytes": 9874,
+        "sha256": "600c95726eae3404127a8b2110f35468ff2ba02943cae0fbcd3ea43c66109d3e"
       },
       "mirrors": {
         "hf_space": {
           "path": "hf_space:data/project_status.json",
           "exists": true,
-          "bytes": 9169,
-          "sha256": "50d3c87b774c8375dcb897bd363d25e392e5fd6571571c41d56e623df15063f8"
+          "bytes": 9874,
+          "sha256": "600c95726eae3404127a8b2110f35468ff2ba02943cae0fbcd3ea43c66109d3e"
         },
         "hf_artifacts": {
           "path": "hf_artifacts:docs/data/project_status.json",
           "exists": true,
-          "bytes": 9169,
-          "sha256": "50d3c87b774c8375dcb897bd363d25e392e5fd6571571c41d56e623df15063f8"
+          "bytes": 9874,
+          "sha256": "600c95726eae3404127a8b2110f35468ff2ba02943cae0fbcd3ea43c66109d3e"
         },
         "hf_model": {
           "path": "hf_model:metrics/project_status.json",
           "exists": true,
-          "bytes": 9169,
-          "sha256": "50d3c87b774c8375dcb897bd363d25e392e5fd6571571c41d56e623df15063f8"
+          "bytes": 9874,
+          "sha256": "600c95726eae3404127a8b2110f35468ff2ba02943cae0fbcd3ea43c66109d3e"
         }
       },
       "failures": []
@@ -444,26 +444,26 @@
         "path": "repo:docs/data/publication_audit.json",
         "exists": true,
         "bytes": 7237,
-        "sha256": "a95c93592ba70709b2fad24a911d19329e6823f25862cd4fcb256788190dd0f2"
+        "sha256": "7fbb19f8990b1a4d902e282c010d27e4391755564fa68af97d96c298c6b054f8"
       },
       "mirrors": {
         "hf_space": {
           "path": "hf_space:data/publication_audit.json",
           "exists": true,
           "bytes": 7237,
-          "sha256": "a95c93592ba70709b2fad24a911d19329e6823f25862cd4fcb256788190dd0f2"
+          "sha256": "7fbb19f8990b1a4d902e282c010d27e4391755564fa68af97d96c298c6b054f8"
         },
         "hf_artifacts": {
           "path": "hf_artifacts:docs/data/publication_audit.json",
           "exists": true,
           "bytes": 7237,
-          "sha256": "a95c93592ba70709b2fad24a911d19329e6823f25862cd4fcb256788190dd0f2"
+          "sha256": "7fbb19f8990b1a4d902e282c010d27e4391755564fa68af97d96c298c6b054f8"
         },
         "hf_model": {
           "path": "hf_model:metrics/publication_audit.json",
           "exists": true,
           "bytes": 7237,
-          "sha256": "a95c93592ba70709b2fad24a911d19329e6823f25862cd4fcb256788190dd0f2"
+          "sha256": "7fbb19f8990b1a4d902e282c010d27e4391755564fa68af97d96c298c6b054f8"
         }
       },
       "failures": []
@@ -598,27 +598,27 @@
       "local": {
         "path": "repo:docs/data/research_roadmap.json",
         "exists": true,
-        "bytes": 5758,
-        "sha256": "54657eb8824416d2128d6e5710543bdaf9e41d7c2fa46dd14ad6b58fede3b5db"
+        "bytes": 7161,
+        "sha256": "cc96118c2c05108c831616151bc027441f7545495adeeb6a4a6a6bffe8da7801"
       },
       "mirrors": {
         "hf_space": {
           "path": "hf_space:data/research_roadmap.json",
           "exists": true,
-          "bytes": 5758,
-          "sha256": "54657eb8824416d2128d6e5710543bdaf9e41d7c2fa46dd14ad6b58fede3b5db"
+          "bytes": 7161,
+          "sha256": "cc96118c2c05108c831616151bc027441f7545495adeeb6a4a6a6bffe8da7801"
         },
         "hf_artifacts": {
           "path": "hf_artifacts:docs/data/research_roadmap.json",
           "exists": true,
-          "bytes": 5758,
-          "sha256": "54657eb8824416d2128d6e5710543bdaf9e41d7c2fa46dd14ad6b58fede3b5db"
+          "bytes": 7161,
+          "sha256": "cc96118c2c05108c831616151bc027441f7545495adeeb6a4a6a6bffe8da7801"
         },
         "hf_model": {
           "path": "hf_model:metrics/research_roadmap.json",
           "exists": true,
-          "bytes": 5758,
-          "sha256": "54657eb8824416d2128d6e5710543bdaf9e41d7c2fa46dd14ad6b58fede3b5db"
+          "bytes": 7161,
+          "sha256": "cc96118c2c05108c831616151bc027441f7545495adeeb6a4a6a6bffe8da7801"
         }
       },
       "failures": []
@@ -629,27 +629,27 @@
       "local": {
         "path": "repo:docs/data/research_roadmap_interactive.json",
         "exists": true,
-        "bytes": 131519,
-        "sha256": "004fbcc7a3582da88dd66504d686604ecb0f04f65c9c8166bb0583e0fc174274"
+        "bytes": 134282,
+        "sha256": "ff37219a9f1d9b386a9d4c42766e4aa28f10ce6ef338dceeedd6bdb4a1b2c40a"
       },
       "mirrors": {
         "hf_space": {
           "path": "hf_space:data/research_roadmap_interactive.json",
           "exists": true,
-          "bytes": 131519,
-          "sha256": "004fbcc7a3582da88dd66504d686604ecb0f04f65c9c8166bb0583e0fc174274"
+          "bytes": 134282,
+          "sha256": "ff37219a9f1d9b386a9d4c42766e4aa28f10ce6ef338dceeedd6bdb4a1b2c40a"
         },
         "hf_artifacts": {
           "path": "hf_artifacts:docs/data/research_roadmap_interactive.json",
           "exists": true,
-          "bytes": 131519,
-          "sha256": "004fbcc7a3582da88dd66504d686604ecb0f04f65c9c8166bb0583e0fc174274"
+          "bytes": 134282,
+          "sha256": "ff37219a9f1d9b386a9d4c42766e4aa28f10ce6ef338dceeedd6bdb4a1b2c40a"
         },
         "hf_model": {
           "path": "hf_model:metrics/research_roadmap_interactive.json",
           "exists": true,
-          "bytes": 131519,
-          "sha256": "004fbcc7a3582da88dd66504d686604ecb0f04f65c9c8166bb0583e0fc174274"
+          "bytes": 134282,
+          "sha256": "ff37219a9f1d9b386a9d4c42766e4aa28f10ce6ef338dceeedd6bdb4a1b2c40a"
         }
       },
       "failures": []
@@ -1692,21 +1692,21 @@
       "local": {
         "path": "repo:scripts/build_artifact_index.py",
         "exists": true,
-        "bytes": 26568,
-        "sha256": "a611b399e858560f6afb41e121f033724753c5167d04e0d7bf243e569de88f04"
+        "bytes": 27020,
+        "sha256": "0ca7ed96f24caecbab31687cffa99f0eba8471258986412a294614e688c5aff5"
       },
       "mirrors": {
         "hf_artifacts": {
           "path": "hf_artifacts:scripts/build_artifact_index.py",
           "exists": true,
-          "bytes": 26568,
-          "sha256": "a611b399e858560f6afb41e121f033724753c5167d04e0d7bf243e569de88f04"
+          "bytes": 27020,
+          "sha256": "0ca7ed96f24caecbab31687cffa99f0eba8471258986412a294614e688c5aff5"
         },
         "hf_model": {
           "path": "hf_model:scripts/build_artifact_index.py",
           "exists": true,
-          "bytes": 26568,
-          "sha256": "a611b399e858560f6afb41e121f033724753c5167d04e0d7bf243e569de88f04"
+          "bytes": 27020,
+          "sha256": "0ca7ed96f24caecbab31687cffa99f0eba8471258986412a294614e688c5aff5"
         }
       },
       "failures": []
@@ -2017,21 +2017,21 @@
       "local": {
         "path": "repo:scripts/validate_publication_package.py",
         "exists": true,
-        "bytes": 17125,
-        "sha256": "51febee7a4caa4e3cbb3833c0c13ac502bd7106fdb3df06e868ed00bc8f9fd9e"
+        "bytes": 17197,
+        "sha256": "2a617f3204ffb8c59d1c5bc1828b4441a4d014bb531655fd0613e128a6d9abc2"
       },
       "mirrors": {
         "hf_artifacts": {
           "path": "hf_artifacts:scripts/validate_publication_package.py",
           "exists": true,
-          "bytes": 17125,
-          "sha256": "51febee7a4caa4e3cbb3833c0c13ac502bd7106fdb3df06e868ed00bc8f9fd9e"
+          "bytes": 17197,
+          "sha256": "2a617f3204ffb8c59d1c5bc1828b4441a4d014bb531655fd0613e128a6d9abc2"
         },
         "hf_model": {
           "path": "hf_model:scripts/validate_publication_package.py",
           "exists": true,
-          "bytes": 17125,
-          "sha256": "51febee7a4caa4e3cbb3833c0c13ac502bd7106fdb3df06e868ed00bc8f9fd9e"
+          "bytes": 17197,
+          "sha256": "2a617f3204ffb8c59d1c5bc1828b4441a4d014bb531655fd0613e128a6d9abc2"
         }
       },
       "failures": []
@@ -2217,21 +2217,21 @@
       "local": {
         "path": "repo:docs/index.html",
         "exists": true,
-        "bytes": 172286,
-        "sha256": "a736850416c0061adddbb6ced5897efd1add499ec26e510b6fe21a4945b341c8"
+        "bytes": 174923,
+        "sha256": "099fcc01cbb4d50f62c508b10f343f05b1c883962b85bda294bcede99af2a0f1"
       },
       "mirrors": {
         "hf_space": {
           "path": "hf_space:index.html",
           "exists": true,
-          "bytes": 172286,
-          "sha256": "a736850416c0061adddbb6ced5897efd1add499ec26e510b6fe21a4945b341c8"
+          "bytes": 174923,
+          "sha256": "099fcc01cbb4d50f62c508b10f343f05b1c883962b85bda294bcede99af2a0f1"
         },
         "hf_artifacts_docs": {
           "path": "hf_artifacts:docs/index.html",
           "exists": true,
-          "bytes": 172286,
-          "sha256": "a736850416c0061adddbb6ced5897efd1add499ec26e510b6fe21a4945b341c8"
+          "bytes": 174923,
+          "sha256": "099fcc01cbb4d50f62c508b10f343f05b1c883962b85bda294bcede99af2a0f1"
         }
       },
       "failures": []
@@ -2242,21 +2242,21 @@
       "local": {
         "path": "repo:docs/research_roadmap.html",
         "exists": true,
-        "bytes": 31554,
-        "sha256": "f51e83a4495f2d2012ec4c48191d66ca4456a00d7fcb335a427b7d86afc66109"
+        "bytes": 31702,
+        "sha256": "1b20a5cc342b3ba59ad808eed9f5bf978e2d9ac438c88b5c3eeba01f4e14b883"
       },
       "mirrors": {
         "hf_space": {
           "path": "hf_space:research_roadmap.html",
           "exists": true,
-          "bytes": 31554,
-          "sha256": "f51e83a4495f2d2012ec4c48191d66ca4456a00d7fcb335a427b7d86afc66109"
+          "bytes": 31702,
+          "sha256": "1b20a5cc342b3ba59ad808eed9f5bf978e2d9ac438c88b5c3eeba01f4e14b883"
         },
         "hf_artifacts_docs": {
           "path": "hf_artifacts:docs/research_roadmap.html",
           "exists": true,
-          "bytes": 31554,
-          "sha256": "f51e83a4495f2d2012ec4c48191d66ca4456a00d7fcb335a427b7d86afc66109"
+          "bytes": 31702,
+          "sha256": "1b20a5cc342b3ba59ad808eed9f5bf978e2d9ac438c88b5c3eeba01f4e14b883"
         }
       },
       "failures": []
@@ -2844,27 +2844,27 @@
       "local": {
         "path": "repo:FOUNDATION_MODEL_PLAN.md",
         "exists": true,
-        "bytes": 6559,
-        "sha256": "955be6559b554f1c6c4141dd6ca2818127d89585df3940c2bd9b975ad9047926"
+        "bytes": 9075,
+        "sha256": "444d13ab556d2e16a199a7fca191b87c85ab8685d167aab357bc6341839299a2"
       },
       "mirrors": {
         "hf_space": {
           "path": "hf_space:FOUNDATION_MODEL_PLAN.md",
           "exists": true,
-          "bytes": 6559,
-          "sha256": "955be6559b554f1c6c4141dd6ca2818127d89585df3940c2bd9b975ad9047926"
+          "bytes": 9075,
+          "sha256": "444d13ab556d2e16a199a7fca191b87c85ab8685d167aab357bc6341839299a2"
         },
         "hf_artifacts": {
           "path": "hf_artifacts:FOUNDATION_MODEL_PLAN.md",
           "exists": true,
-          "bytes": 6559,
-          "sha256": "955be6559b554f1c6c4141dd6ca2818127d89585df3940c2bd9b975ad9047926"
+          "bytes": 9075,
+          "sha256": "444d13ab556d2e16a199a7fca191b87c85ab8685d167aab357bc6341839299a2"
         },
         "hf_model": {
           "path": "hf_model:FOUNDATION_MODEL_PLAN.md",
           "exists": true,
-          "bytes": 6559,
-          "sha256": "955be6559b554f1c6c4141dd6ca2818127d89585df3940c2bd9b975ad9047926"
+          "bytes": 9075,
+          "sha256": "444d13ab556d2e16a199a7fca191b87c85ab8685d167aab357bc6341839299a2"
         }
       },
       "failures": []
@@ -2937,27 +2937,27 @@
       "local": {
         "path": "repo:RESEARCH_ROADMAP.md",
         "exists": true,
-        "bytes": 6677,
-        "sha256": "58491bfb68ad3e6b7569bdb1a3cac3de7682a49beb9de368a114d58ebf0b118b"
+        "bytes": 8388,
+        "sha256": "0b3e3356076998ad94dc39f708cc783a4ebeab76c9da661cdd37ea12a3bb3665"
       },
       "mirrors": {
         "hf_space": {
           "path": "hf_space:RESEARCH_ROADMAP.md",
           "exists": true,
-          "bytes": 6677,
-          "sha256": "58491bfb68ad3e6b7569bdb1a3cac3de7682a49beb9de368a114d58ebf0b118b"
+          "bytes": 8388,
+          "sha256": "0b3e3356076998ad94dc39f708cc783a4ebeab76c9da661cdd37ea12a3bb3665"
         },
         "hf_artifacts": {
           "path": "hf_artifacts:RESEARCH_ROADMAP.md",
           "exists": true,
-          "bytes": 6677,
-          "sha256": "58491bfb68ad3e6b7569bdb1a3cac3de7682a49beb9de368a114d58ebf0b118b"
+          "bytes": 8388,
+          "sha256": "0b3e3356076998ad94dc39f708cc783a4ebeab76c9da661cdd37ea12a3bb3665"
         },
         "hf_model": {
           "path": "hf_model:RESEARCH_ROADMAP.md",
           "exists": true,
-          "bytes": 6677,
-          "sha256": "58491bfb68ad3e6b7569bdb1a3cac3de7682a49beb9de368a114d58ebf0b118b"
+          "bytes": 8388,
+          "sha256": "0b3e3356076998ad94dc39f708cc783a4ebeab76c9da661cdd37ea12a3bb3665"
         }
       },
       "failures": []
@@ -2968,27 +2968,27 @@
       "local": {
         "path": "repo:PROJECT_STATUS.md",
         "exists": true,
-        "bytes": 6648,
-        "sha256": "b052c725472f1d59232918a4d5b0f3668534c1e25e24189307159f5a0157d58f"
+        "bytes": 7207,
+        "sha256": "7baaba976ccc254da1a03ee2653057d1e08f3fb0c0cad035886c362442828720"
       },
       "mirrors": {
         "hf_space": {
           "path": "hf_space:PROJECT_STATUS.md",
           "exists": true,
-          "bytes": 6648,
-          "sha256": "b052c725472f1d59232918a4d5b0f3668534c1e25e24189307159f5a0157d58f"
+          "bytes": 7207,
+          "sha256": "7baaba976ccc254da1a03ee2653057d1e08f3fb0c0cad035886c362442828720"
         },
         "hf_artifacts": {
           "path": "hf_artifacts:PROJECT_STATUS.md",
           "exists": true,
-          "bytes": 6648,
-          "sha256": "b052c725472f1d59232918a4d5b0f3668534c1e25e24189307159f5a0157d58f"
+          "bytes": 7207,
+          "sha256": "7baaba976ccc254da1a03ee2653057d1e08f3fb0c0cad035886c362442828720"
         },
         "hf_model": {
           "path": "hf_model:PROJECT_STATUS.md",
           "exists": true,
-          "bytes": 6648,
-          "sha256": "b052c725472f1d59232918a4d5b0f3668534c1e25e24189307159f5a0157d58f"
+          "bytes": 7207,
+          "sha256": "7baaba976ccc254da1a03ee2653057d1e08f3fb0c0cad035886c362442828720"
         }
       },
       "failures": []

data/project_status.json

@@ -82,7 +82,7 @@
                 "RESEARCH_ROADMAP.md",
                 "docs/data/research_roadmap.json"
             ],
-            "readout": "The roadmap connects public-sample task development to 128-episode data preparation, Qwen3-Omni LoRA, foundation-model selection, robustness runs, and larger omni/world-model extensions."
+            "readout": "The roadmap connects public-sample task development to 128-episode data preparation, Qwen3-Omni LoRA, foundation-model selection, robustness runs, world/policy branches, and the future Xperience-native pretraining goal."
         },
         {
             "area": "Foundation-model plan",
@@ -93,6 +93,14 @@
             ],
             "readout": "Qwen3-Omni remains the first trainable held-out LoRA baseline; Cosmos 3 is added as the first world-model/action-generation branch; OpenVLA/openpi/GR00T are policy candidates after action targets are explicit."
         },
+        {
+            "area": "Xperience Embodied Foundation Model",
+            "status": "future_goal",
+            "evidence": [
+                "XPERIENCE_EMBODIED_FOUNDATION_MODEL_PRETRAINING.md"
+            ],
+            "readout": "A future full-corpus pretraining plan describes target modules, objectives, staged scale-up, hardware ranges, and evaluation for a domain-specific embodied foundation model."
+        },
         {
             "area": "Official dataset wording",
             "status": "verified",
@@ -167,6 +175,7 @@
         "Inspect RESEARCH_TAKEAWAYS.md and docs/data/research_takeaways.json before interpreting model scores.",
         "Inspect RESEARCH_ROADMAP.md and docs/data/research_roadmap.json for the path from public-sample task work to multi-episode modeling.",
         "Inspect FOUNDATION_MODEL_PLAN.md and docs/data/foundation_model_plan.json before choosing a backbone branch.",
+        "Inspect XPERIENCE_EMBODIED_FOUNDATION_MODEL_PRETRAINING.md for the long-term full-corpus pretraining goal.",
         "Inspect docs/data/summary_metrics.json and results/episode_task_suite/neural_mlp/ to check the 12-task outputs.",
         "Inspect results/audio_ablation/AUDIO_ABLATION_SUMMARY.md before judging whether audio helps the current task suite.",
         "Inspect EVALUATION_PROTOCOL.md before judging task metrics or leakage controls.",
@@ -180,6 +189,7 @@
     "The current reconstruction task reconstructs feature vectors, not pixel-depth, mesh, NeRF, or Gaussian reconstruction.",
     "Audio is one of the synchronized source modalities in the current task representation.",
     "The audio ablation report compares audio/no-audio variants across all 12 task contracts in results/audio_ablation/.",
-    "Foundation-model selection is explicit: Qwen3-Omni is the immediate trainable pilot, Cosmos 3 is the first world-model branch, and policy models such as OpenVLA/openpi/GR00T wait for action-target conversion."
+    "Foundation-model selection is explicit: Qwen3-Omni is the immediate trainable pilot, Cosmos 3 is the first world-model branch, and policy models such as OpenVLA/openpi/GR00T wait for action-target conversion.",
+    "The Xperience Embodied Foundation Model is a future native-pretraining goal, not a completed model or current benchmark."
   ]
 }

data/publication_audit.json

@@ -1,6 +1,6 @@
 {
   "status": "pass",
-  "generated_at_utc": "2026-06-04T18:32:51+00:00",
+  "generated_at_utc": "2026-06-04T20:43:37+00:00",
   "checks": [
     {
       "name": "required_publication_assets_present",
@@ -182,8 +182,8 @@
     "github_repo": {
       "root": "repo",
       "exists": true,
-      "file_count": 386,
-      "text_file_count": 320,
+      "file_count": 396,
+      "text_file_count": 330,
       "largest_file": {
         "path": "results/episode_task_suite/modality_reconstruction/predictions.npz",
         "bytes": 55702978
@@ -193,8 +193,8 @@
     "hf_space_bundle": {
       "root": "hf_publish/space",
       "exists": true,
-      "file_count": 316,
-      "text_file_count": 250,
+      "file_count": 317,
+      "text_file_count": 251,
       "largest_file": {
         "path": "results/episode_task_suite/modality_reconstruction/predictions.npz",
         "bytes": 55702978
@@ -204,8 +204,8 @@
     "hf_artifact_bundle": {
       "root": "hf_publish/artifacts",
       "exists": true,
-      "file_count": 417,
-      "text_file_count": 329,
+      "file_count": 418,
+      "text_file_count": 330,
       "largest_file": {
         "path": "results/episode_task_suite/modality_reconstruction/predictions.npz",
         "bytes": 55702978
@@ -215,8 +215,8 @@
     "hf_model_bundle": {
       "root": "hf_publish/model",
       "exists": true,
-      "file_count": 643,
-      "text_file_count": 518,
+      "file_count": 644,
+      "text_file_count": 519,
       "largest_file": {
         "path": "pytorch_model.bin",
         "bytes": 93495480

data/research_roadmap.json

@@ -1,7 +1,7 @@
 {
   "title": "Ropedia Xperience-10M Research Roadmap",
-  "summary": "Staged path from the public-sample task lab to multi-episode held-out evaluation, foundation-model selection, and larger omni/world-model extensions.",
-  "current_decision_point": "Keep the public-sample task suite as the development harness, prepare the selected official Xperience-10M episodes for the held-out Qwen3-Omni pilot, then branch into Cosmos 3 world modeling and policy-model experiments after the data preparation path is stable.",
+  "summary": "Staged path from the public-sample task lab to multi-episode held-out evaluation, foundation-model selection, world/policy branches, and a future Xperience-native embodied foundation model.",
+  "current_decision_point": "Keep the public-sample task suite as the development harness, prepare the selected official Xperience-10M episodes for the held-out Qwen3-Omni pilot, then branch into Cosmos 3 world modeling and policy-model experiments after the data preparation path is stable. The Xperience Embodied Foundation Model is a later full-corpus pretraining goal, not a current result.",
   "phases": [
     {
       "id": "public_sample_task_lab",
@@ -126,6 +126,30 @@
         "updated model cards"
       ],
       "reader_takeaway": "The long-term direction is richer multimodal representation learning for embodied-AI reasoning, with model branches chosen by task fit rather than by a single default backbone."
+    },
+    {
+      "id": "xperience_embodied_foundation_pretraining",
+      "name": "Xperience Embodied Foundation Model Pretraining",
+      "status": "future",
+      "entry_condition": "Full-corpus access, PB-scale storage path, high-throughput data loading, multi-node compute, and positive scaling evidence from smaller multi-episode runs.",
+      "deliverables": [
+        "full-corpus episode and split manifests",
+        "pretraining shard and provenance manifests",
+        "0.3B-1B and 1B-3B scaling pilots",
+        "3B-7B Xperience-native domain model target",
+        "held-out episode/session/activity/object evaluations",
+        "missing-modality robustness report",
+        "model card and data-boundary report"
+      ],
+      "completion_evidence": [
+        "pretraining metadata",
+        "checkpoint inventory",
+        "scaling curves",
+        "held-out evaluation reports",
+        "qualitative retrieval or future-state examples",
+        "safety and data-boundary report"
+      ],
+      "reader_takeaway": "The final research direction is a domain-specific embodied foundation model trained directly on Xperience-10M, after smaller pilots justify the cost and infrastructure."
     }
   ],
   "public_surfaces_to_update": [
@@ -134,6 +158,7 @@
     "RESEARCH_TAKEAWAYS.md",
     "EVALUATION_PROTOCOL.md",
     "ARTIFACT_GUIDE.md",
+    "XPERIENCE_EMBODIED_FOUNDATION_MODEL_PRETRAINING.md",
     "docs/index.html",
     "docs/data/research_roadmap.json",
     "Hugging Face Space card",

data/research_roadmap_interactive.json

@@ -1837,7 +1837,8 @@
         "NVIDIA GR00T"
       ],
       "first_world_model_branch": "Cosmos 3",
-      "immediate_trainable_backbone": "Qwen3-Omni"
+      "immediate_trainable_backbone": "Qwen3-Omni",
+      "long_term_native_pretraining_goal": "Xperience Embodied Foundation Model"
     },
     "evaluation_additions": [
       {
@@ -1921,6 +1922,11 @@
         "action": "Publish branch results only with real manifests, predictions, metrics, and qualitative examples.",
         "name": "Publishing threshold",
         "step": 6
+      },
+      {
+        "action": "Start a from-scratch Xperience Embodied Foundation Model only after smaller scaling stages, full-corpus storage, multi-node compute, and held-out evaluation protocols are in place.",
+        "name": "Xperience-native pretraining",
+        "step": 7
       }
     ],
     "model_families": [
@@ -2023,6 +2029,21 @@
           "Useful after action target design.",
           "Less directly omni-modal than Qwen3-Omni or Cosmos 3."
         ]
+      },
+      {
+        "best_role": "Domain model over synchronized embodied experience.",
+        "category": "xperience_native_pretraining_goal",
+        "current_decision": "future_goal_after_scaling_evidence",
+        "entry_condition": "Full-corpus data path, PB-scale storage, multi-node compute, and positive smaller-run scaling evidence.",
+        "family": "Xperience Embodied Foundation Model",
+        "openness": "future project-specific model if full-corpus access and compute exist",
+        "priority": 8,
+        "public_source": "XPERIENCE_EMBODIED_FOUNDATION_MODEL_PRETRAINING.md",
+        "xperience10m_fit": [
+          "Uses the full aligned modality stack rather than treating sensors as auxiliary metadata.",
+          "Targets temporal embodied representation learning across perception, motion, geometry, audio, and language.",
+          "Can become the shared pretraining backbone for Qwen-style instruction tasks, Cosmos-style world modeling, and policy/action branches."
+        ]
       }
     ],
     "source_links": [
@@ -2057,11 +2078,15 @@
       {
         "label": "LeRobot / SmolVLA",
         "url": "https://github.com/huggingface/lerobot"
+      },
+      {
+        "label": "Xperience Embodied Foundation Model pretraining plan",
+        "url": "XPERIENCE_EMBODIED_FOUNDATION_MODEL_PRETRAINING.md"
       }
     ],
     "status": "planning_artifact"
   },
-  "generated_at_utc": "2026-06-04T16:42:13+00:00",
+  "generated_at_utc": "2026-06-04T20:40:29+00:00",
   "omni_plan": {
     "adapter": "LoRA rank 16, alpha 32, dropout 0.05",
     "backbone": "Qwen/Qwen3-Omni-30B-A3B-Instruct",
@@ -2208,6 +2233,31 @@
       "reader_takeaway": "The long-term direction is richer multimodal representation learning for embodied-AI reasoning, with model branches chosen by task fit rather than by a single default backbone.",
       "stage": "future",
       "status": "planned"
+    },
+    {
+      "completion_evidence": [
+        "pretraining metadata",
+        "checkpoint inventory",
+        "scaling curves",
+        "held-out evaluation reports",
+        "qualitative retrieval or future-state examples",
+        "safety and data-boundary report"
+      ],
+      "deliverables": [
+        "full-corpus episode and split manifests",
+        "pretraining shard and provenance manifests",
+        "0.3B-1B and 1B-3B scaling pilots",
+        "3B-7B Xperience-native domain model target",
+        "held-out episode/session/activity/object evaluations",
+        "missing-modality robustness report",
+        "model card and data-boundary report"
+      ],
+      "entry_condition": "Full-corpus access, PB-scale storage path, high-throughput data loading, multi-node compute, and positive scaling evidence from smaller multi-episode runs.",
+      "id": "xperience_embodied_foundation_pretraining",
+      "name": "Xperience Embodied Foundation Model Pretraining",
+      "reader_takeaway": "The final research direction is a domain-specific embodied foundation model trained directly on Xperience-10M, after smaller pilots justify the cost and infrastructure.",
+      "stage": "future",
+      "status": "future"
     }
   ],
   "scale_up": {

docs/data/artifact_index.json

@@ -1,11 +1,11 @@
 {
   "title": "Ropedia Xperience-10M Task Suite Artifact Index",
-  "generated_at_utc": "2026-06-04T16:42:13+00:00",
+  "generated_at_utc": "2026-06-04T20:40:52+00:00",
   "status": "pass",
-  "artifact_count": 72,
+  "artifact_count": 73,
   "missing": [],
   "by_kind": {
-    "project_path": 11,
+    "project_path": 12,
     "project_scope": 1,
     "source_alignment": 5,
     "publication_workflow": 1,
@@ -62,8 +62,8 @@
       "surface": "repo_hf",
       "shows": "Gives a compact current-state table for first-pass readers.",
       "exists": true,
-      "bytes": 7138,
-      "sha256": "67d85a198ee90082e47d790bd0f4d9dafbc97625cd39b17cc94b9785ec25104a"
+      "bytes": 7207,
+      "sha256": "7baaba976ccc254da1a03ee2653057d1e08f3fb0c0cad035886c362442828720"
     },
     {
       "id": "project_status_json",
@@ -73,8 +73,8 @@
       "surface": "website_hf",
       "shows": "Machine-readable copy of the current project status for website and HF mirrors.",
       "exists": true,
-      "bytes": 9169,
-      "sha256": "50d3c87b774c8375dcb897bd363d25e392e5fd6571571c41d56e623df15063f8"
+      "bytes": 9874,
+      "sha256": "600c95726eae3404127a8b2110f35468ff2ba02943cae0fbcd3ea43c66109d3e"
     },
     {
       "id": "research_roadmap",
@@ -84,8 +84,8 @@
       "surface": "repo_hf",
       "shows": "Defines the path from public-sample task development to multi-episode held-out evaluation and larger omni-model extensions.",
       "exists": true,
-      "bytes": 6677,
-      "sha256": "58491bfb68ad3e6b7569bdb1a3cac3de7682a49beb9de368a114d58ebf0b118b"
+      "bytes": 8388,
+      "sha256": "0b3e3356076998ad94dc39f708cc783a4ebeab76c9da661cdd37ea12a3bb3665"
     },
     {
       "id": "research_roadmap_json",
@@ -95,8 +95,8 @@
       "surface": "website_hf",
       "shows": "Machine-readable research roadmap for the website and Hugging Face mirrors.",
       "exists": true,
-      "bytes": 5758,
-      "sha256": "54657eb8824416d2128d6e5710543bdaf9e41d7c2fa46dd14ad6b58fede3b5db"
+      "bytes": 7161,
+      "sha256": "cc96118c2c05108c831616151bc027441f7545495adeeb6a4a6a6bffe8da7801"
     },
     {
       "id": "foundation_model_plan",
@@ -106,8 +106,8 @@
       "surface": "repo_hf",
       "shows": "Defines the post-data-gate backbone choices: Qwen3-Omni first, Cosmos 3 for world modeling, and VLA/policy models after action-target conversion.",
       "exists": true,
-      "bytes": 6559,
-      "sha256": "955be6559b554f1c6c4141dd6ca2818127d89585df3940c2bd9b975ad9047926"
+      "bytes": 9075,
+      "sha256": "444d13ab556d2e16a199a7fca191b87c85ab8685d167aab357bc6341839299a2"
     },
     {
       "id": "foundation_model_plan_json",
@@ -117,8 +117,19 @@
       "surface": "website_hf",
       "shows": "Machine-readable foundation-model selection matrix with source links, entry conditions, and evaluation additions.",
       "exists": true,
-      "bytes": 8889,
-      "sha256": "e9b11114fa290253000b921575586780ccc3ba17665235259d4326c524f6ce97"
+      "bytes": 12981,
+      "sha256": "9cce52025a2e2f8afb4660e2af3353aea6ad0a1af380849218dd74c0acc271bb"
+    },
+    {
+      "id": "xperience_embodied_foundation_pretraining",
+      "title": "Xperience Embodied Foundation Model pretraining goal",
+      "path": "XPERIENCE_EMBODIED_FOUNDATION_MODEL_PRETRAINING.md",
+      "kind": "project_path",
+      "surface": "repo_hf",
+      "shows": "Describes the future full-corpus Xperience-native pretraining goal, target modules, objectives, staged scale-up, hardware ranges, and evaluation protocol.",
+      "exists": true,
+      "bytes": 9182,
+      "sha256": "b5a6ddc58647cd895a4772b110ecc9f4d685427fb37b81b22c6c02d2b9b323f1"
     },
     {
       "id": "evidence_contract",
@@ -150,8 +161,8 @@
       "surface": "repo_hf",
       "shows": "Gives the human-readable map from project scope to data, tasks, platform mirrors, and scale-up status.",
       "exists": true,
-      "bytes": 16890,
-      "sha256": "8bce9a773daf36214e377a7154b72a4493efd0f7d1a1941d5e0fc9bf784a29e5"
+      "bytes": 11440,
+      "sha256": "9b8821a9b14fe1744f2e6b5c419b2c5daaf70b57f1944caf1105c36c0c66c119"
     },
     {
       "id": "official_dataset_card_alignment",
@@ -195,7 +206,7 @@
       "shows": "Machine-readable source-alignment pass/fail check for repo, website, and HF surfaces.",
       "exists": true,
       "bytes": 4432,
-      "sha256": "96c7adc61c869fab71ef34ec2f6ec4f5f88af844509bd3d51d3818732d1f84b6"
+      "sha256": "06c6e2d111c72df01ed127fd288e6675b63e35a21ae12a2523931a072bd0bc49"
     },
     {
       "id": "source_alignment_validator",
@@ -573,8 +584,8 @@
       "surface": "repo_hf",
       "shows": "Generates the selective artifact catalog from local files.",
       "exists": true,
-      "bytes": 26568,
-      "sha256": "a611b399e858560f6afb41e121f033724753c5167d04e0d7bf243e569de88f04"
+      "bytes": 27020,
+      "sha256": "0ca7ed96f24caecbab31687cffa99f0eba8471258986412a294614e688c5aff5"
     },
     {
       "id": "publication_audit",
@@ -585,7 +596,7 @@
       "volatile": true,
       "shows": "Confirms public bundles exclude raw data, caches, heavy archives, and credential text.",
       "exists": true,
-      "bytes": 7289,
+      "bytes": 11811,
       "hash_policy": "existence_and_size_only"
     },
     {
@@ -597,7 +608,7 @@
       "volatile": true,
       "shows": "Separates setup paths from completed held-out-episode results.",
       "exists": true,
-      "bytes": 19505,
+      "bytes": 18981,
       "hash_policy": "existence_and_size_only"
     },
     {
@@ -609,7 +620,7 @@
       "volatile": true,
       "shows": "Confirms prepared GitHub/HF Space/artifact/model mirrors share the same critical data, figure, website HTML, and validator files.",
       "exists": true,
-      "bytes": 108617,
+      "bytes": 108621,
       "hash_policy": "existence_and_size_only"
     },
     {
@@ -621,7 +632,7 @@
       "volatile": true,
       "shows": "Confirms local website links, anchors, JSON data files, and referenced images resolve.",
       "exists": true,
-      "bytes": 14923,
+      "bytes": 14891,
       "hash_policy": "existence_and_size_only"
     },
     {

docs/data/foundation_model_plan.json

@@ -2,6 +2,16 @@
   "title": "Xperience-10M Foundation Model Plan",
   "status": "planning_artifact",
   "current_boundary": "No held-out multi-episode foundation-model result has been completed in this repo. The current foundation-model artifacts are setup-stage until enough valid episodes are prepared and evaluated.",
+  "backbone_registry": {
+    "config_dir": "configs/omni_backbones",
+    "validator": "scripts/omni/backbone_registry.py --validate --json",
+    "extension_contract": "OMNI_MODEL_EXTENSION_CONTRACT.md",
+    "implemented_backbone": "qwen3_omni_lora",
+    "planned_backbones": [
+      "cosmos_world_model",
+      "policy_vla_branch"
+    ]
+  },
   "decision": {
     "immediate_trainable_backbone": "Qwen3-Omni",
     "first_world_model_branch": "Cosmos 3",
@@ -10,7 +20,65 @@
       "openpi pi0/pi0.5",
       "NVIDIA GR00T"
     ],
-    "external_reasoning_reference": "Gemini Robotics"
+    "external_reasoning_reference": "Gemini Robotics",
+    "long_term_native_pretraining_goal": "Xperience Embodied Foundation Model"
+  },
+  "future_pretraining_goal": {
+    "name": "Xperience Embodied Foundation Model",
+    "status": "future_planning_goal",
+    "role": "Domain-specific embodied foundation model pretrained on full Xperience-10M if full-corpus data, storage, and compute become available.",
+    "not_current_result": true,
+    "document": "XPERIENCE_EMBODIED_FOUNDATION_MODEL_PRETRAINING.md",
+    "entry_conditions": [
+      "Selected multi-episode Qwen3-Omni pilot trains and evaluates cleanly.",
+      "Scaling from 128 episodes to thousands of episodes shows measurable value.",
+      "Full-corpus storage, derived-shard storage, and fast active-cache capacity are available.",
+      "Distributed training, checkpoint/restart, and provenance tracking are reliable.",
+      "Evaluation covers held-out episodes, sessions, activities, objects, and missing-modality robustness."
+    ],
+    "target_modules": [
+      "multi-view video encoder",
+      "audio encoder",
+      "depth and geometry encoder",
+      "pose/SLAM encoder",
+      "hand/body mocap encoder",
+      "IMU encoder",
+      "language encoder/decoder",
+      "temporal fusion transformer",
+      "task heads and decoders"
+    ],
+    "pretraining_objectives": [
+      "masked multimodal modeling",
+      "cross-modal contrastive alignment",
+      "future-state prediction",
+      "ego-motion and hand-motion forecasting",
+      "action and procedure prediction",
+      "language grounding and captioning",
+      "contact and affordance prediction",
+      "optional policy-style targets after action conversion"
+    ],
+    "hardware_ranges": [
+      {
+        "goal": "0.3B-1B pilot",
+        "compute": "8-32 modern 80GB-class data-center GPUs",
+        "use": "prove objectives and data loaders"
+      },
+      {
+        "goal": "1B-3B domain model",
+        "compute": "32-128 GPUs",
+        "use": "research-scale Xperience representation learning"
+      },
+      {
+        "goal": "3B-7B full-corpus domain model",
+        "compute": "128-512 GPUs",
+        "use": "first realistic full Xperience-native foundation model"
+      },
+      {
+        "goal": "30B-class omni model from scratch",
+        "compute": "512-2000+ GPUs",
+        "use": "lab-scale project after scaling curves justify cost"
+      }
+    ]
   },
   "model_families": [
     {
@@ -112,6 +180,21 @@
       "current_decision": "optional_baseline_after_data_staging",
       "entry_condition": "Action labels and baseline protocol exist.",
       "public_source": "https://github.com/huggingface/lerobot"
+    },
+    {
+      "priority": 8,
+      "family": "Xperience Embodied Foundation Model",
+      "category": "xperience_native_pretraining_goal",
+      "openness": "future project-specific model if full-corpus access and compute exist",
+      "best_role": "Domain model over synchronized embodied experience.",
+      "xperience10m_fit": [
+        "Uses the full aligned modality stack rather than treating sensors as auxiliary metadata.",
+        "Targets temporal embodied representation learning across perception, motion, geometry, audio, and language.",
+        "Can become the shared pretraining backbone for Qwen-style instruction tasks, Cosmos-style world modeling, and policy/action branches."
+      ],
+      "current_decision": "future_goal_after_scaling_evidence",
+      "entry_condition": "Full-corpus data path, PB-scale storage, multi-node compute, and positive smaller-run scaling evidence.",
+      "public_source": "XPERIENCE_EMBODIED_FOUNDATION_MODEL_PRETRAINING.md"
     }
   ],
   "execution_order": [
@@ -144,6 +227,11 @@
       "step": 6,
       "name": "Publishing threshold",
       "action": "Publish branch results only with real manifests, predictions, metrics, and qualitative examples."
+    },
+    {
+      "step": 7,
+      "name": "Xperience-native pretraining",
+      "action": "Start a from-scratch Xperience Embodied Foundation Model only after smaller scaling stages, full-corpus storage, multi-node compute, and held-out evaluation protocols are in place."
     }
   ],
   "evaluation_additions": [
@@ -230,6 +318,10 @@
     {
       "label": "LeRobot / SmolVLA",
       "url": "https://github.com/huggingface/lerobot"
+    },
+    {
+      "label": "Xperience Embodied Foundation Model pretraining plan",
+      "url": "XPERIENCE_EMBODIED_FOUNDATION_MODEL_PRETRAINING.md"
     }
   ]
 }

docs/data/mirror_parity.json

@@ -1,6 +1,6 @@
 {
   "status": "pass",
-  "generated_at_utc": "2026-06-04T16:49:59+00:00",
+  "generated_at_utc": "2026-06-04T20:45:22+00:00",
   "hf_root": "hf_publish",
   "summary": {
     "group_count": 101,
@@ -71,27 +71,27 @@
       "local": {
         "path": "repo:docs/data/artifact_index.json",
         "exists": true,
-        "bytes": 32296,
-        "sha256": "5494e5ee1e40bc50d44a9cd6f77c8de694175939bda4f174fb5b1554e53ec508"
+        "bytes": 32864,
+        "sha256": "ec7d17898c42fd76109567c201f9638059b6a9a11a48817b32677a0eb2662178"
       },
       "mirrors": {
         "hf_space": {
           "path": "hf_space:data/artifact_index.json",
           "exists": true,
-          "bytes": 32296,
-          "sha256": "5494e5ee1e40bc50d44a9cd6f77c8de694175939bda4f174fb5b1554e53ec508"
+          "bytes": 32864,
+          "sha256": "ec7d17898c42fd76109567c201f9638059b6a9a11a48817b32677a0eb2662178"
         },
         "hf_artifacts": {
           "path": "hf_artifacts:docs/data/artifact_index.json",
           "exists": true,
-          "bytes": 32296,
-          "sha256": "5494e5ee1e40bc50d44a9cd6f77c8de694175939bda4f174fb5b1554e53ec508"
+          "bytes": 32864,
+          "sha256": "ec7d17898c42fd76109567c201f9638059b6a9a11a48817b32677a0eb2662178"
         },
         "hf_model": {
           "path": "hf_model:metrics/artifact_index.json",
           "exists": true,
-          "bytes": 32296,
-          "sha256": "5494e5ee1e40bc50d44a9cd6f77c8de694175939bda4f174fb5b1554e53ec508"
+          "bytes": 32864,
+          "sha256": "ec7d17898c42fd76109567c201f9638059b6a9a11a48817b32677a0eb2662178"
         }
       },
       "failures": []
@@ -226,27 +226,27 @@
       "local": {
         "path": "repo:docs/data/foundation_model_plan.json",
         "exists": true,
-        "bytes": 8889,
-        "sha256": "e9b11114fa290253000b921575586780ccc3ba17665235259d4326c524f6ce97"
+        "bytes": 12981,
+        "sha256": "9cce52025a2e2f8afb4660e2af3353aea6ad0a1af380849218dd74c0acc271bb"
       },
       "mirrors": {
         "hf_space": {
           "path": "hf_space:data/foundation_model_plan.json",
           "exists": true,
-          "bytes": 8889,
-          "sha256": "e9b11114fa290253000b921575586780ccc3ba17665235259d4326c524f6ce97"
+          "bytes": 12981,
+          "sha256": "9cce52025a2e2f8afb4660e2af3353aea6ad0a1af380849218dd74c0acc271bb"
         },
         "hf_artifacts": {
           "path": "hf_artifacts:docs/data/foundation_model_plan.json",
           "exists": true,
-          "bytes": 8889,
-          "sha256": "e9b11114fa290253000b921575586780ccc3ba17665235259d4326c524f6ce97"
+          "bytes": 12981,
+          "sha256": "9cce52025a2e2f8afb4660e2af3353aea6ad0a1af380849218dd74c0acc271bb"
         },
         "hf_model": {
           "path": "hf_model:metrics/foundation_model_plan.json",
           "exists": true,
-          "bytes": 8889,
-          "sha256": "e9b11114fa290253000b921575586780ccc3ba17665235259d4326c524f6ce97"
+          "bytes": 12981,
+          "sha256": "9cce52025a2e2f8afb4660e2af3353aea6ad0a1af380849218dd74c0acc271bb"
         }
       },
       "failures": []
@@ -412,27 +412,27 @@
       "local": {
         "path": "repo:docs/data/project_status.json",
         "exists": true,
-        "bytes": 9169,
-        "sha256": "50d3c87b774c8375dcb897bd363d25e392e5fd6571571c41d56e623df15063f8"
+        "bytes": 9874,
+        "sha256": "600c95726eae3404127a8b2110f35468ff2ba02943cae0fbcd3ea43c66109d3e"
       },
       "mirrors": {
         "hf_space": {
           "path": "hf_space:data/project_status.json",
           "exists": true,
-          "bytes": 9169,
-          "sha256": "50d3c87b774c8375dcb897bd363d25e392e5fd6571571c41d56e623df15063f8"
+          "bytes": 9874,
+          "sha256": "600c95726eae3404127a8b2110f35468ff2ba02943cae0fbcd3ea43c66109d3e"
         },
         "hf_artifacts": {
           "path": "hf_artifacts:docs/data/project_status.json",
           "exists": true,
-          "bytes": 9169,
-          "sha256": "50d3c87b774c8375dcb897bd363d25e392e5fd6571571c41d56e623df15063f8"
+          "bytes": 9874,
+          "sha256": "600c95726eae3404127a8b2110f35468ff2ba02943cae0fbcd3ea43c66109d3e"
         },
         "hf_model": {
           "path": "hf_model:metrics/project_status.json",
           "exists": true,
-          "bytes": 9169,
-          "sha256": "50d3c87b774c8375dcb897bd363d25e392e5fd6571571c41d56e623df15063f8"
+          "bytes": 9874,
+          "sha256": "600c95726eae3404127a8b2110f35468ff2ba02943cae0fbcd3ea43c66109d3e"
         }
       },
       "failures": []
@@ -443,27 +443,27 @@
       "local": {
         "path": "repo:docs/data/publication_audit.json",
         "exists": true,
-        "bytes": 7289,
-        "sha256": "cd84a10ddbfb13943820c8e6113ca377a9ab1215f45df2b3384e752cbcac190b"
+        "bytes": 7237,
+        "sha256": "7fbb19f8990b1a4d902e282c010d27e4391755564fa68af97d96c298c6b054f8"
       },
       "mirrors": {
         "hf_space": {
           "path": "hf_space:data/publication_audit.json",
           "exists": true,
-          "bytes": 7289,
-          "sha256": "cd84a10ddbfb13943820c8e6113ca377a9ab1215f45df2b3384e752cbcac190b"
+          "bytes": 7237,
+          "sha256": "7fbb19f8990b1a4d902e282c010d27e4391755564fa68af97d96c298c6b054f8"
         },
         "hf_artifacts": {
           "path": "hf_artifacts:docs/data/publication_audit.json",
           "exists": true,
-          "bytes": 7289,
-          "sha256": "cd84a10ddbfb13943820c8e6113ca377a9ab1215f45df2b3384e752cbcac190b"
+          "bytes": 7237,
+          "sha256": "7fbb19f8990b1a4d902e282c010d27e4391755564fa68af97d96c298c6b054f8"
         },
         "hf_model": {
           "path": "hf_model:metrics/publication_audit.json",
           "exists": true,
-          "bytes": 7289,
-          "sha256": "cd84a10ddbfb13943820c8e6113ca377a9ab1215f45df2b3384e752cbcac190b"
+          "bytes": 7237,
+          "sha256": "7fbb19f8990b1a4d902e282c010d27e4391755564fa68af97d96c298c6b054f8"
         }
       },
       "failures": []
@@ -598,27 +598,27 @@
       "local": {
         "path": "repo:docs/data/research_roadmap.json",
         "exists": true,
-        "bytes": 5758,
-        "sha256": "54657eb8824416d2128d6e5710543bdaf9e41d7c2fa46dd14ad6b58fede3b5db"
+        "bytes": 7161,
+        "sha256": "cc96118c2c05108c831616151bc027441f7545495adeeb6a4a6a6bffe8da7801"
       },
       "mirrors": {
         "hf_space": {
           "path": "hf_space:data/research_roadmap.json",
           "exists": true,
-          "bytes": 5758,
-          "sha256": "54657eb8824416d2128d6e5710543bdaf9e41d7c2fa46dd14ad6b58fede3b5db"
+          "bytes": 7161,
+          "sha256": "cc96118c2c05108c831616151bc027441f7545495adeeb6a4a6a6bffe8da7801"
         },
         "hf_artifacts": {
           "path": "hf_artifacts:docs/data/research_roadmap.json",
           "exists": true,
-          "bytes": 5758,
-          "sha256": "54657eb8824416d2128d6e5710543bdaf9e41d7c2fa46dd14ad6b58fede3b5db"
+          "bytes": 7161,
+          "sha256": "cc96118c2c05108c831616151bc027441f7545495adeeb6a4a6a6bffe8da7801"
         },
         "hf_model": {
           "path": "hf_model:metrics/research_roadmap.json",
           "exists": true,
-          "bytes": 5758,
-          "sha256": "54657eb8824416d2128d6e5710543bdaf9e41d7c2fa46dd14ad6b58fede3b5db"
+          "bytes": 7161,
+          "sha256": "cc96118c2c05108c831616151bc027441f7545495adeeb6a4a6a6bffe8da7801"
         }
       },
       "failures": []
@@ -629,27 +629,27 @@
       "local": {
         "path": "repo:docs/data/research_roadmap_interactive.json",
         "exists": true,
-        "bytes": 131519,
-        "sha256": "004fbcc7a3582da88dd66504d686604ecb0f04f65c9c8166bb0583e0fc174274"
+        "bytes": 134282,
+        "sha256": "ff37219a9f1d9b386a9d4c42766e4aa28f10ce6ef338dceeedd6bdb4a1b2c40a"
       },
       "mirrors": {
         "hf_space": {
           "path": "hf_space:data/research_roadmap_interactive.json",
           "exists": true,
-          "bytes": 131519,
-          "sha256": "004fbcc7a3582da88dd66504d686604ecb0f04f65c9c8166bb0583e0fc174274"
+          "bytes": 134282,
+          "sha256": "ff37219a9f1d9b386a9d4c42766e4aa28f10ce6ef338dceeedd6bdb4a1b2c40a"
         },
         "hf_artifacts": {
           "path": "hf_artifacts:docs/data/research_roadmap_interactive.json",
           "exists": true,
-          "bytes": 131519,
-          "sha256": "004fbcc7a3582da88dd66504d686604ecb0f04f65c9c8166bb0583e0fc174274"
+          "bytes": 134282,
+          "sha256": "ff37219a9f1d9b386a9d4c42766e4aa28f10ce6ef338dceeedd6bdb4a1b2c40a"
         },
         "hf_model": {
           "path": "hf_model:metrics/research_roadmap_interactive.json",
           "exists": true,
-          "bytes": 131519,
-          "sha256": "004fbcc7a3582da88dd66504d686604ecb0f04f65c9c8166bb0583e0fc174274"
+          "bytes": 134282,
+          "sha256": "ff37219a9f1d9b386a9d4c42766e4aa28f10ce6ef338dceeedd6bdb4a1b2c40a"
         }
       },
       "failures": []
@@ -939,27 +939,27 @@
       "local": {
         "path": "repo:docs/data/website_integrity.json",
         "exists": true,
-        "bytes": 14923,
-        "sha256": "23a03838502e8d43ee2b41e313634ec46a4b329792883aa12fc03b044c4e9b0e"
+        "bytes": 14891,
+        "sha256": "9ba1cfe02568fc9b08209902ce037c445a9a8c3954d20eea4351b04c65ca0a0c"
       },
       "mirrors": {
         "hf_space": {
           "path": "hf_space:data/website_integrity.json",
           "exists": true,
-          "bytes": 14923,
-          "sha256": "23a03838502e8d43ee2b41e313634ec46a4b329792883aa12fc03b044c4e9b0e"
+          "bytes": 14891,
+          "sha256": "9ba1cfe02568fc9b08209902ce037c445a9a8c3954d20eea4351b04c65ca0a0c"
         },
         "hf_artifacts": {
           "path": "hf_artifacts:docs/data/website_integrity.json",
           "exists": true,
-          "bytes": 14923,
-          "sha256": "23a03838502e8d43ee2b41e313634ec46a4b329792883aa12fc03b044c4e9b0e"
+          "bytes": 14891,
+          "sha256": "9ba1cfe02568fc9b08209902ce037c445a9a8c3954d20eea4351b04c65ca0a0c"
         },
         "hf_model": {
           "path": "hf_model:metrics/website_integrity.json",
           "exists": true,
-          "bytes": 14923,
-          "sha256": "23a03838502e8d43ee2b41e313634ec46a4b329792883aa12fc03b044c4e9b0e"
+          "bytes": 14891,
+          "sha256": "9ba1cfe02568fc9b08209902ce037c445a9a8c3954d20eea4351b04c65ca0a0c"
         }
       },
       "failures": []
@@ -1692,21 +1692,21 @@
       "local": {
         "path": "repo:scripts/build_artifact_index.py",
         "exists": true,
-        "bytes": 26568,
-        "sha256": "a611b399e858560f6afb41e121f033724753c5167d04e0d7bf243e569de88f04"
+        "bytes": 27020,
+        "sha256": "0ca7ed96f24caecbab31687cffa99f0eba8471258986412a294614e688c5aff5"
       },
       "mirrors": {
         "hf_artifacts": {
           "path": "hf_artifacts:scripts/build_artifact_index.py",
           "exists": true,
-          "bytes": 26568,
-          "sha256": "a611b399e858560f6afb41e121f033724753c5167d04e0d7bf243e569de88f04"
+          "bytes": 27020,
+          "sha256": "0ca7ed96f24caecbab31687cffa99f0eba8471258986412a294614e688c5aff5"
         },
         "hf_model": {
           "path": "hf_model:scripts/build_artifact_index.py",
           "exists": true,
-          "bytes": 26568,
-          "sha256": "a611b399e858560f6afb41e121f033724753c5167d04e0d7bf243e569de88f04"
+          "bytes": 27020,
+          "sha256": "0ca7ed96f24caecbab31687cffa99f0eba8471258986412a294614e688c5aff5"
         }
       },
       "failures": []
@@ -2017,21 +2017,21 @@
       "local": {
         "path": "repo:scripts/validate_publication_package.py",
         "exists": true,
-        "bytes": 19267,
-        "sha256": "0db7f9a376ac4dfb1bb083a5f35051e2cb18a0d9db5788e7d707d8dc084ad231"
+        "bytes": 17197,
+        "sha256": "2a617f3204ffb8c59d1c5bc1828b4441a4d014bb531655fd0613e128a6d9abc2"
       },
       "mirrors": {
         "hf_artifacts": {
           "path": "hf_artifacts:scripts/validate_publication_package.py",
           "exists": true,
-          "bytes": 19267,
-          "sha256": "0db7f9a376ac4dfb1bb083a5f35051e2cb18a0d9db5788e7d707d8dc084ad231"
+          "bytes": 17197,
+          "sha256": "2a617f3204ffb8c59d1c5bc1828b4441a4d014bb531655fd0613e128a6d9abc2"
         },
         "hf_model": {
           "path": "hf_model:scripts/validate_publication_package.py",
           "exists": true,
-          "bytes": 19267,
-          "sha256": "0db7f9a376ac4dfb1bb083a5f35051e2cb18a0d9db5788e7d707d8dc084ad231"
+          "bytes": 17197,
+          "sha256": "2a617f3204ffb8c59d1c5bc1828b4441a4d014bb531655fd0613e128a6d9abc2"
         }
       },
       "failures": []
@@ -2117,21 +2117,21 @@
       "local": {
         "path": "repo:scripts/validate_website_integrity.py",
         "exists": true,
-        "bytes": 24396,
-        "sha256": "3b4af15250f79827e3010e93636836c3a0c768ba0188a9a7e55e439233988c72"
+        "bytes": 24481,
+        "sha256": "31d85a4674e8005a916e759d820178287e297e0ec08774fe3a70aa3b61b07cf7"
       },
       "mirrors": {
         "hf_artifacts": {
           "path": "hf_artifacts:scripts/validate_website_integrity.py",
           "exists": true,
-          "bytes": 24396,
-          "sha256": "3b4af15250f79827e3010e93636836c3a0c768ba0188a9a7e55e439233988c72"
+          "bytes": 24481,
+          "sha256": "31d85a4674e8005a916e759d820178287e297e0ec08774fe3a70aa3b61b07cf7"
         },
         "hf_model": {
           "path": "hf_model:scripts/validate_website_integrity.py",
           "exists": true,
-          "bytes": 24396,
-          "sha256": "3b4af15250f79827e3010e93636836c3a0c768ba0188a9a7e55e439233988c72"
+          "bytes": 24481,
+          "sha256": "31d85a4674e8005a916e759d820178287e297e0ec08774fe3a70aa3b61b07cf7"
         }
       },
       "failures": []
@@ -2217,21 +2217,21 @@
       "local": {
         "path": "repo:docs/index.html",
         "exists": true,
-        "bytes": 173425,
-        "sha256": "26ac1e7976c11f21f4fd2f3623ac8d339a57b511f6cc8f5e68300062e9def2b0"
+        "bytes": 174923,
+        "sha256": "099fcc01cbb4d50f62c508b10f343f05b1c883962b85bda294bcede99af2a0f1"
       },
       "mirrors": {
         "hf_space": {
           "path": "hf_space:index.html",
           "exists": true,
-          "bytes": 173425,
-          "sha256": "26ac1e7976c11f21f4fd2f3623ac8d339a57b511f6cc8f5e68300062e9def2b0"
+          "bytes": 174923,
+          "sha256": "099fcc01cbb4d50f62c508b10f343f05b1c883962b85bda294bcede99af2a0f1"
         },
         "hf_artifacts_docs": {
           "path": "hf_artifacts:docs/index.html",
           "exists": true,
-          "bytes": 173425,
-          "sha256": "26ac1e7976c11f21f4fd2f3623ac8d339a57b511f6cc8f5e68300062e9def2b0"
+          "bytes": 174923,
+          "sha256": "099fcc01cbb4d50f62c508b10f343f05b1c883962b85bda294bcede99af2a0f1"
         }
       },
       "failures": []
@@ -2242,21 +2242,21 @@
       "local": {
         "path": "repo:docs/research_roadmap.html",
         "exists": true,
-        "bytes": 31554,
-        "sha256": "f51e83a4495f2d2012ec4c48191d66ca4456a00d7fcb335a427b7d86afc66109"
+        "bytes": 31702,
+        "sha256": "1b20a5cc342b3ba59ad808eed9f5bf978e2d9ac438c88b5c3eeba01f4e14b883"
       },
       "mirrors": {
         "hf_space": {
           "path": "hf_space:research_roadmap.html",
           "exists": true,
-          "bytes": 31554,
-          "sha256": "f51e83a4495f2d2012ec4c48191d66ca4456a00d7fcb335a427b7d86afc66109"
+          "bytes": 31702,
+          "sha256": "1b20a5cc342b3ba59ad808eed9f5bf978e2d9ac438c88b5c3eeba01f4e14b883"
         },
         "hf_artifacts_docs": {
           "path": "hf_artifacts:docs/research_roadmap.html",
           "exists": true,
-          "bytes": 31554,
-          "sha256": "f51e83a4495f2d2012ec4c48191d66ca4456a00d7fcb335a427b7d86afc66109"
+          "bytes": 31702,
+          "sha256": "1b20a5cc342b3ba59ad808eed9f5bf978e2d9ac438c88b5c3eeba01f4e14b883"
         }
       },
       "failures": []
@@ -2844,27 +2844,27 @@
       "local": {
         "path": "repo:FOUNDATION_MODEL_PLAN.md",
         "exists": true,
-        "bytes": 6559,
-        "sha256": "955be6559b554f1c6c4141dd6ca2818127d89585df3940c2bd9b975ad9047926"
+        "bytes": 9075,
+        "sha256": "444d13ab556d2e16a199a7fca191b87c85ab8685d167aab357bc6341839299a2"
       },
       "mirrors": {
         "hf_space": {
           "path": "hf_space:FOUNDATION_MODEL_PLAN.md",
           "exists": true,
-          "bytes": 6559,
-          "sha256": "955be6559b554f1c6c4141dd6ca2818127d89585df3940c2bd9b975ad9047926"
+          "bytes": 9075,
+          "sha256": "444d13ab556d2e16a199a7fca191b87c85ab8685d167aab357bc6341839299a2"
         },
         "hf_artifacts": {
           "path": "hf_artifacts:FOUNDATION_MODEL_PLAN.md",
           "exists": true,
-          "bytes": 6559,
-          "sha256": "955be6559b554f1c6c4141dd6ca2818127d89585df3940c2bd9b975ad9047926"
+          "bytes": 9075,
+          "sha256": "444d13ab556d2e16a199a7fca191b87c85ab8685d167aab357bc6341839299a2"
         },
         "hf_model": {
           "path": "hf_model:FOUNDATION_MODEL_PLAN.md",
           "exists": true,
-          "bytes": 6559,
-          "sha256": "955be6559b554f1c6c4141dd6ca2818127d89585df3940c2bd9b975ad9047926"
+          "bytes": 9075,
+          "sha256": "444d13ab556d2e16a199a7fca191b87c85ab8685d167aab357bc6341839299a2"
         }
       },
       "failures": []
@@ -2937,27 +2937,27 @@
       "local": {
         "path": "repo:RESEARCH_ROADMAP.md",
         "exists": true,
-        "bytes": 6677,
-        "sha256": "58491bfb68ad3e6b7569bdb1a3cac3de7682a49beb9de368a114d58ebf0b118b"
+        "bytes": 8388,
+        "sha256": "0b3e3356076998ad94dc39f708cc783a4ebeab76c9da661cdd37ea12a3bb3665"
       },
       "mirrors": {
         "hf_space": {
           "path": "hf_space:RESEARCH_ROADMAP.md",
           "exists": true,
-          "bytes": 6677,
-          "sha256": "58491bfb68ad3e6b7569bdb1a3cac3de7682a49beb9de368a114d58ebf0b118b"
+          "bytes": 8388,
+          "sha256": "0b3e3356076998ad94dc39f708cc783a4ebeab76c9da661cdd37ea12a3bb3665"
         },
         "hf_artifacts": {
           "path": "hf_artifacts:RESEARCH_ROADMAP.md",
           "exists": true,
-          "bytes": 6677,
-          "sha256": "58491bfb68ad3e6b7569bdb1a3cac3de7682a49beb9de368a114d58ebf0b118b"
+          "bytes": 8388,
+          "sha256": "0b3e3356076998ad94dc39f708cc783a4ebeab76c9da661cdd37ea12a3bb3665"
         },
         "hf_model": {
           "path": "hf_model:RESEARCH_ROADMAP.md",
           "exists": true,
-          "bytes": 6677,
-          "sha256": "58491bfb68ad3e6b7569bdb1a3cac3de7682a49beb9de368a114d58ebf0b118b"
+          "bytes": 8388,
+          "sha256": "0b3e3356076998ad94dc39f708cc783a4ebeab76c9da661cdd37ea12a3bb3665"
         }
       },
       "failures": []
@@ -2968,27 +2968,27 @@
       "local": {
         "path": "repo:PROJECT_STATUS.md",
         "exists": true,
-        "bytes": 7138,
-        "sha256": "67d85a198ee90082e47d790bd0f4d9dafbc97625cd39b17cc94b9785ec25104a"
+        "bytes": 7207,
+        "sha256": "7baaba976ccc254da1a03ee2653057d1e08f3fb0c0cad035886c362442828720"
       },
       "mirrors": {
         "hf_space": {
           "path": "hf_space:PROJECT_STATUS.md",
           "exists": true,
-          "bytes": 7138,
-          "sha256": "67d85a198ee90082e47d790bd0f4d9dafbc97625cd39b17cc94b9785ec25104a"
+          "bytes": 7207,
+          "sha256": "7baaba976ccc254da1a03ee2653057d1e08f3fb0c0cad035886c362442828720"
         },
         "hf_artifacts": {
           "path": "hf_artifacts:PROJECT_STATUS.md",
           "exists": true,
-          "bytes": 7138,
-          "sha256": "67d85a198ee90082e47d790bd0f4d9dafbc97625cd39b17cc94b9785ec25104a"
+          "bytes": 7207,
+          "sha256": "7baaba976ccc254da1a03ee2653057d1e08f3fb0c0cad035886c362442828720"
         },
         "hf_model": {
           "path": "hf_model:PROJECT_STATUS.md",
           "exists": true,
-          "bytes": 7138,
-          "sha256": "67d85a198ee90082e47d790bd0f4d9dafbc97625cd39b17cc94b9785ec25104a"
+          "bytes": 7207,
+          "sha256": "7baaba976ccc254da1a03ee2653057d1e08f3fb0c0cad035886c362442828720"
         }
       },
       "failures": []

docs/data/project_status.json

@@ -82,7 +82,7 @@
                 "RESEARCH_ROADMAP.md",
                 "docs/data/research_roadmap.json"
             ],
-            "readout": "The roadmap connects public-sample task development to 128-episode data preparation, Qwen3-Omni LoRA, foundation-model selection, robustness runs, and larger omni/world-model extensions."
+            "readout": "The roadmap connects public-sample task development to 128-episode data preparation, Qwen3-Omni LoRA, foundation-model selection, robustness runs, world/policy branches, and the future Xperience-native pretraining goal."
         },
         {
             "area": "Foundation-model plan",
@@ -93,6 +93,14 @@
             ],
             "readout": "Qwen3-Omni remains the first trainable held-out LoRA baseline; Cosmos 3 is added as the first world-model/action-generation branch; OpenVLA/openpi/GR00T are policy candidates after action targets are explicit."
         },
+        {
+            "area": "Xperience Embodied Foundation Model",
+            "status": "future_goal",
+            "evidence": [
+                "XPERIENCE_EMBODIED_FOUNDATION_MODEL_PRETRAINING.md"
+            ],
+            "readout": "A future full-corpus pretraining plan describes target modules, objectives, staged scale-up, hardware ranges, and evaluation for a domain-specific embodied foundation model."
+        },
         {
             "area": "Official dataset wording",
             "status": "verified",
@@ -167,6 +175,7 @@
         "Inspect RESEARCH_TAKEAWAYS.md and docs/data/research_takeaways.json before interpreting model scores.",
         "Inspect RESEARCH_ROADMAP.md and docs/data/research_roadmap.json for the path from public-sample task work to multi-episode modeling.",
         "Inspect FOUNDATION_MODEL_PLAN.md and docs/data/foundation_model_plan.json before choosing a backbone branch.",
+        "Inspect XPERIENCE_EMBODIED_FOUNDATION_MODEL_PRETRAINING.md for the long-term full-corpus pretraining goal.",
         "Inspect docs/data/summary_metrics.json and results/episode_task_suite/neural_mlp/ to check the 12-task outputs.",
         "Inspect results/audio_ablation/AUDIO_ABLATION_SUMMARY.md before judging whether audio helps the current task suite.",
         "Inspect EVALUATION_PROTOCOL.md before judging task metrics or leakage controls.",
@@ -180,6 +189,7 @@
     "The current reconstruction task reconstructs feature vectors, not pixel-depth, mesh, NeRF, or Gaussian reconstruction.",
     "Audio is one of the synchronized source modalities in the current task representation.",
     "The audio ablation report compares audio/no-audio variants across all 12 task contracts in results/audio_ablation/.",
-    "Foundation-model selection is explicit: Qwen3-Omni is the immediate trainable pilot, Cosmos 3 is the first world-model branch, and policy models such as OpenVLA/openpi/GR00T wait for action-target conversion."
+    "Foundation-model selection is explicit: Qwen3-Omni is the immediate trainable pilot, Cosmos 3 is the first world-model branch, and policy models such as OpenVLA/openpi/GR00T wait for action-target conversion.",
+    "The Xperience Embodied Foundation Model is a future native-pretraining goal, not a completed model or current benchmark."
   ]
 }

docs/data/publication_audit.json

@@ -1,6 +1,6 @@
 {
   "status": "pass",
-  "generated_at_utc": "2026-06-04T16:49:00+00:00",
+  "generated_at_utc": "2026-06-04T20:43:37+00:00",
   "checks": [
     {
       "name": "required_publication_assets_present",
@@ -141,7 +141,7 @@
       "surface": "github_repo",
       "path": "README.md",
       "exists": true,
-      "required_marker_count": 20,
+      "required_marker_count": 10,
       "missing_markers": [],
       "status": "pass"
     },
@@ -149,7 +149,7 @@
       "surface": "hf_space_bundle",
       "path": "README.md",
       "exists": true,
-      "required_marker_count": 20,
+      "required_marker_count": 10,
       "missing_markers": [],
       "status": "pass"
     },
@@ -157,7 +157,7 @@
       "surface": "hf_artifact_bundle",
       "path": "README.md",
       "exists": true,
-      "required_marker_count": 19,
+      "required_marker_count": 7,
       "missing_markers": [],
       "status": "pass"
     },
@@ -165,7 +165,7 @@
       "surface": "hf_artifact_bundle",
       "path": "PROJECT_README.md",
       "exists": true,
-      "required_marker_count": 20,
+      "required_marker_count": 10,
       "missing_markers": [],
       "status": "pass"
     },
@@ -173,7 +173,7 @@
       "surface": "hf_model_bundle",
       "path": "README.md",
       "exists": true,
-      "required_marker_count": 20,
+      "required_marker_count": 10,
       "missing_markers": [],
       "status": "pass"
     }
@@ -182,8 +182,8 @@
     "github_repo": {
       "root": "repo",
       "exists": true,
-      "file_count": 386,
-      "text_file_count": 320,
+      "file_count": 396,
+      "text_file_count": 330,
       "largest_file": {
         "path": "results/episode_task_suite/modality_reconstruction/predictions.npz",
         "bytes": 55702978
@@ -193,8 +193,8 @@
     "hf_space_bundle": {
       "root": "hf_publish/space",
       "exists": true,
-      "file_count": 316,
-      "text_file_count": 250,
+      "file_count": 317,
+      "text_file_count": 251,
       "largest_file": {
         "path": "results/episode_task_suite/modality_reconstruction/predictions.npz",
         "bytes": 55702978
@@ -204,8 +204,8 @@
     "hf_artifact_bundle": {
       "root": "hf_publish/artifacts",
       "exists": true,
-      "file_count": 417,
-      "text_file_count": 329,
+      "file_count": 418,
+      "text_file_count": 330,
       "largest_file": {
         "path": "results/episode_task_suite/modality_reconstruction/predictions.npz",
         "bytes": 55702978
@@ -215,11 +215,11 @@
     "hf_model_bundle": {
       "root": "hf_publish/model",
       "exists": true,
-      "file_count": 640,
-      "text_file_count": 516,
+      "file_count": 644,
+      "text_file_count": 519,
       "largest_file": {
-        "path": "artifacts/episode_task_suite/modality_reconstruction/predictions.npz",
-        "bytes": 55702978
+        "path": "pytorch_model.bin",
+        "bytes": 93495480
       },
       "violations": []
     }

docs/data/research_roadmap.json

@@ -1,7 +1,7 @@
 {
   "title": "Ropedia Xperience-10M Research Roadmap",
-  "summary": "Staged path from the public-sample task lab to multi-episode held-out evaluation, foundation-model selection, and larger omni/world-model extensions.",
-  "current_decision_point": "Keep the public-sample task suite as the development harness, prepare the selected official Xperience-10M episodes for the held-out Qwen3-Omni pilot, then branch into Cosmos 3 world modeling and policy-model experiments after the data preparation path is stable.",
+  "summary": "Staged path from the public-sample task lab to multi-episode held-out evaluation, foundation-model selection, world/policy branches, and a future Xperience-native embodied foundation model.",
+  "current_decision_point": "Keep the public-sample task suite as the development harness, prepare the selected official Xperience-10M episodes for the held-out Qwen3-Omni pilot, then branch into Cosmos 3 world modeling and policy-model experiments after the data preparation path is stable. The Xperience Embodied Foundation Model is a later full-corpus pretraining goal, not a current result.",
   "phases": [
     {
       "id": "public_sample_task_lab",
@@ -126,6 +126,30 @@
         "updated model cards"
       ],
       "reader_takeaway": "The long-term direction is richer multimodal representation learning for embodied-AI reasoning, with model branches chosen by task fit rather than by a single default backbone."
+    },
+    {
+      "id": "xperience_embodied_foundation_pretraining",
+      "name": "Xperience Embodied Foundation Model Pretraining",
+      "status": "future",
+      "entry_condition": "Full-corpus access, PB-scale storage path, high-throughput data loading, multi-node compute, and positive scaling evidence from smaller multi-episode runs.",
+      "deliverables": [
+        "full-corpus episode and split manifests",
+        "pretraining shard and provenance manifests",
+        "0.3B-1B and 1B-3B scaling pilots",
+        "3B-7B Xperience-native domain model target",
+        "held-out episode/session/activity/object evaluations",
+        "missing-modality robustness report",
+        "model card and data-boundary report"
+      ],
+      "completion_evidence": [
+        "pretraining metadata",
+        "checkpoint inventory",
+        "scaling curves",
+        "held-out evaluation reports",
+        "qualitative retrieval or future-state examples",
+        "safety and data-boundary report"
+      ],
+      "reader_takeaway": "The final research direction is a domain-specific embodied foundation model trained directly on Xperience-10M, after smaller pilots justify the cost and infrastructure."
     }
   ],
   "public_surfaces_to_update": [
@@ -134,6 +158,7 @@
     "RESEARCH_TAKEAWAYS.md",
     "EVALUATION_PROTOCOL.md",
     "ARTIFACT_GUIDE.md",
+    "XPERIENCE_EMBODIED_FOUNDATION_MODEL_PRETRAINING.md",
     "docs/index.html",
     "docs/data/research_roadmap.json",
     "Hugging Face Space card",

docs/data/research_roadmap_interactive.json

@@ -1837,7 +1837,8 @@
         "NVIDIA GR00T"
       ],
       "first_world_model_branch": "Cosmos 3",
-      "immediate_trainable_backbone": "Qwen3-Omni"
+      "immediate_trainable_backbone": "Qwen3-Omni",
+      "long_term_native_pretraining_goal": "Xperience Embodied Foundation Model"
     },
     "evaluation_additions": [
       {
@@ -1921,6 +1922,11 @@
         "action": "Publish branch results only with real manifests, predictions, metrics, and qualitative examples.",
         "name": "Publishing threshold",
         "step": 6
+      },
+      {
+        "action": "Start a from-scratch Xperience Embodied Foundation Model only after smaller scaling stages, full-corpus storage, multi-node compute, and held-out evaluation protocols are in place.",
+        "name": "Xperience-native pretraining",
+        "step": 7
       }
     ],
     "model_families": [
@@ -2023,6 +2029,21 @@
           "Useful after action target design.",
           "Less directly omni-modal than Qwen3-Omni or Cosmos 3."
         ]
+      },
+      {
+        "best_role": "Domain model over synchronized embodied experience.",
+        "category": "xperience_native_pretraining_goal",
+        "current_decision": "future_goal_after_scaling_evidence",
+        "entry_condition": "Full-corpus data path, PB-scale storage, multi-node compute, and positive smaller-run scaling evidence.",
+        "family": "Xperience Embodied Foundation Model",
+        "openness": "future project-specific model if full-corpus access and compute exist",
+        "priority": 8,
+        "public_source": "XPERIENCE_EMBODIED_FOUNDATION_MODEL_PRETRAINING.md",
+        "xperience10m_fit": [
+          "Uses the full aligned modality stack rather than treating sensors as auxiliary metadata.",
+          "Targets temporal embodied representation learning across perception, motion, geometry, audio, and language.",
+          "Can become the shared pretraining backbone for Qwen-style instruction tasks, Cosmos-style world modeling, and policy/action branches."
+        ]
       }
     ],
     "source_links": [
@@ -2057,11 +2078,15 @@
       {
         "label": "LeRobot / SmolVLA",
         "url": "https://github.com/huggingface/lerobot"
+      },
+      {
+        "label": "Xperience Embodied Foundation Model pretraining plan",
+        "url": "XPERIENCE_EMBODIED_FOUNDATION_MODEL_PRETRAINING.md"
       }
     ],
     "status": "planning_artifact"
   },
-  "generated_at_utc": "2026-06-04T16:42:13+00:00",
+  "generated_at_utc": "2026-06-04T20:40:29+00:00",
   "omni_plan": {
     "adapter": "LoRA rank 16, alpha 32, dropout 0.05",
     "backbone": "Qwen/Qwen3-Omni-30B-A3B-Instruct",
@@ -2208,6 +2233,31 @@
       "reader_takeaway": "The long-term direction is richer multimodal representation learning for embodied-AI reasoning, with model branches chosen by task fit rather than by a single default backbone.",
       "stage": "future",
       "status": "planned"
+    },
+    {
+      "completion_evidence": [
+        "pretraining metadata",
+        "checkpoint inventory",
+        "scaling curves",
+        "held-out evaluation reports",
+        "qualitative retrieval or future-state examples",
+        "safety and data-boundary report"
+      ],
+      "deliverables": [
+        "full-corpus episode and split manifests",
+        "pretraining shard and provenance manifests",
+        "0.3B-1B and 1B-3B scaling pilots",
+        "3B-7B Xperience-native domain model target",
+        "held-out episode/session/activity/object evaluations",
+        "missing-modality robustness report",
+        "model card and data-boundary report"
+      ],
+      "entry_condition": "Full-corpus access, PB-scale storage path, high-throughput data loading, multi-node compute, and positive scaling evidence from smaller multi-episode runs.",
+      "id": "xperience_embodied_foundation_pretraining",
+      "name": "Xperience Embodied Foundation Model Pretraining",
+      "reader_takeaway": "The final research direction is a domain-specific embodied foundation model trained directly on Xperience-10M, after smaller pilots justify the cost and infrastructure.",
+      "stage": "future",
+      "status": "future"
     }
   ],
   "scale_up": {

docs/index.html

@@ -2141,9 +2141,11 @@
         <p class="hero-copy">
           This project uses the public Xperience-10M sample from Ropedia to explore
           embodied-AI task design, multimodal feature construction, lightweight
-          baselines, and future Omni-model fine-tuning. It starts from the sample
-          episode available now, then keeps the same data contracts ready for
-          held-out multi-episode training when more Xperience-10M data is prepared.
+          baselines, future Omni-model fine-tuning, and the long-term path toward
+          an Xperience-native embodied foundation model. It starts from the
+          sample episode available now, then keeps the same data contracts ready
+          for held-out multi-episode training when more Xperience-10M data is
+          prepared.
         </p>
         <div class="hero-actions">
           <a class="button primary" href="research_roadmap.html">Open roadmap</a>
@@ -2252,7 +2254,7 @@
             </article>
             <article class="brief-card">
               <strong>Scale-up readiness</strong>
-              <p>Connects the same data contract to 32/128-episode held-out pilots, Qwen3-Omni LoRA, Cosmos-style world modeling, and later policy-model branches.</p>
+              <p>Connects the same data contract to 32/128-episode held-out pilots, Qwen3-Omni LoRA, Cosmos-style world modeling, policy-model branches, and the later Xperience-native pretraining goal.</p>
             </article>
           </div>
           <div class="brief-actions">
@@ -2356,7 +2358,7 @@
       <div class="wrap">
         <div class="section-head">
           <h2>Research roadmap.</h2>
-          <p>The project path moves from the current public-sample task lab to multi-episode data preparation, held-out Qwen3-Omni fine-tuning, robustness runs, and larger foundation/world-model extensions.</p>
+          <p>The project path moves from the current public-sample task lab to multi-episode data preparation, held-out Qwen3-Omni fine-tuning, robustness runs, world/policy branches, and the future Xperience Embodied Foundation Model pretraining goal.</p>
         </div>
         <div class="roadmap-grid" aria-label="Research roadmap stages">
           <article class="roadmap-card" data-status="implemented">
@@ -2413,12 +2415,22 @@
               <strong>Evidence</strong><p>Task-specific held-out evaluations, qualitative inspection, and updated model cards.</p>
             </div>
           </article>
+          <article class="roadmap-card" data-status="planned">
+            <span class="roadmap-status">future</span>
+            <h3>Xperience Embodied Foundation Model</h3>
+            <p>Pretrain an Xperience-native domain model over synchronized video, audio, depth, pose, mocap, IMU, and language after smaller scaling stages prove value.</p>
+            <div class="roadmap-meta">
+              <strong>Entry</strong><p>Full-corpus access, PB-scale storage path, multi-node compute, and positive scaling evidence.</p>
+              <strong>Evidence</strong><p>Pretraining manifests, scaling curves, held-out evaluations, checkpoint inventory, model card, and data-boundary report.</p>
+            </div>
+          </article>
         </div>
         <div class="roadmap-links">
           <a href="research_roadmap.html">interactive roadmap</a>
           <a href="https://github.com/ChaoYue0307/ropedia-xperience-10m-task-suite/blob/main/RESEARCH_ROADMAP.md">roadmap document</a>
           <a href="data/research_roadmap.json">roadmap stages</a>
           <a href="data/foundation_model_plan.json">foundation model plan</a>
+          <a href="https://github.com/ChaoYue0307/ropedia-xperience-10m-task-suite/blob/main/XPERIENCE_EMBODIED_FOUNDATION_MODEL_PRETRAINING.md">native pretraining plan</a>
           <a href="data/research_roadmap_interactive.json">interactive map</a>
           <a href="https://github.com/ChaoYue0307/ropedia-xperience-10m-task-suite/blob/main/results/omni_finetune/DATA_ACCESS_STATUS.md">scale-up status</a>
           <a href="https://github.com/ChaoYue0307/ropedia-xperience-10m-task-suite/blob/main/PROJECT_STATUS.md">project status</a>
@@ -2438,7 +2450,7 @@
           <article class="artifact"><h3>Metric contract</h3><p>All 12 tasks list input, target, primary metric, minimal baseline score, and neural MLP score from committed result files.</p><a href="data/summary_metrics.json">summary metrics</a></article>
           <article class="artifact"><h3>Leakage controls</h3><p>Scalers fit on train windows only; future labels, target-side signals, caption/object labels, and contact labels stay on the target side unless explicitly queried.</p><a href="https://github.com/ChaoYue0307/ropedia-xperience-10m-task-suite/blob/main/scripts/build_evaluation_protocol.py">builder script</a></article>
           <article class="artifact"><h3>Audio ablation</h3><p>Audio and no-audio variants are evaluated across all 12 task contracts under the same chronological split.</p><a href="data/audio_ablation_summary.json">audio summary</a></article>
-          <article class="artifact"><h3>Foundation branch selection</h3><p>Qwen3-Omni is the first trainable baseline, Cosmos 3 becomes the world-model branch, and policy models wait for explicit action targets.</p><a href="data/foundation_model_plan.json">backbone plan</a></article>
+          <article class="artifact"><h3>Foundation branch selection</h3><p>Qwen3-Omni is the first trainable baseline, Cosmos 3 becomes the world-model branch, policy models wait for explicit action targets, and Xperience-native pretraining remains a later full-corpus goal.</p><a href="data/foundation_model_plan.json">backbone plan</a></article>
           <article class="artifact"><h3>Next evaluation stage</h3><p>This public-sample run covers single-episode task development. Cross-episode generalization, audio-visual learning, world modeling, policy targets, and held-out Qwen3-Omni training move to the multi-episode stage after selected data is prepared.</p><a href="https://github.com/ChaoYue0307/ropedia-xperience-10m-task-suite/blob/main/results/omni_finetune/DATA_ACCESS_STATUS.md">next-stage plan</a></article>
           <article class="artifact"><h3>Scale-up requirement</h3><p>The Omni pilot requires selected prepared episodes, held-out episode splits, no train/test episode leakage, training metadata, predictions, metrics, and a run report.</p><a href="https://github.com/ChaoYue0307/ropedia-xperience-10m-task-suite/blob/main/results/omni_finetune/DATA_ACCESS_STATUS.md">scale-up status</a></article>
         </div>
@@ -2492,10 +2504,11 @@
           <article class="evidence-card">
             <span class="status-pill">current plan</span>
             <h3>Foundation backbones are separated by role</h3>
-            <p>Qwen3-Omni stays first for held-out LoRA; Cosmos 3 is the world-model branch; OpenVLA/openpi/GR00T are policy candidates after action-space conversion.</p>
+            <p>Qwen3-Omni stays first for held-out LoRA; Cosmos 3 is the world-model branch; OpenVLA/openpi/GR00T are policy candidates after action-space conversion; Xperience-native pretraining is the later full-corpus goal.</p>
             <div class="evidence-links">
               <a href="data/foundation_model_plan.json">foundation model plan</a>
               <a href="https://github.com/ChaoYue0307/ropedia-xperience-10m-task-suite/blob/main/FOUNDATION_MODEL_PLAN.md">plan doc</a>
+              <a href="https://github.com/ChaoYue0307/ropedia-xperience-10m-task-suite/blob/main/XPERIENCE_EMBODIED_FOUNDATION_MODEL_PRETRAINING.md">pretraining plan</a>
             </div>
           </article>
           <article class="evidence-card">
@@ -2628,10 +2641,11 @@
           <article class="reading-card">
             <span class="step-index">04</span>
             <h3>Check the scale-up gate</h3>
-            <p>The multi-episode Qwen3-Omni path is prepared. The selected 128-episode result will be added after staging, preprocessing, training, and held-out evaluation pass.</p>
+            <p>The multi-episode Qwen3-Omni path is prepared. The selected 128-episode result will be added after staging, preprocessing, training, and held-out evaluation pass. The native-pretraining plan shows how this can grow into a full-corpus research direction.</p>
             <div class="reading-links">
               <a href="https://github.com/ChaoYue0307/ropedia-xperience-10m-task-suite/blob/main/results/omni_finetune/DATA_ACCESS_STATUS.md">scale-up status</a>
               <a href="https://github.com/ChaoYue0307/ropedia-xperience-10m-task-suite/blob/main/results/omni_finetune/MULTI_EPISODE_ACCESS_STATUS.md">data access</a>
+              <a href="https://github.com/ChaoYue0307/ropedia-xperience-10m-task-suite/blob/main/XPERIENCE_EMBODIED_FOUNDATION_MODEL_PRETRAINING.md">native pretraining</a>
               <a href="data/project_packet.json">reader path</a>
             </div>
           </article>
@@ -2659,7 +2673,7 @@
           <article class="artifact"><h3>Current project subset</h3><p>One public sample episode, 5,821 frames, 1,161 aligned windows, 8,546-dimensional task inputs, and no raw-data redistribution.</p><a href="data/modality_atlas.json">modality atlas</a></article>
           <article class="artifact"><h3>Covered now</h3><p>Action/subtask labels, next-action prediction, temporal diagnostics, hand trajectory, contact, object relevance, caption grounding, retrieval, reconstruction, and misalignment.</p><a href="data/summary_metrics.json">summary metrics</a></article>
           <article class="artifact"><h3>Responsible use</h3><p>This project is for research exploration and excludes identity recognition, surveillance, biometric profiling, sensitive-attribute inference, and safety-critical deployment.</p><a href="https://github.com/ChaoYue0307/ropedia-xperience-10m-task-suite/blob/main/DATA_NOTICE.md">use notes</a></article>
-          <article class="artifact"><h3>Later milestones</h3><p>Full audio-visual learning, caption generation, depth-pixel prediction, SLAM estimation, neural rendering, policy learning, cross-episode generalization, and held-out Qwen3-Omni evaluation.</p><a href="https://github.com/ChaoYue0307/ropedia-xperience-10m-task-suite/blob/main/results/omni_finetune/DATA_ACCESS_STATUS.md">scale-up status</a></article>
+          <article class="artifact"><h3>Later milestones</h3><p>Full audio-visual learning, caption generation, depth-pixel prediction, SLAM estimation, neural rendering, policy learning, cross-episode generalization, held-out Qwen3-Omni evaluation, and future Xperience-native pretraining.</p><a href="https://github.com/ChaoYue0307/ropedia-xperience-10m-task-suite/blob/main/XPERIENCE_EMBODIED_FOUNDATION_MODEL_PRETRAINING.md">native pretraining</a></article>
         </div>
       </div>
     </section>
@@ -3103,10 +3117,11 @@
             </div>
             <div class="artifact-grid">
               <article class="artifact primary-artifact"><div><h3>Project scope</h3><p>Connects implemented single-episode artifacts, setup-stage Omni work, the selected 128-episode pilot, and later multi-episode milestones.</p></div><a href="https://github.com/ChaoYue0307/ropedia-xperience-10m-task-suite/blob/main/EVIDENCE_CONTRACT.md">evidence contract</a></article>
-              <article class="artifact"><h3>Foundation-model plan</h3><p>Backbone selection matrix covering Qwen3-Omni, Cosmos 3, GR00T, OpenVLA/openpi, Gemini Robotics, Octo, and SmolVLA-style policy candidates.</p><a href="data/foundation_model_plan.json">foundation model plan</a></article>
+              <article class="artifact"><h3>Foundation-model plan</h3><p>Backbone selection matrix covering Qwen3-Omni, Cosmos 3, GR00T, OpenVLA/openpi, Gemini Robotics, Octo, SmolVLA-style policy candidates, and the future Xperience-native pretraining goal.</p><a href="data/foundation_model_plan.json">foundation model plan</a></article>
               <article class="artifact"><h3>Multi-episode data access</h3><p>Public data-access path, selected 128-episode pilot plan, and preparation requirements.</p><a href="https://github.com/ChaoYue0307/ropedia-xperience-10m-task-suite/blob/main/results/omni_finetune/MULTI_EPISODE_ACCESS_STATUS.md">data access</a></article>
               <article class="artifact"><h3>Qwen3-Omni preparation</h3><p>Episode selection and manifest preparation for the current scale-up path.</p><a href="https://github.com/ChaoYue0307/ropedia-xperience-10m-task-suite/blob/main/results/omni_finetune/episode_manifest.json">preparation details</a></article>
               <article class="artifact"><h3>Scale-up requirement</h3><p>What must be available before full pilot training and held-out metrics.</p><a href="https://github.com/ChaoYue0307/ropedia-xperience-10m-task-suite/blob/main/results/omni_finetune/DATA_ACCESS_STATUS.md">training requirements</a></article>
+              <article class="artifact"><h3>Xperience-native pretraining</h3><p>Future plan for a domain-specific embodied foundation model trained from scratch over full-corpus video, audio, geometry, motion, inertial, and language streams.</p><a href="https://github.com/ChaoYue0307/ropedia-xperience-10m-task-suite/blob/main/XPERIENCE_EMBODIED_FOUNDATION_MODEL_PRETRAINING.md">pretraining plan</a></article>
             </div>
           </section>
 
@@ -3123,7 +3138,7 @@
               <article class="artifact"><h3>Dataset notes</h3><p>Official dataset links, public sample source, modalities, access boundary, and current project subset.</p><a href="https://github.com/ChaoYue0307/ropedia-xperience-10m-task-suite/blob/main/XPERIENCE10M_DATASET_CARD_ALIGNMENT.md">dataset notes</a></article>
               <article class="artifact"><h3>Reproducibility</h3><p>Commands and expected outputs for rebuilding the public-sample task suite and visual artifacts.</p><a href="https://github.com/ChaoYue0307/ropedia-xperience-10m-task-suite/blob/main/REPRODUCIBILITY.md">reproduce</a></article>
               <article class="artifact"><h3>Qwen3-Omni status</h3><p>Data requirements and evaluation boundary for the selected multi-episode LoRA pilot.</p><a href="https://github.com/ChaoYue0307/ropedia-xperience-10m-task-suite/blob/main/results/omni_finetune/DATA_ACCESS_STATUS.md">training status</a></article>
-              <article class="artifact"><h3>Foundation-model plan</h3><p>Qwen3-Omni, Cosmos 3, GR00T, OpenVLA/openpi, Gemini Robotics, Octo, and SmolVLA-style branches by role.</p><a href="data/foundation_model_plan.json">model plan</a></article>
+              <article class="artifact"><h3>Foundation-model plan</h3><p>Qwen3-Omni, Cosmos 3, GR00T, OpenVLA/openpi, Gemini Robotics, Octo, SmolVLA-style branches, and the Xperience-native pretraining goal by role.</p><a href="data/foundation_model_plan.json">model plan</a></article>
               <article class="artifact"><h3>Hub artifacts</h3><p>Derived CSV/JSON/Markdown/figure artifacts without redistributing raw Xperience-10M data.</p><a href="https://huggingface.co/datasets/cy0307/ropedia-xperience-10m-task-suite-artifacts">artifact dataset</a></article>
               <article class="artifact"><h3>Baseline models</h3><p>Lightweight minimal and neural task-head model files for the 12 task contracts.</p><a href="https://huggingface.co/cy0307/ropedia-xperience-10m-task-baselines">model repo</a></article>
             </div>
@@ -3143,6 +3158,7 @@
           <article class="artifact"><h3>Transfer</h3><p>Download raw episodes only from official gated sources, exclude visualization.rrd, validate files, then stage them for training.</p></article>
           <article class="artifact"><h3>Current LoRA artifact</h3><p>The current LoRA artifact uses the locally available sample data. The multi-episode result begins after selected data is prepared, preprocessed, trained, and evaluated on held-out sessions.</p></article>
           <article class="artifact"><h3>Backbone branches</h3><p>Qwen3-Omni is the immediate LoRA path; Cosmos 3 is the first world-model branch; GR00T/OpenVLA/openpi become policy branches after action targets are well-defined.</p><a href="data/foundation_model_plan.json">backbone plan</a></article>
+          <article class="artifact"><h3>Native foundation model</h3><p>The long-term goal is a full-corpus Xperience Embodied Foundation Model trained on synchronized perception, geometry, motion, inertial, audio, and language streams after smaller scaling stages validate the approach.</p><a href="https://github.com/ChaoYue0307/ropedia-xperience-10m-task-suite/blob/main/XPERIENCE_EMBODIED_FOUNDATION_MODEL_PRETRAINING.md">pretraining plan</a></article>
         </div>
       </div>
     </section>

docs/research_roadmap.html

@@ -605,8 +605,9 @@
         <h1>Interactive Research Roadmap.</h1>
         <p class="hero-copy">
           This page connects the current public-sample task lab to the four research
-          directions, the next multi-episode Qwen3-Omni fine-tuning path, and
-          the later Cosmos 3 / policy-model branch choices. It loads
+          directions, the next multi-episode Qwen3-Omni fine-tuning path, the
+          later Cosmos 3 / policy-model branch choices, and the future
+          Xperience-native foundation-model pretraining goal. It loads
           directly from generated project artifacts, so the track and task views stay
           tied to the real sample metrics and scale-up status.
         </p>
@@ -630,7 +631,7 @@
           </div>
           <div class="route-step">
             <strong>03</strong>
-            <div><b>Omni + branches</b><span>Qwen3-Omni first, Cosmos 3 and policy models after data preparation</span></div>
+            <div><b>Omni + branches</b><span>Qwen3-Omni first, Cosmos 3 and policy models next, native pretraining later</span></div>
             <em id="routeOmni">pending data</em>
           </div>
         </div>
@@ -701,7 +702,7 @@
       },
       omni: {
         title: "Omni pilot and foundation branches",
-        summary: "Run Qwen3-Omni first for the held-out LoRA pilot, then evaluate Cosmos 3 for world modeling and policy candidates after action targets are explicit.",
+        summary: "Run Qwen3-Omni first for the held-out LoRA pilot, evaluate Cosmos 3 for world modeling and policy candidates after action targets are explicit, then treat Xperience-native pretraining as the full-corpus future goal.",
       }
     };
 

index.html

@@ -2141,9 +2141,11 @@
         <p class="hero-copy">
           This project uses the public Xperience-10M sample from Ropedia to explore
           embodied-AI task design, multimodal feature construction, lightweight
-          baselines, and future Omni-model fine-tuning. It starts from the sample
-          episode available now, then keeps the same data contracts ready for
-          held-out multi-episode training when more Xperience-10M data is prepared.
+          baselines, future Omni-model fine-tuning, and the long-term path toward
+          an Xperience-native embodied foundation model. It starts from the
+          sample episode available now, then keeps the same data contracts ready
+          for held-out multi-episode training when more Xperience-10M data is
+          prepared.
         </p>
         <div class="hero-actions">
           <a class="button primary" href="research_roadmap.html">Open roadmap</a>
@@ -2252,7 +2254,7 @@
             </article>
             <article class="brief-card">
               <strong>Scale-up readiness</strong>
-              <p>Connects the same data contract to 32/128-episode held-out pilots, Qwen3-Omni LoRA, Cosmos-style world modeling, and later policy-model branches.</p>
+              <p>Connects the same data contract to 32/128-episode held-out pilots, Qwen3-Omni LoRA, Cosmos-style world modeling, policy-model branches, and the later Xperience-native pretraining goal.</p>
             </article>
           </div>
           <div class="brief-actions">
@@ -2356,7 +2358,7 @@
       <div class="wrap">
         <div class="section-head">
           <h2>Research roadmap.</h2>
-          <p>The project path moves from the current public-sample task lab to multi-episode data preparation, held-out Qwen3-Omni fine-tuning, robustness runs, and larger foundation/world-model extensions.</p>
+          <p>The project path moves from the current public-sample task lab to multi-episode data preparation, held-out Qwen3-Omni fine-tuning, robustness runs, world/policy branches, and the future Xperience Embodied Foundation Model pretraining goal.</p>
         </div>
         <div class="roadmap-grid" aria-label="Research roadmap stages">
           <article class="roadmap-card" data-status="implemented">
@@ -2413,12 +2415,22 @@
               <strong>Evidence</strong><p>Task-specific held-out evaluations, qualitative inspection, and updated model cards.</p>
             </div>
           </article>
+          <article class="roadmap-card" data-status="planned">
+            <span class="roadmap-status">future</span>
+            <h3>Xperience Embodied Foundation Model</h3>
+            <p>Pretrain an Xperience-native domain model over synchronized video, audio, depth, pose, mocap, IMU, and language after smaller scaling stages prove value.</p>
+            <div class="roadmap-meta">
+              <strong>Entry</strong><p>Full-corpus access, PB-scale storage path, multi-node compute, and positive scaling evidence.</p>
+              <strong>Evidence</strong><p>Pretraining manifests, scaling curves, held-out evaluations, checkpoint inventory, model card, and data-boundary report.</p>
+            </div>
+          </article>
         </div>
         <div class="roadmap-links">
           <a href="research_roadmap.html">interactive roadmap</a>
           <a href="https://github.com/ChaoYue0307/ropedia-xperience-10m-task-suite/blob/main/RESEARCH_ROADMAP.md">roadmap document</a>
           <a href="data/research_roadmap.json">roadmap stages</a>
           <a href="data/foundation_model_plan.json">foundation model plan</a>
+          <a href="https://github.com/ChaoYue0307/ropedia-xperience-10m-task-suite/blob/main/XPERIENCE_EMBODIED_FOUNDATION_MODEL_PRETRAINING.md">native pretraining plan</a>
           <a href="data/research_roadmap_interactive.json">interactive map</a>
           <a href="https://github.com/ChaoYue0307/ropedia-xperience-10m-task-suite/blob/main/results/omni_finetune/DATA_ACCESS_STATUS.md">scale-up status</a>
           <a href="https://github.com/ChaoYue0307/ropedia-xperience-10m-task-suite/blob/main/PROJECT_STATUS.md">project status</a>
@@ -2438,7 +2450,7 @@
           <article class="artifact"><h3>Metric contract</h3><p>All 12 tasks list input, target, primary metric, minimal baseline score, and neural MLP score from committed result files.</p><a href="data/summary_metrics.json">summary metrics</a></article>
           <article class="artifact"><h3>Leakage controls</h3><p>Scalers fit on train windows only; future labels, target-side signals, caption/object labels, and contact labels stay on the target side unless explicitly queried.</p><a href="https://github.com/ChaoYue0307/ropedia-xperience-10m-task-suite/blob/main/scripts/build_evaluation_protocol.py">builder script</a></article>
           <article class="artifact"><h3>Audio ablation</h3><p>Audio and no-audio variants are evaluated across all 12 task contracts under the same chronological split.</p><a href="data/audio_ablation_summary.json">audio summary</a></article>
-          <article class="artifact"><h3>Foundation branch selection</h3><p>Qwen3-Omni is the first trainable baseline, Cosmos 3 becomes the world-model branch, and policy models wait for explicit action targets.</p><a href="data/foundation_model_plan.json">backbone plan</a></article>
+          <article class="artifact"><h3>Foundation branch selection</h3><p>Qwen3-Omni is the first trainable baseline, Cosmos 3 becomes the world-model branch, policy models wait for explicit action targets, and Xperience-native pretraining remains a later full-corpus goal.</p><a href="data/foundation_model_plan.json">backbone plan</a></article>
           <article class="artifact"><h3>Next evaluation stage</h3><p>This public-sample run covers single-episode task development. Cross-episode generalization, audio-visual learning, world modeling, policy targets, and held-out Qwen3-Omni training move to the multi-episode stage after selected data is prepared.</p><a href="https://github.com/ChaoYue0307/ropedia-xperience-10m-task-suite/blob/main/results/omni_finetune/DATA_ACCESS_STATUS.md">next-stage plan</a></article>
           <article class="artifact"><h3>Scale-up requirement</h3><p>The Omni pilot requires selected prepared episodes, held-out episode splits, no train/test episode leakage, training metadata, predictions, metrics, and a run report.</p><a href="https://github.com/ChaoYue0307/ropedia-xperience-10m-task-suite/blob/main/results/omni_finetune/DATA_ACCESS_STATUS.md">scale-up status</a></article>
         </div>
@@ -2492,10 +2504,11 @@
           <article class="evidence-card">
             <span class="status-pill">current plan</span>
             <h3>Foundation backbones are separated by role</h3>
-            <p>Qwen3-Omni stays first for held-out LoRA; Cosmos 3 is the world-model branch; OpenVLA/openpi/GR00T are policy candidates after action-space conversion.</p>
+            <p>Qwen3-Omni stays first for held-out LoRA; Cosmos 3 is the world-model branch; OpenVLA/openpi/GR00T are policy candidates after action-space conversion; Xperience-native pretraining is the later full-corpus goal.</p>
             <div class="evidence-links">
               <a href="data/foundation_model_plan.json">foundation model plan</a>
               <a href="https://github.com/ChaoYue0307/ropedia-xperience-10m-task-suite/blob/main/FOUNDATION_MODEL_PLAN.md">plan doc</a>
+              <a href="https://github.com/ChaoYue0307/ropedia-xperience-10m-task-suite/blob/main/XPERIENCE_EMBODIED_FOUNDATION_MODEL_PRETRAINING.md">pretraining plan</a>
             </div>
           </article>
           <article class="evidence-card">
@@ -2628,10 +2641,11 @@
           <article class="reading-card">
             <span class="step-index">04</span>
             <h3>Check the scale-up gate</h3>
-            <p>The multi-episode Qwen3-Omni path is prepared. The selected 128-episode result will be added after staging, preprocessing, training, and held-out evaluation pass.</p>
+            <p>The multi-episode Qwen3-Omni path is prepared. The selected 128-episode result will be added after staging, preprocessing, training, and held-out evaluation pass. The native-pretraining plan shows how this can grow into a full-corpus research direction.</p>
             <div class="reading-links">
               <a href="https://github.com/ChaoYue0307/ropedia-xperience-10m-task-suite/blob/main/results/omni_finetune/DATA_ACCESS_STATUS.md">scale-up status</a>
               <a href="https://github.com/ChaoYue0307/ropedia-xperience-10m-task-suite/blob/main/results/omni_finetune/MULTI_EPISODE_ACCESS_STATUS.md">data access</a>
+              <a href="https://github.com/ChaoYue0307/ropedia-xperience-10m-task-suite/blob/main/XPERIENCE_EMBODIED_FOUNDATION_MODEL_PRETRAINING.md">native pretraining</a>
               <a href="data/project_packet.json">reader path</a>
             </div>
           </article>
@@ -2659,7 +2673,7 @@
           <article class="artifact"><h3>Current project subset</h3><p>One public sample episode, 5,821 frames, 1,161 aligned windows, 8,546-dimensional task inputs, and no raw-data redistribution.</p><a href="data/modality_atlas.json">modality atlas</a></article>
           <article class="artifact"><h3>Covered now</h3><p>Action/subtask labels, next-action prediction, temporal diagnostics, hand trajectory, contact, object relevance, caption grounding, retrieval, reconstruction, and misalignment.</p><a href="data/summary_metrics.json">summary metrics</a></article>
           <article class="artifact"><h3>Responsible use</h3><p>This project is for research exploration and excludes identity recognition, surveillance, biometric profiling, sensitive-attribute inference, and safety-critical deployment.</p><a href="https://github.com/ChaoYue0307/ropedia-xperience-10m-task-suite/blob/main/DATA_NOTICE.md">use notes</a></article>
-          <article class="artifact"><h3>Later milestones</h3><p>Full audio-visual learning, caption generation, depth-pixel prediction, SLAM estimation, neural rendering, policy learning, cross-episode generalization, and held-out Qwen3-Omni evaluation.</p><a href="https://github.com/ChaoYue0307/ropedia-xperience-10m-task-suite/blob/main/results/omni_finetune/DATA_ACCESS_STATUS.md">scale-up status</a></article>
+          <article class="artifact"><h3>Later milestones</h3><p>Full audio-visual learning, caption generation, depth-pixel prediction, SLAM estimation, neural rendering, policy learning, cross-episode generalization, held-out Qwen3-Omni evaluation, and future Xperience-native pretraining.</p><a href="https://github.com/ChaoYue0307/ropedia-xperience-10m-task-suite/blob/main/XPERIENCE_EMBODIED_FOUNDATION_MODEL_PRETRAINING.md">native pretraining</a></article>
         </div>
       </div>
     </section>
@@ -3103,10 +3117,11 @@
             </div>
             <div class="artifact-grid">
               <article class="artifact primary-artifact"><div><h3>Project scope</h3><p>Connects implemented single-episode artifacts, setup-stage Omni work, the selected 128-episode pilot, and later multi-episode milestones.</p></div><a href="https://github.com/ChaoYue0307/ropedia-xperience-10m-task-suite/blob/main/EVIDENCE_CONTRACT.md">evidence contract</a></article>
-              <article class="artifact"><h3>Foundation-model plan</h3><p>Backbone selection matrix covering Qwen3-Omni, Cosmos 3, GR00T, OpenVLA/openpi, Gemini Robotics, Octo, and SmolVLA-style policy candidates.</p><a href="data/foundation_model_plan.json">foundation model plan</a></article>
+              <article class="artifact"><h3>Foundation-model plan</h3><p>Backbone selection matrix covering Qwen3-Omni, Cosmos 3, GR00T, OpenVLA/openpi, Gemini Robotics, Octo, SmolVLA-style policy candidates, and the future Xperience-native pretraining goal.</p><a href="data/foundation_model_plan.json">foundation model plan</a></article>
               <article class="artifact"><h3>Multi-episode data access</h3><p>Public data-access path, selected 128-episode pilot plan, and preparation requirements.</p><a href="https://github.com/ChaoYue0307/ropedia-xperience-10m-task-suite/blob/main/results/omni_finetune/MULTI_EPISODE_ACCESS_STATUS.md">data access</a></article>
               <article class="artifact"><h3>Qwen3-Omni preparation</h3><p>Episode selection and manifest preparation for the current scale-up path.</p><a href="https://github.com/ChaoYue0307/ropedia-xperience-10m-task-suite/blob/main/results/omni_finetune/episode_manifest.json">preparation details</a></article>
               <article class="artifact"><h3>Scale-up requirement</h3><p>What must be available before full pilot training and held-out metrics.</p><a href="https://github.com/ChaoYue0307/ropedia-xperience-10m-task-suite/blob/main/results/omni_finetune/DATA_ACCESS_STATUS.md">training requirements</a></article>
+              <article class="artifact"><h3>Xperience-native pretraining</h3><p>Future plan for a domain-specific embodied foundation model trained from scratch over full-corpus video, audio, geometry, motion, inertial, and language streams.</p><a href="https://github.com/ChaoYue0307/ropedia-xperience-10m-task-suite/blob/main/XPERIENCE_EMBODIED_FOUNDATION_MODEL_PRETRAINING.md">pretraining plan</a></article>
             </div>
           </section>
 
@@ -3123,7 +3138,7 @@
               <article class="artifact"><h3>Dataset notes</h3><p>Official dataset links, public sample source, modalities, access boundary, and current project subset.</p><a href="https://github.com/ChaoYue0307/ropedia-xperience-10m-task-suite/blob/main/XPERIENCE10M_DATASET_CARD_ALIGNMENT.md">dataset notes</a></article>
               <article class="artifact"><h3>Reproducibility</h3><p>Commands and expected outputs for rebuilding the public-sample task suite and visual artifacts.</p><a href="https://github.com/ChaoYue0307/ropedia-xperience-10m-task-suite/blob/main/REPRODUCIBILITY.md">reproduce</a></article>
               <article class="artifact"><h3>Qwen3-Omni status</h3><p>Data requirements and evaluation boundary for the selected multi-episode LoRA pilot.</p><a href="https://github.com/ChaoYue0307/ropedia-xperience-10m-task-suite/blob/main/results/omni_finetune/DATA_ACCESS_STATUS.md">training status</a></article>
-              <article class="artifact"><h3>Foundation-model plan</h3><p>Qwen3-Omni, Cosmos 3, GR00T, OpenVLA/openpi, Gemini Robotics, Octo, and SmolVLA-style branches by role.</p><a href="data/foundation_model_plan.json">model plan</a></article>
+              <article class="artifact"><h3>Foundation-model plan</h3><p>Qwen3-Omni, Cosmos 3, GR00T, OpenVLA/openpi, Gemini Robotics, Octo, SmolVLA-style branches, and the Xperience-native pretraining goal by role.</p><a href="data/foundation_model_plan.json">model plan</a></article>
               <article class="artifact"><h3>Hub artifacts</h3><p>Derived CSV/JSON/Markdown/figure artifacts without redistributing raw Xperience-10M data.</p><a href="https://huggingface.co/datasets/cy0307/ropedia-xperience-10m-task-suite-artifacts">artifact dataset</a></article>
               <article class="artifact"><h3>Baseline models</h3><p>Lightweight minimal and neural task-head model files for the 12 task contracts.</p><a href="https://huggingface.co/cy0307/ropedia-xperience-10m-task-baselines">model repo</a></article>
             </div>
@@ -3143,6 +3158,7 @@
           <article class="artifact"><h3>Transfer</h3><p>Download raw episodes only from official gated sources, exclude visualization.rrd, validate files, then stage them for training.</p></article>
           <article class="artifact"><h3>Current LoRA artifact</h3><p>The current LoRA artifact uses the locally available sample data. The multi-episode result begins after selected data is prepared, preprocessed, trained, and evaluated on held-out sessions.</p></article>
           <article class="artifact"><h3>Backbone branches</h3><p>Qwen3-Omni is the immediate LoRA path; Cosmos 3 is the first world-model branch; GR00T/OpenVLA/openpi become policy branches after action targets are well-defined.</p><a href="data/foundation_model_plan.json">backbone plan</a></article>
+          <article class="artifact"><h3>Native foundation model</h3><p>The long-term goal is a full-corpus Xperience Embodied Foundation Model trained on synchronized perception, geometry, motion, inertial, audio, and language streams after smaller scaling stages validate the approach.</p><a href="https://github.com/ChaoYue0307/ropedia-xperience-10m-task-suite/blob/main/XPERIENCE_EMBODIED_FOUNDATION_MODEL_PRETRAINING.md">pretraining plan</a></article>
         </div>
       </div>
     </section>

metrics/artifact_index.json

@@ -1,11 +1,11 @@
 {
   "title": "Ropedia Xperience-10M Task Suite Artifact Index",
-  "generated_at_utc": "2026-06-04T16:42:13+00:00",
+  "generated_at_utc": "2026-06-04T20:40:52+00:00",
   "status": "pass",
-  "artifact_count": 72,
+  "artifact_count": 73,
   "missing": [],
   "by_kind": {
-    "project_path": 11,
+    "project_path": 12,
     "project_scope": 1,
     "source_alignment": 5,
     "publication_workflow": 1,
@@ -62,8 +62,8 @@
       "surface": "repo_hf",
       "shows": "Gives a compact current-state table for first-pass readers.",
       "exists": true,
-      "bytes": 7138,
-      "sha256": "67d85a198ee90082e47d790bd0f4d9dafbc97625cd39b17cc94b9785ec25104a"
+      "bytes": 7207,
+      "sha256": "7baaba976ccc254da1a03ee2653057d1e08f3fb0c0cad035886c362442828720"
     },
     {
       "id": "project_status_json",
@@ -73,8 +73,8 @@
       "surface": "website_hf",
       "shows": "Machine-readable copy of the current project status for website and HF mirrors.",
       "exists": true,
-      "bytes": 9169,
-      "sha256": "50d3c87b774c8375dcb897bd363d25e392e5fd6571571c41d56e623df15063f8"
+      "bytes": 9874,
+      "sha256": "600c95726eae3404127a8b2110f35468ff2ba02943cae0fbcd3ea43c66109d3e"
     },
     {
       "id": "research_roadmap",
@@ -84,8 +84,8 @@
       "surface": "repo_hf",
       "shows": "Defines the path from public-sample task development to multi-episode held-out evaluation and larger omni-model extensions.",
       "exists": true,
-      "bytes": 6677,
-      "sha256": "58491bfb68ad3e6b7569bdb1a3cac3de7682a49beb9de368a114d58ebf0b118b"
+      "bytes": 8388,
+      "sha256": "0b3e3356076998ad94dc39f708cc783a4ebeab76c9da661cdd37ea12a3bb3665"
     },
     {
       "id": "research_roadmap_json",
@@ -95,8 +95,8 @@
       "surface": "website_hf",
       "shows": "Machine-readable research roadmap for the website and Hugging Face mirrors.",
       "exists": true,
-      "bytes": 5758,
-      "sha256": "54657eb8824416d2128d6e5710543bdaf9e41d7c2fa46dd14ad6b58fede3b5db"
+      "bytes": 7161,
+      "sha256": "cc96118c2c05108c831616151bc027441f7545495adeeb6a4a6a6bffe8da7801"
     },
     {
       "id": "foundation_model_plan",
@@ -106,8 +106,8 @@
       "surface": "repo_hf",
       "shows": "Defines the post-data-gate backbone choices: Qwen3-Omni first, Cosmos 3 for world modeling, and VLA/policy models after action-target conversion.",
       "exists": true,
-      "bytes": 6559,
-      "sha256": "955be6559b554f1c6c4141dd6ca2818127d89585df3940c2bd9b975ad9047926"
+      "bytes": 9075,
+      "sha256": "444d13ab556d2e16a199a7fca191b87c85ab8685d167aab357bc6341839299a2"
     },
     {
       "id": "foundation_model_plan_json",
@@ -117,8 +117,19 @@
       "surface": "website_hf",
       "shows": "Machine-readable foundation-model selection matrix with source links, entry conditions, and evaluation additions.",
       "exists": true,
-      "bytes": 8889,
-      "sha256": "e9b11114fa290253000b921575586780ccc3ba17665235259d4326c524f6ce97"
+      "bytes": 12981,
+      "sha256": "9cce52025a2e2f8afb4660e2af3353aea6ad0a1af380849218dd74c0acc271bb"
+    },
+    {
+      "id": "xperience_embodied_foundation_pretraining",
+      "title": "Xperience Embodied Foundation Model pretraining goal",
+      "path": "XPERIENCE_EMBODIED_FOUNDATION_MODEL_PRETRAINING.md",
+      "kind": "project_path",
+      "surface": "repo_hf",
+      "shows": "Describes the future full-corpus Xperience-native pretraining goal, target modules, objectives, staged scale-up, hardware ranges, and evaluation protocol.",
+      "exists": true,
+      "bytes": 9182,
+      "sha256": "b5a6ddc58647cd895a4772b110ecc9f4d685427fb37b81b22c6c02d2b9b323f1"
     },
     {
       "id": "evidence_contract",
@@ -150,8 +161,8 @@
       "surface": "repo_hf",
       "shows": "Gives the human-readable map from project scope to data, tasks, platform mirrors, and scale-up status.",
       "exists": true,
-      "bytes": 16890,
-      "sha256": "8bce9a773daf36214e377a7154b72a4493efd0f7d1a1941d5e0fc9bf784a29e5"
+      "bytes": 11440,
+      "sha256": "9b8821a9b14fe1744f2e6b5c419b2c5daaf70b57f1944caf1105c36c0c66c119"
     },
     {
       "id": "official_dataset_card_alignment",
@@ -195,7 +206,7 @@
       "shows": "Machine-readable source-alignment pass/fail check for repo, website, and HF surfaces.",
       "exists": true,
       "bytes": 4432,
-      "sha256": "96c7adc61c869fab71ef34ec2f6ec4f5f88af844509bd3d51d3818732d1f84b6"
+      "sha256": "06c6e2d111c72df01ed127fd288e6675b63e35a21ae12a2523931a072bd0bc49"
     },
     {
       "id": "source_alignment_validator",
@@ -573,8 +584,8 @@
       "surface": "repo_hf",
       "shows": "Generates the selective artifact catalog from local files.",
       "exists": true,
-      "bytes": 26568,
-      "sha256": "a611b399e858560f6afb41e121f033724753c5167d04e0d7bf243e569de88f04"
+      "bytes": 27020,
+      "sha256": "0ca7ed96f24caecbab31687cffa99f0eba8471258986412a294614e688c5aff5"
     },
     {
       "id": "publication_audit",
@@ -585,7 +596,7 @@
       "volatile": true,
       "shows": "Confirms public bundles exclude raw data, caches, heavy archives, and credential text.",
       "exists": true,
-      "bytes": 7289,
+      "bytes": 11811,
       "hash_policy": "existence_and_size_only"
     },
     {
@@ -597,7 +608,7 @@
       "volatile": true,
       "shows": "Separates setup paths from completed held-out-episode results.",
       "exists": true,
-      "bytes": 19505,
+      "bytes": 18981,
       "hash_policy": "existence_and_size_only"
     },
     {
@@ -609,7 +620,7 @@
       "volatile": true,
       "shows": "Confirms prepared GitHub/HF Space/artifact/model mirrors share the same critical data, figure, website HTML, and validator files.",
       "exists": true,
-      "bytes": 108617,
+      "bytes": 108621,
       "hash_policy": "existence_and_size_only"
     },
     {
@@ -621,7 +632,7 @@
       "volatile": true,
       "shows": "Confirms local website links, anchors, JSON data files, and referenced images resolve.",
       "exists": true,
-      "bytes": 14923,
+      "bytes": 14891,
       "hash_policy": "existence_and_size_only"
     },
     {

metrics/foundation_model_plan.json

@@ -2,6 +2,16 @@
   "title": "Xperience-10M Foundation Model Plan",
   "status": "planning_artifact",
   "current_boundary": "No held-out multi-episode foundation-model result has been completed in this repo. The current foundation-model artifacts are setup-stage until enough valid episodes are prepared and evaluated.",
+  "backbone_registry": {
+    "config_dir": "configs/omni_backbones",
+    "validator": "scripts/omni/backbone_registry.py --validate --json",
+    "extension_contract": "OMNI_MODEL_EXTENSION_CONTRACT.md",
+    "implemented_backbone": "qwen3_omni_lora",
+    "planned_backbones": [
+      "cosmos_world_model",
+      "policy_vla_branch"
+    ]
+  },
   "decision": {
     "immediate_trainable_backbone": "Qwen3-Omni",
     "first_world_model_branch": "Cosmos 3",
@@ -10,7 +20,65 @@
       "openpi pi0/pi0.5",
       "NVIDIA GR00T"
     ],
-    "external_reasoning_reference": "Gemini Robotics"
+    "external_reasoning_reference": "Gemini Robotics",
+    "long_term_native_pretraining_goal": "Xperience Embodied Foundation Model"
+  },
+  "future_pretraining_goal": {
+    "name": "Xperience Embodied Foundation Model",
+    "status": "future_planning_goal",
+    "role": "Domain-specific embodied foundation model pretrained on full Xperience-10M if full-corpus data, storage, and compute become available.",
+    "not_current_result": true,
+    "document": "XPERIENCE_EMBODIED_FOUNDATION_MODEL_PRETRAINING.md",
+    "entry_conditions": [
+      "Selected multi-episode Qwen3-Omni pilot trains and evaluates cleanly.",
+      "Scaling from 128 episodes to thousands of episodes shows measurable value.",
+      "Full-corpus storage, derived-shard storage, and fast active-cache capacity are available.",
+      "Distributed training, checkpoint/restart, and provenance tracking are reliable.",
+      "Evaluation covers held-out episodes, sessions, activities, objects, and missing-modality robustness."
+    ],
+    "target_modules": [
+      "multi-view video encoder",
+      "audio encoder",
+      "depth and geometry encoder",
+      "pose/SLAM encoder",
+      "hand/body mocap encoder",
+      "IMU encoder",
+      "language encoder/decoder",
+      "temporal fusion transformer",
+      "task heads and decoders"
+    ],
+    "pretraining_objectives": [
+      "masked multimodal modeling",
+      "cross-modal contrastive alignment",
+      "future-state prediction",
+      "ego-motion and hand-motion forecasting",
+      "action and procedure prediction",
+      "language grounding and captioning",
+      "contact and affordance prediction",
+      "optional policy-style targets after action conversion"
+    ],
+    "hardware_ranges": [
+      {
+        "goal": "0.3B-1B pilot",
+        "compute": "8-32 modern 80GB-class data-center GPUs",
+        "use": "prove objectives and data loaders"
+      },
+      {
+        "goal": "1B-3B domain model",
+        "compute": "32-128 GPUs",
+        "use": "research-scale Xperience representation learning"
+      },
+      {
+        "goal": "3B-7B full-corpus domain model",
+        "compute": "128-512 GPUs",
+        "use": "first realistic full Xperience-native foundation model"
+      },
+      {
+        "goal": "30B-class omni model from scratch",
+        "compute": "512-2000+ GPUs",
+        "use": "lab-scale project after scaling curves justify cost"
+      }
+    ]
   },
   "model_families": [
     {
@@ -112,6 +180,21 @@
       "current_decision": "optional_baseline_after_data_staging",
       "entry_condition": "Action labels and baseline protocol exist.",
       "public_source": "https://github.com/huggingface/lerobot"
+    },
+    {
+      "priority": 8,
+      "family": "Xperience Embodied Foundation Model",
+      "category": "xperience_native_pretraining_goal",
+      "openness": "future project-specific model if full-corpus access and compute exist",
+      "best_role": "Domain model over synchronized embodied experience.",
+      "xperience10m_fit": [
+        "Uses the full aligned modality stack rather than treating sensors as auxiliary metadata.",
+        "Targets temporal embodied representation learning across perception, motion, geometry, audio, and language.",
+        "Can become the shared pretraining backbone for Qwen-style instruction tasks, Cosmos-style world modeling, and policy/action branches."
+      ],
+      "current_decision": "future_goal_after_scaling_evidence",
+      "entry_condition": "Full-corpus data path, PB-scale storage, multi-node compute, and positive smaller-run scaling evidence.",
+      "public_source": "XPERIENCE_EMBODIED_FOUNDATION_MODEL_PRETRAINING.md"
     }
   ],
   "execution_order": [
@@ -144,6 +227,11 @@
       "step": 6,
       "name": "Publishing threshold",
       "action": "Publish branch results only with real manifests, predictions, metrics, and qualitative examples."
+    },
+    {
+      "step": 7,
+      "name": "Xperience-native pretraining",
+      "action": "Start a from-scratch Xperience Embodied Foundation Model only after smaller scaling stages, full-corpus storage, multi-node compute, and held-out evaluation protocols are in place."
     }
   ],
   "evaluation_additions": [
@@ -230,6 +318,10 @@
     {
       "label": "LeRobot / SmolVLA",
       "url": "https://github.com/huggingface/lerobot"
+    },
+    {
+      "label": "Xperience Embodied Foundation Model pretraining plan",
+      "url": "XPERIENCE_EMBODIED_FOUNDATION_MODEL_PRETRAINING.md"
     }
   ]
 }

metrics/mirror_parity.json

@@ -1,6 +1,6 @@
 {
   "status": "pass",
-  "generated_at_utc": "2026-06-04T18:33:44+00:00",
+  "generated_at_utc": "2026-06-04T20:45:22+00:00",
   "hf_root": "hf_publish",
   "summary": {
     "group_count": 101,
@@ -71,27 +71,27 @@
       "local": {
         "path": "repo:docs/data/artifact_index.json",
         "exists": true,
-        "bytes": 32296,
-        "sha256": "5494e5ee1e40bc50d44a9cd6f77c8de694175939bda4f174fb5b1554e53ec508"
+        "bytes": 32864,
+        "sha256": "ec7d17898c42fd76109567c201f9638059b6a9a11a48817b32677a0eb2662178"
       },
       "mirrors": {
         "hf_space": {
           "path": "hf_space:data/artifact_index.json",
           "exists": true,
-          "bytes": 32296,
-          "sha256": "5494e5ee1e40bc50d44a9cd6f77c8de694175939bda4f174fb5b1554e53ec508"
+          "bytes": 32864,
+          "sha256": "ec7d17898c42fd76109567c201f9638059b6a9a11a48817b32677a0eb2662178"
         },
         "hf_artifacts": {
           "path": "hf_artifacts:docs/data/artifact_index.json",
           "exists": true,
-          "bytes": 32296,
-          "sha256": "5494e5ee1e40bc50d44a9cd6f77c8de694175939bda4f174fb5b1554e53ec508"
+          "bytes": 32864,
+          "sha256": "ec7d17898c42fd76109567c201f9638059b6a9a11a48817b32677a0eb2662178"
         },
         "hf_model": {
           "path": "hf_model:metrics/artifact_index.json",
           "exists": true,
-          "bytes": 32296,
-          "sha256": "5494e5ee1e40bc50d44a9cd6f77c8de694175939bda4f174fb5b1554e53ec508"
+          "bytes": 32864,
+          "sha256": "ec7d17898c42fd76109567c201f9638059b6a9a11a48817b32677a0eb2662178"
         }
       },
       "failures": []
@@ -226,27 +226,27 @@
       "local": {
         "path": "repo:docs/data/foundation_model_plan.json",
         "exists": true,
-        "bytes": 8889,
-        "sha256": "e9b11114fa290253000b921575586780ccc3ba17665235259d4326c524f6ce97"
+        "bytes": 12981,
+        "sha256": "9cce52025a2e2f8afb4660e2af3353aea6ad0a1af380849218dd74c0acc271bb"
       },
       "mirrors": {
         "hf_space": {
           "path": "hf_space:data/foundation_model_plan.json",
           "exists": true,
-          "bytes": 8889,
-          "sha256": "e9b11114fa290253000b921575586780ccc3ba17665235259d4326c524f6ce97"
+          "bytes": 12981,
+          "sha256": "9cce52025a2e2f8afb4660e2af3353aea6ad0a1af380849218dd74c0acc271bb"
         },
         "hf_artifacts": {
           "path": "hf_artifacts:docs/data/foundation_model_plan.json",
           "exists": true,
-          "bytes": 8889,
-          "sha256": "e9b11114fa290253000b921575586780ccc3ba17665235259d4326c524f6ce97"
+          "bytes": 12981,
+          "sha256": "9cce52025a2e2f8afb4660e2af3353aea6ad0a1af380849218dd74c0acc271bb"
         },
         "hf_model": {
           "path": "hf_model:metrics/foundation_model_plan.json",
           "exists": true,
-          "bytes": 8889,
-          "sha256": "e9b11114fa290253000b921575586780ccc3ba17665235259d4326c524f6ce97"
+          "bytes": 12981,
+          "sha256": "9cce52025a2e2f8afb4660e2af3353aea6ad0a1af380849218dd74c0acc271bb"
         }
       },
       "failures": []
@@ -412,27 +412,27 @@
       "local": {
         "path": "repo:docs/data/project_status.json",
         "exists": true,
-        "bytes": 9169,
-        "sha256": "50d3c87b774c8375dcb897bd363d25e392e5fd6571571c41d56e623df15063f8"
+        "bytes": 9874,
+        "sha256": "600c95726eae3404127a8b2110f35468ff2ba02943cae0fbcd3ea43c66109d3e"
       },
       "mirrors": {
         "hf_space": {
           "path": "hf_space:data/project_status.json",
           "exists": true,
-          "bytes": 9169,
-          "sha256": "50d3c87b774c8375dcb897bd363d25e392e5fd6571571c41d56e623df15063f8"
+          "bytes": 9874,
+          "sha256": "600c95726eae3404127a8b2110f35468ff2ba02943cae0fbcd3ea43c66109d3e"
         },
         "hf_artifacts": {
           "path": "hf_artifacts:docs/data/project_status.json",
           "exists": true,
-          "bytes": 9169,
-          "sha256": "50d3c87b774c8375dcb897bd363d25e392e5fd6571571c41d56e623df15063f8"
+          "bytes": 9874,
+          "sha256": "600c95726eae3404127a8b2110f35468ff2ba02943cae0fbcd3ea43c66109d3e"
         },
         "hf_model": {
           "path": "hf_model:metrics/project_status.json",
           "exists": true,
-          "bytes": 9169,
-          "sha256": "50d3c87b774c8375dcb897bd363d25e392e5fd6571571c41d56e623df15063f8"
+          "bytes": 9874,
+          "sha256": "600c95726eae3404127a8b2110f35468ff2ba02943cae0fbcd3ea43c66109d3e"
         }
       },
       "failures": []
@@ -444,26 +444,26 @@
         "path": "repo:docs/data/publication_audit.json",
         "exists": true,
         "bytes": 7237,
-        "sha256": "a95c93592ba70709b2fad24a911d19329e6823f25862cd4fcb256788190dd0f2"
+        "sha256": "7fbb19f8990b1a4d902e282c010d27e4391755564fa68af97d96c298c6b054f8"
       },
       "mirrors": {
         "hf_space": {
           "path": "hf_space:data/publication_audit.json",
           "exists": true,
           "bytes": 7237,
-          "sha256": "a95c93592ba70709b2fad24a911d19329e6823f25862cd4fcb256788190dd0f2"
+          "sha256": "7fbb19f8990b1a4d902e282c010d27e4391755564fa68af97d96c298c6b054f8"
         },
         "hf_artifacts": {
           "path": "hf_artifacts:docs/data/publication_audit.json",
           "exists": true,
           "bytes": 7237,
-          "sha256": "a95c93592ba70709b2fad24a911d19329e6823f25862cd4fcb256788190dd0f2"
+          "sha256": "7fbb19f8990b1a4d902e282c010d27e4391755564fa68af97d96c298c6b054f8"
         },
         "hf_model": {
           "path": "hf_model:metrics/publication_audit.json",
           "exists": true,
           "bytes": 7237,
-          "sha256": "a95c93592ba70709b2fad24a911d19329e6823f25862cd4fcb256788190dd0f2"
+          "sha256": "7fbb19f8990b1a4d902e282c010d27e4391755564fa68af97d96c298c6b054f8"
         }
       },
       "failures": []
@@ -598,27 +598,27 @@
       "local": {
         "path": "repo:docs/data/research_roadmap.json",
         "exists": true,
-        "bytes": 5758,
-        "sha256": "54657eb8824416d2128d6e5710543bdaf9e41d7c2fa46dd14ad6b58fede3b5db"
+        "bytes": 7161,
+        "sha256": "cc96118c2c05108c831616151bc027441f7545495adeeb6a4a6a6bffe8da7801"
       },
       "mirrors": {
         "hf_space": {
           "path": "hf_space:data/research_roadmap.json",
           "exists": true,
-          "bytes": 5758,
-          "sha256": "54657eb8824416d2128d6e5710543bdaf9e41d7c2fa46dd14ad6b58fede3b5db"
+          "bytes": 7161,
+          "sha256": "cc96118c2c05108c831616151bc027441f7545495adeeb6a4a6a6bffe8da7801"
         },
         "hf_artifacts": {
           "path": "hf_artifacts:docs/data/research_roadmap.json",
           "exists": true,
-          "bytes": 5758,
-          "sha256": "54657eb8824416d2128d6e5710543bdaf9e41d7c2fa46dd14ad6b58fede3b5db"
+          "bytes": 7161,
+          "sha256": "cc96118c2c05108c831616151bc027441f7545495adeeb6a4a6a6bffe8da7801"
         },
         "hf_model": {
           "path": "hf_model:metrics/research_roadmap.json",
           "exists": true,
-          "bytes": 5758,
-          "sha256": "54657eb8824416d2128d6e5710543bdaf9e41d7c2fa46dd14ad6b58fede3b5db"
+          "bytes": 7161,
+          "sha256": "cc96118c2c05108c831616151bc027441f7545495adeeb6a4a6a6bffe8da7801"
         }
       },
       "failures": []
@@ -629,27 +629,27 @@
       "local": {
         "path": "repo:docs/data/research_roadmap_interactive.json",
         "exists": true,
-        "bytes": 131519,
-        "sha256": "004fbcc7a3582da88dd66504d686604ecb0f04f65c9c8166bb0583e0fc174274"
+        "bytes": 134282,
+        "sha256": "ff37219a9f1d9b386a9d4c42766e4aa28f10ce6ef338dceeedd6bdb4a1b2c40a"
       },
       "mirrors": {
         "hf_space": {
           "path": "hf_space:data/research_roadmap_interactive.json",
           "exists": true,
-          "bytes": 131519,
-          "sha256": "004fbcc7a3582da88dd66504d686604ecb0f04f65c9c8166bb0583e0fc174274"
+          "bytes": 134282,
+          "sha256": "ff37219a9f1d9b386a9d4c42766e4aa28f10ce6ef338dceeedd6bdb4a1b2c40a"
         },
         "hf_artifacts": {
           "path": "hf_artifacts:docs/data/research_roadmap_interactive.json",
           "exists": true,
-          "bytes": 131519,
-          "sha256": "004fbcc7a3582da88dd66504d686604ecb0f04f65c9c8166bb0583e0fc174274"
+          "bytes": 134282,
+          "sha256": "ff37219a9f1d9b386a9d4c42766e4aa28f10ce6ef338dceeedd6bdb4a1b2c40a"
         },
         "hf_model": {
           "path": "hf_model:metrics/research_roadmap_interactive.json",
           "exists": true,
-          "bytes": 131519,
-          "sha256": "004fbcc7a3582da88dd66504d686604ecb0f04f65c9c8166bb0583e0fc174274"
+          "bytes": 134282,
+          "sha256": "ff37219a9f1d9b386a9d4c42766e4aa28f10ce6ef338dceeedd6bdb4a1b2c40a"
         }
       },
       "failures": []
@@ -1692,21 +1692,21 @@
       "local": {
         "path": "repo:scripts/build_artifact_index.py",
         "exists": true,
-        "bytes": 26568,
-        "sha256": "a611b399e858560f6afb41e121f033724753c5167d04e0d7bf243e569de88f04"
+        "bytes": 27020,
+        "sha256": "0ca7ed96f24caecbab31687cffa99f0eba8471258986412a294614e688c5aff5"
       },
       "mirrors": {
         "hf_artifacts": {
           "path": "hf_artifacts:scripts/build_artifact_index.py",
           "exists": true,
-          "bytes": 26568,
-          "sha256": "a611b399e858560f6afb41e121f033724753c5167d04e0d7bf243e569de88f04"
+          "bytes": 27020,
+          "sha256": "0ca7ed96f24caecbab31687cffa99f0eba8471258986412a294614e688c5aff5"
         },
         "hf_model": {
           "path": "hf_model:scripts/build_artifact_index.py",
           "exists": true,
-          "bytes": 26568,
-          "sha256": "a611b399e858560f6afb41e121f033724753c5167d04e0d7bf243e569de88f04"
+          "bytes": 27020,
+          "sha256": "0ca7ed96f24caecbab31687cffa99f0eba8471258986412a294614e688c5aff5"
         }
       },
       "failures": []
@@ -2017,21 +2017,21 @@
       "local": {
         "path": "repo:scripts/validate_publication_package.py",
         "exists": true,
-        "bytes": 17125,
-        "sha256": "51febee7a4caa4e3cbb3833c0c13ac502bd7106fdb3df06e868ed00bc8f9fd9e"
+        "bytes": 17197,
+        "sha256": "2a617f3204ffb8c59d1c5bc1828b4441a4d014bb531655fd0613e128a6d9abc2"
       },
       "mirrors": {
         "hf_artifacts": {
           "path": "hf_artifacts:scripts/validate_publication_package.py",
           "exists": true,
-          "bytes": 17125,
-          "sha256": "51febee7a4caa4e3cbb3833c0c13ac502bd7106fdb3df06e868ed00bc8f9fd9e"
+          "bytes": 17197,
+          "sha256": "2a617f3204ffb8c59d1c5bc1828b4441a4d014bb531655fd0613e128a6d9abc2"
         },
         "hf_model": {
           "path": "hf_model:scripts/validate_publication_package.py",
           "exists": true,
-          "bytes": 17125,
-          "sha256": "51febee7a4caa4e3cbb3833c0c13ac502bd7106fdb3df06e868ed00bc8f9fd9e"
+          "bytes": 17197,
+          "sha256": "2a617f3204ffb8c59d1c5bc1828b4441a4d014bb531655fd0613e128a6d9abc2"
         }
       },
       "failures": []
@@ -2217,21 +2217,21 @@
       "local": {
         "path": "repo:docs/index.html",
         "exists": true,
-        "bytes": 172286,
-        "sha256": "a736850416c0061adddbb6ced5897efd1add499ec26e510b6fe21a4945b341c8"
+        "bytes": 174923,
+        "sha256": "099fcc01cbb4d50f62c508b10f343f05b1c883962b85bda294bcede99af2a0f1"
       },
       "mirrors": {
         "hf_space": {
           "path": "hf_space:index.html",
           "exists": true,
-          "bytes": 172286,
-          "sha256": "a736850416c0061adddbb6ced5897efd1add499ec26e510b6fe21a4945b341c8"
+          "bytes": 174923,
+          "sha256": "099fcc01cbb4d50f62c508b10f343f05b1c883962b85bda294bcede99af2a0f1"
         },
         "hf_artifacts_docs": {
           "path": "hf_artifacts:docs/index.html",
           "exists": true,
-          "bytes": 172286,
-          "sha256": "a736850416c0061adddbb6ced5897efd1add499ec26e510b6fe21a4945b341c8"
+          "bytes": 174923,
+          "sha256": "099fcc01cbb4d50f62c508b10f343f05b1c883962b85bda294bcede99af2a0f1"
         }
       },
       "failures": []
@@ -2242,21 +2242,21 @@
       "local": {
         "path": "repo:docs/research_roadmap.html",
         "exists": true,
-        "bytes": 31554,
-        "sha256": "f51e83a4495f2d2012ec4c48191d66ca4456a00d7fcb335a427b7d86afc66109"
+        "bytes": 31702,
+        "sha256": "1b20a5cc342b3ba59ad808eed9f5bf978e2d9ac438c88b5c3eeba01f4e14b883"
       },
       "mirrors": {
         "hf_space": {
           "path": "hf_space:research_roadmap.html",
           "exists": true,
-          "bytes": 31554,
-          "sha256": "f51e83a4495f2d2012ec4c48191d66ca4456a00d7fcb335a427b7d86afc66109"
+          "bytes": 31702,
+          "sha256": "1b20a5cc342b3ba59ad808eed9f5bf978e2d9ac438c88b5c3eeba01f4e14b883"
         },
         "hf_artifacts_docs": {
           "path": "hf_artifacts:docs/research_roadmap.html",
           "exists": true,
-          "bytes": 31554,
-          "sha256": "f51e83a4495f2d2012ec4c48191d66ca4456a00d7fcb335a427b7d86afc66109"
+          "bytes": 31702,
+          "sha256": "1b20a5cc342b3ba59ad808eed9f5bf978e2d9ac438c88b5c3eeba01f4e14b883"
         }
       },
       "failures": []
@@ -2844,27 +2844,27 @@
       "local": {
         "path": "repo:FOUNDATION_MODEL_PLAN.md",
         "exists": true,
-        "bytes": 6559,
-        "sha256": "955be6559b554f1c6c4141dd6ca2818127d89585df3940c2bd9b975ad9047926"
+        "bytes": 9075,
+        "sha256": "444d13ab556d2e16a199a7fca191b87c85ab8685d167aab357bc6341839299a2"
       },
       "mirrors": {
         "hf_space": {
           "path": "hf_space:FOUNDATION_MODEL_PLAN.md",
           "exists": true,
-          "bytes": 6559,
-          "sha256": "955be6559b554f1c6c4141dd6ca2818127d89585df3940c2bd9b975ad9047926"
+          "bytes": 9075,
+          "sha256": "444d13ab556d2e16a199a7fca191b87c85ab8685d167aab357bc6341839299a2"
         },
         "hf_artifacts": {
           "path": "hf_artifacts:FOUNDATION_MODEL_PLAN.md",
           "exists": true,
-          "bytes": 6559,
-          "sha256": "955be6559b554f1c6c4141dd6ca2818127d89585df3940c2bd9b975ad9047926"
+          "bytes": 9075,
+          "sha256": "444d13ab556d2e16a199a7fca191b87c85ab8685d167aab357bc6341839299a2"
         },
         "hf_model": {
           "path": "hf_model:FOUNDATION_MODEL_PLAN.md",
           "exists": true,
-          "bytes": 6559,
-          "sha256": "955be6559b554f1c6c4141dd6ca2818127d89585df3940c2bd9b975ad9047926"
+          "bytes": 9075,
+          "sha256": "444d13ab556d2e16a199a7fca191b87c85ab8685d167aab357bc6341839299a2"
         }
       },
       "failures": []
@@ -2937,27 +2937,27 @@
       "local": {
         "path": "repo:RESEARCH_ROADMAP.md",
         "exists": true,
-        "bytes": 6677,
-        "sha256": "58491bfb68ad3e6b7569bdb1a3cac3de7682a49beb9de368a114d58ebf0b118b"
+        "bytes": 8388,
+        "sha256": "0b3e3356076998ad94dc39f708cc783a4ebeab76c9da661cdd37ea12a3bb3665"
       },
       "mirrors": {
         "hf_space": {
           "path": "hf_space:RESEARCH_ROADMAP.md",
           "exists": true,
-          "bytes": 6677,
-          "sha256": "58491bfb68ad3e6b7569bdb1a3cac3de7682a49beb9de368a114d58ebf0b118b"
+          "bytes": 8388,
+          "sha256": "0b3e3356076998ad94dc39f708cc783a4ebeab76c9da661cdd37ea12a3bb3665"
         },
         "hf_artifacts": {
           "path": "hf_artifacts:RESEARCH_ROADMAP.md",
           "exists": true,
-          "bytes": 6677,
-          "sha256": "58491bfb68ad3e6b7569bdb1a3cac3de7682a49beb9de368a114d58ebf0b118b"
+          "bytes": 8388,
+          "sha256": "0b3e3356076998ad94dc39f708cc783a4ebeab76c9da661cdd37ea12a3bb3665"
         },
         "hf_model": {
           "path": "hf_model:RESEARCH_ROADMAP.md",
           "exists": true,
-          "bytes": 6677,
-          "sha256": "58491bfb68ad3e6b7569bdb1a3cac3de7682a49beb9de368a114d58ebf0b118b"
+          "bytes": 8388,
+          "sha256": "0b3e3356076998ad94dc39f708cc783a4ebeab76c9da661cdd37ea12a3bb3665"
         }
       },
       "failures": []
@@ -2968,27 +2968,27 @@
       "local": {
         "path": "repo:PROJECT_STATUS.md",
         "exists": true,
-        "bytes": 6648,
-        "sha256": "b052c725472f1d59232918a4d5b0f3668534c1e25e24189307159f5a0157d58f"
+        "bytes": 7207,
+        "sha256": "7baaba976ccc254da1a03ee2653057d1e08f3fb0c0cad035886c362442828720"
       },
       "mirrors": {
         "hf_space": {
           "path": "hf_space:PROJECT_STATUS.md",
           "exists": true,
-          "bytes": 6648,
-          "sha256": "b052c725472f1d59232918a4d5b0f3668534c1e25e24189307159f5a0157d58f"
+          "bytes": 7207,
+          "sha256": "7baaba976ccc254da1a03ee2653057d1e08f3fb0c0cad035886c362442828720"
         },
         "hf_artifacts": {
           "path": "hf_artifacts:PROJECT_STATUS.md",
           "exists": true,
-          "bytes": 6648,
-          "sha256": "b052c725472f1d59232918a4d5b0f3668534c1e25e24189307159f5a0157d58f"
+          "bytes": 7207,
+          "sha256": "7baaba976ccc254da1a03ee2653057d1e08f3fb0c0cad035886c362442828720"
         },
         "hf_model": {
           "path": "hf_model:PROJECT_STATUS.md",
           "exists": true,
-          "bytes": 6648,
-          "sha256": "b052c725472f1d59232918a4d5b0f3668534c1e25e24189307159f5a0157d58f"
+          "bytes": 7207,
+          "sha256": "7baaba976ccc254da1a03ee2653057d1e08f3fb0c0cad035886c362442828720"
         }
       },
       "failures": []

metrics/project_status.json

@@ -82,7 +82,7 @@
                 "RESEARCH_ROADMAP.md",
                 "docs/data/research_roadmap.json"
             ],
-            "readout": "The roadmap connects public-sample task development to 128-episode data preparation, Qwen3-Omni LoRA, foundation-model selection, robustness runs, and larger omni/world-model extensions."
+            "readout": "The roadmap connects public-sample task development to 128-episode data preparation, Qwen3-Omni LoRA, foundation-model selection, robustness runs, world/policy branches, and the future Xperience-native pretraining goal."
         },
         {
             "area": "Foundation-model plan",
@@ -93,6 +93,14 @@
             ],
             "readout": "Qwen3-Omni remains the first trainable held-out LoRA baseline; Cosmos 3 is added as the first world-model/action-generation branch; OpenVLA/openpi/GR00T are policy candidates after action targets are explicit."
         },
+        {
+            "area": "Xperience Embodied Foundation Model",
+            "status": "future_goal",
+            "evidence": [
+                "XPERIENCE_EMBODIED_FOUNDATION_MODEL_PRETRAINING.md"
+            ],
+            "readout": "A future full-corpus pretraining plan describes target modules, objectives, staged scale-up, hardware ranges, and evaluation for a domain-specific embodied foundation model."
+        },
         {
             "area": "Official dataset wording",
             "status": "verified",
@@ -167,6 +175,7 @@
         "Inspect RESEARCH_TAKEAWAYS.md and docs/data/research_takeaways.json before interpreting model scores.",
         "Inspect RESEARCH_ROADMAP.md and docs/data/research_roadmap.json for the path from public-sample task work to multi-episode modeling.",
         "Inspect FOUNDATION_MODEL_PLAN.md and docs/data/foundation_model_plan.json before choosing a backbone branch.",
+        "Inspect XPERIENCE_EMBODIED_FOUNDATION_MODEL_PRETRAINING.md for the long-term full-corpus pretraining goal.",
         "Inspect docs/data/summary_metrics.json and results/episode_task_suite/neural_mlp/ to check the 12-task outputs.",
         "Inspect results/audio_ablation/AUDIO_ABLATION_SUMMARY.md before judging whether audio helps the current task suite.",
         "Inspect EVALUATION_PROTOCOL.md before judging task metrics or leakage controls.",
@@ -180,6 +189,7 @@
     "The current reconstruction task reconstructs feature vectors, not pixel-depth, mesh, NeRF, or Gaussian reconstruction.",
     "Audio is one of the synchronized source modalities in the current task representation.",
     "The audio ablation report compares audio/no-audio variants across all 12 task contracts in results/audio_ablation/.",
-    "Foundation-model selection is explicit: Qwen3-Omni is the immediate trainable pilot, Cosmos 3 is the first world-model branch, and policy models such as OpenVLA/openpi/GR00T wait for action-target conversion."
+    "Foundation-model selection is explicit: Qwen3-Omni is the immediate trainable pilot, Cosmos 3 is the first world-model branch, and policy models such as OpenVLA/openpi/GR00T wait for action-target conversion.",
+    "The Xperience Embodied Foundation Model is a future native-pretraining goal, not a completed model or current benchmark."
   ]
 }

metrics/publication_audit.json

@@ -1,6 +1,6 @@
 {
   "status": "pass",
-  "generated_at_utc": "2026-06-04T18:32:51+00:00",
+  "generated_at_utc": "2026-06-04T20:43:37+00:00",
   "checks": [
     {
       "name": "required_publication_assets_present",
@@ -182,8 +182,8 @@
     "github_repo": {
       "root": "repo",
       "exists": true,
-      "file_count": 386,
-      "text_file_count": 320,
+      "file_count": 396,
+      "text_file_count": 330,
       "largest_file": {
         "path": "results/episode_task_suite/modality_reconstruction/predictions.npz",
         "bytes": 55702978
@@ -193,8 +193,8 @@
     "hf_space_bundle": {
       "root": "hf_publish/space",
       "exists": true,
-      "file_count": 316,
-      "text_file_count": 250,
+      "file_count": 317,
+      "text_file_count": 251,
       "largest_file": {
         "path": "results/episode_task_suite/modality_reconstruction/predictions.npz",
         "bytes": 55702978
@@ -204,8 +204,8 @@
     "hf_artifact_bundle": {
       "root": "hf_publish/artifacts",
       "exists": true,
-      "file_count": 417,
-      "text_file_count": 329,
+      "file_count": 418,
+      "text_file_count": 330,
       "largest_file": {
         "path": "results/episode_task_suite/modality_reconstruction/predictions.npz",
         "bytes": 55702978
@@ -215,8 +215,8 @@
     "hf_model_bundle": {
       "root": "hf_publish/model",
       "exists": true,
-      "file_count": 643,
-      "text_file_count": 518,
+      "file_count": 644,
+      "text_file_count": 519,
       "largest_file": {
         "path": "pytorch_model.bin",
         "bytes": 93495480

metrics/research_roadmap.json

@@ -1,7 +1,7 @@
 {
   "title": "Ropedia Xperience-10M Research Roadmap",
-  "summary": "Staged path from the public-sample task lab to multi-episode held-out evaluation, foundation-model selection, and larger omni/world-model extensions.",
-  "current_decision_point": "Keep the public-sample task suite as the development harness, prepare the selected official Xperience-10M episodes for the held-out Qwen3-Omni pilot, then branch into Cosmos 3 world modeling and policy-model experiments after the data preparation path is stable.",
+  "summary": "Staged path from the public-sample task lab to multi-episode held-out evaluation, foundation-model selection, world/policy branches, and a future Xperience-native embodied foundation model.",
+  "current_decision_point": "Keep the public-sample task suite as the development harness, prepare the selected official Xperience-10M episodes for the held-out Qwen3-Omni pilot, then branch into Cosmos 3 world modeling and policy-model experiments after the data preparation path is stable. The Xperience Embodied Foundation Model is a later full-corpus pretraining goal, not a current result.",
   "phases": [
     {
       "id": "public_sample_task_lab",
@@ -126,6 +126,30 @@
         "updated model cards"
       ],
       "reader_takeaway": "The long-term direction is richer multimodal representation learning for embodied-AI reasoning, with model branches chosen by task fit rather than by a single default backbone."
+    },
+    {
+      "id": "xperience_embodied_foundation_pretraining",
+      "name": "Xperience Embodied Foundation Model Pretraining",
+      "status": "future",
+      "entry_condition": "Full-corpus access, PB-scale storage path, high-throughput data loading, multi-node compute, and positive scaling evidence from smaller multi-episode runs.",
+      "deliverables": [
+        "full-corpus episode and split manifests",
+        "pretraining shard and provenance manifests",
+        "0.3B-1B and 1B-3B scaling pilots",
+        "3B-7B Xperience-native domain model target",
+        "held-out episode/session/activity/object evaluations",
+        "missing-modality robustness report",
+        "model card and data-boundary report"
+      ],
+      "completion_evidence": [
+        "pretraining metadata",
+        "checkpoint inventory",
+        "scaling curves",
+        "held-out evaluation reports",
+        "qualitative retrieval or future-state examples",
+        "safety and data-boundary report"
+      ],
+      "reader_takeaway": "The final research direction is a domain-specific embodied foundation model trained directly on Xperience-10M, after smaller pilots justify the cost and infrastructure."
     }
   ],
   "public_surfaces_to_update": [
@@ -134,6 +158,7 @@
     "RESEARCH_TAKEAWAYS.md",
     "EVALUATION_PROTOCOL.md",
     "ARTIFACT_GUIDE.md",
+    "XPERIENCE_EMBODIED_FOUNDATION_MODEL_PRETRAINING.md",
     "docs/index.html",
     "docs/data/research_roadmap.json",
     "Hugging Face Space card",

metrics/research_roadmap_interactive.json

@@ -1837,7 +1837,8 @@
         "NVIDIA GR00T"
       ],
       "first_world_model_branch": "Cosmos 3",
-      "immediate_trainable_backbone": "Qwen3-Omni"
+      "immediate_trainable_backbone": "Qwen3-Omni",
+      "long_term_native_pretraining_goal": "Xperience Embodied Foundation Model"
     },
     "evaluation_additions": [
       {
@@ -1921,6 +1922,11 @@
         "action": "Publish branch results only with real manifests, predictions, metrics, and qualitative examples.",
         "name": "Publishing threshold",
         "step": 6
+      },
+      {
+        "action": "Start a from-scratch Xperience Embodied Foundation Model only after smaller scaling stages, full-corpus storage, multi-node compute, and held-out evaluation protocols are in place.",
+        "name": "Xperience-native pretraining",
+        "step": 7
       }
     ],
     "model_families": [
@@ -2023,6 +2029,21 @@
           "Useful after action target design.",
           "Less directly omni-modal than Qwen3-Omni or Cosmos 3."
         ]
+      },
+      {
+        "best_role": "Domain model over synchronized embodied experience.",
+        "category": "xperience_native_pretraining_goal",
+        "current_decision": "future_goal_after_scaling_evidence",
+        "entry_condition": "Full-corpus data path, PB-scale storage, multi-node compute, and positive smaller-run scaling evidence.",
+        "family": "Xperience Embodied Foundation Model",
+        "openness": "future project-specific model if full-corpus access and compute exist",
+        "priority": 8,
+        "public_source": "XPERIENCE_EMBODIED_FOUNDATION_MODEL_PRETRAINING.md",
+        "xperience10m_fit": [
+          "Uses the full aligned modality stack rather than treating sensors as auxiliary metadata.",
+          "Targets temporal embodied representation learning across perception, motion, geometry, audio, and language.",
+          "Can become the shared pretraining backbone for Qwen-style instruction tasks, Cosmos-style world modeling, and policy/action branches."
+        ]
       }
     ],
     "source_links": [
@@ -2057,11 +2078,15 @@
       {
         "label": "LeRobot / SmolVLA",
         "url": "https://github.com/huggingface/lerobot"
+      },
+      {
+        "label": "Xperience Embodied Foundation Model pretraining plan",
+        "url": "XPERIENCE_EMBODIED_FOUNDATION_MODEL_PRETRAINING.md"
       }
     ],
     "status": "planning_artifact"
   },
-  "generated_at_utc": "2026-06-04T16:42:13+00:00",
+  "generated_at_utc": "2026-06-04T20:40:29+00:00",
   "omni_plan": {
     "adapter": "LoRA rank 16, alpha 32, dropout 0.05",
     "backbone": "Qwen/Qwen3-Omni-30B-A3B-Instruct",
@@ -2208,6 +2233,31 @@
       "reader_takeaway": "The long-term direction is richer multimodal representation learning for embodied-AI reasoning, with model branches chosen by task fit rather than by a single default backbone.",
       "stage": "future",
       "status": "planned"
+    },
+    {
+      "completion_evidence": [
+        "pretraining metadata",
+        "checkpoint inventory",
+        "scaling curves",
+        "held-out evaluation reports",
+        "qualitative retrieval or future-state examples",
+        "safety and data-boundary report"
+      ],
+      "deliverables": [
+        "full-corpus episode and split manifests",
+        "pretraining shard and provenance manifests",
+        "0.3B-1B and 1B-3B scaling pilots",
+        "3B-7B Xperience-native domain model target",
+        "held-out episode/session/activity/object evaluations",
+        "missing-modality robustness report",
+        "model card and data-boundary report"
+      ],
+      "entry_condition": "Full-corpus access, PB-scale storage path, high-throughput data loading, multi-node compute, and positive scaling evidence from smaller multi-episode runs.",
+      "id": "xperience_embodied_foundation_pretraining",
+      "name": "Xperience Embodied Foundation Model Pretraining",
+      "reader_takeaway": "The final research direction is a domain-specific embodied foundation model trained directly on Xperience-10M, after smaller pilots justify the cost and infrastructure.",
+      "stage": "future",
+      "status": "future"
     }
   ],
   "scale_up": {

research_roadmap.html

@@ -605,8 +605,9 @@
         <h1>Interactive Research Roadmap.</h1>
         <p class="hero-copy">
           This page connects the current public-sample task lab to the four research
-          directions, the next multi-episode Qwen3-Omni fine-tuning path, and
-          the later Cosmos 3 / policy-model branch choices. It loads
+          directions, the next multi-episode Qwen3-Omni fine-tuning path, the
+          later Cosmos 3 / policy-model branch choices, and the future
+          Xperience-native foundation-model pretraining goal. It loads
           directly from generated project artifacts, so the track and task views stay
           tied to the real sample metrics and scale-up status.
         </p>
@@ -630,7 +631,7 @@
           </div>
           <div class="route-step">
             <strong>03</strong>
-            <div><b>Omni + branches</b><span>Qwen3-Omni first, Cosmos 3 and policy models after data preparation</span></div>
+            <div><b>Omni + branches</b><span>Qwen3-Omni first, Cosmos 3 and policy models next, native pretraining later</span></div>
             <em id="routeOmni">pending data</em>
           </div>
         </div>
@@ -701,7 +702,7 @@
       },
       omni: {
         title: "Omni pilot and foundation branches",
-        summary: "Run Qwen3-Omni first for the held-out LoRA pilot, then evaluate Cosmos 3 for world modeling and policy candidates after action targets are explicit.",
+        summary: "Run Qwen3-Omni first for the held-out LoRA pilot, evaluate Cosmos 3 for world modeling and policy candidates after action targets are explicit, then treat Xperience-native pretraining as the full-corpus future goal.",
       }
     };
 

scripts/build_artifact_index.py

@@ -81,6 +81,14 @@ ARTIFACTS = [
         "surface": "website_hf",
         "shows": "Machine-readable foundation-model selection matrix with source links, entry conditions, and evaluation additions.",
     },
+    {
+        "id": "xperience_embodied_foundation_pretraining",
+        "title": "Xperience Embodied Foundation Model pretraining goal",
+        "path": "XPERIENCE_EMBODIED_FOUNDATION_MODEL_PRETRAINING.md",
+        "kind": "project_path",
+        "surface": "repo_hf",
+        "shows": "Describes the future full-corpus Xperience-native pretraining goal, target modules, objectives, staged scale-up, hardware ranges, and evaluation protocol.",
+    },
     {
         "id": "evidence_contract",
         "title": "Evidence contract",

scripts/validate_publication_package.py

@@ -221,6 +221,8 @@ def scan(root: Path, *, paths: list[Path] | None = None, display_root: str | Non
                         "detail": reason,
                     })
             for needle, reason in STALE_PRESENTATION_STRINGS.items():
+                if path_rel == ".mailmap":
+                    continue
                 if needle in text:
                     violations.append({
                         "kind": "stale_presentation_copy",