Publish validation-aware Qwen3-Omni diagnostic package

EVALUATION_PROTOCOL.md

@@ -70,25 +70,25 @@ are not foundation models.
 
 ## Current Limitations
 
-- Cross-episode generalization is evaluated in the later multi-episode stage.
+- Cross-episode generalization for Qwen3-Omni has a first verified diagnostic pilot, but strong model quality is not yet shown.
 - Feature-vector reconstruction is separate from pixel depth, mesh, NeRF, or Gaussian reconstruction.
-- Qwen3-Omni setup artifacts are preparation artifacts until the selected held-out pilot runs.
+- The verified validation-aware Qwen3-Omni diagnostic pilot has weak held-out metrics and needs structured-output and task-quality improvements before larger model-quality claims.
 - Full audio-visual representation learning still needs multi-episode training; the current report includes single-episode audio/no-audio ablations.
 
 ## Scale-Up Gate
 
-The full Qwen3-Omni fine-tuning pilot requires all of the following before
-reporting held-out model metrics:
+The next Qwen3-Omni quality pilot requires all of the following before
+claiming improved held-out model quality:
 
 - selected prepared Xperience-10M episodes
 - held-out episode split with no train/test episode leakage
+- validation samples during training
 - manifest, training metadata, progress logs, metrics, predictions, and run report
 - held-out evaluation on test episodes rather than train windows
 
-Current status: prepared; selected data relay in progress. Read
-`results/omni_finetune/DATA_ACCESS_STATUS.md` and
-`results/omni_finetune/MULTI_EPISODE_ACCESS_STATUS.md` before interpreting any
-Qwen3-Omni artifact.
+Current status: verified diagnostic pilot; quality target not met. Read
+`docs/data/omni_finetune_verified_result.json` before interpreting any
+Qwen3-Omni metric.
 
 ## Machine-Readable Copy
 

PROJECT_STATUS.md

@@ -2,8 +2,8 @@
 
 This is the fastest way to understand the current research project state.
 It summarizes what has already been implemented from the public
-Xperience-10M sample, what is being prepared for multi-episode training, and
-which artifacts support the next development step.
+Xperience-10M sample, what the first multi-episode Qwen3-Omni diagnostic pilot
+shows, and which artifacts support the next development step.
 
 ## Research Positioning
 
@@ -21,15 +21,15 @@ 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, world/policy branches, and the future Xperience-native pretraining goal. |
+| Research roadmap | Current | `RESEARCH_ROADMAP.md`, `docs/data/research_roadmap.json` | The roadmap connects public-sample task development to the verified validation-aware Qwen3-Omni diagnostic baseline, structured-output improvement pass, 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. |
+| 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 local infrastructure details. |
 | Public package policy | Verified | `DATA_NOTICE.md`, `REPRODUCIBILITY.md` | Raw Xperience-10M data, private gated files, large archives, credentials, and full Qwen weights are not redistributed. |
 | Reproducibility | Verified for the public sample | `REPRODUCIBILITY.md`, `docs/data/reproducibility_matrix.json`, `notes/reproducibility_audit.md` | The public sample workflow has explicit commands, expected outputs, and exact-match reproduction evidence. |
-| Qwen3-Omni fine-tuning | Data preparation; full metrics pending | `results/omni_finetune/DATA_ACCESS_STATUS.md`, `results/omni_finetune/MULTI_EPISODE_ACCESS_STATUS.md` | The gated full dataset is available for a selected 128-episode pilot; final held-out metrics require completed preprocessing, manifest construction, training, and evaluation. |
+| Qwen3-Omni fine-tuning | Verified validation-aware diagnostic held-out pilot; quality target not met | `docs/data/omni_finetune_verified_result.json`, `results/omni_finetune/verified_public/`, `scripts/omni/package_verified_omni_result.py`, `scripts/omni/audit_verified_omni_package.py` | The selected 96/16/16 episode split produced a validation-aware public-safe held-out package with 3,808 exported windows, 512 validation windows, and 448 test predictions. JSON validity is 87.50%, below the 98% target, so the result is a diagnostic baseline and the next pass should focus on structured-output improvements and error analysis. |
 | Raw Xperience-10M redistribution | Not included | `DATA_NOTICE.md`, `docs/data/publication_audit.json` | Raw MP4, HDF5, RRD files, private gated data, and full Qwen weights are intentionally excluded. |
 
 ## Fast Research Route
@@ -53,15 +53,18 @@ scale-up readiness; it is not presented as final full-dataset model quality.
    controls.
 10. Inspect `XPERIENCE10M_DATASET_CARD_ALIGNMENT.md` only if you need the
    detailed upstream dataset-card context.
-11. Inspect `results/omni_finetune/DATA_ACCESS_STATUS.md` before judging
-   Qwen3-Omni scale-up status.
+11. Inspect `docs/data/omni_finetune_verified_result.json` before judging the
+   Qwen3-Omni diagnostic pilot.
 
 ## Current Reading Notes
 
 - Cross-episode generalization is a later multi-episode evaluation target; the
   current results use one public sample episode.
-- Older pilot path names refer to setup files, not completed held-out
-  training results.
+- Public-facing fine-tuning results should come from the verified result
+  package, not from live process logs or setup-only artifacts.
+- The first Qwen3-Omni held-out package verifies the pipeline, not strong model
+  quality: JSON validity is 87.50%, action macro-F1 is 0.0027, and subtask
+  accuracy is 0.0067.
 - The current reconstruction task reconstructs feature vectors, not pixel
   depth, meshes, NeRF outputs, or Gaussian splats.
 - Audio is part of the current 8,546-dimensional baseline feature vector.

README.md

@@ -1,110 +1,1124 @@
----
-license: other
-pretty_name: Ropedia Xperience-10M Task Suite Artifacts
-tags:
-  - robotics
-  - embodied-ai
-  - multimodal
-  - ropedia
-  - xperience-10m
-  - evaluation
-  - baseline
-  - neural-network
-  - pytorch
-  - retrieval
-  - audio
-task_categories:
-  - robotics
-  - time-series-forecasting
-  - text-retrieval
-language:
-  - en
-size_categories:
-  - 1K<n<10K
-models:
-  - cy0307/ropedia-xperience-10m-task-baselines
-  - cy0307/ropedia-qwen3-omni-lora-smoke
 configs:
   - config_name: episode_sample
     data_files:
       - split: public_sample
         path: viewer/episode_windows.jsonl
----
 
-# Ropedia Xperience-10M Task Suite Artifacts
+# Ropedia Xperience-10M Task Suite
 
-This dataset repo stores derived artifacts for the Ropedia Xperience-10M
-sample research project: manifests, metrics, predictions, figures, notes,
-scripts, website data, compact baseline task-head files, and audio contribution
-results. It does not redistribute raw
-Xperience-10M MP4/HDF5/RRD data or full Qwen weights.
+[![Website](https://img.shields.io/badge/site-GitHub%20Pages-1f63e9)](https://chaoyue0307.github.io/ropedia-xperience-10m-task-suite/)
+[![HF Space](https://img.shields.io/badge/Hugging%20Face-Space-ffb000)](https://huggingface.co/spaces/cy0307/ropedia-xperience-10m-task-suite)
+[![Dataset](https://img.shields.io/badge/dataset-Xperience--10M%20by%20Ropedia-008b9a)](https://huggingface.co/datasets/ropedia-ai/xperience-10m)
+[![GitHub Package](https://img.shields.io/badge/package-GHCR-2496ed)](https://github.com/ChaoYue0307/ropedia-xperience-10m-task-suite/pkgs/container/ropedia-xperience-10m-task-suite)
+[![Scope](https://img.shields.io/badge/scope-single%20public%20sample-b65b04)](#scope)
+[![Citation](https://img.shields.io/badge/citation-CFF-7ae5c3)](CITATION.cff)
+[![License](https://img.shields.io/badge/license-code%20MIT%20%2B%20data%20terms-ccffa0)](LICENSE)
 
-Current public-sample contract:
+<p align="center">
+  <img src="docs/assets/brand/xperience10m-logo-social-card.png" alt="Ropedia Xperience-10M Task Suite logo card" width="760">
+</p>
 
-- 1 public Xperience-10M sample episode
-- 5,821 frames and 1,161 aligned 20-frame windows
-- 8,546-dimensional multimodal task representation
-- 12 minimal task heads, 12 compact neural MLP heads, and 4 extension probes
-- single-episode chronological split; selected multi-episode pilot preparation is underway
-- the gated full dataset is available for a selected 128-episode pilot before held-out evaluation
-- future Xperience Embodied Foundation Model pretraining is documented as a long-term full-corpus goal, not as a completed model
+A research-development project built on the public Xperience-10M sample episode
+released by Ropedia. The goal is to make one richly multimodal egocentric
+episode understandable, turn it into concrete embodied-AI task definitions, and
+prepare the same pipeline for future held-out multi-episode training.
 
-![Ropedia Xperience-10M Task Suite logo](assets/brand/xperience10m-logo-social-card.png)
+## What To Open First
 
-![12-task infographic](assets/task_suite_infographic.png?v=xperience10m-taskfirst-v13-modality-xl)
+Start with `PROJECT_README.md` for the full project walkthrough, `PROJECT_STATUS.md`
+for the current evidence boundary, and `docs/index.html` for the static dashboard
+copy. The current Qwen3-Omni validation-aware diagnostic result is summarized in
+`docs/data/omni_finetune_verified_result.json`.
 
-![Qwen3-Omni LoRA training pipeline](assets/qwen3_omni_lora_pipeline.png?v=qwen3-lora-v1)
+## Dataset Boundary
 
-The LoRA pipeline figure shows how prepared valid Xperience-10M episodes become
-split-safe Qwen3-Omni training records, adapter inputs, predictions, metrics,
-run reports, and upload-ready LoRA artifacts. It documents the training path;
-final held-out model-quality metrics remain pending completed multi-episode
-training and evaluation.
+This artifact dataset contains derived windows, manifests, metrics, predictions,
+figures, reports, and public-safe Qwen3-Omni diagnostic summaries. Raw Xperience-10M videos,
+HDF5 annotations, RRD visualizations, private gated data, Qwen base weights, and
+LoRA adapter weights are not redistributed here.
 
-## What To Open First
+## Related Hub Repositories
+
+- Space: https://huggingface.co/spaces/cy0307/ropedia-xperience-10m-task-suite
+- Baseline model repo: https://huggingface.co/cy0307/ropedia-xperience-10m-task-baselines
+- Qwen3-Omni diagnostic LoRA repo: https://huggingface.co/cy0307/ropedia-qwen3-omni-lora-smoke
+- Official gated dataset: https://huggingface.co/datasets/ropedia-ai/xperience-10m
+- Public sample dataset: https://huggingface.co/datasets/ropedia-ai/xperience-10m-sample
+
+The central research questions are:
+
+- What can be learned from one aligned Xperience-10M episode while separating
+  sample-specific observations from later multi-episode questions?
+- Which input/output tasks are meaningful for embodied AI when video, depth,
+  pose, mocap, IMU, and language annotations are synchronized?
+- What baseline models and evaluation files should exist before scaling to
+  Qwen3-Omni or other multimodal foundation-model fine-tuning?
 
-| Reader goal | Start here |
+## Why This Project Exists
+
+This project is organized as a compact research artifact around Xperience-10M:
+start from a real public episode, make every modality and label path inspectable,
+turn the data into concrete embodied-AI tasks, and keep the evaluation boundary
+clear while preparing the next multi-episode experiments. The emphasis is on
+research judgment as much as implementation: what the sample can show, what it
+cannot show, and what evidence should exist before claiming model quality.
+
+The work is designed to demonstrate four capabilities that matter for
+embodied-AI research infrastructure:
+
+| Capability | What this project shows |
 | --- | --- |
-| Understand the whole project | `PROJECT_BRIEF.md`, `PROJECT_STATUS.md`, `PROJECT_README.md` |
-| Inspect the 12 task results | `results/episode_task_suite/summary_report.json`, `RESEARCH_TAKEAWAYS.md` |
-| Compare baseline model outputs | `results/episode_task_suite/`, `results/episode_task_suite/neural_mlp/` |
-| Explore one episode visually | `docs/single_episode_explorer.html`, `docs/research_roadmap.html` |
-| Understand audio's contribution | `results/audio_ablation/AUDIO_ABLATION_SUMMARY.md` |
-| Follow the Qwen3-Omni scale-up path | `FOUNDATION_MODEL_PLAN.md`, `results/omni_finetune/DATA_ACCESS_STATUS.md` |
-| Explore additional development directions | `ADDITIONAL_DEVELOPMENT_DIRECTIONS.md`, `docs/data/additional_development_directions.json` |
-| Read the future native pretraining goal | `XPERIENCE_EMBODIED_FOUNDATION_MODEL_PRETRAINING.md` |
+| 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, policy-model branches, and the future Xperience-native foundation-model pretraining goal |
 
-## Related Hub Repositories
+## Start Here
 
-| Repository | Role |
+For a first pass, use [`PROJECT_BRIEF.md`](PROJECT_BRIEF.md) or the
+machine-readable [`docs/data/project_brief.json`](docs/data/project_brief.json).
+They give the project shape in one page: what exists now, what the public
+sample can support, where the 12 tasks and baselines live, and what must happen
+before the multi-episode omni-model stage becomes a real held-out evaluation.
+
+| Reader goal | Best entry point |
 | --- | --- |
-| [`cy0307/ropedia-xperience-10m-task-baselines`](https://huggingface.co/cy0307/ropedia-xperience-10m-task-baselines) | Minimal and neural baseline task-head weights for these artifacts |
-| [`cy0307/ropedia-qwen3-omni-lora-smoke`](https://huggingface.co/cy0307/ropedia-qwen3-omni-lora-smoke) | Qwen3-Omni PEFT LoRA smoke adapter from the initial end-to-end readiness run |
-| [`cy0307/ropedia-xperience-10m-task-suite`](https://huggingface.co/spaces/cy0307/ropedia-xperience-10m-task-suite) | Interactive project dashboard |
-| [`ghcr.io/chaoyue0307/ropedia-xperience-10m-task-suite`](https://github.com/ChaoYue0307/ropedia-xperience-10m-task-suite/pkgs/container/ropedia-xperience-10m-task-suite) | GitHub Container Registry package for the static dashboard |
-| [`ropedia-ai/xperience-10m`](https://huggingface.co/datasets/ropedia-ai/xperience-10m) | Official gated upstream Xperience-10M dataset |
-| [`ropedia-ai/xperience-10m-sample`](https://huggingface.co/datasets/ropedia-ai/xperience-10m-sample) | Public sample episode source |
+| Understand the whole project quickly | [`PROJECT_BRIEF.md`](PROJECT_BRIEF.md) |
+| See the visual research dashboard | [GitHub Pages dashboard](https://chaoyue0307.github.io/ropedia-xperience-10m-task-suite/) |
+| 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) |
+| See additional concrete project directions | [`ADDITIONAL_DEVELOPMENT_DIRECTIONS.md`](ADDITIONAL_DEVELOPMENT_DIRECTIONS.md), [`docs/data/additional_development_directions.json`](docs/data/additional_development_directions.json) |
+| 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) |
 
-## Dataset Boundary
+## Research Project Overview
+
+| Theme | Current implementation |
+| --- | --- |
+| Dataset slice | One public Xperience-10M sample episode, 5,821 frames, 1,161 windows, and an 8,546-dimensional representation |
+| Modalities | Video, audio, depth, camera pose/SLAM, hand/body mocap, IMU, calibration, and language annotations |
+| 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 | A first selected-episode Qwen3-Omni LoRA diagnostic pilot has completed on the 96/16/16 split; it proves the multi-episode export/train/eval/package loop, but the weak held-out metrics make it a baseline for error analysis rather than a strong model. Cosmos 3/world-model and VLA/policy branches reuse the same split and package contract after their targets are implemented. |
+| Public surfaces | GitHub repo, GitHub Pages dashboard, GHCR static-site package, HF Space, HF artifact dataset, HF baseline-model repo, and HF collection |
+
+For the fastest interpretation of the current metrics, start with
+[`RESEARCH_TAKEAWAYS.md`](RESEARCH_TAKEAWAYS.md) and
+[`docs/data/research_takeaways.json`](docs/data/research_takeaways.json).
+They summarize what the public sample results actually show: class shift under
+chronological splits, neural gains on dynamics/order/alignment, harder
+retrieval/reconstruction probes, and why the next model-quality step needs
+held-out episodes.
+
+Current contributions:
+
+- manifested sliding-window features over the currently extracted modalities,
+- motion-only and current all-feature baseline models,
+- 12 end-to-end episode-level tasks,
+- lightweight neural MLP heads for the same 12 task contracts,
+- a generated four-direction research taxonomy matching the Ropedia job tracks,
+- four additional direction-extension probes with minimal and neural baselines,
+- 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 project is best read as a staged embodied-AI research study:
+
+| Layer | Current scope | Where to start |
+| --- | --- | --- |
+| 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 | The selected 128-episode Qwen3-Omni LoRA diagnostic pilot has a verified validation-aware held-out package: 96/16/16 selected episodes, 3,808 exported windows, 512 validation windows, 448 held-out test windows, and public-safe metrics/predictions. JSON validity is 87.50%, below the 98% target, so the next pass focuses on structured-output reliability and task-quality error analysis. | [`RESEARCH_ROADMAP.md`](RESEARCH_ROADMAP.md), [`FOUNDATION_MODEL_PLAN.md`](FOUNDATION_MODEL_PLAN.md), [`docs/data/omni_finetune_verified_result.json`](docs/data/omni_finetune_verified_result.json), [`results/omni_finetune/verified_public/`](results/omni_finetune/verified_public/) |
+
+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
+
+If you only have one minute, use
+[`PROJECT_STATUS.md`](PROJECT_STATUS.md) and
+[`docs/data/project_status.json`](docs/data/project_status.json).
+They give the current research state in one compact table:
+
+| Area | Current decision |
+| --- | --- |
+| 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 |
+| 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 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 | Verified diagnostic result package exists for the selected 96/16/16 episode split; current held-out metrics are weak and below the JSON-validity quality target |
+| Raw Xperience-10M data / full Qwen weights | Not redistributed |
+
+## 90-Second Research Project Path
+
+If you are reading the project cold, open these in order:
+
+| Step | Question | Primary artifacts | What should be true |
+| --- | --- | --- | --- |
+| 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? | [`docs/data/omni_finetune_verified_result.json`](docs/data/omni_finetune_verified_result.json), [`DATA_ACCESS_STATUS.md`](results/omni_finetune/DATA_ACCESS_STATUS.md), [`MULTI_EPISODE_ACCESS_STATUS.md`](results/omni_finetune/MULTI_EPISODE_ACCESS_STATUS.md) | The first held-out diagnostic pilot is verified; strong model quality remains pending because JSON validity is 87.50% and action/subtask metrics remain weak. |
+
+A compact reader-path summary is available at
+[`docs/data/project_packet.json`](docs/data/project_packet.json).
+
+## Supporting Files
+
+[`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.
+
+[`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
+
+[`EVALUATION_PROTOCOL.md`](EVALUATION_PROTOCOL.md) and
+[`docs/data/evaluation_protocol.json`](docs/data/evaluation_protocol.json) are
+generated from committed metric artifacts. They define:
+
+- the 20-frame window unit, stride, feature dimension, and raw-data policy,
+- the chronological 70/30 single-episode split and its generalization limit,
+- the per-task input, target, primary metric, minimal score, and neural score,
+- leakage controls for future labels, target-side signals, caption/object
+  labels, and train-only normalization,
+- current limitations, including cross-episode generalization,
+  audio-visual learning, pixel-depth reconstruction, and real held-out
+  multi-episode Qwen3-Omni quality.
+
+## Dataset Context
+
+The official [`ropedia-ai/xperience-10m`](https://huggingface.co/datasets/ropedia-ai/xperience-10m)
+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.
+
+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 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,
+- `annotation.hdf5` carrying depth, SLAM/camera pose, hand/body mocap, IMU,
+  language/caption annotations, calibration, metadata, and timing records,
+- an 8,546-dimensional baseline representation using video, audio, depth,
+  pose/SLAM, mocap, IMU, calibration, and language-derived signals.
+
+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 task-suite results come from the public
+sample, and the first multi-episode Qwen3-Omni diagnostic pilot is verified but
+not yet strong model quality.
+
+Start with the visual dashboard:
+
+**[chaoyue0307.github.io/ropedia-xperience-10m-task-suite](https://chaoyue0307.github.io/ropedia-xperience-10m-task-suite/)**
+
+Hugging Face Space app:
+
+**[cy0307-ropedia-xperience-10m-task-suite.static.hf.space](https://cy0307-ropedia-xperience-10m-task-suite.static.hf.space/)**
+
+## Read This Project In Three Layers
+
+| Layer | What to inspect | Why it matters |
+| --- | --- | --- |
+| 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 |
+| 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 work, larger model branches, and the future native-pretraining goal |
+| Additional development directions | `ADDITIONAL_DEVELOPMENT_DIRECTIONS.md`, `docs/data/additional_development_directions.json` | Records concrete non-backbone tracks: taxonomy, benchmark protocol, representation learning, skill graphs, affordances, 3D/4D memory, QA, and policy transfer |
+| 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 |
+| 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 |
+
+## Links
+
+| Resource | Link |
+| --- | --- |
+| This GitHub repo | [github.com/ChaoYue0307/ropedia-xperience-10m-task-suite](https://github.com/ChaoYue0307/ropedia-xperience-10m-task-suite) |
+| This project website | [chaoyue0307.github.io/ropedia-xperience-10m-task-suite](https://chaoyue0307.github.io/ropedia-xperience-10m-task-suite/) |
+| This Hugging Face Space | [huggingface.co/spaces/cy0307/ropedia-xperience-10m-task-suite](https://huggingface.co/spaces/cy0307/ropedia-xperience-10m-task-suite) |
+| Live Hugging Face static app | [cy0307-ropedia-xperience-10m-task-suite.static.hf.space](https://cy0307-ropedia-xperience-10m-task-suite.static.hf.space/) |
+| GitHub Container package | [ghcr.io/chaoyue0307/ropedia-xperience-10m-task-suite](https://github.com/ChaoYue0307/ropedia-xperience-10m-task-suite/pkgs/container/ropedia-xperience-10m-task-suite) |
+| Derived artifacts on Hugging Face | [huggingface.co/datasets/cy0307/ropedia-xperience-10m-task-suite-artifacts](https://huggingface.co/datasets/cy0307/ropedia-xperience-10m-task-suite-artifacts) |
+| Minimal and neural task baselines on Hugging Face | [huggingface.co/cy0307/ropedia-xperience-10m-task-baselines](https://huggingface.co/cy0307/ropedia-xperience-10m-task-baselines) |
+| Hugging Face collection | [huggingface.co/collections/cy0307/ropedia-xperience-10m-task-suite](https://huggingface.co/collections/cy0307/ropedia-xperience-10m-task-suite) |
+| Xperience-10M dataset website | [ropedia.com/dataset](https://ropedia.com/dataset) |
+| Xperience-10M release page | [ropedia.com/blog/20260316_xperience_10m](https://ropedia.com/blog/20260316_xperience_10m) |
+| Ropedia GitHub organization | [github.com/Ropedia](https://github.com/Ropedia) |
+| HOMIE Toolkit | [github.com/Ropedia/HOMIE-toolkit](https://github.com/Ropedia/HOMIE-toolkit) |
+| Xperience-10M Hugging Face dataset | [huggingface.co/datasets/ropedia-ai/xperience-10m](https://huggingface.co/datasets/ropedia-ai/xperience-10m) |
+| Xperience-10M sample on Hugging Face | [huggingface.co/datasets/ropedia-ai/xperience-10m-sample](https://huggingface.co/datasets/ropedia-ai/xperience-10m-sample) |
+| Ropedia Hugging Face organization | [huggingface.co/ropedia-ai](https://huggingface.co/ropedia-ai) |
+
+## Citation, License, And Metadata
+
+Use [`CITATION.cff`](CITATION.cff) when citing this project. The repository
+also includes [`codemeta.json`](codemeta.json) for machine-readable software
+metadata and [`docs/data/project_manifest.json`](docs/data/project_manifest.json)
+for website/Hugging Face surface metadata.
+
+The code files are MIT-licensed. Raw Xperience-10M data is not redistributed
+here, and dataset use remains governed by the official Ropedia/Xperience-10M
+terms. See [`LICENSE`](LICENSE) and [`DATA_NOTICE.md`](DATA_NOTICE.md).
+
+![Ropedia Xperience-10M 12-task infographic](docs/assets/task_suite_infographic.png?v=xperience10m-taskfirst-v13-modality-xl)
+
+The infographic uses a custom text-free research background and puts the shared
+processing contract plus all 12 task families before the modality atlas.
+Public-sample modality thumbnails remain enlarged below the task map. The task
+names, input/output summaries, and metrics are overlaid from
+[`results/episode_task_suite/summary_report.json`](results/episode_task_suite/summary_report.json)
+with [`scripts/render_task_suite_infographic.py`](scripts/render_task_suite_infographic.py),
+so the published PNG is a presentation graphic with verified labels and metrics,
+not a hallucinated metric sheet.
+
+The website also includes a responsive native modality atlas backed by
+[`docs/data/modality_atlas.json`](docs/data/modality_atlas.json) and
+[`docs/assets/modalities/`](docs/assets/modalities/). Those assets are small
+derived thumbnails from the public sample, not raw Xperience-10M files.
+
+![Verified Pipeline](docs/assets/pipeline_diagram.png?v=xperience10m-nn)
+
+![Qwen3-Omni LoRA training pipeline](docs/assets/qwen3_omni_lora_pipeline.png?v=qwen3-lora-v1)
+
+![Minimal and neural 12-task model architectures](docs/assets/task_architectures.png?v=xperience10m-nn)
+
+The pipeline and architecture figures use the same pattern: text-free visual
+backgrounds carry the composition, while
+[`scripts/render_overview_figures.py`](scripts/render_overview_figures.py)
+overlays exact labels, dimensions, and metrics from the committed result files.
+
+## Scope
+
+This is a learning, inspection, and pipeline-validation repo built from one
+public sample episode. The next model-quality stage is to run the same suite
+over many episodes and split train/test by held-out episode.
+
+## What Is Inside
+
+```text
+scripts/
+  train_min_action_model.py         # motion/IMU baseline
+  train_all_modalities_model.py     # current all-feature lightweight baseline
+  episode_task_suite.py             # 12 end-to-end task definitions
+  neural_task_models.py             # optional PyTorch MLP heads for all 12 tasks
+  research_direction_taxonomy.py    # maps 12 tasks to the four research tracks
+  research_direction_extension_tasks.py # one extra data-backed probe per track
+  task_walkthroughs.py              # human-readable task-card and walkthrough-storyboard metadata
+  generate_visualizations.py        # refreshes SVG charts + summary JSON
+  render_task_suite_infographic.py  # renders the task-suite presentation PNG
+  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 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          # 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, and images
+  validate_publication_package.py   # checks public repo + HF bundle contents
+  publish_hf_bundles.py             # uploads prepared HF Space/artifact/model bundles
+  omni/
+    download_sample_modelscope.py   # ModelScope sample download helper
+    build_episode_manifest.py       # metadata-only multi-episode scanner
+    plan_finetune_sample_budget.py  # storage/sample-count planner
+    qwen3_omni_adapter_smoke.py     # real-data Qwen3-Omni adapter setup check
+
+results/
+  min_action_model/                 # motion-only action baseline artifacts
+  min_subtask_model/                # motion-only subtask baseline artifacts
+  min_all_modalities_action_model/  # current all-feature action artifacts
+  min_all_modalities_subtask_model/ # current all-feature subtask artifacts
+  episode_task_suite/               # 12-task suite metrics and predictions
+    neural_mlp/                     # optional neural baseline artifacts per task
+    research_directions/            # four-track taxonomy, CSV, and summary
+    research_direction_extensions/  # four extra direction probes + predictions
+    task_walkthroughs/              # case-study walkthroughs for all 12 tasks
+  omni_exploration/                 # ModelScope readiness-check artifacts
+
+docs/
+  index.html                        # GitHub Pages dashboard
+  data/additional_development_directions.json # concrete non-backbone project directions
+  data/summary_metrics.json         # website-readable metrics bundle
+  data/evidence_contract.json       # machine-readable project scope
+  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/task_surface_integrity.json  # machine-readable task-card/storyboard integrity check
+  data/project_manifest.json        # machine-readable public-surface metadata
+  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
+  data/task_walkthroughs.json       # human-readable task-card and walkthrough-storyboard data
+  data/modality_atlas.json          # responsive modality-card data
+  assets/brand/*.png                # project logo, favicon, social card
+  assets/task_suite_infographic.png # 12-task presentation graphic
+  assets/modalities/                # public-sample derived modality thumbnails
+  assets/pipeline_diagram.png       # verified episode pipeline graphic
+  assets/qwen3_omni_lora_pipeline.png # Qwen3-Omni LoRA training-flow figure
+  assets/task_architectures.png     # verified 12-task minimal architecture map
+  assets/charts/*.svg               # regenerated visualizations
+
+notes/
+  min_action_model.md
+  all_modalities_model.md
+  episode_task_suite.md
+```
+
+Raw Xperience-10M data is **not** committed. Download it from the official
+Ropedia distribution and follow the dataset terms.
+
+## GitHub Package
+
+The public dashboard is packaged as a static-site container on GitHub Container
+Registry. It contains the `docs/` site plus the main reader documents; it does
+not include raw Xperience-10M videos, raw annotations, gated data, or model
+weights.
+
+```bash
+docker pull ghcr.io/chaoyue0307/ropedia-xperience-10m-task-suite:latest
+docker run --rm -p 8080:80 ghcr.io/chaoyue0307/ropedia-xperience-10m-task-suite:latest
+```
+
+Then open `http://localhost:8080`.
+
+## Data Expected
+
+The scripts expect a workspace with the Ropedia HOMIE toolkit and the
+Xperience-10M sample episode:
+
+```text
+<workspace>/
+  HOMIE-toolkit/
+  data/sample/xperience-10m-sample/
+    annotation.hdf5
+    fisheye_cam0.mp4
+    fisheye_cam1.mp4
+    fisheye_cam2.mp4
+    fisheye_cam3.mp4
+    stereo_left.mp4
+    stereo_right.mp4
+```
+
+The public sample dataset identifier is:
+
+```text
+ropedia-ai/xperience-10m-sample
+```
+
+Hugging Face URL:
+
+```text
+https://huggingface.co/datasets/ropedia-ai/xperience-10m-sample
+```
+
+## Quickstart
+
+From a workspace folder:
+
+```bash
+git clone https://github.com/Ropedia/HOMIE-toolkit.git
+python3.12 -m venv .venv
+source .venv/bin/activate
+pip install -r HOMIE-toolkit/requirements.txt huggingface_hub hf_xet
+```
+
+Download the sample:
+
+```bash
+hf download ropedia-ai/xperience-10m-sample \
+  --repo-type dataset \
+  --local-dir data/sample/xperience-10m-sample
+```
+
+If Hugging Face access is unavailable in your environment, use ModelScope:
+
+```bash
+python scripts/omni/download_sample_modelscope.py \
+  --output-dir data/sample/xperience-10m-sample \
+  --mode minimal
+```
+
+`--mode minimal` downloads `annotation.hdf5`, `README.md`, and
+`fisheye_cam0.mp4`. Use `--mode all-training` to add all six MP4 streams while
+still skipping `visualization.rrd`.
+
+Clone and run this repo:
+
+```bash
+git clone https://github.com/ChaoYue0307/ropedia-xperience-10m-task-suite.git
+cd ropedia-xperience-10m-task-suite
+python scripts/episode_task_suite.py --workspace /path/to/workspace
+```
+
+Run the same 12-task suite with lightweight neural heads:
+
+```bash
+pip install torch
+python scripts/episode_task_suite.py \
+  --workspace /path/to/workspace \
+  --include-neural
+```
+
+Run the smaller baselines:
+
+```bash
+python scripts/train_min_action_model.py --workspace /path/to/workspace
+python scripts/train_all_modalities_model.py --workspace /path/to/workspace
+```
+
+## Xperience-10M Fine-Tuning Exploration
+
+This repo includes a first Qwen3-Omni fine-tuning path over Xperience-10M. The
+repository separates public-sample evidence from multi-episode fine-tuning
+artifacts. The validation-aware selected-episode held-out package is now verified as a
+diagnostic pilot, not a strong final model.
+The useful distinction is:
+
+- direct Qwen3-Omni inputs: RGB/fisheye video, embedded MP4 audio, and language
+  prompts,
+- adapter-required Xperience-10M sensor inputs: depth, pose/SLAM, hand/body
+  mocap, contacts, and IMU.
+
+![Xperience-10M to Qwen3-Omni LoRA training flow](docs/assets/qwen3_omni_lora_pipeline.png?v=qwen3-lora-v1)
+
+The figure shows the intended end-to-end training flow: raw valid episodes enter
+episode-level split validation, parallel media/sensor export creates Qwen-style
+JSONL records, Qwen3-Omni receives video/audio/text directly, the sensor bridge
+adds depth/pose/mocap/IMU features, LoRA adapters are trained on prepared
+train/val episodes, and sealed held-out test evaluation produces predictions,
+metrics, run reports, and upload-ready adapter artifacts.
+
+The scale-up path requires valid prepared episodes, held-out episode splits,
+training metadata, predictions, metrics, and a run report. A result is ready
+for public README, website, or Hugging Face updates only after the validator
+passes and `scripts/omni/package_verified_omni_result.py` creates a
+public-safe derived-artifact package. The current verified package is listed in
+[`docs/data/omni_finetune_verified_result.json`](docs/data/omni_finetune_verified_result.json).
+
+### Sample Count Decision
+
+Do not treat "10M" as a reason to start with the entire dataset. The engineering
+unit that matters first is diverse held-out episodes, not adjacent windows from
+one session.
+
+| Phase | Episodes/samples | Approx windows at stride 5 | Purpose |
+| --- | ---: | ---: | --- |
+| Readiness | 1-3 | 1k-3k | Verify loaders, token alignment, and task heads |
+| Pilot | 16-32 | 18k-37k | First held-out-episode evaluation |
+| Useful LoRA run | 64-128 | 74k-149k | Train sensor adapters plus selected Qwen3-Omni LoRA |
+| Storage-heavy run | 256+ | 297k+ | Only after download layout and checkpoint size are stable |
+
+Use the budget helper before downloading:
+
+```bash
+python scripts/omni/plan_finetune_sample_budget.py \
+  --storage-root /path/to/storage \
+  --target-free-after-download-gb 800 \
+  --all-training-per-episode-gb 2.4 \
+  --full-preview-per-episode-gb 5.1
+```
+
+### Multi-Episode Readiness Gate
+
+```bash
+python scripts/omni/discover_xperience10m_sources.py \
+  --workspace /path/to/ropedia-xperience-10m-task-suite \
+  --data-root /path/to/xperience10m_data \
+  --output results/omni_finetune/source_discovery.json
+```
+
+Current status in this repo:
+
+- public_sample_valid_episodes: 1 (degraded-valid: annotation + fisheye_cam0.mp4)
+- gated_metadata_audit: 12,102 complete visible episodes across 802 complete sessions
+- selected_episode_plan: 128 source-balanced episodes, 96/16/16 train/val/test
+- selected_download_size: 277.71 GiB excluding `visualization.rrd`
+- verified_validation_aware_diagnostic_package: true
+- selected_split: 96 train / 16 validation / 16 held-out test episodes
+- exported_windows: 2,848 train / 512 validation / 448 test
+- validation_samples_used: 512
+- held_out_eval: 448 test windows from 14 exported test episodes
+- train_loss / val_loss: 0.4130 / 0.0331
+- current_quality_target: JSON validity 87.50%, below the 98% target
+- gated dataset: available for selected multi-episode data preparation
+- source_discovery: `results/omni_finetune/source_discovery.json`
+- data_status: `results/omni_finetune/DATA_ACCESS_STATUS.md`
+- access_status: `results/omni_finetune/MULTI_EPISODE_ACCESS_STATUS.md`
+
+Use this gate before scheduling any full fine-tune run. The pilot should use
+balanced held-out selection, not the first paths in repository order. The
+current 128-episode selection filters for complete leaf episodes, excludes
+`visualization.rrd`, balances episode-size bands, and preserves one selected
+episode per top-level session UUID.
+
+### Progressive Train/Validation Pilot
+
+The selected 128-episode plan can be used before every episode has arrived by
+training only on prepared `train` episodes and monitoring prepared `val` episodes.
+The final `test` episodes stay sealed until the end, so early development does
+not contaminate held-out evaluation.
+
+```bash
+python scripts/omni/build_selection_episode_manifest.py \
+  --workspace /path/to/ropedia-xperience-10m-task-suite \
+  --data-root /path/to/xperience10m_128 \
+  --selection-json results/omni_finetune/xperience10m_128_episode_selection.json \
+  --output results/omni_finetune/trainval_progressive/episode_manifest_trainval.json \
+  --include-split train \
+  --include-split val
+```
+
+`scripts/omni/run_trainval_progressive_128.sh` wraps the same guard, exports a
+train/val-only Qwen3-Omni JSONL dataset, and launches LoRA training without
+running final test evaluation. The exporter uses session-qualified episode IDs
+and path-based split matching so repeated folder names such as `ep1` cannot
+collide across different sessions.
+
+For larger prepared subsets, `scripts/omni/run_trainval_parallel_export_8gpu.sh`
+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 episodes while optionally monitoring
+validation loss, 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 \
+TRAIN_VAL_SPLIT=val \
+MAX_VAL_SAMPLES=512 \
+scripts/omni/run_128_fullsplit_parallel_export_8gpu.sh
+```
+
+The current verified diagnostic package uses the same selected split and 8-GPU
+training path, records validation loss over 512 validation windows, and keeps
+the held-out test split sealed for final evaluation. The next pass should keep
+this package contract while tightening JSON decoding, target formatting, and
+action/subtask error analysis.
+
+Monitor the run with:
+
+```bash
+python scripts/omni/monitor_omni_progress.py \
+  --run-id xperience10m_qwen3_omni_128ep_fullsplit_fast8gpu
+```
+
+The monitor reads training `progress.jsonl`, new evaluator partial-prediction
+progress, and legacy generation logs, so long held-out evals can still expose
+sample-level progress even before final metrics are written.
+
+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 the eval validator passes, create the public-safe result package:
+
+```bash
+python scripts/omni/package_verified_omni_result.py \
+  --dataset-run-id xperience10m_qwen3_omni_128ep_fullsplit_fast8gpu \
+  --train-run-id <train_run_id> \
+  --eval-run-id <eval_run_id>
+```
+
+For long-running remote jobs, the packaging step can be watched automatically:
+
+```bash
+python scripts/omni/watch_verified_omni_package.py \
+  --dataset-run-id xperience10m_qwen3_omni_128ep_fullsplit_fast8gpu \
+  --train-run-id <train_run_id> \
+  --eval-run-id <eval_run_id>
+```
+
+While waiting, the watcher can append `eval_progress_observed` events from
+partial prediction files or legacy generation logs. This keeps the package
+status file useful during long held-out evaluations.
+
+The package copies only small derived artifacts such as metrics, predictions,
+confusion matrices, run reports, manifests, validation summaries, and training
+metadata. The exact required eval files and primary metrics come from the
+selected backbone contract in `configs/omni_backbones`, so Qwen3-Omni,
+Cosmos-style world models, and VLA/policy branches can share the same verified
+publication gate once their model-specific evaluators exist. The package
+excludes raw Xperience-10M files, base-model weights, adapter or checkpoint
+weights, full checkpoints, and large archives.
+
+For hardware setups that can run multiple eval workers, the Qwen evaluator also
+supports deterministic sample shards:
+
+```bash
+python scripts/omni/eval_qwen3_omni_lora.py \
+  --dataset-jsonl results/omni_finetune/xperience10m_qwen3_omni_128ep_fullsplit_fast8gpu_dataset/dataset.jsonl \
+  --adapter-dir checkpoints/<train_run_id>/adapter_lora \
+  --run-id <eval_shard_0> \
+  --eval-split test \
+  --sample-offset 0 \
+  --sample-stride 4
+
+python scripts/omni/merge_qwen3_omni_eval_shards.py \
+  --dataset-jsonl results/omni_finetune/xperience10m_qwen3_omni_128ep_fullsplit_fast8gpu_dataset/dataset.jsonl \
+  --output-dir results/omni_finetune/<merged_eval_run_id> \
+  --shard-dir results/omni_finetune/<eval_shard_0> \
+  --shard-dir results/omni_finetune/<eval_shard_1> \
+  --shard-dir results/omni_finetune/<eval_shard_2> \
+  --shard-dir results/omni_finetune/<eval_shard_3>
+```
+
+Only the merged eval directory should be validated and reported publicly,
+because the merger checks coverage and recomputes the metrics from all
+held-out predictions.
+
+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 Qwen3-Omni LoRA artifacts
+
+The public-safe verified package intentionally excludes raw data, base Qwen
+weights, LoRA weights, and full checkpoints. Adapter upload is a separate step:
+use it only when the intended adapter directory is present and the model card
+clearly distinguishes older smoke weights from the selected-episode diagnostic
+or validation-aware run.
+
+```bash
+python3 scripts/omni/upload_qwen3_omni_lora_to_hf.py \
+  --repo-id cy0307/ropedia-qwen3-omni-lora-smoke \
+  --source-dir /path/to/adapter_upload_package \
+  --message "Upload Xperience-10M Qwen3-Omni LoRA pilot"
+```
+
+This script requires a valid Hugging Face token via `HF_TOKEN` or `--token`.
+Network availability to `huggingface.co` is required.
+
+### Foundation Backbone Plan
+
+The next modeling plan tracks several foundation-model branches instead of
+assuming one backbone solves every Xperience-10M objective.
+
+| Branch | Current role | When to use it |
+| --- | --- | --- |
+| Qwen3-Omni | First trainable multimodal LoRA pilot | Use for the selected 128-episode held-out baseline over video/audio/language plus sensor-bridge features. |
+| Cosmos 3 | First world-model/action-generation branch | Use after data preparation for future-window prediction, action-conditioned world modeling, and synthetic-data usefulness tests. |
+| 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. 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
+```
+
+Verify that every configured backbone can pass the public-safe packaging
+contract on synthetic derived artifacts:
+
+```bash
+python scripts/omni/smoke_test_backbone_packaging.py
+```
+
+After a real held-out package is created, audit it before updating README,
+website, or Hugging Face pages:
+
+```bash
+python scripts/omni/audit_verified_omni_package.py \
+  --package-dir results/omni_finetune/verified_public/<eval_run_id>
+```
+
+Create a new planned backbone branch from an existing contract template with:
+
+```bash
+python scripts/omni/scaffold_omni_backbone.py \
+  --template-backbone policy_vla_branch \
+  --id new_policy_branch \
+  --display-name "New Policy Branch" \
+  --model-family "Model family name" \
+  --dataset-contract xperience10m_observation_action_v1 \
+  --training-objective observation_to_action_policy \
+  --checkpoint-gate policy_checkpoint_action_space_and_normalizer \
+  --dry-run
+```
+
+Each backbone config declares the checkpoint gate, required train/eval files,
+allowed public artifacts, and forbidden private or heavyweight artifacts. This
+keeps Qwen3-Omni, Cosmos-style world models, and policy/VLA branches on the same
+split, validation, and publication discipline even though their training targets
+are different.
+
+## Additional Development Directions
+
+Beyond backbone selection and fine-tuning, Xperience-10M supports several
+concrete research-development tracks:
+
+| Direction | First useful artifact | Role in the project |
+| --- | --- | --- |
+| Episode taxonomy and data engine | Episode atlas, balance report, and split builder | Select representative data before training. |
+| Standardized benchmark protocol | Versioned train/val/test manifests and metric scripts | Make future model results comparable. |
+| Multimodal representation learning | Contrastive and masked-window encoder objectives | Learn reusable video/audio/depth/pose/mocap/IMU/language features. |
+| Skill and procedure graph mining | Step graph, transitions, preconditions, and effects | Connect perception to planning and long-horizon reasoning. |
+| Human-object affordance modeling | Contact, reachable-object, tool-use, and next-affordance tasks | Model what actions the scene makes possible. |
+| 3D/4D scene and object memory | Persistent scene/object maps from depth, pose, multiview video, and objects | Track world state beyond single frames. |
+| Data-quality and synchronization diagnostics | Per-episode QA for drift, missing streams, calibration, and corrupted files | Keep large multimodal training trustworthy. |
+| Policy, retargeting, and simulation transfer | Action-token conversion and robot-compatible imitation examples | Bridge human egocentric experience to robot policy work. |
+
+See [`ADDITIONAL_DEVELOPMENT_DIRECTIONS.md`](ADDITIONAL_DEVELOPMENT_DIRECTIONS.md)
+and [`docs/data/additional_development_directions.json`](docs/data/additional_development_directions.json).
+
+## Four Research Directions
+
+The 12 tasks are now organized against the four Ropedia research directions in
+a generated artifact, not only in prose:
+
+- [`research_direction_taxonomy.json`](results/episode_task_suite/research_directions/research_direction_taxonomy.json)
+- [`research_direction_task_map.csv`](results/episode_task_suite/research_directions/research_direction_task_map.csv)
+- [`research_direction_summary.md`](results/episode_task_suite/research_directions/research_direction_summary.md)
+- [`docs/data/research_directions.json`](docs/data/research_directions.json)
+
+The taxonomy uses two current baselines for every task:
+
+| Baseline | Role |
+| --- | --- |
+| Minimal interpretable heads | Softmax, logistic, ridge, and retrieval heads over the 8,546-dimensional multimodal representation. These expose the input/output contract cleanly. |
+| Neural MLP heads | Small PyTorch MLP classifiers/regressors on the same features and splits. These check whether nonlinear heads help before moving to Qwen/Omni fine-tuning. |
+
+Current direction-level coverage:
+
+| Direction | Current status | Covered task evidence | What is not solved yet |
+| --- | --- | --- | --- |
+| A. Human Modeling & Motion Understanding | Partially implemented | Hand Trajectory Forecasting and Contact State Prediction are direct; Action Recognition and Object Relevance Prediction are proxies. Neural MLP improves hand forecasting from `0.8647` to `0.1079` MPJPE. | No full body/shape model, SMPL/MANO target, deformation prior, or multi-episode motion-generation evaluation yet. |
+| B. 3D/4D Reconstruction & Neural Rendering | Proxy tasks only | Cross-Modal Retrieval, Cross-Modal Reconstruction, and Multimodal Synchronization Detection test alignment/reconstruction prerequisites. | No NeRF, Gaussian Splatting, TSDF, mesh, novel-view synthesis, or calibrated 4D reconstruction model yet. |
+| C. Egocentric Vision & Interaction | Strongest implemented track | 6 direct tasks: action, subtask, transition, next-action, object relevance, and caption grounding, plus alignment/order diagnostics and audio ablation. | Single-episode chronological split limits generalization; stronger audio and video-language backbones still need multi-episode testing. |
+| D. Scene Reconstruction & World Modeling | Early proxy tasks | Procedure Step Recognition, Next-Action Prediction, Object Relevance Prediction, Cross-Modal Retrieval, Cross-Modal Reconstruction, Temporal Order Verification, and Multimodal Synchronization Detection provide state/world-model probes. | No persistent scene graph, object permanence task, long-term map, or held-out-episode world model yet. |
+
+The important interpretation is that all four directions can be **started** from
+the Xperience-10M sample modalities, but only direction C is strongly represented
+by the current 12-task suite. Directions A, B, and D need additional targets and
+multi-episode training before they become full research deliverables.
+
+## Four Direction-Extension Probes
+
+Beyond the original 12 core tasks, the repo now includes one extra data-backed
+probe for each research direction. These probes are computed from the same
+`shared_windows.npz`, `windows.csv`, and `feature_manifest.json` artifacts, so
+the reported numbers are computed from sample-derived features and saved metric artifacts.
+
+- [`research_direction_extension_results.json`](results/episode_task_suite/research_direction_extensions/research_direction_extension_results.json)
+- [`research_direction_extension_summary.md`](results/episode_task_suite/research_direction_extensions/research_direction_extension_summary.md)
+- [`docs/data/research_direction_extensions.json`](docs/data/research_direction_extensions.json)
+- [`research_direction_extension_tasks.svg`](docs/assets/charts/research_direction_extension_tasks.svg)
+
+![Four direction extension probes](docs/assets/charts/research_direction_extension_tasks.svg)
+
+| Direction | New extension task | Input | Output | Minimal | Neural MLP | Why it matters |
+| --- | --- | --- | --- | ---: | ---: | --- |
+| A. Human Modeling & Motion Understanding | Body and Hand Motion Intensity | non-mocap video/depth/pose/IMU/SLAM/language features | high vs low body/hand motion | `0.7827` macro-F1 | `0.7986` macro-F1 | Starts a human-motion-energy target without leaking mocap input. |
+| B. 3D/4D Reconstruction & Neural Rendering | Multi-View Consistency Retrieval | fisheye camera feature query | synchronized stereo-left view rank | `0.5534` MRR | `0.3469` MRR | Tests whether multi-view features preserve synchronized 4D scene identity. |
+| C. Egocentric Vision & Interaction | Action Phase Progress Estimation | non-caption multimodal window | progress inside current action segment | `0.3416` MAE | `0.3038` MAE | Adds a task-structure/intent-style target beyond class labels. |
+| D. Scene Reconstruction & World Modeling | Short-Horizon Ego-Motion Forecasting | current sensors excluding camera translation and captions | future camera-translation delta | `0.1989` MAE | `0.0989` MAE | Starts a short-horizon world-model target over wearer motion. |
+
+Run:
+
+```bash
+python scripts/research_direction_extension_tasks.py
+```
+
+These four probes make the four-direction mapping more concrete, but they are
+still single-episode extension baselines. Full research conclusions still require
+multi-episode training, held-out episode evaluation, and stronger task-specific
+models.
+
+## Task Walkthroughs For Juniors
+
+Every task now has a beginner-facing explanation with:
+
+- a concrete coffee-episode case study,
+- exact input contract,
+- middle process modules,
+- output contract,
+- minimal and neural metric,
+- one important limitation.
+
+Primary files:
+
+- [`TASK_WALKTHROUGHS.md`](results/episode_task_suite/task_walkthroughs/TASK_WALKTHROUGHS.md)
+- [`task_walkthroughs.json`](results/episode_task_suite/task_walkthroughs/task_walkthroughs.json)
+- [`docs/data/task_walkthroughs.json`](docs/data/task_walkthroughs.json)
+- [`docs/data/task_surface_integrity.json`](docs/data/task_surface_integrity.json)
+
+Compact map:
+
+| Task | Case study | Input -> process -> output |
+| --- | --- | --- |
+| Action Recognition | A pouring window should be named as the current action. | all-modality window -> action label builder + classifier -> action class |
+| Procedure Step Recognition | A fine action is grouped into a broader drink-preparation stage. | all-modality window -> subtask label builder + classifier -> subtask label |
+| Action Boundary Detection | Detect the change from preparing to pouring. | window -> boundary builder + binary classifier -> boundary/steady |
+| Next-Action Prediction | A preparing window predicts what happens 20 frames later. | current window -> future-label shift + classifier -> next action |
+| Hand Trajectory Forecasting | A hand moving toward a cup becomes a future 3D hand path. | current window -> future mocap target + regressor -> hand trajectory |
+| Contact State Prediction | Decide whether hand/body contact is happening. | non-contact features -> contact target + binary classifier -> contact label |
+| Object Relevance Prediction | Infer milk, cup, coffee, or related objects during pouring. | non-caption features -> multi-hot object target + sigmoid heads -> object set |
+| Language Grounding | Query Pour milk into coffee and retrieve the matching moment. | text-like query + candidates -> projection + cosine ranker -> ranked windows |
+| Cross-Modal Retrieval | Motion/IMU from pouring retrieves matching depth/video. | motion/IMU/camera -> projection + candidate index -> ranked depth/video windows |
+| Cross-Modal Reconstruction | Infer depth/video features from motion, IMU, and camera pose. | source modalities -> scaler + regressor -> target modality vector |
+| Temporal Order Verification | Tell whether reaching then pouring was reversed. | adjacent window pair -> pair combiner + binary classifier -> correct/reversed |
+| Multimodal Synchronization Detection | Catch motion paired with visual/depth features shifted in time. | motion side + visual side -> aligned/shifted pair builder + classifier -> aligned/shifted |
+
+## Minimal 12-Task Architectures
+
+These are deliberately minimal baselines. They are useful because every
+input/output contract is explicit, not because they are strong embodied-AI
+models.
+
+Shared setup:
+
+```text
+raw episode -> 20-frame windows, stride 5 -> 8,546-dimensional multimodal representation
+chronological split: first 70% train, last 30% test
+scalers are fit on train windows only
+```
+
+There are four reusable head families:
+
+| Head family | Used by | What it means |
+| --- | --- | --- |
+| Linear softmax classifier | Action Recognition, Procedure Step Recognition, Action Boundary Detection, Next-Action Prediction, Contact State Prediction, Temporal Order Verification, Multimodal Synchronization Detection | z-score features, then `XW+b`, softmax, cross-entropy, L2 |
+| Dual ridge regression/projection | Hand Trajectory Forecasting, Cross-Modal Reconstruction | z-score input/target, solve ridge regression with L2=10 |
+| Ridge + cosine ranking | Language Grounding, Cross-Modal Retrieval | project one modality into another feature space, then rank candidates by cosine |
+| Multi-label logistic regression | Object Relevance Prediction | z-score non-caption features, sigmoid object heads, threshold at 0.5 |
+
+The optional neural run keeps the same window representation, leakage filters,
+chronological splits, and metrics, but replaces the task heads with small
+PyTorch MLP classifiers or regressors. Its outputs live under
+[`results/episode_task_suite/neural_mlp/`](results/episode_task_suite/neural_mlp/),
+and the rollup is stored in the `neural_tasks` section of
+[`results/episode_task_suite/summary_report.json`](results/episode_task_suite/summary_report.json).
+
+The task-specific heads are:
+
+| Task | Input | Minimal head | Output |
+| --- | --- | --- | --- |
+| Action Recognition | all featurized modalities | linear softmax | current action class |
+| Procedure Step Recognition | all featurized modalities | linear softmax | current subtask class |
+| Action Boundary Detection | all featurized modalities | linear softmax | steady vs action boundary |
+| Next-Action Prediction | all featurized modalities at `t` | linear softmax | action at `t+20` frames |
+| Hand Trajectory Forecasting | all featurized modalities at `t` | ridge regression | future 10-frame left/right hand joints |
+| Contact State Prediction | non-contact and non-caption signals | linear softmax | any body contact |
+| Object Relevance Prediction | non-caption signals | multi-label logistic | relevant object set |
+| Language Grounding | sensor windows projected to text space | ridge projection + cosine ranking | matching time window for text query |
+| Cross-Modal Retrieval | motion/IMU/camera projected to visual space | ridge projection + cosine ranking | matching depth/video window |
+| Cross-Modal Reconstruction | motion/IMU/camera | ridge regression | compressed depth/video target |
+| Temporal Order Verification | `[x_t, x_t+1, x_t+1-x_t]` | binary linear softmax | correct vs reversed order |
+| Multimodal Synchronization Detection | motion plus visual pair | binary linear softmax | aligned vs shifted by 8 windows |
+
+## Key Results
+
+| Experiment | Main score | Accuracy | Notes |
+| --- | ---: | ---: | --- |
+| Motion-only action | 0.9688 macro-F1 | 0.9828 | Uses motion/IMU features only |
+| Current all-feature action | 0.9829 macro-F1 | 0.9863 | 8,546-dimensional multimodal representation |
+| Motion-only subtask | 0.9528 macro-F1 | 0.9759 | Strong within-episode subtask signal |
+| Current all-feature subtask | 0.9173 macro-F1 | 0.9828 | High accuracy, lower class-balanced score |
+| Cross-modal retrieval | 0.3678 top-5 | n/a | Motion/IMU/camera/audio retrieves matching depth/video |
+| Transition detection | 0.6118 macro-F1 | 0.9080 | Boundary F1 is 0.1250 |
+| Hand trajectory forecast | 0.8647 MPJPE | n/a | Predicts future hand-joint trajectory |
+| Neural MLP hand forecast | 0.1079 MPJPE | n/a | Same features/split, nonlinear regression head |
+| Neural MLP temporal order | 0.8520 F1 | 0.8578 | Strong improvement on adjacent-window ordering |
+| Neural MLP misalignment | 0.7153 F1 | 0.7009 | Detects shifted motion/visual/audio pairs better than the linear head |
+| Audio ablation | +0.0418 mean delta | n/a | Current audio variant improves the primary metric on 6 of 12 task contracts |
+| Alternate audio representation | +0.0936 mean delta | n/a | Alternate audio-window representation improves over the baseline audio variant on 6 of 12 task contracts |
+
+## Audio Contribution Study
+
+The audio ablation keeps the same windows and task labels, then compares input
+variants under the same chronological split. The script
+[`scripts/audio_ablation_and_raw_upgrade.py`](scripts/audio_ablation_and_raw_upgrade.py)
+reuses the real task-suite windows and evaluates six variants for
+every task: current inputs, no audio, audio-only, alternate audio-only, audio
+representation replacement, and all inputs plus the alternate audio representation.
+
+The measured single-episode result is task-specific:
+
+| Readout | Value |
+| --- | ---: |
+| Tasks where current audio improves the primary metric | 6 / 12 |
+| Mean current-audio delta | +0.0418 |
+| Tasks where alternate audio representation improves over baseline audio | 6 / 12 |
+| Mean alternate-representation delta vs baseline audio | +0.0936 |
+
+Full files:
+
+- [`results/audio_ablation/AUDIO_ABLATION_SUMMARY.md`](results/audio_ablation/AUDIO_ABLATION_SUMMARY.md)
+- [`results/audio_ablation/audio_ablation_metrics.csv`](results/audio_ablation/audio_ablation_metrics.csv)
+- [`results/audio_ablation/audio_delta_summary.csv`](results/audio_ablation/audio_delta_summary.csv)
+- [`docs/data/audio_ablation_summary.json`](docs/data/audio_ablation_summary.json)
+- [`docs/assets/charts/audio_ablation_delta.svg`](docs/assets/charts/audio_ablation_delta.svg)
+
+## Neural MLP Results
+
+The neural baseline was run locally with `--include-neural` for all 12 tasks
+using 80 epochs, hidden size 128, batch size 128, and CPU execution. It is not a
+foundation model result; it is a controlled nonlinear-head comparison over the
+same 8,546-dimensional multimodal representation.
+
+| Task | Neural metric | Minimal metric | Readout |
+| --- | ---: | ---: | --- |
+| Action Recognition | 0.0148 macro-F1 | 0.0500 macro-F1 | Still blocked by unseen future classes |
+| Procedure Step Recognition | 0.0281 macro-F1 | 0.0506 macro-F1 | Same single-episode split limitation |
+| Action Boundary Detection | 0.5862 macro-F1 | 0.6118 macro-F1 | Similar to the linear baseline |
+| Next-Action Prediction | 0.0419 macro-F1 | 0.0593 macro-F1 | Same unseen-label issue |
+| Hand Trajectory Forecasting | 0.1079 MPJPE | 0.8647 MPJPE | Neural regression improves this target |
+| Contact State Prediction | 1.0000 macro-F1 | 1.0000 macro-F1 | Degenerate one-class sample |
+| Object Relevance Prediction | 0.1679 micro-F1 | 0.1803 micro-F1 | Similar weak object signal |
+| Language Grounding | 0.0168 MRR | 0.0160 MRR | Similar ranking behavior |
+| Cross-Modal Retrieval | 0.1300 MRR | 0.2693 MRR | Linear ridge remains stronger here |
+| Cross-Modal Reconstruction | -0.0102 R2 | -0.0153 R2 | Small improvement but still weak |
+| Temporal Order Verification | 0.8520 F1 | 0.5400 F1 | Neural head captures local temporal structure |
+| Multimodal Synchronization Detection | 0.7153 F1 | 0.5052 F1 | Neural head improves alignment detection |
+
+The strongest single-episode self-supervised signal is cross-modal retrieval:
+motion/IMU/camera/audio features retrieve matching depth/video windows substantially
+better than random.
+
+## Single-Episode Diagnostics and Explorer
+
+While waiting for broader Xperience-10M access, the repo now includes an
+artifact-driven diagnostics pass over the public sample episode:
+
+- `results/single_episode_diagnostics/object_labels/window_object_labels.csv`
+  exports 1,161 real window-level object-label sets from `annotation.hdf5`.
+- `results/single_episode_diagnostics/modality_ablation/ablation_metrics.csv`
+  recomputes all 96 task/modality cells, including object relevance.
+- `results/single_episode_diagnostics/timeline_overlay/timeline_overlay.csv`
+  aligns 2,079 existing prediction rows back to the episode timeline.
+- `results/single_episode_diagnostics/alignment_stress/alignment_shift_metrics.csv`
+  evaluates cross-modal retrieval under explicit time shifts.
+- `docs/single_episode_explorer.html` is a static interactive page for
+  inspecting window labels, objects, predictions, modality statistics, and
+  diagnostic scores.
+
+These are single-episode research diagnostics. They are useful for studying
+task definitions, feature behavior, and model errors before scaling to more
+episodes; they are not reported as multi-episode benchmark results.
+
+## Reproducibility Check
+
+I re-ran the full pipeline from the local raw public sample into a temporary
+local workspace and compared regenerated metrics with the committed
+artifacts. The baseline metrics, 12 task metrics, feature manifest, and
+available modality manifest matched exactly after float normalization.
+
+See [`notes/reproducibility_audit.md`](notes/reproducibility_audit.md) for the
+commands and verification evidence.
+
+## Why Some Scores Are Low
+
+The task suite intentionally uses a chronological split:
+
+```text
+first 70% of the episode -> train
+last 30% of the episode  -> test
+```
+
+The test segment contains some action/subtask labels never seen during training.
+Timeline and next-action classifiers therefore expose the core limitation of
+single-episode learning instead of hiding it behind random splits.
+
+## Modalities Used
+
+The current public-sample pipeline uses:
+
+- hand/body mocap joints and contact labels,
+- camera translation and rotation,
+- IMU acceleration and gyroscope traces,
+- depth confidence features,
+- six video streams,
+- audio from the sample MP4 stream,
+- caption/object/interaction text features,
+- SLAM point-cloud summary features,
+- calibration parameters.
+
+The full technical source manifest is stored in
+[`results/episode_task_suite/feature_manifest.json`](results/episode_task_suite/feature_manifest.json).
+
+## Data Notice
 
-The official full dataset is
-[`ropedia-ai/xperience-10m`](https://huggingface.co/datasets/ropedia-ai/xperience-10m),
-and the public sample source is
-[`ropedia-ai/xperience-10m-sample`](https://huggingface.co/datasets/ropedia-ai/xperience-10m-sample).
-This artifact repo contains derived research outputs from the sample episode and
-selected-pilot preparation files. Raw Xperience-10M videos, annotations, RRD
-visualizations, gated data, and full Qwen weights are not redistributed.
-
-For readers who need source provenance, the detailed dataset-card alignment
-notes are included in `XPERIENCE10M_DATASET_CARD_ALIGNMENT.md` and
-`SOURCE_ALIGNMENT_AUDIT.md`. Most readers can start with the dashboard,
-task results, and related model repositories above.
-
-## Future Foundation-Model Goal
-
-The included `XPERIENCE_EMBODIED_FOUNDATION_MODEL_PRETRAINING.md` describes a
-future **Xperience Embodied Foundation Model**: a domain-specific model
-pretrained from scratch over full-corpus Xperience-10M video, audio, depth,
-pose/SLAM, mocap, IMU, and language streams after smaller multi-episode pilots
-prove value and the required storage/compute path exists. This is a research
-roadmap item, not a completed artifact in this dataset repo.
+Xperience-10M data belongs to its original authors and is subject to the
+official Ropedia dataset license and access terms. This repo contains code and
+derived experiment artifacts only; it does not redistribute the raw videos or
+raw annotation dataset.

RESEARCH_ROADMAP.md

@@ -10,26 +10,29 @@ should exist before the stage is treated as complete.
 | Stage | Status | Entry condition | Research deliverables | Completion evidence |
 | --- | --- | --- | --- | --- |
 | Public-Sample Task Lab | Implemented | One public Xperience-10M sample episode is available. | 1,161 aligned windows, 12 task contracts, minimal heads, neural MLP heads, modality atlas, task walkthroughs, and derived figures. | `PROJECT_STATUS.md`, `EVALUATION_PROTOCOL.md`, `RESEARCH_TAKEAWAYS.md`, `docs/data/summary_metrics.json`, `results/episode_task_suite/summary_report.json` |
-| Multi-Episode Data Preparation | Active | Gated dataset availability and enough storage for selected episodes. | 128 selected episodes, episode manifest, missing-view manifest, held-out episode split, and source-discovery report. | `results/omni_finetune/DATA_ACCESS_STATUS.md`, `results/omni_finetune/MULTI_EPISODE_ACCESS_STATUS.md`, `results/omni_finetune/source_discovery.json` |
-| Qwen3-Omni LoRA Pilot | Next | Selected episodes prepared locally with no train/test episode leakage. | Dataset JSONL/media manifests, LoRA adapter checkpoint, progress logs, held-out predictions, metrics, confusion matrices, and run report. | `dataset_manifest.json`, `training_metadata.json`, `progress.jsonl`, `metrics.json`, `predictions.jsonl`, `RUN_REPORT.md` |
-| 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. |
+| Multi-Episode Data Preparation | Implemented for first selected pilot | Gated dataset availability and enough storage for selected episodes. | 128 selected episodes, episode manifest, missing-view manifest, held-out episode split, and source-discovery report. | `results/omni_finetune/DATA_ACCESS_STATUS.md`, `results/omni_finetune/MULTI_EPISODE_ACCESS_STATUS.md`, `results/omni_finetune/xperience10m_128_episode_selection.json` |
+| Qwen3-Omni LoRA Validation-Aware Diagnostic Pilot | Verified baseline | Selected episodes prepared locally with no train/test episode leakage. | Dataset JSONL/media manifests, LoRA adapter checkpoint, progress logs, validation monitoring, held-out predictions, metrics, confusion matrices, and run report. | `docs/data/omni_finetune_verified_result.json`, `results/omni_finetune/verified_public/`, `metrics.json`, `predictions.jsonl`, `RUN_REPORT.md` |
+| Structured-Output And Error-Analysis Pass | Active next step | The validation-aware diagnostic package exists and shows weak held-out quality. | Same 96/16/16 split, stricter JSON decoding or target formatting, action/subtask error analysis, held-out test evaluation, and comparison to the verified validation-aware baseline. | Updated quality-target report, error-analysis tables, held-out metrics, and verified public package. |
+| Foundation-Model Selection Matrix | Current | 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 validation-aware 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
 
-The useful next decision is data scale plus backbone fit: keep the public-sample
-task suite as the development harness, stage enough official Xperience-10M
-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. 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.
+The useful next decision is model-quality improvement plus backbone fit: keep
+the public-sample task suite as the development harness, use the verified
+Qwen3-Omni validation-aware diagnostic pilot as the first cross-episode
+baseline, then improve format reliability and task quality before claiming
+model quality.
+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. 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. The first multi-episode pilot is enough to verify the
+end-to-end training loop, but its weak metrics are not final model quality.
 
 ## Additional Concrete Development Directions
 
@@ -38,7 +41,7 @@ depend on immediately training a larger foundation model:
 
 | Direction | First artifact | Research value |
 | --- | --- | --- |
-| Episode taxonomy and data engine | Episode atlas, category tags, balance report, and split builder. | Makes episode selection representative and auditable. |
+| Episode taxonomy and data engine | Episode atlas, category tags, balance report, and split builder. | Makes episode selection representative and measurable. |
 | Standardized benchmark protocol | Fixed splits, task cards, metric scripts, and leakage checks. | Makes future model comparisons fair. |
 | Multimodal representation learning | Contrastive and masked-window objectives over synchronized modalities. | Learns reusable encoders before expensive large-model training. |
 | Skill and procedure graph mining | Steps, transitions, preconditions, effects, and temporal skill graphs. | Connects perception to planning and long-horizon reasoning. |
@@ -72,7 +75,8 @@ Evidence to inspect:
 This stage expands the same data contract to official gated episodes. The key
 research requirement is episode-level separation: training and test examples
 must come from different episodes, not different windows inside the same
-episode.
+episode. The first selected 96/16/16 split has been used for a verified
+Qwen3-Omni diagnostic pilot.
 
 Evidence to inspect:
 
@@ -84,8 +88,11 @@ Evidence to inspect:
 ### 3. Qwen3-Omni LoRA Pilot
 
 This stage uses Qwen3-Omni as the multimodal backbone and trains lightweight
-LoRA adapters. The first target is a complete held-out-episode training and
-evaluation loop with inspectable manifests, predictions, and metrics.
+LoRA adapters. The first held-out diagnostic package now exists. It proves the
+export, training, evaluation, validation, and public-safe packaging loop, but
+the metrics are weak: JSON validity is 87.50%, action macro-F1 is 0.0027, and
+subtask accuracy is 0.0067. Treat it as a baseline and error-analysis starting
+point.
 
 Expected outputs:
 

RESEARCH_TAKEAWAYS.md

@@ -97,17 +97,18 @@ Current scope: This is a single-episode ablation over fixed ridge heads. It vali
 
 ### The next scientific unit is held-out episodes, not more adjacent windows
 
-The prepared Qwen3-Omni path now targets a selected 128-episode pilot; held-out metrics will be reported after staging, training, and evaluation complete.
+The selected Qwen3-Omni path now has a verified validation-aware held-out diagnostic pilot. It proves the cross-episode train/validation/eval loop, but the weak metrics show that structured-output reliability and task-quality error analysis are the next modeling problems.
 
 | Metric | Value |
 | --- | ---: |
-| `target_episodes` | 128 |
-| `selected_sessions` | 128 |
-| `valid_candidates` | 12,102 |
+| `selected_episodes` | 128 |
+| `held_out_test_windows` | 448 |
+| `json_validity_rate` | 0.8750 |
+| `action_macro_f1` | 0.0027 |
 
-Source: `results/omni_finetune/MULTI_EPISODE_ACCESS_STATUS.md`.
+Source: `docs/data/omni_finetune_verified_result.json`.
 
-Current scope: The selected-episode Qwen3-Omni fine-tune requires completed data preparation and held-out evaluation; the 32-episode Qwen3-Omni fine-tune requires gated data preparation before any real held-out metric is reported.
+Current scope: The selected-episode Qwen3-Omni validation-aware diagnostic pilot is verified, but held-out quality is still weak and JSON validity remains below the 98% target.
 
 ## How To Read These Results
 

docs/data/evaluation_protocol.json

@@ -2,7 +2,7 @@
   "title": "Ropedia Xperience-10M Task Suite Evaluation Protocol",
   "status": "pass",
   "version": "2026-06-01",
-  "generated_at_utc": "2026-06-04T16:42:13+00:00",
+  "generated_at_utc": "2026-06-06T13:49:32+00:00",
   "source_files": [
     "docs/data/summary_metrics.json",
     "results/episode_task_suite/summary_report.json",
@@ -303,22 +303,23 @@
     "Report unseen test classes when the chronological split exposes labels absent from the train segment."
   ],
   "current_limitations": [
-    "Cross-episode generalization is evaluated in the later multi-episode stage.",
+    "Cross-episode generalization for Qwen3-Omni has a first verified diagnostic pilot, but strong model quality is not yet shown.",
     "Feature-vector reconstruction is separate from pixel depth, mesh, NeRF, or Gaussian reconstruction.",
-    "Qwen3-Omni setup artifacts are preparation artifacts until the selected held-out pilot runs.",
+    "The verified validation-aware Qwen3-Omni diagnostic pilot has weak held-out metrics and needs structured-output and task-quality improvements before larger model-quality claims.",
     "Full audio-visual representation learning still needs multi-episode training; the current report includes single-episode audio/no-audio ablations."
   ],
   "scale_up_gate": {
-    "required_before_full_omni_pilot": [
+    "required_before_next_omni_quality_pilot": [
       "selected prepared Xperience-10M episodes",
       "held-out episode split with no train/test episode leakage",
+      "validation samples during training",
       "manifest, training metadata, progress logs, metrics, predictions, and run report",
       "held-out evaluation on test episodes rather than train windows"
     ],
-    "current_status": "prepared; selected data relay in progress",
+    "current_status": "verified diagnostic pilot; quality target not met",
     "evidence": [
-      "results/omni_finetune/DATA_ACCESS_STATUS.md",
-      "results/omni_finetune/MULTI_EPISODE_ACCESS_STATUS.md"
+      "docs/data/omni_finetune_verified_result.json",
+      "results/omni_finetune/verified_public/"
     ]
   }
 }

docs/data/foundation_model_plan.json

@@ -1,7 +1,7 @@
 {
   "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.",
+  "current_boundary": "A first held-out multi-episode Qwen3-Omni diagnostic pilot is verified in this repo, but it is not a strong model result. The current foundation-model work should treat it as the baseline train/eval/package loop before validation-aware Qwen reruns, Cosmos-style world modeling, or policy/VLA branches.",
   "backbone_registry": {
     "config_dir": "configs/omni_backbones",
     "validator": "scripts/omni/backbone_registry.py --validate --json",
@@ -206,7 +206,7 @@
     {
       "step": 2,
       "name": "First held-out baseline",
-      "action": "Run Qwen3-Omni LoRA to establish the full train/eval loop."
+      "action": "Run validation-aware Qwen3-Omni LoRA to improve the verified diagnostic baseline."
     },
     {
       "step": 3,

docs/data/omni_finetune_verified_result.json

@@ -0,0 +1,78 @@
+{
+  "title": "Verified Qwen3-Omni LoRA Validation-Aware Held-Out Pilot",
+  "status": "verified_validation_aware_diagnostic_pilot",
+  "status_date": "2026-06-06",
+  "backbone": "Qwen/Qwen3-Omni-30B-A3B-Instruct",
+  "adapter": "Qwen3-Omni LoRA",
+  "dataset": "Ropedia Xperience-10M selected 128-episode pilot",
+  "split_policy": {
+    "unit": "episode",
+    "selected_episode_counts": {
+      "train": 96,
+      "val": 16,
+      "test": 16
+    },
+    "exported_window_counts": {
+      "train": 2848,
+      "val": 512,
+      "test": 448
+    },
+    "exported_episode_counts": {
+      "train": 89,
+      "val": 16,
+      "test": 14
+    },
+    "skipped_selected_episodes": 9,
+    "leakage_policy": "Train, validation, and test are separated by episode/session; test windows are used only for held-out evaluation."
+  },
+  "training": {
+    "num_processes": 8,
+    "epochs": 1,
+    "lora_rank": 16,
+    "lora_alpha": 32,
+    "lora_dropout": 0.05,
+    "num_train_samples": 2848,
+    "num_val_samples": 512,
+    "history": [
+      {
+        "epoch": 1,
+        "train_loss": 0.41304643672440994,
+        "val_loss": 0.0330660454928875,
+        "global_step": 356
+      }
+    ],
+    "loss": "answer-token cross entropy over supervised JSON tokens",
+    "note": "This validation-aware run uses the selected validation split during training and preserves the held-out test split for final evaluation."
+  },
+  "evaluation": {
+    "split": "test",
+    "num_samples": 448,
+    "held_out_episode_count": 14,
+    "json_validity_rate": 0.875,
+    "action_macro_f1": 0.0026621494447581404,
+    "subtask_accuracy": 0.006696428571428571,
+    "transition_accuracy": 0.8504464285714286,
+    "next_action_accuracy": 0.024553571428571428,
+    "contact_accuracy": 0.6450892857142857,
+    "object_micro_f1": 0.22299431459254582,
+    "quality_target": {
+      "json_validity_rate": 0.98,
+      "status": "not_met"
+    },
+    "previous_diagnostic_json_validity_rate": 0.8526785714285714
+  },
+  "interpretation": "This is a real held-out multi-episode validation-aware diagnostic pilot proving the export, LoRA training with validation monitoring, evaluation, validation, and public-safe packaging loop. JSON validity improved over the earlier no-validation diagnostic run, but task-quality metrics remain weak, so it should be used as a baseline and error-analysis starting point rather than a strong Xperience-10M model.",
+  "public_package": {
+    "path": "results/omni_finetune/verified_public/xperience10m_qwen3_omni_128ep_96train_16val_16test_valmon_20260605_eval",
+    "audit_status": "pass",
+    "contains_raw_xperience10m_data": false,
+    "contains_qwen_base_weights": false,
+    "contains_lora_weights": false
+  },
+  "required_next_steps": [
+    "Improve JSON-format reliability through prompt, decoding, constrained parsing, or target formatting changes.",
+    "Add error analysis by episode, action family, object category, and missing-modality state.",
+    "Run a second validation-aware Qwen3-Omni pass only after the JSON/output contract is tightened.",
+    "Keep the same verified package contract for Cosmos-style world-model and VLA/policy branches."
+  ]
+}

docs/data/project_status.json

@@ -1,8 +1,8 @@
 {
   "title": "Ropedia Xperience-10M Task Suite Project Status",
   "version": "2026-06-01",
-  "decision": "public_sample_pipeline_verified_multi_episode_omni_data_staging",
-  "research_positioning": "A research-engineering study that makes one public Xperience-10M sample episode inspectable, defines embodied-AI tasks over synchronized modalities, records baseline behavior, and keeps later multi-episode model-quality claims separate from current single-episode evidence.",
+  "decision": "public_sample_pipeline_verified_qwen3_omni_validation_aware_diagnostic_pilot",
+  "research_positioning": "A research-engineering study that makes one public Xperience-10M sample episode inspectable, defines embodied-AI tasks over synchronized modalities, records baseline behavior, and uses the selected-episode Qwen3-Omni validation-aware diagnostic pilot as a verified but weak cross-episode baseline.",
   "scope_boundary": {
     "validated_episode_count": 1,
     "aligned_frames": 5821,
@@ -13,7 +13,20 @@
     "direction_extension_probe_count": 4,
     "audio_featurized": true,
     "raw_xperience10m_data_redistributed": false,
-    "qwen3_omni_32_episode_claim": false
+    "qwen3_omni_32_episode_claim": false,
+    "qwen3_omni_verified_diagnostic_pilot": true,
+    "qwen3_omni_selected_episode_counts": {
+      "train": 96,
+      "val": 16,
+      "test": 16
+    },
+    "qwen3_omni_exported_window_counts": {
+      "train": 2848,
+      "val": 512,
+      "test": 448
+    },
+    "qwen3_omni_json_validity_rate": 0.875,
+    "qwen3_omni_validation_aware": true
   },
   "rows": [
     {
@@ -36,92 +49,92 @@
       ],
       "readout": "All 12 task contracts have committed metrics, predictions, and minimal baseline outputs."
     },
-        {
-            "area": "Neural heads",
-            "status": "verified",
-            "evidence": [
-                "scripts/neural_task_models.py",
-                "results/episode_task_suite/neural_mlp/"
-            ],
-            "readout": "Each task also has a compact PyTorch MLP run over the same feature tensor and chronological split."
-        },
-        {
-            "area": "Audio contribution study",
-            "status": "verified",
-            "evidence": [
-                "scripts/audio_ablation_and_raw_upgrade.py",
-                "results/audio_ablation/",
-                "docs/data/audio_ablation_summary.json"
-            ],
-            "readout": "Audio variants improve the primary metric on 6 of 12 task contracts in this single-episode setting."
-        },
-        {
-            "area": "Evaluation protocol",
-            "status": "verified",
-            "evidence": [
-                "EVALUATION_PROTOCOL.md",
-                "docs/data/evaluation_protocol.json",
-                "scripts/build_evaluation_protocol.py"
-            ],
-            "readout": "Windowing, chronological split, per-task metrics, leakage controls, and current limitations are generated from committed metric artifacts."
-        },
-        {
-            "area": "Research takeaways",
-            "status": "verified",
-            "evidence": [
-                "RESEARCH_TAKEAWAYS.md",
-                "docs/data/research_takeaways.json",
-                "scripts/build_research_takeaways.py"
-            ],
-            "readout": "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."
-        },
-        {
-            "area": "Research roadmap",
-            "status": "current",
-            "evidence": [
-                "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, world/policy branches, and the future Xperience-native pretraining goal."
-        },
-        {
-            "area": "Foundation-model plan",
-            "status": "current",
-            "evidence": [
-                "FOUNDATION_MODEL_PLAN.md",
-                "docs/data/foundation_model_plan.json"
-            ],
-            "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",
+    {
+      "area": "Neural heads",
+      "status": "verified",
+      "evidence": [
+        "scripts/neural_task_models.py",
+        "results/episode_task_suite/neural_mlp/"
+      ],
+      "readout": "Each task also has a compact PyTorch MLP run over the same feature tensor and chronological split."
+    },
+    {
+      "area": "Audio contribution study",
+      "status": "verified",
+      "evidence": [
+        "scripts/audio_ablation_and_raw_upgrade.py",
+        "results/audio_ablation/",
+        "docs/data/audio_ablation_summary.json"
+      ],
+      "readout": "Audio variants improve the primary metric on 6 of 12 task contracts in this single-episode setting."
+    },
+    {
+      "area": "Evaluation protocol",
+      "status": "verified",
+      "evidence": [
+        "EVALUATION_PROTOCOL.md",
+        "docs/data/evaluation_protocol.json",
+        "scripts/build_evaluation_protocol.py"
+      ],
+      "readout": "Windowing, chronological split, per-task metrics, leakage controls, and current limitations are generated from committed metric artifacts."
+    },
+    {
+      "area": "Research takeaways",
+      "status": "verified",
+      "evidence": [
+        "RESEARCH_TAKEAWAYS.md",
+        "docs/data/research_takeaways.json",
+        "scripts/build_research_takeaways.py"
+      ],
+      "readout": "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."
+    },
+    {
+      "area": "Research roadmap",
+      "status": "current",
+      "evidence": [
+        "RESEARCH_ROADMAP.md",
+        "docs/data/research_roadmap.json"
+      ],
+      "readout": "The roadmap connects public-sample task development to the verified Qwen3-Omni diagnostic pilot, validation-aware diagnostics, foundation-model selection, robustness runs, world/policy branches, and the future Xperience-native pretraining goal."
+    },
+    {
+      "area": "Foundation-model plan",
+      "status": "current",
+      "evidence": [
+        "FOUNDATION_MODEL_PLAN.md",
+        "docs/data/foundation_model_plan.json"
+      ],
+      "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",
       "evidence": [
         "XPERIENCE10M_DATASET_CARD_ALIGNMENT.md",
         "docs/data/xperience10m_dataset_card_alignment.json"
       ],
-            "readout": "Public wording is aligned to the official gated Xperience-10M dataset card, public sample card, and HF API metadata, including modalities, scale, access path, sample license/tooling, and current project coverage."
-        },
-        {
-            "area": "Source alignment",
-            "status": "verified",
-            "evidence": [
-                "SOURCE_ALIGNMENT_AUDIT.md",
-                "docs/data/source_alignment_audit.json",
-                "scripts/validate_source_alignment.py"
-            ],
-            "readout": "Source facts, sample details, API-listing notes, and project coverage are checked across repo docs, website, and HF cards."
-        },
-        {
-            "area": "Website and HF mirrors",
+      "readout": "Public wording is aligned to the official gated Xperience-10M dataset card, public sample card, and HF API metadata, including modalities, scale, access path, sample license/tooling, and current project coverage."
+    },
+    {
+      "area": "Source alignment",
+      "status": "verified",
+      "evidence": [
+        "SOURCE_ALIGNMENT_AUDIT.md",
+        "docs/data/source_alignment_audit.json",
+        "scripts/validate_source_alignment.py"
+      ],
+      "readout": "Source facts, sample details, API-listing notes, and project coverage are checked across repo docs, website, and HF cards."
+    },
+    {
+      "area": "Website and HF mirrors",
       "status": "verified",
       "evidence": [
         "docs/data/website_integrity.json",
@@ -152,12 +165,14 @@
     },
     {
       "area": "Qwen3-Omni fine-tuning",
-      "status": "data_preparation_full_metrics_pending",
+      "status": "verified_validation_aware_diagnostic_pilot_quality_target_not_met",
       "evidence": [
-        "results/omni_finetune/DATA_ACCESS_STATUS.md",
-        "results/omni_finetune/MULTI_EPISODE_ACCESS_STATUS.md"
+        "docs/data/omni_finetune_verified_result.json",
+        "results/omni_finetune/verified_public/",
+        "scripts/omni/package_verified_omni_result.py",
+        "scripts/omni/audit_verified_omni_package.py"
       ],
-      "readout": "The gated full dataset is available for a selected 128-episode pilot; final held-out metrics require completed preprocessing, manifest construction, training, and held-out evaluation."
+      "readout": "The selected 96/16/16 episode split produced a validation-aware public-safe held-out package with 3,808 exported windows, 512 validation windows, and 448 test predictions. JSON validity is 87.50%, below the 98% target, so it is a stronger diagnostic baseline but not a strong model-quality result."
     },
     {
       "area": "Raw Xperience-10M redistribution",
@@ -171,21 +186,21 @@
   ],
   "fast_research_route": [
     "Read PROJECT_STATUS.md and EVIDENCE_CONTRACT.md to establish what is implemented.",
-        "Open docs/data/project_packet.json for the machine-readable project path.",
-        "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.",
-        "Inspect SOURCE_ALIGNMENT_AUDIT.md before judging source-card consistency across public surfaces.",
-        "Inspect XPERIENCE10M_DATASET_CARD_ALIGNMENT.md before judging dataset wording.",
-    "Inspect results/omni_finetune/DATA_ACCESS_STATUS.md before judging Qwen3-Omni scale-up status."
+    "Open docs/data/project_packet.json for the machine-readable project path.",
+    "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.",
+    "Inspect SOURCE_ALIGNMENT_AUDIT.md before judging source-card consistency across public surfaces.",
+    "Inspect XPERIENCE10M_DATASET_CARD_ALIGNMENT.md before judging dataset wording.",
+    "Inspect docs/data/omni_finetune_verified_result.json before judging the Qwen3-Omni diagnostic pilot."
   ],
   "current_reading_notes": [
-    "Cross-episode generalization is evaluated in the later multi-episode stage.",
-    "Older pilot path names refer to setup files, not completed held-out training results.",
+    "The validation-aware Qwen3-Omni diagnostic pilot is verified, but current held-out quality is still weak.",
+    "Use docs/data/omni_finetune_verified_result.json and the latest verified_public validation-aware package for current held-out results.",
     "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/.",

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, 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.",
+  "summary": "Staged path from the public-sample task lab to a verified validation-aware Qwen3-Omni diagnostic pilot, structured-output improvement pass, 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, use the verified selected-episode Qwen3-Omni validation-aware diagnostic pilot as the first cross-episode baseline, improve structured-output reliability and task-quality error analysis, then branch into Cosmos 3 world modeling and policy-model experiments after their targets are implemented. The Xperience Embodied Foundation Model is a later full-corpus pretraining goal, not a current result.",
   "additional_development_directions": {
     "source_document": "ADDITIONAL_DEVELOPMENT_DIRECTIONS.md",
     "source_json": "docs/data/additional_development_directions.json",
@@ -33,8 +33,8 @@
     },
     {
       "id": "multi_episode_data_staging",
-      "name": "Multi-Episode Data Staging",
-      "status": "active",
+      "name": "Multi-Episode Data Preparation",
+      "status": "implemented_for_first_pilot",
       "entry_condition": "Gated dataset availability and enough storage for selected episodes.",
       "deliverables": [
         "128 selected episodes",
@@ -48,23 +48,26 @@
         "results/omni_finetune/MULTI_EPISODE_ACCESS_STATUS.md",
         "results/omni_finetune/source_discovery.json"
       ],
-      "reader_takeaway": "The next scale decision is data preparation, with train/test separation at the episode level."
+      "reader_takeaway": "The first selected split is available for Qwen3-Omni diagnostics, with train/test separation at the episode level."
     },
     {
-      "id": "qwen3_omni_lora_pilot",
-      "name": "Qwen3-Omni LoRA Pilot",
-      "status": "next",
+      "id": "qwen3_omni_lora_diagnostic_pilot",
+      "name": "Qwen3-Omni LoRA Validation-Aware Diagnostic Pilot",
+      "status": "verified_baseline",
       "entry_condition": "Selected episodes are prepared locally with no train/test episode leakage.",
       "deliverables": [
         "dataset JSONL/media manifests",
         "LoRA adapter checkpoint",
         "progress logs",
+        "validation monitoring",
         "held-out predictions",
         "metrics",
         "confusion matrices",
         "run report"
       ],
       "completion_evidence": [
+        "docs/data/omni_finetune_verified_result.json",
+        "results/omni_finetune/verified_public/",
         "dataset_manifest.json",
         "training_metadata.json",
         "progress.jsonl",
@@ -72,7 +75,27 @@
         "predictions.jsonl",
         "RUN_REPORT.md"
       ],
-      "reader_takeaway": "The first omni-model pilot should establish a complete held-out-episode training and evaluation loop."
+      "reader_takeaway": "The first omni-model pilot establishes the full held-out training/validation/evaluation loop, but the weak metrics make it a diagnostic baseline."
+    },
+    {
+      "id": "qwen3_omni_structured_output_error_analysis",
+      "name": "Structured-Output And Error-Analysis Pass",
+      "status": "active_next_step",
+      "entry_condition": "The validation-aware diagnostic package exists and shows weak held-out quality.",
+      "deliverables": [
+        "same 96/16/16 episode split",
+        "stricter JSON decoding or target formatting",
+        "episode/action/object error analysis",
+        "held-out test evaluation",
+        "comparison to the verified validation-aware baseline"
+      ],
+      "completion_evidence": [
+        "quality-target report",
+        "error-analysis tables",
+        "held-out metrics",
+        "verified public-safe package"
+      ],
+      "reader_takeaway": "The next pass should improve output reliability and task metrics before larger model-quality claims."
     },
     {
       "id": "foundation_model_selection_matrix",

docs/data/research_roadmap_interactive.json

@@ -127,7 +127,7 @@
       "Build the episode taxonomy and data-quality diagnostics first.",
       "Lock the benchmark protocol and split manifests before reporting model scores.",
       "Add representation-learning and skill-graph objectives once enough episodes are staged.",
-      "Add affordance, 3D/4D memory, and policy-retargeting branches after labels and action targets are auditable."
+      "Add affordance, 3D/4D memory, and policy-retargeting branches after labels and action targets are measurable."
     ],
     "public_boundary": "These are proposed development tracks. They are not reported as completed held-out benchmark results.",
     "source_document": "ADDITIONAL_DEVELOPMENT_DIRECTIONS.md",
@@ -2035,7 +2035,7 @@
         "step": 1
       },
       {
-        "action": "Run Qwen3-Omni LoRA to establish the full train/eval loop.",
+        "action": "Run validation-aware Qwen3-Omni LoRA to improve the verified diagnostic baseline.",
         "name": "First held-out baseline",
         "step": 2
       },
@@ -2222,7 +2222,7 @@
     ],
     "status": "planning_artifact"
   },
-  "generated_at_utc": "2026-06-04T21:22:15+00:00",
+  "generated_at_utc": "2026-06-06T13:49:32+00:00",
   "omni_plan": {
     "adapter": "LoRA rank 16, alpha 32, dropout 0.05",
     "backbone": "Qwen/Qwen3-Omni-30B-A3B-Instruct",
@@ -2279,13 +2279,15 @@
       ],
       "entry_condition": "Gated dataset availability and enough storage for selected episodes.",
       "id": "multi_episode_data_staging",
-      "name": "Multi-Episode Data Staging",
-      "reader_takeaway": "The next scale decision is data preparation, with train/test separation at the episode level.",
-      "stage": "scale_up",
-      "status": "active"
+      "name": "Multi-Episode Data Preparation",
+      "reader_takeaway": "The first selected split is available for Qwen3-Omni diagnostics, with train/test separation at the episode level.",
+      "stage": "future",
+      "status": "implemented_for_first_pilot"
     },
     {
       "completion_evidence": [
+        "docs/data/omni_finetune_verified_result.json",
+        "results/omni_finetune/verified_public/",
         "dataset_manifest.json",
         "training_metadata.json",
         "progress.jsonl",
@@ -2297,17 +2299,39 @@
         "dataset JSONL/media manifests",
         "LoRA adapter checkpoint",
         "progress logs",
+        "validation monitoring",
         "held-out predictions",
         "metrics",
         "confusion matrices",
         "run report"
       ],
       "entry_condition": "Selected episodes are prepared locally with no train/test episode leakage.",
-      "id": "qwen3_omni_lora_pilot",
-      "name": "Qwen3-Omni LoRA Pilot",
-      "reader_takeaway": "The first omni-model pilot should establish a complete held-out-episode training and evaluation loop.",
-      "stage": "omni",
-      "status": "next"
+      "id": "qwen3_omni_lora_diagnostic_pilot",
+      "name": "Qwen3-Omni LoRA Validation-Aware Diagnostic Pilot",
+      "reader_takeaway": "The first omni-model pilot establishes the full held-out training/validation/evaluation loop, but the weak metrics make it a diagnostic baseline.",
+      "stage": "future",
+      "status": "verified_baseline"
+    },
+    {
+      "completion_evidence": [
+        "quality-target report",
+        "error-analysis tables",
+        "held-out metrics",
+        "verified public-safe package"
+      ],
+      "deliverables": [
+        "same 96/16/16 episode split",
+        "stricter JSON decoding or target formatting",
+        "episode/action/object error analysis",
+        "held-out test evaluation",
+        "comparison to the verified validation-aware baseline"
+      ],
+      "entry_condition": "The validation-aware diagnostic package exists and shows weak held-out quality.",
+      "id": "qwen3_omni_structured_output_error_analysis",
+      "name": "Structured-Output And Error-Analysis Pass",
+      "reader_takeaway": "The next pass should improve output reliability and task metrics before larger model-quality claims.",
+      "stage": "future",
+      "status": "active_next_step"
     },
     {
       "completion_evidence": [
@@ -2404,7 +2428,7 @@
       "visualization.rrd"
     ],
     "selection_strategy": "stratified_round_robin_by_top_level_session",
-    "status": "selected_episode_preparation",
+    "status": "verified_validation_aware_diagnostic_pilot",
     "target_episodes": 128,
     "valid_candidates": 12102
   },

docs/data/research_takeaways.json

@@ -1,7 +1,7 @@
 {
   "title": "Ropedia Xperience-10M Research Takeaways",
   "status": "pass",
-  "generated_at_utc": "2026-06-04T16:42:13+00:00",
+  "generated_at_utc": "2026-06-06T13:49:32+00:00",
   "source_files": [
     "docs/data/summary_metrics.json",
     "results/episode_task_suite/summary_report.json",
@@ -166,23 +166,27 @@
     {
       "id": "scale_requires_episodes",
       "title": "The next scientific unit is held-out episodes, not more adjacent windows",
-      "readout": "The prepared Qwen3-Omni path now targets a selected 128-episode pilot; held-out metrics will be reported after staging, training, and evaluation complete.",
+      "readout": "The selected Qwen3-Omni path now has a verified validation-aware held-out diagnostic pilot. It proves the cross-episode train/validation/eval loop, but the weak metrics show that structured-output reliability and task-quality error analysis are the next modeling problems.",
       "evidence": [
         {
-          "label": "target_episodes",
+          "label": "selected_episodes",
           "value": 128
         },
         {
-          "label": "selected_sessions",
-          "value": 128
+          "label": "held_out_test_windows",
+          "value": 448
+        },
+        {
+          "label": "json_validity_rate",
+          "value": 0.875
         },
         {
-          "label": "valid_candidates",
-          "value": 12102
+          "label": "action_macro_f1",
+          "value": 0.0026621494447581404
         }
       ],
-      "source": "results/omni_finetune/MULTI_EPISODE_ACCESS_STATUS.md",
-      "current_scope": "The selected-episode Qwen3-Omni fine-tune requires completed data preparation and held-out evaluation; the 32-episode Qwen3-Omni fine-tune requires gated data preparation before any real held-out metric is reported."
+      "source": "docs/data/omni_finetune_verified_result.json",
+      "current_scope": "The selected-episode Qwen3-Omni validation-aware diagnostic pilot is verified, but held-out quality is still weak and JSON validity remains below the 98% target."
     }
   ]
 }

results/omni_finetune/verified_public/xperience10m_qwen3_omni_128ep_96train_16val_16test_valmon_20260605_eval/PUBLIC_RESULT_SUMMARY.md

@@ -0,0 +1,25 @@
+# Verified Omni Fine-Tuning Result
+
+- Backbone: `qwen3_omni_lora`
+- Dataset run: `xperience10m_qwen3_omni_128ep_96train_16val_16test_valmon_20260605`
+- Training run: `xperience10m_qwen3_omni_128ep_96train_16val_16test_valmon_20260605_lora`
+- Evaluation run: `xperience10m_qwen3_omni_128ep_96train_16val_16test_valmon_20260605_eval`
+- Validation status: `verified`
+- Held-out eval split: `test`
+- Held-out episodes: `14`
+- Prediction rows: `448`
+
+## Primary Metrics
+
+- json_validity_rate: `0.875`
+- action_macro_f1: `0.0026621494447581404`
+- subtask_accuracy: `0.006696428571428571`
+- transition_accuracy: `0.8504464285714286`
+- next_action_accuracy: `0.024553571428571428`
+- contact_accuracy: `0.6450892857142857`
+- object_micro_f1: `0.22299431459254582`
+- held_out_episode_count: `14`
+
+Raw Xperience-10M files, base-model weights, adapter or checkpoint weights, full checkpoints, and large archives are not included.
+
+Use this package as the source for README, website, and Hugging Face updates.

results/omni_finetune/verified_public/xperience10m_qwen3_omni_128ep_96train_16val_16test_valmon_20260605_eval/eval/RUN_REPORT.md

@@ -0,0 +1,13 @@
+# Qwen3-Omni LoRA Evaluation
+
+- Base model: `<workspace-parent>/modelscope_models/Qwen__Qwen3-Omni-30B-A3B-Instruct`
+- Adapter: `<project>/checkpoints/xperience10m_qwen3_omni_128ep_96train_16val_16test_valmon_20260605_lora/adapter_lora`
+- Dataset: `<project>/results/omni_finetune/xperience10m_qwen3_omni_128ep_96train_16val_16test_valmon_20260605_dataset/dataset.jsonl`
+- Eval split: `test`
+- Samples: `448`
+- Episodes: `14`
+- Accuracy: `0.0246`
+- Macro-F1: `0.0027`
+- Unseen eval labels: `144`
+
+Artifacts include `metrics.json`, `predictions.csv`, `per_class_metrics.csv`, and `confusion_matrix.csv`.

results/omni_finetune/verified_public/xperience10m_qwen3_omni_128ep_96train_16val_16test_valmon_20260605_eval/eval/metrics.json

@@ -0,0 +1,1578 @@
+{
+  "num_samples": 448,
+  "accuracy": 0.024553571428571428,
+  "macro_f1": 0.0026621494447581404,
+  "labels": [
+    "Adjust Mahjong tile",
+    "Adjust Mahjong tile alignment",
+    "Adjust Mahjong tile on the stack",
+    "Adjust Mahjong tiles",
+    "Adjust bead piles",
+    "Adjust canned food on shelf",
+    "Adjust cans in bin",
+    "Adjust cans in container",
+    "Adjust cans in tray",
+    "Adjust cardboard",
+    "Adjust cardboard divider",
+    "Adjust cardboard position",
+    "Adjust container on shelf",
+    "Adjust container position",
+    "Adjust containers on shelf",
+    "Adjust foam strip",
+    "Adjust grip",
+    "Adjust grip on container",
+    "Adjust hand position",
+    "Adjust item on shelf",
+    "Adjust lantern shape",
+    "Adjust lantern string",
+    "Adjust paper",
+    "Adjust paper strip",
+    "Adjust perspective",
+    "Adjust placement on shelf",
+    "Adjust position",
+    "Adjust pot position",
+    "Adjust puzzle piece",
+    "Adjust red button",
+    "Adjust red button in row",
+    "Adjust red button position",
+    "Adjust retail item position",
+    "Adjust retail items on shelf",
+    "Adjust ruler position",
+    "Adjust smartphone and sort pieces",
+    "Adjust snack package",
+    "Adjust tile row alignment",
+    "Adjust vacuum cleaner position",
+    "Adjusting a puzzle piece",
+    "Adjusting canned goods on shelf",
+    "Adjusting fabric for cutting",
+    "Adjusting fabric position",
+    "Adjusting puzzle piece",
+    "Align Mahjong tiles",
+    "Align and place retail item",
+    "Align blue strip",
+    "Align button",
+    "Align button in row",
+    "Align button row",
+    "Align buttons",
+    "Align canned food on shelf",
+    "Align canned goods on shelf",
+    "Align cardboard piece",
+    "Align cardboard strip",
+    "Align charging cable",
+    "Align edges of paper lantern",
+    "Align foam piece",
+    "Align foam strip",
+    "Align paper lantern edges",
+    "Align paper strip",
+    "Align plastic containers",
+    "Align red button in row",
+    "Align red buttons",
+    "Align ruler",
+    "Align ruler and mark cardboard",
+    "Align ruler on cardboard",
+    "Align ruler with crease",
+    "Align scissors",
+    "Apply adhesive tape to lantern",
+    "Approach boxes",
+    "Approach desk",
+    "Approach packing area",
+    "Approach restocking supplies",
+    "Approach table",
+    "Approach work table",
+    "Approach workstation",
+    "Approaching and pressing the door switch",
+    "Approaching the table",
+    "Approaching work table",
+    "Arrange Mahjong tiles",
+    "Arrange beads by color",
+    "Arrange black buttons",
+    "Arrange button cluster",
+    "Arrange buttons",
+    "Arrange buttons in a line",
+    "Arrange buttons in row",
+    "Arrange buttons on table",
+    "Arrange buttons on the table",
+    "Arrange canned products on shelf",
+    "Arrange cans in box",
+    "Arrange cans on shelf",
+    "Arrange cardboard",
+    "Arrange cardboard piece",
+    "Arrange cardboard pieces",
+    "Arrange cardboard squares",
+    "Arrange container on shelf",
+    "Arrange items on shelf",
+    "Arrange orange buttons",
+    "Arrange paper stars",
+    "Arrange paper strips",
+    "Arrange plastic containers",
+    "Arrange red buttons",
+    "Arrange small buttons",
+    "Arrange star beads",
+    "Arrange star beads for counting",
+    "Arrange star-shaped beads",
+    "Arrange tiles into row",
+    "Arrive at a different workstation",
+    "Assemble cardboard pieces",
+    "Assemble foam strips",
+    "Assess shelf arrangement",
+    "Attach foam strip",
+    "Attach material to paper strip",
+    "Attempt to fit puzzle piece",
+    "Begin folding paper strip",
+    "Begin rolling quilling strip",
+    "Bend and manipulate plastic strip",
+    "Browse and interact with phone interface",
+    "Browse mobile phone",
+    "Browse smartphone screen",
+    "Browsing mobile phone",
+    "Browsing smartphone content",
+    "Bundle display hooks",
+    "Cap marker",
+    "Carry cardboard piece",
+    "Carry cereal boxes",
+    "Carry cereal towards aisle",
+    "Carry container",
+    "Carry crate of cans",
+    "Carry item to shelf",
+    "Carry pasta box towards aisle",
+    "Carry plastic container",
+    "Carry stool to next shelf",
+    "Check phone",
+    "Check smart watch",
+    "Check watch",
+    "Clean shelf",
+    "Close cardboard box",
+    "Closing the door",
+    "Combine bead piles",
+    "Complete the cut",
+    "Connect cable to device",
+    "Continue cutting fabric",
+    "Continue cutting newspaper",
+    "Continue folding paper strip",
+    "Count and arrange paper stars",
+    "Count and record paper stars",
+    "Count paper stars",
+    "Counting and organizing beads",
+    "Counting star beads",
+    "Curve foam strip into loop",
+    "Cut along the edge of the newspaper",
+    "Cut along the line",
+    "Cut along the marked line",
+    "Cut along the newspaper edge",
+    "Cut cardboard",
+    "Cut cardboard along line",
+    "Cut cardboard grid",
+    "Cut cardboard into triangles",
+    "Cut cardboard pattern",
+    "Cut cardboard piece",
+    "Cut cardboard piece with scissors",
+    "Cut cardboard pieces with scissors",
+    "Cut cardboard shape",
+    "Cut cardboard sheet",
+    "Cut cardboard sheet with scissors",
+    "Cut cardboard square",
+    "Cut cardboard strip",
+    "Cut cardboard strip with scissors",
+    "Cut cardboard strip with utility knife",
+    "Cut cardboard triangle",
+    "Cut cardboard tube",
+    "Cut cardboard with scissors",
+    "Cut cardboard with utility knife",
+    "Cut fabric with scissors",
+    "Cut light green fabric",
+    "Cut newspaper",
+    "Cut newspaper with scissors",
+    "Cut out cardboard pattern",
+    "Cut section from newspaper",
+    "Cutting fabric",
+    "Deposit beads into box",
+    "Deposit cardboard squares",
+    "Discard item into bin",
+    "Discard paper towel",
+    "Draw grid line",
+    "Draw grid line with pen",
+    "Draw line",
+    "Draw line along ruler",
+    "Draw line on cardboard",
+    "Draw line with marker",
+    "Draw line with pen",
+    "Draw lines on cardboard",
+    "Draw lines with pen and ruler",
+    "Draw lines with ruler",
+    "Draw straight line",
+    "Draw straight lines on cardboard",
+    "Drawing grid line",
+    "Drawing grid line with pen and ruler",
+    "Drawing grid line with ruler",
+    "Drawing lines on cardboard",
+    "Drop cardboard square into box",
+    "Dry hands",
+    "Enter the room",
+    "Enter workspace",
+    "Entering the VR training room",
+    "Examine canned goods",
+    "Examine item",
+    "Examine labels",
+    "Examine product",
+    "Expand paper lantern",
+    "Extract wire hangers from box",
+    "Finish placing cardboard cutouts",
+    "Finish washing hands",
+    "Finish wiping and inspect jar",
+    "Finishing coil",
+    "Fold and manipulate ribbon",
+    "Fold and organize paper strips",
+    "Fold blue strip",
+    "Fold cardboard",
+    "Fold cardboard edge",
+    "Fold cardboard shape",
+    "Fold cardboard sheet",
+    "Fold cut cardboard",
+    "Fold foam piece",
+    "Fold lucky star",
+    "Fold newspaper",
+    "Fold paper lantern",
+    "Fold paper star",
+    "Fold paper strip",
+    "Fold paper strip into a star",
+    "Fold paper strip into knot",
+    "Fold paper strip into lucky star",
+    "Fold paper strip into star",
+    "Fold plastic strip",
+    "Fold purple paper",
+    "Fold purple paper strip",
+    "Fold ribbon",
+    "Folding cardboard",
+    "Folding paper strip",
+    "Forming quilled paper shape",
+    "Gather cardboard pieces",
+    "Gather pieces",
+    "Gather pieces into box",
+    "Gather star beads",
+    "Gathering colored beads",
+    "Gathering items",
+    "Gathering star beads",
+    "Gesturing",
+    "Grasp and retrieve item",
+    "Grasp cardboard sheet",
+    "Grasp cleaning bottle",
+    "Grasp door handle",
+    "Grasp electronic object",
+    "Grasp item",
+    "Grasp lantern",
+    "Grasp lantern component",
+    "Grasp next item",
+    "Grasp origami stars",
+    "Grasp package",
+    "Grasp paper strip",
+    "Grasp plastic bag on shelf",
+    "Grasp product from box",
+    "Grasp product from shelf",
+    "Grasp retail item",
+    "Grasp shopping bag",
+    "Grasp snack package",
+    "Grasping cleaning cloth",
+    "Greeting/acknowledging participants",
+    "Guide utility knife along ruler",
+    "Handle paper lantern component",
+    "Hold and align cardboard",
+    "Hold and align newspaper",
+    "Hold and align paper strip",
+    "Hold and bend paper strip",
+    "Hold and bend plastic strip",
+    "Hold and crease purple paper",
+    "Hold and examine item",
+    "Hold and inspect can",
+    "Hold and manipulate paper strip",
+    "Hold and mark cardboard piece",
+    "Hold and rotate paper strip",
+    "Hold and view phone",
+    "Hold and wipe product",
+    "Hold beads",
+    "Hold bin and move through aisle",
+    "Hold blue product box",
+    "Hold blue strip",
+    "Hold canned food",
+    "Hold cardboard",
+    "Hold cardboard piece",
+    "Hold cardboard pieces",
+    "Hold cardboard strip",
+    "Hold cardboard with ruler",
+    "Hold charger",
+    "Hold charger and cable",
+    "Hold charging cable",
+    "Hold cleaning cloth",
+    "Hold container",
+    "Hold container lid",
+    "Hold container of canned food",
+    "Hold craft tool",
+    "Hold device and cable",
+    "Hold earbud case",
+    "Hold electronic accessory",
+    "Hold electronic item",
+    "Hold empty container",
+    "Hold foam pieces",
+    "Hold instructional sign",
+    "Hold item",
+    "Hold item and adjust posture",
+    "Hold items",
+    "Hold items and inspect shelf",
+    "Hold items in hand",
+    "Hold newspaper",
+    "Hold paper lantern",
+    "Hold paper strip",
+    "Hold pen and paper",
+    "Hold phone",
+    "Hold pickle jar",
+    "Hold portable charger",
+    "Hold power adapter",
+    "Hold power bank and cable",
+    "Hold product",
+    "Hold product labels",
+    "Hold product package",
+    "Hold quilled paper coil",
+    "Hold quilled paper piece",
+    "Hold quilling paper",
+    "Hold recording sheet and pen",
+    "Hold ruler",
+    "Hold ruler and draw line",
+    "Hold ruler and mark cardboard",
+    "Hold ruler and marker",
+    "Hold ruler and pen steady",
+    "Hold ruler on cardboard",
+    "Hold ruler steady",
+    "Hold scissors",
+    "Hold small cardboard pieces",
+    "Hold small object",
+    "Hold small piece of ribbon",
+    "Hold small product bag",
+    "Hold small white box",
+    "Hold smartphone",
+    "Hold smartphone box",
+    "Hold snack package",
+    "Hold snack packages",
+    "Hold supplement bottle",
+    "Hold tray of canned goods",
+    "Hold utility knife",
+    "Hold water bottle",
+    "Holding marker",
+    "Identify next cardboard piece",
+    "Inflate paper star",
+    "Initiate star folding",
+    "Insert charging cable",
+    "Insert charging cable into power bank",
+    "Insert plug into power adapter",
+    "Inspect Dior gift box",
+    "Inspect almond package",
+    "Inspect and place item on shelf",
+    "Inspect bottle",
+    "Inspect cardboard piece",
+    "Inspect cardboard strip",
+    "Inspect charging case",
+    "Inspect electronic item",
+    "Inspect jar",
+    "Inspect product",
+    "Inspect product lid",
+    "Inspect shelf",
+    "Inspect shelf and organize stock",
+    "Inspect shelf condition",
+    "Inspect smartphone box",
+    "Inspect strip",
+    "Inspect supplement bottle",
+    "Interact with colleagues",
+    "Interact with phone",
+    "Interact with smartphone",
+    "Interact with smartphone screen",
+    "Interacting with phone screen",
+    "Interaction with coworker",
+    "Interlock paper strips",
+    "Labeling cardboard piece",
+    "Labeling cardboard square",
+    "Labeling cardboard squares",
+    "Lift blue strip",
+    "Lift pen and shift ruler",
+    "Lift pot lid",
+    "Lift utility knife",
+    "Lock phone",
+    "Look around the table",
+    "Look away",
+    "Manipulate adhesive strip",
+    "Manipulate and inspect colorful pieces",
+    "Manipulate bead",
+    "Manipulate beads",
+    "Manipulate cardboard piece",
+    "Manipulate cardboard shape",
+    "Manipulate cardboard sheet",
+    "Manipulate colorful pieces",
+    "Manipulate component",
+    "Manipulate component on strip",
+    "Manipulate craft paper strips",
+    "Manipulate craft piece",
+    "Manipulate folded paper star",
+    "Manipulate light blue strip",
+    "Manipulate material",
+    "Manipulate paper decoration",
+    "Manipulate paper edge",
+    "Manipulate paper piece",
+    "Manipulate paper quilling piece",
+    "Manipulate paper star",
+    "Manipulate paper stars",
+    "Manipulate paper strip",
+    "Manipulate paper strips",
+    "Manipulate plastic strip",
+    "Manipulate plastic strips",
+    "Manipulate power cable plug",
+    "Manipulate puzzle piece",
+    "Manipulate puzzle pieces",
+    "Manipulate quilled paper",
+    "Manipulate quilled paper shape",
+    "Manipulate quilled paper strip",
+    "Manipulate quilled paper strips",
+    "Manipulate quilling paper",
+    "Manipulate quilling strip",
+    "Manipulate ribbon knot",
+    "Manipulate ribbon piece",
+    "Manipulate small component",
+    "Manipulate small object",
+    "Manipulate small paper segment",
+    "Manipulate star",
+    "Manipulate yellow strip",
+    "Manipulating paper strips",
+    "Mark cardboard",
+    "Mark cardboard piece",
+    "Mark cardboard strip with pen",
+    "Mark cardboard with marker",
+    "Mark cardboard with pen",
+    "Mark cardboard with pen and ruler",
+    "Mark cardboard with ruler",
+    "Mark cardboard with ruler and pen",
+    "Mark fabric",
+    "Mark fabric with pen",
+    "Mark fabric with pen and ruler",
+    "Mark line on cardboard",
+    "Mark lines on cardboard",
+    "Mark lines with marker",
+    "Mark lines with pen along ruler",
+    "Mark list with pen",
+    "Mark paper list",
+    "Mark straight line",
+    "Marking cardboard piece",
+    "Marking cardboard with pen",
+    "Marking lines on cardboard",
+    "Measure and mark cardboard",
+    "Measure cardboard with ruler",
+    "Move Mahjong tile",
+    "Move along shelf",
+    "Move along the shelf",
+    "Move along the shelves",
+    "Move along the supermarket aisle",
+    "Move and place black buttons",
+    "Move away from collection box",
+    "Move away from desk",
+    "Move away from shelf",
+    "Move away from table",
+    "Move away from workstation",
+    "Move bin",
+    "Move bin to shelf area",
+    "Move black button",
+    "Move blue beads",
+    "Move box to next position",
+    "Move button to line",
+    "Move camera over surface",
+    "Move can towards shelf",
+    "Move canned goods container",
+    "Move cardboard",
+    "Move cardboard box",
+    "Move cardboard piece",
+    "Move cardboard sheet",
+    "Move cardboard to pile",
+    "Move container toward shelf",
+    "Move dustpan to side",
+    "Move hand",
+    "Move hand away",
+    "Move hand away from shelf",
+    "Move hand away from workspace",
+    "Move hand back to box",
+    "Move hand over button pile",
+    "Move hand to paper stars",
+    "Move hand toward craft materials",
+    "Move item to bag",
+    "Move marker and adjust hand",
+    "Move marker and ruler",
+    "Move marker away",
+    "Move orange buttons",
+    "Move origami stars",
+    "Move pen",
+    "Move pen aside",
+    "Move pen away",
+    "Move phone",
+    "Move piece to pile",
+    "Move pieces into box",
+    "Move pineapple chips",
+    "Move plastic storage bin",
+    "Move plush toy",
+    "Move pot",
+    "Move product to box",
+    "Move product to shelf",
+    "Move product towards shelf",
+    "Move puzzle piece",
+    "Move ruler",
+    "Move ruler and tools",
+    "Move scissors away",
+    "Move small blue foam piece towards the strip",
+    "Move smartphone",
+    "Move storage bin",
+    "Move through aisle",
+    "Move through the training room",
+    "Move to box",
+    "Move to desk",
+    "Move to next section",
+    "Move to shelf",
+    "Move to shelf base",
+    "Move to stock products",
+    "Move towards aisle",
+    "Move towards box",
+    "Move towards kitchen area",
+    "Move towards shelf",
+    "Move towards table",
+    "Move towards the stove",
+    "Move tray towards packing area",
+    "Move utility knife along ruler",
+    "Move vacuum cleaner",
+    "Move vacuum cleaner hose",
+    "Moving cardboard square",
+    "Moving hand",
+    "Moving hand towards cardboard stack",
+    "Moving ruler",
+    "Observe and pause",
+    "Observe and walk through store",
+    "Observe colleague and workspace",
+    "Observe craft layout",
+    "Observe desktop layout",
+    "Observe paper and count objects",
+    "Observe paper quilling station",
+    "Observe puzzle progress",
+    "Observe room",
+    "Observe shelf",
+    "Observe shelf status",
+    "Observe sorting progress",
+    "Observe stocking",
+    "Observe surroundings",
+    "Observe workspace",
+    "Open cardboard box",
+    "Open door",
+    "Open earbud case",
+    "Open folded paper lantern",
+    "Open paper lantern",
+    "Open paper lantern component",
+    "Open small case",
+    "Open stove pot lid",
+    "Open supplement bottle",
+    "Operate smartphone",
+    "Organize bag contents",
+    "Organize cardboard pieces",
+    "Organize item on shelf",
+    "Organize products",
+    "Organize snacks in box",
+    "Organize tools and materials",
+    "Pack beads into box",
+    "Peel blue strip",
+    "Peel foam strip",
+    "Pick up Dior gift box",
+    "Pick up Mahjong tile",
+    "Pick up accessory",
+    "Pick up and sort cardboard",
+    "Pick up another bottle",
+    "Pick up another canned item",
+    "Pick up another item",
+    "Pick up beads",
+    "Pick up black button",
+    "Pick up blue foam piece",
+    "Pick up blue paper strip",
+    "Pick up bottle",
+    "Pick up bottled sauce",
+    "Pick up button",
+    "Pick up can",
+    "Pick up canned food",
+    "Pick up canned good",
+    "Pick up canned goods",
+    "Pick up canned item",
+    "Pick up canned product",
+    "Pick up cardboard",
+    "Pick up cardboard cutout",
+    "Pick up cardboard piece",
+    "Pick up cardboard square",
+    "Pick up cardboard stack",
+    "Pick up cardboard strip",
+    "Pick up cardboard tray",
+    "Pick up cereal boxes",
+    "Pick up charging cable",
+    "Pick up charging case",
+    "Pick up cleaning cloth",
+    "Pick up colored tile",
+    "Pick up container",
+    "Pick up container from box",
+    "Pick up craft material",
+    "Pick up cut cardboard piece",
+    "Pick up dustpan",
+    "Pick up electronic accessory",
+    "Pick up electronic accessory from box",
+    "Pick up electronic device",
+    "Pick up electronic item",
+    "Pick up electronic product",
+    "Pick up food item",
+    "Pick up gift box",
+    "Pick up grocery item",
+    "Pick up item",
+    "Pick up item from bin",
+    "Pick up item from box",
+    "Pick up item from shelf",
+    "Pick up items from the shopping bag",
+    "Pick up jar",
+    "Pick up light blue strip",
+    "Pick up marker",
+    "Pick up metal ruler",
+    "Pick up new cardboard piece",
+    "Pick up new electronic product",
+    "Pick up new product from box",
+    "Pick up next gift box",
+    "Pick up next item from bin",
+    "Pick up next product from bin",
+    "Pick up nut bar box",
+    "Pick up object",
+    "Pick up oil bottle",
+    "Pick up orange button",
+    "Pick up pack from shelf",
+    "Pick up packaged paper lantern component",
+    "Pick up paper star",
+    "Pick up paper strip",
+    "Pick up paper towel",
+    "Pick up pasta box",
+    "Pick up pen",
+    "Pick up phone",
+    "Pick up pickle jar",
+    "Pick up pink water bottle",
+    "Pick up plastic bin",
+    "Pick up plastic container",
+    "Pick up plush toy",
+    "Pick up portable charger",
+    "Pick up power bank",
+    "Pick up product",
+    "Pick up product box",
+    "Pick up product from bin",
+    "Pick up product from box",
+    "Pick up product from shelf",
+    "Pick up puzzle piece",
+    "Pick up red button",
+    "Pick up retail item",
+    "Pick up sauce bottle",
+    "Pick up scissors",
+    "Pick up shopping bag",
+    "Pick up small cardboard piece",
+    "Pick up small item",
+    "Pick up small object",
+    "Pick up small piece of material",
+    "Pick up smartphone",
+    "Pick up snack package",
+    "Pick up snack packages",
+    "Pick up snack packs",
+    "Pick up snack pouch",
+    "Pick up spice jar",
+    "Pick up stapler",
+    "Pick up star",
+    "Pick up star bead",
+    "Pick up star-shaped bead",
+    "Pick up storage container",
+    "Pick up supplement bottle",
+    "Pick up supplies from box",
+    "Pick up tin can",
+    "Pick up tool",
+    "Pick up utility knife",
+    "Pick up water bottle",
+    "Pick up yellow item",
+    "Pick up yellow paper strip",
+    "Picking up bottle",
+    "Picking up crafting material",
+    "Picking up stock",
+    "Place Mahjong tile on stack",
+    "Place Mahjong tile on the stack",
+    "Place accessory box",
+    "Place accessory into box",
+    "Place accessory on shelf",
+    "Place and align button",
+    "Place and count bead",
+    "Place another canned food on shelf",
+    "Place back Dior gift box",
+    "Place bead on table",
+    "Place bottle back on shelf",
+    "Place box on shelf",
+    "Place button",
+    "Place button in group",
+    "Place button in row",
+    "Place can on shelf",
+    "Place canned food in bin",
+    "Place canned food in container",
+    "Place canned food on shelf",
+    "Place canned good on shelf",
+    "Place canned goods in container",
+    "Place canned product on shelf",
+    "Place cans into box",
+    "Place cardboard",
+    "Place cardboard piece",
+    "Place cardboard piece on stack",
+    "Place cardboard square",
+    "Place cardboard square on stack",
+    "Place cardboard strip",
+    "Place charger on table",
+    "Place charging case down",
+    "Place cloth on floor",
+    "Place colored tile",
+    "Place container in bin",
+    "Place container on floor",
+    "Place container on shelf",
+    "Place controller on table",
+    "Place crate on floor",
+    "Place device on lap",
+    "Place down paper pieces",
+    "Place down paper segment",
+    "Place down pen",
+    "Place down pink water bottle",
+    "Place down ruler and pen",
+    "Place down scissors",
+    "Place down strip",
+    "Place finished star on table",
+    "Place gift box into bin",
+    "Place gift box on shelf",
+    "Place hand on table",
+    "Place item back",
+    "Place item back on shelf",
+    "Place item in bag",
+    "Place item in container",
+    "Place item in shopping bag",
+    "Place item into bag",
+    "Place item into shopping bag",
+    "Place item on shelf",
+    "Place item on table",
+    "Place items on shelf",
+    "Place items on table",
+    "Place items on the shelf",
+    "Place jar in box",
+    "Place jar into shelf box",
+    "Place jar on shelf",
+    "Place ketchup bottle on shelf",
+    "Place knife down",
+    "Place lid back",
+    "Place marked piece down",
+    "Place marker down",
+    "Place material",
+    "Place oil in container",
+    "Place paper star",
+    "Place paper star in row",
+    "Place pen on cardboard",
+    "Place pen on table",
+    "Place phone down",
+    "Place phone on desk",
+    "Place phone on shelf",
+    "Place phone on table",
+    "Place pickle jar in box",
+    "Place piece into puzzle",
+    "Place plush toy into bag",
+    "Place plush toy on shelf",
+    "Place product in box",
+    "Place product on shelf",
+    "Place puzzle piece",
+    "Place quilled paper shape",
+    "Place red button",
+    "Place ribbon onto project",
+    "Place ruler on cardboard",
+    "Place sauce bottle on shelf",
+    "Place sauce in container",
+    "Place scissors aside",
+    "Place scissors down",
+    "Place scissors on table",
+    "Place smartphone down",
+    "Place smartphone on cardboard",
+    "Place smartphone on desk",
+    "Place smartphone on stand",
+    "Place smartphone on table",
+    "Place snack in box",
+    "Place snack on shelf",
+    "Place snack package in box",
+    "Place snack package on shelf",
+    "Place snack packages on shelf",
+    "Place snack pouch in container",
+    "Place snack pouch on shelf",
+    "Place spice jar in container",
+    "Place star",
+    "Place star in row",
+    "Place star on table",
+    "Place stars in container",
+    "Place stool on floor",
+    "Place storage container on floor",
+    "Place strip on table",
+    "Place supplement bottle in container",
+    "Place tool on table",
+    "Place towel",
+    "Place water bottle on table",
+    "Place white box on table",
+    "Placing labeled cardboard square",
+    "Placing labeled square",
+    "Placing paper strip",
+    "Placing pen on table",
+    "Placing phone down",
+    "Placing piece on stack",
+    "Placing stock on shelf",
+    "Plug cable into portable charger",
+    "Position cardboard for cutting",
+    "Position cardboard piece",
+    "Position cardboard strip",
+    "Position cardboard tray",
+    "Position cardboard tube",
+    "Position container near shelf",
+    "Position container on shelf",
+    "Position hands for work",
+    "Position ribbon piece",
+    "Position ruler and mark cardboard",
+    "Position ruler on cardboard",
+    "Position scissors",
+    "Position scissors for next cut",
+    "Position scissors to cut cardboard",
+    "Position shelving divider",
+    "Position the ruler",
+    "Position tray",
+    "Position utility knife",
+    "Position utility knife on cardboard",
+    "Positioning cardboard on workspace",
+    "Positioning paper strip",
+    "Positioning puzzle piece",
+    "Positioning ruler on cardboard",
+    "Prepare paper strip",
+    "Prepare to cut cardboard",
+    "Prepare to draw lines",
+    "Prepare to pick up item",
+    "Prepare to place bottle on shelf",
+    "Prepare to place cardboard",
+    "Prepare to place item in bag",
+    "Prepare to place product",
+    "Prepare to resume cutting",
+    "Prepare to sort beads",
+    "Preparing to craft",
+    "Press fold",
+    "Pull back hand",
+    "Pull paper strip",
+    "Push vacuum cleaner",
+    "Put down phone",
+    "Put down scissors",
+    "Put down smartphone",
+    "Put down utility knife",
+    "Put down water bottle",
+    "Putting away smartphone",
+    "Reach and sort buttons",
+    "Reach for Mahjong tiles",
+    "Reach for additional items",
+    "Reach for and examine canned goods",
+    "Reach for and pick up smartphone",
+    "Reach for another container",
+    "Reach for another item",
+    "Reach for beads",
+    "Reach for black button",
+    "Reach for button",
+    "Reach for can",
+    "Reach for canned food",
+    "Reach for canned goods",
+    "Reach for cardboard box",
+    "Reach for cardboard piece",
+    "Reach for cleaning supplies",
+    "Reach for container",
+    "Reach for craft items",
+    "Reach for empty shelf space",
+    "Reach for item",
+    "Reach for item in box",
+    "Reach for item on shelf",
+    "Reach for items",
+    "Reach for items in box",
+    "Reach for more pieces",
+    "Reach for next can",
+    "Reach for next canned food",
+    "Reach for next canned food item",
+    "Reach for next canned product",
+    "Reach for next item",
+    "Reach for next piece",
+    "Reach for next product",
+    "Reach for object",
+    "Reach for paper strip",
+    "Reach for paper strips",
+    "Reach for phone",
+    "Reach for product",
+    "Reach for product labels",
+    "Reach for product on shelf",
+    "Reach for puzzle piece",
+    "Reach for retail item",
+    "Reach for shelf",
+    "Reach for shelving divider",
+    "Reach for snack package",
+    "Reach for snack pouch",
+    "Reach for star",
+    "Reach for stars",
+    "Reach for utility knife",
+    "Reach for water bottle",
+    "Reach for wire hangers",
+    "Reach into bag",
+    "Reach into box",
+    "Reach towards shelf",
+    "Reaching for beads",
+    "Realign Mahjong tiles",
+    "Rearrange Mahjong tile",
+    "Rearrange Mahjong tiles",
+    "Rearrange shelf item",
+    "Record count",
+    "Record count on notepad",
+    "Record star count",
+    "Record star count on paper",
+    "Release and prepare new strip",
+    "Release bottle",
+    "Release cardboard",
+    "Release cardboard piece",
+    "Release cardboard piece and gesture",
+    "Release cardboard shape",
+    "Release container",
+    "Release folded paper",
+    "Release food item",
+    "Release hook",
+    "Release label",
+    "Release lantern",
+    "Release paper",
+    "Release paper coil",
+    "Release paper star",
+    "Release paper strip",
+    "Release pickle jar",
+    "Release product on shelf",
+    "Release puzzle piece",
+    "Release quilling strip",
+    "Release scissors",
+    "Release smartphone",
+    "Remove cardboard flap",
+    "Remove cardboard pattern",
+    "Remove cardboard pattern piece",
+    "Remove cleaning bottle",
+    "Remove item from bag",
+    "Remove item from shelf",
+    "Remove lid from container",
+    "Remove paper lantern part from packaging",
+    "Remove plastic container from shelf",
+    "Remove plastic container from storage box",
+    "Remove plastic packaging",
+    "Remove ruler",
+    "Remove ruler and marker",
+    "Remove shelf label",
+    "Remove storage bin from shelf",
+    "Reorganize bin contents",
+    "Reposition and cut",
+    "Reposition cardboard for cutting",
+    "Reposition hand",
+    "Reposition hands",
+    "Reposition hands and ruler",
+    "Reposition marker",
+    "Reposition newspaper",
+    "Reposition pen and prepare for next line",
+    "Reposition ruler",
+    "Reposition ruler and pen",
+    "Reposition scissors",
+    "Reposition sign and organize beads",
+    "Reposition tools",
+    "Reposition utility knife",
+    "Repositioning ruler",
+    "Repositioning ruler and cardboard",
+    "Resume counting stars",
+    "Resume observation",
+    "Resume sorting blue beads",
+    "Resume writing on paper",
+    "Retract camera/reposition view",
+    "Retract hand",
+    "Retract hand from bag",
+    "Retrieve another container",
+    "Retrieve canned food from box",
+    "Retrieve hand to table",
+    "Retrieve items from bag",
+    "Retrieve next canned food item",
+    "Retrieve paper strip",
+    "Retrieve paper strips",
+    "Retrieve snack from container",
+    "Retrieve star",
+    "Retrieving more beads",
+    "Return to sorting",
+    "Reviewing count record",
+    "Rinse cloth in sink",
+    "Roll quilling paper",
+    "Rolling paper strip",
+    "Rub hands together",
+    "Scan for next piece",
+    "Scan supermarket shelves",
+    "Score cardboard",
+    "Scroll on smartphone",
+    "Scroll smartphone screen",
+    "Scroll through photo gallery",
+    "Scrolling and viewing content on phone",
+    "Scrolling or navigating on phone",
+    "Search for puzzle piece",
+    "Secure paper edges with adhesive",
+    "Secure ribbon with needle",
+    "Securing paper structure",
+    "Select a bottle",
+    "Select and pick up a canned item",
+    "Select another item",
+    "Select paper strip",
+    "Select product from box",
+    "Selecting new paper strip",
+    "Separate cardboard piece",
+    "Set down scissors and pick up power bank",
+    "Set down utility knife",
+    "Slide utility knife along ruler",
+    "Sort Mahjong tiles",
+    "Sort and adjust button line",
+    "Sort and arrange buttons",
+    "Sort and arrange cardboard pieces",
+    "Sort and count beads",
+    "Sort and place buttons",
+    "Sort and place paper star",
+    "Sort and stack cardboard pieces",
+    "Sort beads",
+    "Sort beads and write count",
+    "Sort beads by color",
+    "Sort beads by hand",
+    "Sort beads on table",
+    "Sort beads on the table",
+    "Sort blue beads",
+    "Sort blue star-shaped pieces",
+    "Sort button",
+    "Sort button by color",
+    "Sort buttons",
+    "Sort buttons by color",
+    "Sort canned goods in tray",
+    "Sort colored tiles",
+    "Sort colorful pieces",
+    "Sort craft items",
+    "Sort cut cardboard",
+    "Sort light blue origami stars",
+    "Sort orange button",
+    "Sort orange buttons",
+    "Sort origami stars",
+    "Sort origami stars by color",
+    "Sort paper star",
+    "Sort paper stars",
+    "Sort plastic pieces",
+    "Sort purple beads",
+    "Sort purple star-shaped objects",
+    "Sort puzzle pieces",
+    "Sort quilled paper pieces",
+    "Sort small colorful pieces",
+    "Sort small craft pieces",
+    "Sort small objects",
+    "Sort small plastic pieces",
+    "Sort star-shaped beads",
+    "Sort star-shaped objects",
+    "Sort star-shaped objects by color",
+    "Sort tiles",
+    "Sort tiles by color",
+    "Sort yellow star-shaped objects",
+    "Sorting buttons",
+    "Sorting colorful paper pieces",
+    "Sorting paper stars",
+    "Stabilize cardboard",
+    "Stabilize ruler",
+    "Stack cardboard pieces",
+    "Stack cardboard square",
+    "Stack cardboard squares",
+    "Stacking cardboard pieces",
+    "Stacking cardboard square",
+    "Stacking cardboard squares",
+    "Stand up and walk away",
+    "Start cutting",
+    "Start folding paper strip",
+    "Starting to label next square",
+    "Stir contents",
+    "Stop measuring and put down tools",
+    "Stop sorting stars",
+    "Sweep debris",
+    "Sweep floor debris",
+    "Switch to scissors",
+    "Switching marker",
+    "Tap smartphone screen",
+    "Tapping on smartphone screen",
+    "Tapping smartphone screen",
+    "Tear newspaper",
+    "Tear off cardboard segment",
+    "Touch canned goods",
+    "Touch pieces in box",
+    "Touch shelf edge",
+    "Trace pattern on cardboard",
+    "Transition to cutting",
+    "Transition to standing position",
+    "Trim cardboard",
+    "Trim cardboard piece",
+    "Type on smartphone",
+    "Typing message on smartphone",
+    "Typing on phone",
+    "Typing on smartphone",
+    "Update paper record",
+    "Use phone",
+    "Use phone to check instructions",
+    "Use phone to check stock",
+    "Use phone while crafting",
+    "Use smartphone",
+    "Vacuum edge of carpet",
+    "Vacuum the carpet",
+    "Vacuuming along the wall edge",
+    "Vacuuming carpet corner",
+    "Vacuuming carpet edge",
+    "Vacuuming the carpet edge",
+    "View content on smartphone",
+    "View phone screen",
+    "Viewing phone screen",
+    "Walk across office",
+    "Walk across room",
+    "Walk across the room",
+    "Walk away",
+    "Walk in hallway",
+    "Walk through corridor",
+    "Walk through doorway",
+    "Walk through hallway",
+    "Walk through office",
+    "Walk through store",
+    "Walk through workspace",
+    "Walk towards aisle",
+    "Walk towards desk",
+    "Walk towards next aisle",
+    "Walk towards other aisles",
+    "Walk towards room",
+    "Walk towards shelf",
+    "Walk towards shelves",
+    "Walk towards storage area",
+    "Walk towards table",
+    "Walk towards workspace",
+    "Walk with cardboard",
+    "Walk with cardboard cutout",
+    "Walk with marker",
+    "Walk with shopping bag",
+    "Walking across the room",
+    "Walking along the aisle",
+    "Walking in the hallway",
+    "Walking in the workspace",
+    "Walking through classroom",
+    "Walking through office hallway",
+    "Walking through the office",
+    "Walking to sink",
+    "Walking towards door",
+    "Walking towards workstation",
+    "Washing hands",
+    "Washing hands in sink",
+    "Wipe down shelf",
+    "Wipe electronic item",
+    "Wipe food product",
+    "Wipe grocery shelf",
+    "Wipe item",
+    "Wipe jar",
+    "Wipe ketchup bottle",
+    "Wipe kitchen counter",
+    "Wipe product",
+    "Wipe retail item",
+    "Wipe shelf",
+    "Wipe shelf surface",
+    "Wipe the plastic jar",
+    "Wipe the product jar",
+    "Wipe the shelf",
+    "Wiping countertop",
+    "Withdraw hand",
+    "Write count on paper",
+    "Write on notepad",
+    "Write on paper",
+    "Write on paper record",
+    "Writing on notepad",
+    "fold purple ribbon",
+    "sort craft materials",
+    "{\"action\": \"Pour liquid into bowl\", \"contact\": \"yes\", \"evidence_window\": {\"end_frame\": 19, \"start_frame\": 0}, \"next_action\": \"Pour liquid into bowl\", \"objects\": [\"bottle\", \"bowl\", \"kitchen counter\", \"kitchen sink\", \"kitchen tap\", \"kitchen cabinet\", \"kitchen shelf\", \"kitchen window\", \"kitchen utensils\", \"kitchen appliances\", \"kitchen utensil rack\", \"kitchen towel\", \"kitchen cabinet door\", \"kitchen cabinet drawer\", \"kitchen cabinet shelf\", \"kitchen cabinet handle\", \"kitchen cabinet knob\", \"kitchen cabinet latch\", \"kitchen cabinet hinge\", \"kitchen cabinet panel\", \"kitchen cabinet frame\", \"kitchen cabinet base\", \"kitchen cabinet top\", \"kitchen cabinet side\", \"kitchen cabinet back\", \"kitchen cabinet door panel\", \"kitchen cabinet door frame\", \"kitchen cabinet door handle\", \"kitchen cabinet door knob\", \"kitchen cabinet door latch\", \"kitchen cabinet door hinge\", \"kitchen cabinet door panel frame\", \"kitchen cabinet door panel handle\", \"kitchen cabinet door panel knob\", \"kitchen cabinet door panel latch\", \"kitchen cabinet door panel hinge\", \"kitchen cabinet door panel frame handle\", \"kitchen cabinet door panel frame knob\", \"kitchen cabinet door panel frame latch\", \"kitchen cabinet door panel frame hinge\", \"kitchen cabinet door panel frame panel\", \"kitchen cabinet door panel frame panel handle\", \"kitchen cabinet door panel frame panel knob\", \"kitchen cabinet door panel frame panel latch\", \"kitchen cabinet door panel frame panel hinge\", \"kitchen cabinet door panel frame panel frame\", \"kitchen cabinet door panel frame panel frame handle\", \"kitchen cabinet door panel frame panel frame knob\", \"kitchen cabinet door panel frame panel frame latch\", \"kitchen cabinet door panel frame panel frame hinge\", \"kitchen cabinet door panel frame panel frame panel\", \"kitchen cabinet door panel frame panel frame panel handle\", \"kitchen cabinet door panel frame panel frame panel knob\", \"kitchen cabinet door panel frame panel frame panel latch\", \"kitchen cabinet door panel frame panel frame panel hinge\", \"kitchen cabinet door panel frame panel frame panel frame\", \"kitchen cabinet door panel frame panel frame panel frame handle\", \"kitchen cabinet door panel frame panel frame panel frame knob\", \"kitchen cabinet door panel frame panel frame panel frame latch\", \"kitchen cabinet door panel frame panel frame panel frame hinge\", \"kitchen cabinet door panel frame panel frame panel frame panel\", \"kitchen cabinet door panel frame panel frame panel frame panel handle\", \"kitchen cabinet door panel frame panel frame panel frame panel knob\", \"kitchen cabinet door panel frame panel frame panel frame panel latch\", \"kitchen cabinet door panel frame panel frame panel frame panel hinge\", \"kitchen cabinet door panel frame panel frame panel frame panel frame\", \"kitchen cabinet door panel frame panel frame panel frame panel frame handle\", \"kitchen cabinet door panel frame panel frame panel frame panel frame knob\", \"kitchen cabinet door panel frame panel frame panel frame panel frame latch\", \"kitchen cabinet door panel frame panel frame panel frame panel frame hinge\", \"kitchen cabinet door panel frame panel frame panel frame panel frame panel\", \"kitchen cabinet door panel frame panel frame panel frame panel frame panel handle\", \"kitchen cabinet door panel frame panel frame panel frame panel frame panel knob\", \"kitchen cabinet door panel frame panel frame panel frame panel frame panel latch\", \"kitchen cabinet door panel frame panel frame panel frame panel frame panel hinge\", \"kitchen cabinet door panel frame panel frame panel frame panel frame panel frame\", \"kitchen cabinet door panel frame panel frame panel frame panel frame panel frame handle\", \"kitchen cabinet door panel frame panel frame panel frame panel frame panel frame knob\", \"kitchen cabinet door panel frame panel frame panel frame panel frame panel frame latch\", \"kitchen cabinet door panel frame panel frame panel frame panel frame panel frame hinge\", \"kitchen cabinet door panel frame panel frame panel frame panel frame panel frame panel\", \"kitchen cabinet door panel frame panel frame panel frame panel frame panel frame panel handle\", \"kitchen cabinet door panel frame panel frame panel frame panel frame panel frame panel knob\", \"kitchen cabinet door panel frame panel frame panel frame panel frame panel frame panel latch\", \"kitchen cabinet door panel frame panel frame panel frame panel frame panel frame panel hinge\", \"kitchen cabinet door panel frame panel frame panel frame panel frame panel frame panel frame\", \"kitchen cabinet door panel frame panel frame panel frame panel frame panel frame panel frame handle\", \"kitchen cabinet door panel frame panel frame panel frame panel frame panel frame panel frame knob\", \"kitchen cabinet door panel frame panel frame panel frame panel frame panel frame panel frame latch\", \"kitchen cabinet door panel frame panel frame panel frame panel frame panel frame panel frame hinge\", \"kitchen cabinet",
+    "Close the door",
+    "{\"action\": \"Close door\", \"contact\": \"yes\", \"evidence_window\": {\"end_frame\": 2539, \"start_frame\": 2520}, \"next_action\": \"Close door\", \"objects\": [\"door\", \"door handle\", \"plastic bag\", \"blue bucket\", \"red bucket\", \"blue container\", \"red container\", \"white container\", \"white box\", \"white bag\", \"white cloth\", \"white plastic bag\", \"white plastic container\", \"white cardboard box\", \"white paper\", \"white paper bag\", \"white paper sheet\", \"white paper strip\", \"white plastic sheet\", \"white plastic bag with red label\", \"white plastic container with red label\", \"white cardboard tray\", \"white cardboard piece\", \"white cardboard square\", \"white cardboard shape\", \"white paper lantern\", \"white paper lantern component\", \"white paper star\", \"white paper star shape\", \"white paper lantern piece\", \"white paper lantern part\", \"white paper decoration\", \"white paper cutout\", \"white paper shape\", \"white paper segment\", \"white paper coil\", \"white paper strip with red label\", \"white paper strip with blue label\", \"white paper strip with yellow label\", \"white paper strip with green label\", \"white paper strip with purple label\", \"white paper strip with orange label\", \"white paper strip with black label\", \"white paper strip with pink label\", \"white paper strip with brown label\", \"white paper strip with grey label\", \"white paper strip with turquoise label\", \"white paper strip with maroon label\", \"white paper strip with lavender label\", \"white paper strip with olive green label\", \"white paper strip with coral label\", \"white paper strip with navy blue label\", \"white paper strip with light blue label\", \"white paper strip with dark blue label\", \"white paper strip with light green label\", \"white paper strip with dark green label\", \"white paper strip with yellow-green label\", \"white paper strip with lime green label\", \"white paper strip with forest green label\", \"white paper strip with emerald green label\", \"white paper strip with teal label\", \"white paper strip with turquoise blue label\", \"white paper strip with sky blue label\", \"white paper strip with baby blue label\", \"white paper strip with periwinkle label\", \"white paper strip with lavender blue label\", \"white paper strip with violet label\", \"white paper strip with purple blue label\", \"white paper strip with indigo label\", \"white paper strip with dark purple label\", \"white paper strip with magenta label\", \"white paper strip with pink purple label\", \"white paper strip with rose pink label\", \"white paper strip with coral pink label\", \"white paper strip with salmon pink label\", \"white paper strip with peach label\", \"white paper strip with apricot label\", \"white paper strip with orange yellow label\", \"white paper strip with golden yellow label\", \"white paper strip with lemon yellow label\", \"white paper strip with yellow green label\", \"white paper strip with chartreuse label\", \"white paper strip with olive label\", \"white paper strip with khaki label\", \"white paper strip with tan label\", \"white paper strip with beige label\", \"white paper strip with cream label\", \"white paper strip with off white label\", \"white paper strip with light grey label\", \"white paper strip with grey label\", \"white paper strip with dark grey label\", \"white paper strip with black label\", \"white paper strip with brown label\", \"white paper strip with dark brown label\", \"white paper strip with chocolate label\", \"white paper strip with coffee label\", \"white paper strip with taupe label\", \"white paper strip with maroon label\", \"white paper strip with burgundy label\", \"white paper strip with red label\", \"white paper strip with dark red label\", \"white paper strip with crimson label\", \"white paper strip with cherry red label\", \"white paper strip with pink label\", \"white paper strip with light pink label\", \"white paper strip with pale pink label\", \"white paper strip with rose label\", \"white paper strip with blush label\", \"white paper strip with coral label\", \"white paper strip with salmon label\", \"white paper strip with peach label\", \"white paper strip with apricot label\", \"white paper strip with orange label\", \"white paper strip with light orange label\", \"white paper strip with pale orange label\", \"white paper strip with tangerine label\", \"white paper strip with mango label\", \"white paper strip with yellow label\", \"white paper strip with light yellow label\", \"white paper strip with pale yellow label\", \"white paper strip with lemon label\", \"white paper strip with lime label\", \"white paper strip with green label\", \"white paper strip with light green label\", \"white paper strip with pale green label\", \"white paper strip with mint label\", \"white paper strip with sage label\", \"white paper strip with olive label\", \"white paper strip with khaki",
+    "{\"action\": \"Close the door\", \"contact\": \"yes\", \"evidence_window\": {\"end_frame\": 2559, \"start_frame\": 2540}, \"next_action\": \"Close the door\", \"objects\": [\"door\", \"door handle\", \"door knob\", \"door frame\", \"door hinge\", \"door latch\", \"door lock\", \"door panel\", \"door post\", \"door stop\", \"door weather stripping\", \"doorway\", \"garage door\", \"garage door opener\", \"garage door track\", \"garage door wheel\", \"garage door window\", \"garage door opener button\", \"garage door opener remote\", \"garage door opener receiver\", \"garage door opener motor\", \"garage door opener chain\", \"garage door opener belt\", \"garage door opener rail\", \"garage door opener sensor\", \"garage door opener control panel\", \"garage door opener light\", \"garage door opener emergency release\", \"garage door opener remote control\", \"garage door opener wall switch\", \"garage door opener keypad\", \"garage door opener smartphone app\", \"garage door opener voice control\", \"garage door opener timer\", \"garage door opener remote control\", \"garage door opener wall switch\", \"garage door opener keypad\", \"garage door opener smartphone app\", \"garage door opener voice control\", \"garage door opener timer\", \"garage door opener emergency release\", \"garage door opener sensor\", \"garage door opener control panel\", \"garage door opener light\", \"garage door opener motor\", \"garage door opener chain\", \"garage door opener belt\", \"garage door opener rail\", \"garage door opener remote\", \"garage door opener receiver\", \"garage door opener button\", \"garage door window\", \"garage door track\", \"garage door wheel\", \"garage door\", \"garage door opener\", \"garage door opener button\", \"garage door opener remote\", \"garage door opener receiver\", \"garage door opener motor\", \"garage door opener chain\", \"garage door opener belt\", \"garage door opener rail\", \"garage door opener sensor\", \"garage door opener control panel\", \"garage door opener light\", \"garage door opener emergency release\", \"garage door opener remote control\", \"garage door opener wall switch\", \"garage door opener keypad\", \"garage door opener smartphone app\", \"garage door opener voice control\", \"garage door opener timer\", \"garage door opener remote control\", \"garage door opener wall switch\", \"garage door opener keypad\", \"garage door opener smartphone app\", \"garage door opener voice control\", \"garage door opener timer\", \"garage door opener emergency release\", \"garage door opener sensor\", \"garage door opener control panel\", \"garage door opener light\", \"garage door opener motor\", \"garage door opener chain\", \"garage door opener belt\", \"garage door opener rail\", \"garage door opener remote\", \"garage door opener receiver\", \"garage door opener button\", \"garage door window\", \"garage door track\", \"garage door wheel\", \"garage door\", \"garage door opener\", \"garage door opener button\", \"garage door opener remote\", \"garage door opener receiver\", \"garage door opener motor\", \"garage door opener chain\", \"garage door opener belt\", \"garage door opener rail\", \"garage door opener sensor\", \"garage door opener control panel\", \"garage door opener light\", \"garage door opener emergency release\", \"garage door opener remote control\", \"garage door opener wall switch\", \"garage door opener keypad\", \"garage door opener smartphone app\", \"garage door opener voice control\", \"garage door opener timer\", \"garage door opener remote control\", \"garage door opener wall switch\", \"garage door opener keypad\", \"garage door opener smartphone app\", \"garage door opener voice control\", \"garage door opener timer\", \"garage door opener emergency release\", \"garage door opener sensor\", \"garage door opener control panel\", \"garage door opener light\", \"garage door opener motor\", \"garage door opener chain\", \"garage door opener belt\", \"garage door opener rail\", \"garage door opener remote\", \"garage door opener receiver\", \"garage door opener button\", \"garage door window\", \"garage door track\", \"garage door wheel\", \"garage door\", \"garage door opener\", \"garage door opener button\", \"garage door opener remote\", \"garage door opener receiver\", \"garage door opener motor\", \"garage door opener chain\", \"garage door opener belt\", \"garage door opener rail\", \"garage door opener sensor\", \"garage",
+    "{\"action\": \"Close cabinet door\", \"contact\": \"yes\", \"evidence_window\": {\"end_frame\": 3419, \"start_frame\": 3400}, \"next_action\": \"Close cabinet door\", \"objects\": [\"cabinet\", \"cabinet door\", \"cabinet handle\", \"cabinet shelf\", \"cabinet drawer\", \"cabinet drawer handle\", \"cabinet drawer knob\", \"cabinet drawer pull\", \"cabinet drawer slide\", \"cabinet drawer slide rail\", \"cabinet drawer slide track\", \"cabinet drawer slide roller\", \"cabinet drawer slide bracket\", \"cabinet drawer slide support\", \"cabinet drawer slide guide\", \"cabinet drawer slide stop\", \"cabinet drawer slide clip\", \"cabinet drawer slide clip holder\", \"cabinet drawer slide clip bracket\", \"cabinet drawer slide clip support\", \"cabinet drawer slide clip guide\", \"cabinet drawer slide clip stop\", \"cabinet drawer slide clip holder bracket\", \"cabinet drawer slide clip holder support\", \"cabinet drawer slide clip holder guide\", \"cabinet drawer slide clip holder stop\", \"cabinet drawer slide clip holder clip\", \"cabinet drawer slide clip holder clip holder\", \"cabinet drawer slide clip holder clip bracket\", \"cabinet drawer slide clip holder clip support\", \"cabinet drawer slide clip holder clip guide\", \"cabinet drawer slide clip holder clip stop\", \"cabinet drawer slide clip holder clip holder bracket\", \"cabinet drawer slide clip holder clip holder support\", \"cabinet drawer slide clip holder clip holder guide\", \"cabinet drawer slide clip holder clip holder stop\", \"cabinet drawer slide clip holder clip holder clip\", \"cabinet drawer slide clip holder clip holder clip holder\", \"cabinet drawer slide clip holder clip holder clip bracket\", \"cabinet drawer slide clip holder clip holder clip support\", \"cabinet drawer slide clip holder clip holder clip guide\", \"cabinet drawer slide clip holder clip holder clip stop\", \"cabinet drawer slide clip holder clip holder clip holder bracket\", \"cabinet drawer slide clip holder clip holder clip holder support\", \"cabinet drawer slide clip holder clip holder clip holder guide\", \"cabinet drawer slide clip holder clip holder clip holder stop\", \"cabinet drawer slide clip holder clip holder clip holder clip\", \"cabinet drawer slide clip holder clip holder clip holder clip holder\", \"cabinet drawer slide clip holder clip holder clip holder clip bracket\", \"cabinet drawer slide clip holder clip holder clip holder clip support\", \"cabinet drawer slide clip holder clip holder clip holder clip guide\", \"cabinet drawer slide clip holder clip holder clip holder clip stop\", \"cabinet drawer slide clip holder clip holder clip holder clip holder bracket\", \"cabinet drawer slide clip holder clip holder clip holder clip holder support\", \"cabinet drawer slide clip holder clip holder clip holder clip holder guide\", \"cabinet drawer slide clip holder clip holder clip holder clip holder stop\", \"cabinet drawer slide clip holder clip holder clip holder clip holder clip\", \"cabinet drawer slide clip holder clip holder clip holder clip holder clip holder\", \"cabinet drawer slide clip holder clip holder clip holder clip holder clip bracket\", \"cabinet drawer slide clip holder clip holder clip holder clip holder clip support\", \"cabinet drawer slide clip holder clip holder clip holder clip holder clip guide\", \"cabinet drawer slide clip holder clip holder clip holder clip holder clip stop\", \"cabinet drawer slide clip holder clip holder clip holder clip holder clip holder bracket\", \"cabinet drawer slide clip holder clip holder clip holder clip holder clip holder support\", \"cabinet drawer slide clip holder clip holder clip holder clip holder clip holder guide\", \"cabinet drawer slide clip holder clip holder clip holder clip holder clip holder stop\", \"cabinet drawer slide clip holder clip holder clip holder clip holder clip holder clip\", \"cabinet drawer slide clip holder clip holder clip holder clip holder clip holder clip holder\", \"cabinet drawer slide clip holder clip holder clip holder clip holder clip holder clip bracket\", \"cabinet drawer slide clip holder clip holder clip holder clip holder clip holder clip support\", \"cabinet drawer slide clip holder clip holder clip holder clip holder clip holder clip guide\", \"cabinet drawer slide clip holder clip holder clip holder clip holder clip holder clip stop\", \"cabinet drawer slide clip holder clip holder clip holder clip holder clip holder clip holder bracket\", \"cabinet drawer slide clip holder clip holder clip holder clip holder clip holder clip holder support\", \"cabinet drawer slide clip holder clip holder clip holder clip holder clip holder clip holder guide\", \"cabinet drawer slide clip holder clip holder clip holder clip holder clip holder clip holder stop\", \"cabinet drawer slide clip holder clip holder clip holder clip holder clip holder clip holder clip\", \"cabinet drawer slide clip holder clip holder clip holder clip holder clip holder clip holder clip holder\", \"cabinet drawer slide clip holder clip holder clip holder clip holder clip holder clip holder clip bracket\", \"cabinet drawer slide clip holder clip holder clip holder clip holder clip holder clip holder clip support",
+    "Close cabinet",
+    "Close cabinet door",
+    "Wipe hands",
+    "Hold paper star",
+    "Place pot on the counter",
+    "Rinse pot",
+    "Place marker on table",
+    "Carry snack package",
+    "Place paper strip on table",
+    "Reach for jar",
+    "Hold jar",
+    "Place beads on table",
+    "{\"action\": \"Hold paper star\", \"contact\": \"yes\", \"evidence_window\": {\"end_frame\": 1959, \"start_frame\": 1940}, \"next_action\": \"Hold paper star\", \"objects\": [\"paper star\", \"table\", \"scissors\", \"phone\", \"mouse\", \"beads\", \"pen\", \"light blue paper\", \"yellow paper\", \"purple paper\", \"cardboard\", \"cardboard pieces\", \"cardboard shapes\", \"cardboard square\", \"cardboard tray\", \"cardboard tube\", \"cardboard strip\", \"utility knife\", \"ruler\", \"marker\", \"jar\", \"canned food\", \"tin can\", \"plastic container\", \"plastic packaging\", \"box\", \"shopping bag\", \"gift box\", \"product\", \"product box\", \"product package\", \"supplement bottle\", \"bottle\", \"sauce bottle\", \"pickle jar\", \"water bottle\", \"pink water bottle\", \"cleaning bottle\", \"earbud case\", \"charging case\", \"portable charger\", \"power bank\", \"charger\", \"charging cable\", \"power adapter\", \"smartphone\", \"smartphone box\", \"computer mouse\", \"pen and paper\", \"paper\", \"paper pieces\", \"paper segments\", \"paper lantern\", \"paper lantern component\", \"blue product box\", \"blue strip\", \"light blue strip\", \"yellow item\", \"yellow paper strip\", \"purple paper strip\", \"adhesive tape\", \"ribbon\", \"ribbon piece\", \"plastic strip\", \"foam piece\", \"foam strips\", \"cardboard pattern\", \"cardboard pattern piece\", \"stapler\", \"cleaning cloth\", \"dustpan\", \"box of beads\", \"container\", \"container of canned food\", \"small cardboard piece\", \"small piece of material\", \"small object\", \"small product bag\", \"small white box\", \"bean\", \"bean pile\", \"button\", \"black button\", \"orange button\", \"red button\", \"small button\", \"button cluster\", \"button row\", \"star\", \"star bead\", \"star-shaped bead\", \"purple bead\", \"light blue origami star\", \"yellow star-shaped object\", \"purple star-shaped object\", \"colored tile\", \"colored pieces\", \"jewel\", \"jewel piece\", \"Mahjong tile\", \"Mahjong tiles\", \"puzzle piece\", \"quilled paper\", \"quilled paper piece\", \"quilled paper shape\", \"quilling paper\", \"quilling paper strip\", \"paper quilling piece\", \"paper quilling strip\", \"paper star\", \"lucky star\", \"folded paper\", \"folded paper star\", \"paper strip\", \"paper decoration\", \"colored paper\", \"colorful pieces\", \"cardboard sheet\", \"cardboard square\", \"cardboard pieces\", \"cardboard shapes\", \"cardboard cutout\", \"cardboard tray\", \"cardboard tube\", \"cardboard strip\", \"newspaper\", \"newspaper piece\", \"plastic container\", \"plastic storage bin\", \"tin can\", \"jar\", \"canned good\", \"canned product\", \"packaged paper lantern component\", \"snack package\", \"snack pouch\", \"container of snack\", \"gift box\", \"product package\", \"supplement bottle\", \"bottle\", \"ketchup bottle\", \"sauce bottle\", \"pickle jar\", \"water bottle\", \"pink water bottle\", \"cleaning bottle\", \"earbud case\", \"charging case\", \"portable charger\", \"power bank\", \"charger\", \"charging cable\", \"power adapter\", \"smartphone\", \"smartphone box\", \"computer mouse\", \"pen\", \"notepad\", \"paper\", \"paper pieces\", \"paper segments\", \"pieces\", \"lantern\", \"lantern component\", \"blue product box\", \"blue strip\", \"light blue strip\", \"yellow item\", \"yellow paper strip\", \"purple paper strip\", \"adhesive strip\", \"ribbon knot\", \"small paper segment\", \"foam piece\", \"foam strips\", \"cardboard pattern\", \"cardboard pattern piece\", \"stapler\", \"cleaning cloth\", \"dustpan\", \"box of beads\", \"container\", \"small cardboard piece\", \"small piece of material\", \"small object\", \"small product bag\", \"small white box\", \"bean\", \"bean pile\", \"button\", \"black button\", \"orange button\", \"red button\", \"small button\", \"buttons\", \"button cluster\", \"button row\", \"star\", \"star bead\", \"star-shaped bead\", \"purple bead\", \"light blue origami star\", \"yellow star-shaped object\", \"purple star-shaped object\", \"colored tile\", \"colored pieces\", \"jewel\", \"jewel piece\", \"Mahjong tile\", \"Mahjong tiles\", \"puzzle piece\", \"quilled paper\", \"qu",
+    "Place earbud case on table",
+    "Hold red button",
+    "Adjust beads by color",
+    "Approach workspace",
+    "{\"action\": \"Approach workspace\", \"contact\": \"no\", \"evidence_window\": {\"end_frame\": 39, \"start_frame\": 20}, \"next_action\": \"Approach workspace\", \"objects\": [\"cardboard\", \"cardboard pieces\", \"cardboard stack\", \"cardboard tray\", \"cardboard tube\", \"cardboard pieces\", \"cardboard shapes\", \"cardboard squares\", \"cardboard sheet\", \"cardboard cutouts\", \"cardboard shapes\", \"cardboard pieces\", \"cardboard shapes\", \"cardboard pieces\", \"cardboard shapes\", \"cardboard pieces\", \"cardboard shapes\", \"cardboard pieces\", \"cardboard shapes\", \"cardboard pieces\", \"cardboard shapes\", \"cardboard pieces\", \"cardboard shapes\", \"cardboard pieces\", \"cardboard shapes\", \"cardboard pieces\", \"cardboard shapes\", \"cardboard pieces\", \"cardboard shapes\", \"cardboard pieces\", \"cardboard shapes\", \"cardboard pieces\", \"cardboard shapes\", \"cardboard pieces\", \"cardboard shapes\", \"cardboard pieces\", \"cardboard shapes\", \"cardboard pieces\", \"cardboard shapes\", \"cardboard pieces\", \"cardboard shapes\", \"cardboard pieces\", \"cardboard shapes\", \"cardboard pieces\", \"cardboard shapes\", \"cardboard pieces\", \"cardboard shapes\", \"cardboard pieces\", \"cardboard shapes\", \"cardboard pieces\", \"cardboard shapes\", \"cardboard pieces\", \"cardboard shapes\", \"cardboard pieces\", \"cardboard shapes\", \"cardboard pieces\", \"cardboard shapes\", \"cardboard pieces\", \"cardboard shapes\", \"cardboard pieces\", \"cardboard shapes\", \"cardboard pieces\", \"cardboard shapes\", \"cardboard pieces\", \"cardboard shapes\", \"cardboard pieces\", \"cardboard shapes\", \"cardboard pieces\", \"cardboard shapes\", \"cardboard pieces\", \"cardboard shapes\", \"cardboard pieces\", \"cardboard shapes\", \"cardboard pieces\", \"cardboard shapes\", \"cardboard pieces\", \"cardboard shapes\", \"cardboard pieces\", \"cardboard shapes\", \"cardboard pieces\", \"cardboard shapes\", \"cardboard pieces\", \"cardboard shapes\", \"cardboard pieces\", \"cardboard shapes\", \"cardboard pieces\", \"cardboard shapes\", \"cardboard pieces\", \"cardboard shapes\", \"cardboard pieces\", \"cardboard shapes\", \"cardboard pieces\", \"cardboard shapes\", \"cardboard pieces\", \"cardboard shapes\", \"cardboard pieces\", \"cardboard shapes\", \"cardboard pieces\", \"cardboard shapes\", \"cardboard pieces\", \"cardboard shapes\", \"cardboard pieces\", \"cardboard shapes\", \"cardboard pieces\", \"cardboard shapes\", \"cardboard pieces\", \"cardboard shapes\", \"cardboard pieces\", \"cardboard shapes\", \"cardboard pieces\", \"cardboard shapes\", \"cardboard pieces\", \"cardboard shapes\", \"cardboard pieces\", \"cardboard shapes\", \"cardboard pieces\", \"cardboard shapes\", \"cardboard pieces\", \"cardboard shapes\", \"cardboard pieces\", \"cardboard shapes\", \"cardboard pieces\", \"cardboard shapes\", \"cardboard pieces\", \"cardboard shapes\", \"cardboard pieces\", \"cardboard shapes\", \"cardboard pieces\", \"cardboard shapes\", \"cardboard pieces\", \"cardboard shapes\", \"cardboard pieces\", \"cardboard shapes\", \"cardboard pieces\", \"cardboard shapes\", \"cardboard pieces\", \"cardboard shapes\", \"cardboard pieces\", \"cardboard shapes\", \"cardboard pieces\", \"cardboard shapes\", \"cardboard pieces\", \"cardboard shapes\", \"cardboard pieces\", \"cardboard shapes\", \"cardboard pieces\", \"cardboard shapes\", \"cardboard pieces\", \"cardboard shapes\", \"cardboard pieces\", \"cardboard shapes\", \"cardboard pieces\", \"cardboard shapes\", \"cardboard pieces\", \"cardboard shapes\", \"cardboard pieces\", \"cardboard shapes\", \"cardboard pieces\", \"cardboard shapes\", \"cardboard pieces\", \"cardboard shapes\", \"cardboard pieces\", \"cardboard shapes\", \"cardboard pieces\", \"cardboard shapes\", \"cardboard pieces\", \"cardboard shapes\", \"cardboard pieces\", \"cardboard shapes\", \"cardboard pieces\", \"cardboard shapes\", \"cardboard pieces\", \"cardboard shapes\", \"cardboard pieces\", \"cardboard shapes\", \"cardboard pieces\", \"cardboard shapes\", \"cardboard pieces\", \"cardboard shapes\", \"cardboard pieces\", \"cardboard shapes\", \"cardboard pieces\", \"cardboard shapes\", \"cardboard pieces\", \"cardboard shapes\", \"cardboard pieces\", \"cardboard shapes\", \"cardboard pieces\", \"cardboard shapes\", \"cardboard pieces\", \"cardboard shapes\", \"cardboard pieces\", \"cardboard shapes\", \"cardboard pieces\", \"cardboard shapes\", \"card"
+  ],
+  "model_id": "<workspace-parent>/modelscope_models/Qwen__Qwen3-Omni-30B-A3B-Instruct",
+  "adapter_dir": "<project>/checkpoints/xperience10m_qwen3_omni_128ep_96train_16val_16test_valmon_20260605_lora/adapter_lora",
+  "dataset_jsonl": "<project>/results/omni_finetune/xperience10m_qwen3_omni_128ep_96train_16val_16test_valmon_20260605_dataset/dataset.jsonl",
+  "eval_split": "test",
+  "train_split": "train",
+  "sample_offset": 0,
+  "sample_stride": 1,
+  "num_eval_episodes": 14,
+  "held_out_episode_count": 14,
+  "unseen_eval_labels": [
+    "Adjust canned food on shelf",
+    "Adjust lantern shape",
+    "Adjust lantern string",
+    "Adjust paper",
+    "Adjust pot position",
+    "Align canned food on shelf",
+    "Align edges of paper lantern",
+    "Align paper lantern edges",
+    "Apply adhesive tape to lantern",
+    "Approach boxes",
+    "Approaching and pressing the door switch",
+    "Approaching the table",
+    "Arrange buttons in a line",
+    "Arrange star beads",
+    "Arrange star beads for counting",
+    "Attempt to fit puzzle piece",
+    "Bend and manipulate plastic strip",
+    "Browse smartphone screen",
+    "Bundle display hooks",
+    "Closing the door",
+    "Counting and organizing beads",
+    "Counting star beads",
+    "Cut along the marked line",
+    "Entering the VR training room",
+    "Expand paper lantern",
+    "Extract wire hangers from box",
+    "Fold paper lantern",
+    "Fold plastic strip",
+    "Gather star beads",
+    "Gesturing",
+    "Grasp cleaning bottle",
+    "Grasp lantern",
+    "Grasp lantern component",
+    "Grasping cleaning cloth",
+    "Greeting/acknowledging participants",
+    "Handle paper lantern component",
+    "Hold and bend plastic strip",
+    "Hold and manipulate paper strip",
+    "Hold canned food",
+    "Hold container lid",
+    "Hold earbud case",
+    "Hold paper lantern",
+    "Identify next cardboard piece",
+    "Inspect shelf condition",
+    "Lift pot lid",
+    "Manipulate adhesive strip",
+    "Manipulate bead",
+    "Manipulate beads",
+    "Manipulate craft paper strips",
+    "Manipulate craft piece",
+    "Manipulate material",
+    "Manipulate paper decoration",
+    "Manipulate paper edge",
+    "Manipulate plastic strip",
+    "Manipulate plastic strips",
+    "Manipulate puzzle piece",
+    "Manipulate yellow strip",
+    "Manipulating paper strips",
+    "Move dustpan to side",
+    "Move hand away",
+    "Move hand away from shelf",
+    "Move marker and adjust hand",
+    "Move pot",
+    "Move through aisle",
+    "Move through the training room",
+    "Move towards kitchen area",
+    "Move towards the stove",
+    "Observe and pause",
+    "Observe and walk through store",
+    "Observe colleague and workspace",
+    "Observe puzzle progress",
+    "Open earbud case",
+    "Open folded paper lantern",
+    "Open paper lantern",
+    "Open paper lantern component",
+    "Open stove pot lid",
+    "Operate smartphone",
+    "Pick up dustpan",
+    "Pick up items from the shopping bag",
+    "Pick up packaged paper lantern component",
+    "Pick up puzzle piece",
+    "Pick up small piece of material",
+    "Pick up star bead",
+    "Picking up bottle",
+    "Picking up crafting material",
+    "Place and count bead",
+    "Place another canned food on shelf",
+    "Place cloth on floor",
+    "Place hand on table",
+    "Place items on the shelf",
+    "Place lid back",
+    "Place marked piece down",
+    "Place material",
+    "Place piece into puzzle",
+    "Place smartphone down",
+    "Place smartphone on stand",
+    "Place towel",
+    "Placing paper strip",
+    "Preparing to craft",
+    "Put down smartphone",
+    "Reach for cleaning supplies",
+    "Reach for craft items",
+    "Reach for next can",
+    "Reach for next canned food",
+    "Reach for puzzle piece",
+    "Reach for wire hangers",
+    "Record count",
+    "Release cardboard piece and gesture",
+    "Release hook",
+    "Release lantern",
+    "Release smartphone",
+    "Remove cleaning bottle",
+    "Remove paper lantern part from packaging",
+    "Remove plastic packaging",
+    "Reposition hand",
+    "Resume observation",
+    "Retrieve canned food from box",
+    "Retrieve next canned food item",
+    "Retrieving more beads",
+    "Rinse cloth in sink",
+    "Scroll smartphone screen",
+    "Secure paper edges with adhesive",
+    "Securing paper structure",
+    "Sort and adjust button line",
+    "Sort and arrange buttons",
+    "Sort and count beads",
+    "Sort and place buttons",
+    "Sort beads and write count",
+    "Sort button",
+    "Sort buttons",
+    "Sort craft items",
+    "Sort puzzle pieces",
+    "Sort small craft pieces",
+    "Start cutting",
+    "Stir contents",
+    "Use phone while crafting",
+    "Walk towards other aisles",
+    "Walking across the room",
+    "Walking in the hallway",
+    "Walking towards door",
+    "Washing hands in sink",
+    "Wipe kitchen counter",
+    "Wiping countertop",
+    "sort craft materials"
+  ],
+  "num_unseen_label_samples": 317,
+  "seen_label_accuracy": 0.04580152671755725,
+  "unseen_label_accuracy": 0.015772870662460567,
+  "eval_label_counts": {
+    "Manipulate paper strip": 14,
+    "Move phone": 7,
+    "Cut cardboard": 7,
+    "Cut along the marked line": 7,
+    "Use smartphone": 6,
+    "Pick up utility knife": 6,
+    "Manipulate paper decoration": 5,
+    "Manipulate paper edge": 5,
+    "Manipulate adhesive strip": 5,
+    "Hold smartphone": 4,
+    "Interact with smartphone": 4,
+    "Manipulate craft piece": 4,
+    "Manipulate craft paper strips": 4,
+    "Operate smartphone": 4,
+    "Release paper strip": 4,
+    "Arrange buttons": 4,
+    "Arrange buttons in a line": 4,
+    "Approaching and pressing the door switch": 4,
+    "Bend and manipulate plastic strip": 4,
+    "Placing paper strip": 4,
+    "Securing paper structure": 4,
+    "Secure paper edges with adhesive": 4,
+    "Reposition hand": 3,
+    "Mark cardboard piece": 3,
+    "Hold cardboard piece": 3,
+    "Cut cardboard piece": 3,
+    "Marking cardboard piece": 3,
+    "Move marker and adjust hand": 3,
+    "Identify next cardboard piece": 3,
+    "sort craft materials": 3,
+    "Release scissors": 3,
+    "Release smartphone": 3,
+    "Sort small craft pieces": 3,
+    "Reach for craft items": 3,
+    "Sort craft items": 3,
+    "Move smartphone": 3,
+    "Place hand on table": 3,
+    "Browse smartphone screen": 3,
+    "Scroll smartphone screen": 3,
+    "Put down smartphone": 3,
+    "Place smartphone down": 3,
+    "Pick up puzzle piece": 3,
+    "Place piece into puzzle": 3,
+    "Manipulate puzzle piece": 3,
+    "Manipulate puzzle pieces": 3,
+    "Place puzzle piece": 3,
+    "Observe puzzle progress": 3,
+    "Adjust puzzle piece": 3,
+    "Attempt to fit puzzle piece": 3,
+    "Sort buttons": 3,
+    "Sort and arrange buttons": 3,
+    "Pick up button": 3,
+    "Place button": 3,
+    "Sort button": 3,
+    "Sort and adjust button line": 3,
+    "Sort and place buttons": 3,
+    "Walking in the hallway": 3,
+    "Entering the VR training room": 3,
+    "Greeting/acknowledging participants": 3,
+    "Move through the training room": 3,
+    "Manipulate plastic strips": 3,
+    "Manipulate plastic strip": 3,
+    "Hold and bend plastic strip": 3,
+    "Fold plastic strip": 3,
+    "Hold container lid": 2,
+    "Closing the door": 2,
+    "Grasp cleaning bottle": 2,
+    "Grasping cleaning cloth": 2,
+    "Adjust pot position": 2,
+    "Start cutting": 2,
+    "Observe and walk through store": 2,
+    "Move to shelf": 2,
+    "Inspect shelf condition": 2,
+    "Observe colleague and workspace": 2,
+    "Observe workspace": 2,
+    "Approach boxes": 2,
+    "Extract wire hangers from box": 2,
+    "Bundle display hooks": 2,
+    "Move through aisle": 2,
+    "Pick up items from the shopping bag": 2,
+    "Place items on the shelf": 2,
+    "Place item on shelf": 2,
+    "Adjust item on shelf": 2,
+    "Place marked piece down": 2,
+    "Pick up new cardboard piece": 2,
+    "Organize cardboard pieces": 2,
+    "Release cardboard piece and gesture": 2,
+    "Observe and pause": 2,
+    "Gesturing": 2,
+    "Resume observation": 2,
+    "Pick up can": 2,
+    "Hold canned food": 2,
+    "Pick up canned food": 2,
+    "Align canned food on shelf": 2,
+    "Place another canned food on shelf": 2,
+    "Adjust canned food on shelf": 2,
+    "Move hand away from shelf": 2,
+    "Place can on shelf": 2,
+    "Move hand away": 2,
+    "Hold earbud case": 2,
+    "Open earbud case": 2,
+    "Pick up smartphone": 2,
+    "Open paper lantern": 2,
+    "Fold paper lantern": 2,
+    "Grasp lantern": 2,
+    "Grasp lantern component": 2,
+    "Align paper lantern edges": 2,
+    "Adjust lantern string": 2,
+    "Handle paper lantern component": 2,
+    "Open folded paper lantern": 2,
+    "Adjust lantern shape": 2,
+    "Hold paper lantern": 2,
+    "Apply adhesive tape to lantern": 2,
+    "Open paper lantern component": 2,
+    "Expand paper lantern": 2,
+    "Align edges of paper lantern": 2,
+    "Search for puzzle piece": 2,
+    "Release puzzle piece": 2,
+    "Reach for puzzle piece": 2,
+    "Sort puzzle pieces": 2,
+    "Approaching the table": 2,
+    "Preparing to craft": 2,
+    "Picking up crafting material": 2,
+    "Use phone": 2,
+    "Pick up small piece of material": 2,
+    "Manipulate material": 2,
+    "Place material": 2,
+    "Manipulate yellow strip": 2,
+    "Manipulating paper strips": 2,
+    "Manipulate bead": 2,
+    "Manipulate beads": 2,
+    "Place phone down": 2,
+    "Hold and manipulate paper strip": 2,
+    "Sort beads": 2,
+    "Record count": 2,
+    "Counting and organizing beads": 2,
+    "Pick up star bead": 2,
+    "Place and count bead": 2,
+    "Hold beads": 2,
+    "Arrange star beads": 2,
+    "Pick up pen": 2,
+    "Counting star beads": 2,
+    "Retrieving more beads": 2,
+    "Adjust paper": 2,
+    "Gather star beads": 2,
+    "Arrange star beads for counting": 2,
+    "Sort and count beads": 2,
+    "Place smartphone on stand": 1,
+    "Pick up dustpan": 1,
+    "Move dustpan to side": 1,
+    "Move towards the stove": 1,
+    "Open stove pot lid": 1,
+    "Walking towards door": 1,
+    "Picking up bottle": 1,
+    "Wipe kitchen counter": 1,
+    "Rinse cloth in sink": 1,
+    "Move towards kitchen area": 1,
+    "Place cloth on floor": 1,
+    "Reach for cleaning supplies": 1,
+    "Remove cleaning bottle": 1,
+    "Washing hands in sink": 1,
+    "Wiping countertop": 1,
+    "Lift pot lid": 1,
+    "Stir contents": 1,
+    "Place lid back": 1,
+    "Move pot": 1,
+    "Place towel": 1,
+    "Walk towards shelves": 1,
+    "Reach for wire hangers": 1,
+    "Release hook": 1,
+    "Walk towards other aisles": 1,
+    "Reach for another item": 1,
+    "Reach for next item": 1,
+    "Place canned food on shelf": 1,
+    "Reach for next can": 1,
+    "Reach into box": 1,
+    "Retrieve next canned food item": 1,
+    "Reach for next canned food": 1,
+    "Retrieve canned food from box": 1,
+    "Release lantern": 1,
+    "Pick up packaged paper lantern component": 1,
+    "Remove paper lantern part from packaging": 1,
+    "Remove plastic packaging": 1,
+    "Walking across the room": 1,
+    "Use phone while crafting": 1,
+    "Sort beads and write count": 1,
+    "Sort star-shaped beads": 1,
+    "Write count on paper": 1,
+    "Write on paper": 1
+  },
+  "json_validity_rate": 0.875,
+  "action_macro_f1": 0.0026621494447581404,
+  "subtask_accuracy": 0.006696428571428571,
+  "transition_accuracy": 0.8504464285714286,
+  "next_action_accuracy": 0.024553571428571428,
+  "contact_accuracy": 0.6450892857142857,
+  "object_micro_f1": 0.22299431459254582,
+  "caption_window_grounding": {
+    "mrr": null,
+    "recall_at_5": null,
+    "note": "Grounding ranking requires a retrieval candidate set; JSON evidence_window is stored for later scoring."
+  }
+}

results/omni_finetune/verified_public/xperience10m_qwen3_omni_128ep_96train_16val_16test_valmon_20260605_eval/eval/per_class_metrics.csv

@@ -0,0 +1,1211 @@
+class_name,support,predicted,precision,recall,f1
+Adjust Mahjong tile,0,9,0.0,0.0,0.0
+Adjust Mahjong tile alignment,0,0,0.0,0.0,0.0
+Adjust Mahjong tile on the stack,0,0,0.0,0.0,0.0
+Adjust Mahjong tiles,0,0,0.0,0.0,0.0
+Adjust bead piles,0,0,0.0,0.0,0.0
+Adjust canned food on shelf,2,5,0.4,1.0,0.5714285714285715
+Adjust cans in bin,0,0,0.0,0.0,0.0
+Adjust cans in container,0,0,0.0,0.0,0.0
+Adjust cans in tray,0,0,0.0,0.0,0.0
+Adjust cardboard,0,1,0.0,0.0,0.0
+Adjust cardboard divider,0,0,0.0,0.0,0.0
+Adjust cardboard position,0,0,0.0,0.0,0.0
+Adjust container on shelf,0,0,0.0,0.0,0.0
+Adjust container position,0,0,0.0,0.0,0.0
+Adjust containers on shelf,0,0,0.0,0.0,0.0
+Adjust foam strip,0,0,0.0,0.0,0.0
+Adjust grip,0,0,0.0,0.0,0.0
+Adjust grip on container,0,0,0.0,0.0,0.0
+Adjust hand position,0,0,0.0,0.0,0.0
+Adjust item on shelf,2,0,0.0,0.0,0.0
+Adjust lantern shape,2,3,0.0,0.0,0.0
+Adjust lantern string,2,3,0.3333333333333333,0.5,0.4
+Adjust paper,2,0,0.0,0.0,0.0
+Adjust paper strip,0,2,0.0,0.0,0.0
+Adjust perspective,0,0,0.0,0.0,0.0
+Adjust placement on shelf,0,0,0.0,0.0,0.0
+Adjust position,0,0,0.0,0.0,0.0
+Adjust pot position,2,0,0.0,0.0,0.0
+Adjust puzzle piece,3,0,0.0,0.0,0.0
+Adjust red button,0,4,0.0,0.0,0.0
+Adjust red button in row,0,0,0.0,0.0,0.0
+Adjust red button position,0,0,0.0,0.0,0.0
+Adjust retail item position,0,0,0.0,0.0,0.0
+Adjust retail items on shelf,0,0,0.0,0.0,0.0
+Adjust ruler position,0,0,0.0,0.0,0.0
+Adjust smartphone and sort pieces,0,3,0.0,0.0,0.0
+Adjust snack package,0,0,0.0,0.0,0.0
+Adjust tile row alignment,0,0,0.0,0.0,0.0
+Adjust vacuum cleaner position,0,0,0.0,0.0,0.0
+Adjusting a puzzle piece,0,0,0.0,0.0,0.0
+Adjusting canned goods on shelf,0,0,0.0,0.0,0.0
+Adjusting fabric for cutting,0,0,0.0,0.0,0.0
+Adjusting fabric position,0,0,0.0,0.0,0.0
+Adjusting puzzle piece,0,0,0.0,0.0,0.0
+Align Mahjong tiles,0,0,0.0,0.0,0.0
+Align and place retail item,0,0,0.0,0.0,0.0
+Align blue strip,0,0,0.0,0.0,0.0
+Align button,0,0,0.0,0.0,0.0
+Align button in row,0,0,0.0,0.0,0.0
+Align button row,0,0,0.0,0.0,0.0
+Align buttons,0,0,0.0,0.0,0.0
+Align canned food on shelf,2,0,0.0,0.0,0.0
+Align canned goods on shelf,0,0,0.0,0.0,0.0
+Align cardboard piece,0,0,0.0,0.0,0.0
+Align cardboard strip,0,0,0.0,0.0,0.0
+Align charging cable,0,0,0.0,0.0,0.0
+Align edges of paper lantern,2,0,0.0,0.0,0.0
+Align foam piece,0,0,0.0,0.0,0.0
+Align foam strip,0,0,0.0,0.0,0.0
+Align paper lantern edges,2,0,0.0,0.0,0.0
+Align paper strip,0,0,0.0,0.0,0.0
+Align plastic containers,0,0,0.0,0.0,0.0
+Align red button in row,0,0,0.0,0.0,0.0
+Align red buttons,0,0,0.0,0.0,0.0
+Align ruler,0,0,0.0,0.0,0.0
+Align ruler and mark cardboard,0,0,0.0,0.0,0.0
+Align ruler on cardboard,0,0,0.0,0.0,0.0
+Align ruler with crease,0,0,0.0,0.0,0.0
+Align scissors,0,0,0.0,0.0,0.0
+Apply adhesive tape to lantern,2,0,0.0,0.0,0.0
+Approach boxes,2,0,0.0,0.0,0.0
+Approach desk,0,0,0.0,0.0,0.0
+Approach packing area,0,4,0.0,0.0,0.0
+Approach restocking supplies,0,0,0.0,0.0,0.0
+Approach table,0,7,0.0,0.0,0.0
+Approach work table,0,0,0.0,0.0,0.0
+Approach workstation,0,1,0.0,0.0,0.0
+Approaching and pressing the door switch,4,0,0.0,0.0,0.0
+Approaching the table,2,0,0.0,0.0,0.0
+Approaching work table,0,0,0.0,0.0,0.0
+Arrange Mahjong tiles,0,0,0.0,0.0,0.0
+Arrange beads by color,0,0,0.0,0.0,0.0
+Arrange black buttons,0,0,0.0,0.0,0.0
+Arrange button cluster,0,0,0.0,0.0,0.0
+Arrange buttons,4,0,0.0,0.0,0.0
+Arrange buttons in a line,4,0,0.0,0.0,0.0
+Arrange buttons in row,0,0,0.0,0.0,0.0
+Arrange buttons on table,0,0,0.0,0.0,0.0
+Arrange buttons on the table,0,0,0.0,0.0,0.0
+Arrange canned products on shelf,0,0,0.0,0.0,0.0
+Arrange cans in box,0,0,0.0,0.0,0.0
+Arrange cans on shelf,0,0,0.0,0.0,0.0
+Arrange cardboard,0,0,0.0,0.0,0.0
+Arrange cardboard piece,0,0,0.0,0.0,0.0
+Arrange cardboard pieces,0,0,0.0,0.0,0.0
+Arrange cardboard squares,0,0,0.0,0.0,0.0
+Arrange container on shelf,0,0,0.0,0.0,0.0
+Arrange items on shelf,0,0,0.0,0.0,0.0
+Arrange orange buttons,0,0,0.0,0.0,0.0
+Arrange paper stars,0,0,0.0,0.0,0.0
+Arrange paper strips,0,0,0.0,0.0,0.0
+Arrange plastic containers,0,0,0.0,0.0,0.0
+Arrange red buttons,0,0,0.0,0.0,0.0
+Arrange small buttons,0,0,0.0,0.0,0.0
+Arrange star beads,2,0,0.0,0.0,0.0
+Arrange star beads for counting,2,0,0.0,0.0,0.0
+Arrange star-shaped beads,0,0,0.0,0.0,0.0
+Arrange tiles into row,0,0,0.0,0.0,0.0
+Arrive at a different workstation,0,0,0.0,0.0,0.0
+Assemble cardboard pieces,0,0,0.0,0.0,0.0
+Assemble foam strips,0,0,0.0,0.0,0.0
+Assess shelf arrangement,0,0,0.0,0.0,0.0
+Attach foam strip,0,0,0.0,0.0,0.0
+Attach material to paper strip,0,0,0.0,0.0,0.0
+Attempt to fit puzzle piece,3,0,0.0,0.0,0.0
+Begin folding paper strip,0,0,0.0,0.0,0.0
+Begin rolling quilling strip,0,0,0.0,0.0,0.0
+Bend and manipulate plastic strip,4,0,0.0,0.0,0.0
+Browse and interact with phone interface,0,0,0.0,0.0,0.0
+Browse mobile phone,0,0,0.0,0.0,0.0
+Browse smartphone screen,3,4,0.0,0.0,0.0
+Browsing mobile phone,0,0,0.0,0.0,0.0
+Browsing smartphone content,0,0,0.0,0.0,0.0
+Bundle display hooks,2,0,0.0,0.0,0.0
+Cap marker,0,0,0.0,0.0,0.0
+Carry cardboard piece,0,2,0.0,0.0,0.0
+Carry cereal boxes,0,4,0.0,0.0,0.0
+Carry cereal towards aisle,0,0,0.0,0.0,0.0
+Carry container,0,3,0.0,0.0,0.0
+Carry crate of cans,0,0,0.0,0.0,0.0
+Carry item to shelf,0,0,0.0,0.0,0.0
+Carry pasta box towards aisle,0,0,0.0,0.0,0.0
+Carry plastic container,0,0,0.0,0.0,0.0
+Carry stool to next shelf,0,0,0.0,0.0,0.0
+Check phone,0,2,0.0,0.0,0.0
+Check smart watch,0,1,0.0,0.0,0.0
+Check watch,0,0,0.0,0.0,0.0
+Clean shelf,0,0,0.0,0.0,0.0
+Close cardboard box,0,0,0.0,0.0,0.0
+Closing the door,2,0,0.0,0.0,0.0
+Combine bead piles,0,0,0.0,0.0,0.0
+Complete the cut,0,0,0.0,0.0,0.0
+Connect cable to device,0,0,0.0,0.0,0.0
+Continue cutting fabric,0,0,0.0,0.0,0.0
+Continue cutting newspaper,0,0,0.0,0.0,0.0
+Continue folding paper strip,0,0,0.0,0.0,0.0
+Count and arrange paper stars,0,0,0.0,0.0,0.0
+Count and record paper stars,0,0,0.0,0.0,0.0
+Count paper stars,0,0,0.0,0.0,0.0
+Counting and organizing beads,2,0,0.0,0.0,0.0
+Counting star beads,2,0,0.0,0.0,0.0
+Curve foam strip into loop,0,0,0.0,0.0,0.0
+Cut along the edge of the newspaper,0,0,0.0,0.0,0.0
+Cut along the line,0,0,0.0,0.0,0.0
+Cut along the marked line,7,0,0.0,0.0,0.0
+Cut along the newspaper edge,0,0,0.0,0.0,0.0
+Cut cardboard,7,6,0.0,0.0,0.0
+Cut cardboard along line,0,0,0.0,0.0,0.0
+Cut cardboard grid,0,0,0.0,0.0,0.0
+Cut cardboard into triangles,0,0,0.0,0.0,0.0
+Cut cardboard pattern,0,0,0.0,0.0,0.0
+Cut cardboard piece,3,0,0.0,0.0,0.0
+Cut cardboard piece with scissors,0,0,0.0,0.0,0.0
+Cut cardboard pieces with scissors,0,0,0.0,0.0,0.0
+Cut cardboard shape,0,3,0.0,0.0,0.0
+Cut cardboard sheet,0,0,0.0,0.0,0.0
+Cut cardboard sheet with scissors,0,0,0.0,0.0,0.0
+Cut cardboard square,0,0,0.0,0.0,0.0
+Cut cardboard strip,0,0,0.0,0.0,0.0
+Cut cardboard strip with scissors,0,0,0.0,0.0,0.0
+Cut cardboard strip with utility knife,0,0,0.0,0.0,0.0
+Cut cardboard triangle,0,0,0.0,0.0,0.0
+Cut cardboard tube,0,0,0.0,0.0,0.0
+Cut cardboard with scissors,0,1,0.0,0.0,0.0
+Cut cardboard with utility knife,0,12,0.0,0.0,0.0
+Cut fabric with scissors,0,0,0.0,0.0,0.0
+Cut light green fabric,0,0,0.0,0.0,0.0
+Cut newspaper,0,0,0.0,0.0,0.0
+Cut newspaper with scissors,0,0,0.0,0.0,0.0
+Cut out cardboard pattern,0,0,0.0,0.0,0.0
+Cut section from newspaper,0,0,0.0,0.0,0.0
+Cutting fabric,0,0,0.0,0.0,0.0
+Deposit beads into box,0,0,0.0,0.0,0.0
+Deposit cardboard squares,0,0,0.0,0.0,0.0
+Discard item into bin,0,0,0.0,0.0,0.0
+Discard paper towel,0,0,0.0,0.0,0.0
+Draw grid line,0,0,0.0,0.0,0.0
+Draw grid line with pen,0,0,0.0,0.0,0.0
+Draw line,0,0,0.0,0.0,0.0
+Draw line along ruler,0,0,0.0,0.0,0.0
+Draw line on cardboard,0,0,0.0,0.0,0.0
+Draw line with marker,0,0,0.0,0.0,0.0
+Draw line with pen,0,8,0.0,0.0,0.0
+Draw lines on cardboard,0,0,0.0,0.0,0.0
+Draw lines with pen and ruler,0,0,0.0,0.0,0.0
+Draw lines with ruler,0,0,0.0,0.0,0.0
+Draw straight line,0,0,0.0,0.0,0.0
+Draw straight lines on cardboard,0,0,0.0,0.0,0.0
+Drawing grid line,0,0,0.0,0.0,0.0
+Drawing grid line with pen and ruler,0,0,0.0,0.0,0.0
+Drawing grid line with ruler,0,0,0.0,0.0,0.0
+Drawing lines on cardboard,0,0,0.0,0.0,0.0
+Drop cardboard square into box,0,0,0.0,0.0,0.0
+Dry hands,0,0,0.0,0.0,0.0
+Enter the room,0,2,0.0,0.0,0.0
+Enter workspace,0,0,0.0,0.0,0.0
+Entering the VR training room,3,0,0.0,0.0,0.0
+Examine canned goods,0,0,0.0,0.0,0.0
+Examine item,0,0,0.0,0.0,0.0
+Examine labels,0,0,0.0,0.0,0.0
+Examine product,0,0,0.0,0.0,0.0
+Expand paper lantern,2,2,0.0,0.0,0.0
+Extract wire hangers from box,2,0,0.0,0.0,0.0
+Finish placing cardboard cutouts,0,0,0.0,0.0,0.0
+Finish washing hands,0,0,0.0,0.0,0.0
+Finish wiping and inspect jar,0,0,0.0,0.0,0.0
+Finishing coil,0,0,0.0,0.0,0.0
+Fold and manipulate ribbon,0,0,0.0,0.0,0.0
+Fold and organize paper strips,0,0,0.0,0.0,0.0
+Fold blue strip,0,0,0.0,0.0,0.0
+Fold cardboard,0,0,0.0,0.0,0.0
+Fold cardboard edge,0,0,0.0,0.0,0.0
+Fold cardboard shape,0,0,0.0,0.0,0.0
+Fold cardboard sheet,0,0,0.0,0.0,0.0
+Fold cut cardboard,0,0,0.0,0.0,0.0
+Fold foam piece,0,0,0.0,0.0,0.0
+Fold lucky star,0,0,0.0,0.0,0.0
+Fold newspaper,0,0,0.0,0.0,0.0
+Fold paper lantern,2,1,0.0,0.0,0.0
+Fold paper star,0,1,0.0,0.0,0.0
+Fold paper strip,0,5,0.0,0.0,0.0
+Fold paper strip into a star,0,0,0.0,0.0,0.0
+Fold paper strip into knot,0,0,0.0,0.0,0.0
+Fold paper strip into lucky star,0,0,0.0,0.0,0.0
+Fold paper strip into star,0,0,0.0,0.0,0.0
+Fold plastic strip,3,0,0.0,0.0,0.0
+Fold purple paper,0,0,0.0,0.0,0.0
+Fold purple paper strip,0,0,0.0,0.0,0.0
+Fold ribbon,0,0,0.0,0.0,0.0
+Folding cardboard,0,0,0.0,0.0,0.0
+Folding paper strip,0,0,0.0,0.0,0.0
+Forming quilled paper shape,0,0,0.0,0.0,0.0
+Gather cardboard pieces,0,0,0.0,0.0,0.0
+Gather pieces,0,0,0.0,0.0,0.0
+Gather pieces into box,0,0,0.0,0.0,0.0
+Gather star beads,2,0,0.0,0.0,0.0
+Gathering colored beads,0,0,0.0,0.0,0.0
+Gathering items,0,0,0.0,0.0,0.0
+Gathering star beads,0,0,0.0,0.0,0.0
+Gesturing,2,1,0.0,0.0,0.0
+Grasp and retrieve item,0,0,0.0,0.0,0.0
+Grasp cardboard sheet,0,0,0.0,0.0,0.0
+Grasp cleaning bottle,2,0,0.0,0.0,0.0
+Grasp door handle,0,0,0.0,0.0,0.0
+Grasp electronic object,0,0,0.0,0.0,0.0
+Grasp item,0,0,0.0,0.0,0.0
+Grasp lantern,2,0,0.0,0.0,0.0
+Grasp lantern component,2,0,0.0,0.0,0.0
+Grasp next item,0,0,0.0,0.0,0.0
+Grasp origami stars,0,0,0.0,0.0,0.0
+Grasp package,0,0,0.0,0.0,0.0
+Grasp paper strip,0,0,0.0,0.0,0.0
+Grasp plastic bag on shelf,0,0,0.0,0.0,0.0
+Grasp product from box,0,0,0.0,0.0,0.0
+Grasp product from shelf,0,0,0.0,0.0,0.0
+Grasp retail item,0,0,0.0,0.0,0.0
+Grasp shopping bag,0,0,0.0,0.0,0.0
+Grasp snack package,0,0,0.0,0.0,0.0
+Grasping cleaning cloth,2,0,0.0,0.0,0.0
+Greeting/acknowledging participants,3,0,0.0,0.0,0.0
+Guide utility knife along ruler,0,0,0.0,0.0,0.0
+Handle paper lantern component,2,0,0.0,0.0,0.0
+Hold and align cardboard,0,0,0.0,0.0,0.0
+Hold and align newspaper,0,0,0.0,0.0,0.0
+Hold and align paper strip,0,0,0.0,0.0,0.0
+Hold and bend paper strip,0,0,0.0,0.0,0.0
+Hold and bend plastic strip,3,0,0.0,0.0,0.0
+Hold and crease purple paper,0,0,0.0,0.0,0.0
+Hold and examine item,0,0,0.0,0.0,0.0
+Hold and inspect can,0,0,0.0,0.0,0.0
+Hold and manipulate paper strip,2,0,0.0,0.0,0.0
+Hold and mark cardboard piece,0,0,0.0,0.0,0.0
+Hold and rotate paper strip,0,0,0.0,0.0,0.0
+Hold and view phone,0,0,0.0,0.0,0.0
+Hold and wipe product,0,0,0.0,0.0,0.0
+Hold beads,2,0,0.0,0.0,0.0
+Hold bin and move through aisle,0,0,0.0,0.0,0.0
+Hold blue product box,0,0,0.0,0.0,0.0
+Hold blue strip,0,0,0.0,0.0,0.0
+Hold canned food,2,0,0.0,0.0,0.0
+Hold cardboard,0,0,0.0,0.0,0.0
+Hold cardboard piece,3,0,0.0,0.0,0.0
+Hold cardboard pieces,0,0,0.0,0.0,0.0
+Hold cardboard strip,0,0,0.0,0.0,0.0
+Hold cardboard with ruler,0,0,0.0,0.0,0.0
+Hold charger,0,0,0.0,0.0,0.0
+Hold charger and cable,0,0,0.0,0.0,0.0
+Hold charging cable,0,0,0.0,0.0,0.0
+Hold cleaning cloth,0,0,0.0,0.0,0.0
+Hold container,0,0,0.0,0.0,0.0
+Hold container lid,2,0,0.0,0.0,0.0
+Hold container of canned food,0,0,0.0,0.0,0.0
+Hold craft tool,0,0,0.0,0.0,0.0
+Hold device and cable,0,0,0.0,0.0,0.0
+Hold earbud case,2,0,0.0,0.0,0.0
+Hold electronic accessory,0,0,0.0,0.0,0.0
+Hold electronic item,0,0,0.0,0.0,0.0
+Hold empty container,0,0,0.0,0.0,0.0
+Hold foam pieces,0,0,0.0,0.0,0.0
+Hold instructional sign,0,0,0.0,0.0,0.0
+Hold item,0,0,0.0,0.0,0.0
+Hold item and adjust posture,0,0,0.0,0.0,0.0
+Hold items,0,0,0.0,0.0,0.0
+Hold items and inspect shelf,0,0,0.0,0.0,0.0
+Hold items in hand,0,0,0.0,0.0,0.0
+Hold newspaper,0,0,0.0,0.0,0.0
+Hold paper lantern,2,5,0.2,0.5,0.28571428571428575
+Hold paper strip,0,24,0.0,0.0,0.0
+Hold pen and paper,0,0,0.0,0.0,0.0
+Hold phone,0,0,0.0,0.0,0.0
+Hold pickle jar,0,0,0.0,0.0,0.0
+Hold portable charger,0,0,0.0,0.0,0.0
+Hold power adapter,0,0,0.0,0.0,0.0
+Hold power bank and cable,0,0,0.0,0.0,0.0
+Hold product,0,0,0.0,0.0,0.0
+Hold product labels,0,0,0.0,0.0,0.0
+Hold product package,0,0,0.0,0.0,0.0
+Hold quilled paper coil,0,0,0.0,0.0,0.0
+Hold quilled paper piece,0,0,0.0,0.0,0.0
+Hold quilling paper,0,0,0.0,0.0,0.0
+Hold recording sheet and pen,0,0,0.0,0.0,0.0
+Hold ruler,0,0,0.0,0.0,0.0
+Hold ruler and draw line,0,0,0.0,0.0,0.0
+Hold ruler and mark cardboard,0,0,0.0,0.0,0.0
+Hold ruler and marker,0,0,0.0,0.0,0.0
+Hold ruler and pen steady,0,0,0.0,0.0,0.0
+Hold ruler on cardboard,0,0,0.0,0.0,0.0
+Hold ruler steady,0,0,0.0,0.0,0.0
+Hold scissors,0,0,0.0,0.0,0.0
+Hold small cardboard pieces,0,0,0.0,0.0,0.0
+Hold small object,0,0,0.0,0.0,0.0
+Hold small piece of ribbon,0,0,0.0,0.0,0.0
+Hold small product bag,0,0,0.0,0.0,0.0
+Hold small white box,0,0,0.0,0.0,0.0
+Hold smartphone,4,42,0.07142857142857142,0.75,0.13043478260869565
+Hold smartphone box,0,0,0.0,0.0,0.0
+Hold snack package,0,0,0.0,0.0,0.0
+Hold snack packages,0,0,0.0,0.0,0.0
+Hold supplement bottle,0,0,0.0,0.0,0.0
+Hold tray of canned goods,0,0,0.0,0.0,0.0
+Hold utility knife,0,0,0.0,0.0,0.0
+Hold water bottle,0,0,0.0,0.0,0.0
+Holding marker,0,1,0.0,0.0,0.0
+Identify next cardboard piece,3,0,0.0,0.0,0.0
+Inflate paper star,0,0,0.0,0.0,0.0
+Initiate star folding,0,0,0.0,0.0,0.0
+Insert charging cable,0,0,0.0,0.0,0.0
+Insert charging cable into power bank,0,0,0.0,0.0,0.0
+Insert plug into power adapter,0,0,0.0,0.0,0.0
+Inspect Dior gift box,0,0,0.0,0.0,0.0
+Inspect almond package,0,0,0.0,0.0,0.0
+Inspect and place item on shelf,0,0,0.0,0.0,0.0
+Inspect bottle,0,0,0.0,0.0,0.0
+Inspect cardboard piece,0,0,0.0,0.0,0.0
+Inspect cardboard strip,0,0,0.0,0.0,0.0
+Inspect charging case,0,0,0.0,0.0,0.0
+Inspect electronic item,0,0,0.0,0.0,0.0
+Inspect jar,0,0,0.0,0.0,0.0
+Inspect product,0,0,0.0,0.0,0.0
+Inspect product lid,0,0,0.0,0.0,0.0
+Inspect shelf,0,0,0.0,0.0,0.0
+Inspect shelf and organize stock,0,0,0.0,0.0,0.0
+Inspect shelf condition,2,0,0.0,0.0,0.0
+Inspect smartphone box,0,0,0.0,0.0,0.0
+Inspect strip,0,0,0.0,0.0,0.0
+Inspect supplement bottle,0,0,0.0,0.0,0.0
+Interact with colleagues,0,0,0.0,0.0,0.0
+Interact with phone,0,0,0.0,0.0,0.0
+Interact with smartphone,4,0,0.0,0.0,0.0
+Interact with smartphone screen,0,0,0.0,0.0,0.0
+Interacting with phone screen,0,0,0.0,0.0,0.0
+Interaction with coworker,0,0,0.0,0.0,0.0
+Interlock paper strips,0,0,0.0,0.0,0.0
+Labeling cardboard piece,0,0,0.0,0.0,0.0
+Labeling cardboard square,0,0,0.0,0.0,0.0
+Labeling cardboard squares,0,0,0.0,0.0,0.0
+Lift blue strip,0,0,0.0,0.0,0.0
+Lift pen and shift ruler,0,0,0.0,0.0,0.0
+Lift pot lid,1,1,1.0,1.0,1.0
+Lift utility knife,0,0,0.0,0.0,0.0
+Lock phone,0,0,0.0,0.0,0.0
+Look around the table,0,0,0.0,0.0,0.0
+Look away,0,0,0.0,0.0,0.0
+Manipulate adhesive strip,5,0,0.0,0.0,0.0
+Manipulate and inspect colorful pieces,0,0,0.0,0.0,0.0
+Manipulate bead,2,0,0.0,0.0,0.0
+Manipulate beads,2,0,0.0,0.0,0.0
+Manipulate cardboard piece,0,0,0.0,0.0,0.0
+Manipulate cardboard shape,0,0,0.0,0.0,0.0
+Manipulate cardboard sheet,0,0,0.0,0.0,0.0
+Manipulate colorful pieces,0,0,0.0,0.0,0.0
+Manipulate component,0,0,0.0,0.0,0.0
+Manipulate component on strip,0,0,0.0,0.0,0.0
+Manipulate craft paper strips,4,0,0.0,0.0,0.0
+Manipulate craft piece,4,0,0.0,0.0,0.0
+Manipulate folded paper star,0,0,0.0,0.0,0.0
+Manipulate light blue strip,0,0,0.0,0.0,0.0
+Manipulate material,2,0,0.0,0.0,0.0
+Manipulate paper decoration,5,0,0.0,0.0,0.0
+Manipulate paper edge,5,0,0.0,0.0,0.0
+Manipulate paper piece,0,0,0.0,0.0,0.0
+Manipulate paper quilling piece,0,0,0.0,0.0,0.0
+Manipulate paper star,0,3,0.0,0.0,0.0
+Manipulate paper stars,0,0,0.0,0.0,0.0
+Manipulate paper strip,14,9,0.1111111111111111,0.07142857142857142,0.08695652173913043
+Manipulate paper strips,0,0,0.0,0.0,0.0
+Manipulate plastic strip,3,0,0.0,0.0,0.0
+Manipulate plastic strips,3,0,0.0,0.0,0.0
+Manipulate power cable plug,0,0,0.0,0.0,0.0
+Manipulate puzzle piece,3,0,0.0,0.0,0.0
+Manipulate puzzle pieces,3,0,0.0,0.0,0.0
+Manipulate quilled paper,0,0,0.0,0.0,0.0
+Manipulate quilled paper shape,0,0,0.0,0.0,0.0
+Manipulate quilled paper strip,0,0,0.0,0.0,0.0
+Manipulate quilled paper strips,0,0,0.0,0.0,0.0
+Manipulate quilling paper,0,0,0.0,0.0,0.0
+Manipulate quilling strip,0,0,0.0,0.0,0.0
+Manipulate ribbon knot,0,0,0.0,0.0,0.0
+Manipulate ribbon piece,0,0,0.0,0.0,0.0
+Manipulate small component,0,0,0.0,0.0,0.0
+Manipulate small object,0,0,0.0,0.0,0.0
+Manipulate small paper segment,0,0,0.0,0.0,0.0
+Manipulate star,0,0,0.0,0.0,0.0
+Manipulate yellow strip,2,0,0.0,0.0,0.0
+Manipulating paper strips,2,0,0.0,0.0,0.0
+Mark cardboard,0,2,0.0,0.0,0.0
+Mark cardboard piece,3,0,0.0,0.0,0.0
+Mark cardboard strip with pen,0,0,0.0,0.0,0.0
+Mark cardboard with marker,0,3,0.0,0.0,0.0
+Mark cardboard with pen,0,0,0.0,0.0,0.0
+Mark cardboard with pen and ruler,0,0,0.0,0.0,0.0
+Mark cardboard with ruler,0,0,0.0,0.0,0.0
+Mark cardboard with ruler and pen,0,0,0.0,0.0,0.0
+Mark fabric,0,0,0.0,0.0,0.0
+Mark fabric with pen,0,0,0.0,0.0,0.0
+Mark fabric with pen and ruler,0,0,0.0,0.0,0.0
+Mark line on cardboard,0,0,0.0,0.0,0.0
+Mark lines on cardboard,0,0,0.0,0.0,0.0
+Mark lines with marker,0,0,0.0,0.0,0.0
+Mark lines with pen along ruler,0,0,0.0,0.0,0.0
+Mark list with pen,0,0,0.0,0.0,0.0
+Mark paper list,0,0,0.0,0.0,0.0
+Mark straight line,0,0,0.0,0.0,0.0
+Marking cardboard piece,3,0,0.0,0.0,0.0
+Marking cardboard with pen,0,0,0.0,0.0,0.0
+Marking lines on cardboard,0,0,0.0,0.0,0.0
+Measure and mark cardboard,0,0,0.0,0.0,0.0
+Measure cardboard with ruler,0,0,0.0,0.0,0.0
+Move Mahjong tile,0,0,0.0,0.0,0.0
+Move along shelf,0,0,0.0,0.0,0.0
+Move along the shelf,0,0,0.0,0.0,0.0
+Move along the shelves,0,0,0.0,0.0,0.0
+Move along the supermarket aisle,0,0,0.0,0.0,0.0
+Move and place black buttons,0,0,0.0,0.0,0.0
+Move away from collection box,0,0,0.0,0.0,0.0
+Move away from desk,0,0,0.0,0.0,0.0
+Move away from shelf,0,0,0.0,0.0,0.0
+Move away from table,0,0,0.0,0.0,0.0
+Move away from workstation,0,0,0.0,0.0,0.0
+Move bin,0,0,0.0,0.0,0.0
+Move bin to shelf area,0,0,0.0,0.0,0.0
+Move black button,0,0,0.0,0.0,0.0
+Move blue beads,0,0,0.0,0.0,0.0
+Move box to next position,0,0,0.0,0.0,0.0
+Move button to line,0,0,0.0,0.0,0.0
+Move camera over surface,0,0,0.0,0.0,0.0
+Move can towards shelf,0,0,0.0,0.0,0.0
+Move canned goods container,0,0,0.0,0.0,0.0
+Move cardboard,0,0,0.0,0.0,0.0
+Move cardboard box,0,0,0.0,0.0,0.0
+Move cardboard piece,0,0,0.0,0.0,0.0
+Move cardboard sheet,0,0,0.0,0.0,0.0
+Move cardboard to pile,0,0,0.0,0.0,0.0
+Move container toward shelf,0,0,0.0,0.0,0.0
+Move dustpan to side,1,0,0.0,0.0,0.0
+Move hand,0,0,0.0,0.0,0.0
+Move hand away,2,0,0.0,0.0,0.0
+Move hand away from shelf,2,0,0.0,0.0,0.0
+Move hand away from workspace,0,0,0.0,0.0,0.0
+Move hand back to box,0,0,0.0,0.0,0.0
+Move hand over button pile,0,0,0.0,0.0,0.0
+Move hand to paper stars,0,0,0.0,0.0,0.0
+Move hand toward craft materials,0,0,0.0,0.0,0.0
+Move item to bag,0,0,0.0,0.0,0.0
+Move marker and adjust hand,3,0,0.0,0.0,0.0
+Move marker and ruler,0,0,0.0,0.0,0.0
+Move marker away,0,0,0.0,0.0,0.0
+Move orange buttons,0,0,0.0,0.0,0.0
+Move origami stars,0,0,0.0,0.0,0.0
+Move pen,0,0,0.0,0.0,0.0
+Move pen aside,0,0,0.0,0.0,0.0
+Move pen away,0,0,0.0,0.0,0.0
+Move phone,7,0,0.0,0.0,0.0
+Move piece to pile,0,0,0.0,0.0,0.0
+Move pieces into box,0,0,0.0,0.0,0.0
+Move pineapple chips,0,0,0.0,0.0,0.0
+Move plastic storage bin,0,0,0.0,0.0,0.0
+Move plush toy,0,0,0.0,0.0,0.0
+Move pot,1,0,0.0,0.0,0.0
+Move product to box,0,0,0.0,0.0,0.0
+Move product to shelf,0,0,0.0,0.0,0.0
+Move product towards shelf,0,0,0.0,0.0,0.0
+Move puzzle piece,0,0,0.0,0.0,0.0
+Move ruler,0,0,0.0,0.0,0.0
+Move ruler and tools,0,0,0.0,0.0,0.0
+Move scissors away,0,0,0.0,0.0,0.0
+Move small blue foam piece towards the strip,0,0,0.0,0.0,0.0
+Move smartphone,3,0,0.0,0.0,0.0
+Move storage bin,0,0,0.0,0.0,0.0
+Move through aisle,2,0,0.0,0.0,0.0
+Move through the training room,3,0,0.0,0.0,0.0
+Move to box,0,0,0.0,0.0,0.0
+Move to desk,0,0,0.0,0.0,0.0
+Move to next section,0,0,0.0,0.0,0.0
+Move to shelf,2,0,0.0,0.0,0.0
+Move to shelf base,0,0,0.0,0.0,0.0
+Move to stock products,0,0,0.0,0.0,0.0
+Move towards aisle,0,0,0.0,0.0,0.0
+Move towards box,0,0,0.0,0.0,0.0
+Move towards kitchen area,1,0,0.0,0.0,0.0
+Move towards shelf,0,0,0.0,0.0,0.0
+Move towards table,0,0,0.0,0.0,0.0
+Move towards the stove,1,0,0.0,0.0,0.0
+Move tray towards packing area,0,0,0.0,0.0,0.0
+Move utility knife along ruler,0,0,0.0,0.0,0.0
+Move vacuum cleaner,0,0,0.0,0.0,0.0
+Move vacuum cleaner hose,0,0,0.0,0.0,0.0
+Moving cardboard square,0,0,0.0,0.0,0.0
+Moving hand,0,0,0.0,0.0,0.0
+Moving hand towards cardboard stack,0,0,0.0,0.0,0.0
+Moving ruler,0,0,0.0,0.0,0.0
+Observe and pause,2,1,0.0,0.0,0.0
+Observe and walk through store,2,3,0.0,0.0,0.0
+Observe colleague and workspace,2,0,0.0,0.0,0.0
+Observe craft layout,0,0,0.0,0.0,0.0
+Observe desktop layout,0,0,0.0,0.0,0.0
+Observe paper and count objects,0,0,0.0,0.0,0.0
+Observe paper quilling station,0,0,0.0,0.0,0.0
+Observe puzzle progress,3,0,0.0,0.0,0.0
+Observe room,0,1,0.0,0.0,0.0
+Observe shelf,0,0,0.0,0.0,0.0
+Observe shelf status,0,0,0.0,0.0,0.0
+Observe sorting progress,0,0,0.0,0.0,0.0
+Observe stocking,0,0,0.0,0.0,0.0
+Observe surroundings,0,2,0.0,0.0,0.0
+Observe workspace,2,1,0.0,0.0,0.0
+Open cardboard box,0,0,0.0,0.0,0.0
+Open door,0,0,0.0,0.0,0.0
+Open earbud case,2,0,0.0,0.0,0.0
+Open folded paper lantern,2,0,0.0,0.0,0.0
+Open paper lantern,2,0,0.0,0.0,0.0
+Open paper lantern component,2,0,0.0,0.0,0.0
+Open small case,0,0,0.0,0.0,0.0
+Open stove pot lid,1,0,0.0,0.0,0.0
+Open supplement bottle,0,0,0.0,0.0,0.0
+Operate smartphone,4,0,0.0,0.0,0.0
+Organize bag contents,0,0,0.0,0.0,0.0
+Organize cardboard pieces,2,0,0.0,0.0,0.0
+Organize item on shelf,0,0,0.0,0.0,0.0
+Organize products,0,0,0.0,0.0,0.0
+Organize snacks in box,0,0,0.0,0.0,0.0
+Organize tools and materials,0,0,0.0,0.0,0.0
+Pack beads into box,0,0,0.0,0.0,0.0
+Peel blue strip,0,0,0.0,0.0,0.0
+Peel foam strip,0,0,0.0,0.0,0.0
+Pick up Dior gift box,0,0,0.0,0.0,0.0
+Pick up Mahjong tile,0,0,0.0,0.0,0.0
+Pick up accessory,0,0,0.0,0.0,0.0
+Pick up and sort cardboard,0,0,0.0,0.0,0.0
+Pick up another bottle,0,0,0.0,0.0,0.0
+Pick up another canned item,0,0,0.0,0.0,0.0
+Pick up another item,0,0,0.0,0.0,0.0
+Pick up beads,0,0,0.0,0.0,0.0
+Pick up black button,0,0,0.0,0.0,0.0
+Pick up blue foam piece,0,0,0.0,0.0,0.0
+Pick up blue paper strip,0,0,0.0,0.0,0.0
+Pick up bottle,0,0,0.0,0.0,0.0
+Pick up bottled sauce,0,0,0.0,0.0,0.0
+Pick up button,3,0,0.0,0.0,0.0
+Pick up can,2,0,0.0,0.0,0.0
+Pick up canned food,2,1,0.0,0.0,0.0
+Pick up canned good,0,0,0.0,0.0,0.0
+Pick up canned goods,0,0,0.0,0.0,0.0
+Pick up canned item,0,0,0.0,0.0,0.0
+Pick up canned product,0,0,0.0,0.0,0.0
+Pick up cardboard,0,0,0.0,0.0,0.0
+Pick up cardboard cutout,0,0,0.0,0.0,0.0
+Pick up cardboard piece,0,1,0.0,0.0,0.0
+Pick up cardboard square,0,0,0.0,0.0,0.0
+Pick up cardboard stack,0,0,0.0,0.0,0.0
+Pick up cardboard strip,0,0,0.0,0.0,0.0
+Pick up cardboard tray,0,0,0.0,0.0,0.0
+Pick up cereal boxes,0,0,0.0,0.0,0.0
+Pick up charging cable,0,0,0.0,0.0,0.0
+Pick up charging case,0,0,0.0,0.0,0.0
+Pick up cleaning cloth,0,0,0.0,0.0,0.0
+Pick up colored tile,0,0,0.0,0.0,0.0
+Pick up container,0,0,0.0,0.0,0.0
+Pick up container from box,0,0,0.0,0.0,0.0
+Pick up craft material,0,0,0.0,0.0,0.0
+Pick up cut cardboard piece,0,0,0.0,0.0,0.0
+Pick up dustpan,1,0,0.0,0.0,0.0
+Pick up electronic accessory,0,0,0.0,0.0,0.0
+Pick up electronic accessory from box,0,0,0.0,0.0,0.0
+Pick up electronic device,0,0,0.0,0.0,0.0
+Pick up electronic item,0,0,0.0,0.0,0.0
+Pick up electronic product,0,0,0.0,0.0,0.0
+Pick up food item,0,0,0.0,0.0,0.0
+Pick up gift box,0,0,0.0,0.0,0.0
+Pick up grocery item,0,0,0.0,0.0,0.0
+Pick up item,0,0,0.0,0.0,0.0
+Pick up item from bin,0,0,0.0,0.0,0.0
+Pick up item from box,0,0,0.0,0.0,0.0
+Pick up item from shelf,0,0,0.0,0.0,0.0
+Pick up items from the shopping bag,2,0,0.0,0.0,0.0
+Pick up jar,0,4,0.0,0.0,0.0
+Pick up light blue strip,0,0,0.0,0.0,0.0
+Pick up marker,0,0,0.0,0.0,0.0
+Pick up metal ruler,0,0,0.0,0.0,0.0
+Pick up new cardboard piece,2,0,0.0,0.0,0.0
+Pick up new electronic product,0,0,0.0,0.0,0.0
+Pick up new product from box,0,0,0.0,0.0,0.0
+Pick up next gift box,0,0,0.0,0.0,0.0
+Pick up next item from bin,0,0,0.0,0.0,0.0
+Pick up next product from bin,0,0,0.0,0.0,0.0
+Pick up nut bar box,0,0,0.0,0.0,0.0
+Pick up object,0,0,0.0,0.0,0.0
+Pick up oil bottle,0,0,0.0,0.0,0.0
+Pick up orange button,0,0,0.0,0.0,0.0
+Pick up pack from shelf,0,0,0.0,0.0,0.0
+Pick up packaged paper lantern component,1,0,0.0,0.0,0.0
+Pick up paper star,0,2,0.0,0.0,0.0
+Pick up paper strip,0,0,0.0,0.0,0.0
+Pick up paper towel,0,0,0.0,0.0,0.0
+Pick up pasta box,0,0,0.0,0.0,0.0
+Pick up pen,2,0,0.0,0.0,0.0
+Pick up phone,0,0,0.0,0.0,0.0
+Pick up pickle jar,0,0,0.0,0.0,0.0
+Pick up pink water bottle,0,0,0.0,0.0,0.0
+Pick up plastic bin,0,0,0.0,0.0,0.0
+Pick up plastic container,0,0,0.0,0.0,0.0
+Pick up plush toy,0,0,0.0,0.0,0.0
+Pick up portable charger,0,0,0.0,0.0,0.0
+Pick up power bank,0,0,0.0,0.0,0.0
+Pick up product,0,0,0.0,0.0,0.0
+Pick up product box,0,0,0.0,0.0,0.0
+Pick up product from bin,0,0,0.0,0.0,0.0
+Pick up product from box,0,0,0.0,0.0,0.0
+Pick up product from shelf,0,0,0.0,0.0,0.0
+Pick up puzzle piece,3,0,0.0,0.0,0.0
+Pick up red button,0,0,0.0,0.0,0.0
+Pick up retail item,0,0,0.0,0.0,0.0
+Pick up sauce bottle,0,0,0.0,0.0,0.0
+Pick up scissors,0,0,0.0,0.0,0.0
+Pick up shopping bag,0,0,0.0,0.0,0.0
+Pick up small cardboard piece,0,0,0.0,0.0,0.0
+Pick up small item,0,0,0.0,0.0,0.0
+Pick up small object,0,0,0.0,0.0,0.0
+Pick up small piece of material,2,0,0.0,0.0,0.0
+Pick up smartphone,2,0,0.0,0.0,0.0
+Pick up snack package,0,0,0.0,0.0,0.0
+Pick up snack packages,0,0,0.0,0.0,0.0
+Pick up snack packs,0,0,0.0,0.0,0.0
+Pick up snack pouch,0,0,0.0,0.0,0.0
+Pick up spice jar,0,0,0.0,0.0,0.0
+Pick up stapler,0,0,0.0,0.0,0.0
+Pick up star,0,0,0.0,0.0,0.0
+Pick up star bead,2,0,0.0,0.0,0.0
+Pick up star-shaped bead,0,0,0.0,0.0,0.0
+Pick up storage container,0,0,0.0,0.0,0.0
+Pick up supplement bottle,0,0,0.0,0.0,0.0
+Pick up supplies from box,0,0,0.0,0.0,0.0
+Pick up tin can,0,0,0.0,0.0,0.0
+Pick up tool,0,0,0.0,0.0,0.0
+Pick up utility knife,6,0,0.0,0.0,0.0
+Pick up water bottle,0,0,0.0,0.0,0.0
+Pick up yellow item,0,0,0.0,0.0,0.0
+Pick up yellow paper strip,0,0,0.0,0.0,0.0
+Picking up bottle,1,0,0.0,0.0,0.0
+Picking up crafting material,2,0,0.0,0.0,0.0
+Picking up stock,0,0,0.0,0.0,0.0
+Place Mahjong tile on stack,0,0,0.0,0.0,0.0
+Place Mahjong tile on the stack,0,0,0.0,0.0,0.0
+Place accessory box,0,0,0.0,0.0,0.0
+Place accessory into box,0,0,0.0,0.0,0.0
+Place accessory on shelf,0,0,0.0,0.0,0.0
+Place and align button,0,0,0.0,0.0,0.0
+Place and count bead,2,0,0.0,0.0,0.0
+Place another canned food on shelf,2,0,0.0,0.0,0.0
+Place back Dior gift box,0,0,0.0,0.0,0.0
+Place bead on table,0,0,0.0,0.0,0.0
+Place bottle back on shelf,0,0,0.0,0.0,0.0
+Place box on shelf,0,0,0.0,0.0,0.0
+Place button,3,0,0.0,0.0,0.0
+Place button in group,0,0,0.0,0.0,0.0
+Place button in row,0,0,0.0,0.0,0.0
+Place can on shelf,2,1,1.0,0.5,0.6666666666666666
+Place canned food in bin,0,0,0.0,0.0,0.0
+Place canned food in container,0,0,0.0,0.0,0.0
+Place canned food on shelf,1,0,0.0,0.0,0.0
+Place canned good on shelf,0,0,0.0,0.0,0.0
+Place canned goods in container,0,0,0.0,0.0,0.0
+Place canned product on shelf,0,0,0.0,0.0,0.0
+Place cans into box,0,0,0.0,0.0,0.0
+Place cardboard,0,0,0.0,0.0,0.0
+Place cardboard piece,0,0,0.0,0.0,0.0
+Place cardboard piece on stack,0,6,0.0,0.0,0.0
+Place cardboard square,0,0,0.0,0.0,0.0
+Place cardboard square on stack,0,0,0.0,0.0,0.0
+Place cardboard strip,0,0,0.0,0.0,0.0
+Place charger on table,0,0,0.0,0.0,0.0
+Place charging case down,0,0,0.0,0.0,0.0
+Place cloth on floor,1,0,0.0,0.0,0.0
+Place colored tile,0,0,0.0,0.0,0.0
+Place container in bin,0,0,0.0,0.0,0.0
+Place container on floor,0,0,0.0,0.0,0.0
+Place container on shelf,0,0,0.0,0.0,0.0
+Place controller on table,0,0,0.0,0.0,0.0
+Place crate on floor,0,0,0.0,0.0,0.0
+Place device on lap,0,0,0.0,0.0,0.0
+Place down paper pieces,0,0,0.0,0.0,0.0
+Place down paper segment,0,0,0.0,0.0,0.0
+Place down pen,0,0,0.0,0.0,0.0
+Place down pink water bottle,0,0,0.0,0.0,0.0
+Place down ruler and pen,0,0,0.0,0.0,0.0
+Place down scissors,0,0,0.0,0.0,0.0
+Place down strip,0,0,0.0,0.0,0.0
+Place finished star on table,0,0,0.0,0.0,0.0
+Place gift box into bin,0,0,0.0,0.0,0.0
+Place gift box on shelf,0,0,0.0,0.0,0.0
+Place hand on table,3,0,0.0,0.0,0.0
+Place item back,0,0,0.0,0.0,0.0
+Place item back on shelf,0,0,0.0,0.0,0.0
+Place item in bag,0,0,0.0,0.0,0.0
+Place item in container,0,0,0.0,0.0,0.0
+Place item in shopping bag,0,0,0.0,0.0,0.0
+Place item into bag,0,0,0.0,0.0,0.0
+Place item into shopping bag,0,0,0.0,0.0,0.0
+Place item on shelf,2,0,0.0,0.0,0.0
+Place item on table,0,0,0.0,0.0,0.0
+Place items on shelf,0,2,0.0,0.0,0.0
+Place items on table,0,0,0.0,0.0,0.0
+Place items on the shelf,2,0,0.0,0.0,0.0
+Place jar in box,0,0,0.0,0.0,0.0
+Place jar into shelf box,0,0,0.0,0.0,0.0
+Place jar on shelf,0,22,0.0,0.0,0.0
+Place ketchup bottle on shelf,0,0,0.0,0.0,0.0
+Place knife down,0,0,0.0,0.0,0.0
+Place lid back,1,0,0.0,0.0,0.0
+Place marked piece down,2,0,0.0,0.0,0.0
+Place marker down,0,0,0.0,0.0,0.0
+Place material,2,0,0.0,0.0,0.0
+Place oil in container,0,0,0.0,0.0,0.0
+Place paper star,0,40,0.0,0.0,0.0
+Place paper star in row,0,0,0.0,0.0,0.0
+Place pen on cardboard,0,0,0.0,0.0,0.0
+Place pen on table,0,0,0.0,0.0,0.0
+Place phone down,2,0,0.0,0.0,0.0
+Place phone on desk,0,0,0.0,0.0,0.0
+Place phone on shelf,0,0,0.0,0.0,0.0
+Place phone on table,0,0,0.0,0.0,0.0
+Place pickle jar in box,0,0,0.0,0.0,0.0
+Place piece into puzzle,3,0,0.0,0.0,0.0
+Place plush toy into bag,0,0,0.0,0.0,0.0
+Place plush toy on shelf,0,0,0.0,0.0,0.0
+Place product in box,0,0,0.0,0.0,0.0
+Place product on shelf,0,0,0.0,0.0,0.0
+Place puzzle piece,3,0,0.0,0.0,0.0
+Place quilled paper shape,0,0,0.0,0.0,0.0
+Place red button,0,2,0.0,0.0,0.0
+Place ribbon onto project,0,0,0.0,0.0,0.0
+Place ruler on cardboard,0,0,0.0,0.0,0.0
+Place sauce bottle on shelf,0,0,0.0,0.0,0.0
+Place sauce in container,0,0,0.0,0.0,0.0
+Place scissors aside,0,0,0.0,0.0,0.0
+Place scissors down,0,0,0.0,0.0,0.0
+Place scissors on table,0,0,0.0,0.0,0.0
+Place smartphone down,3,0,0.0,0.0,0.0
+Place smartphone on cardboard,0,0,0.0,0.0,0.0
+Place smartphone on desk,0,0,0.0,0.0,0.0
+Place smartphone on stand,1,0,0.0,0.0,0.0
+Place smartphone on table,0,3,0.0,0.0,0.0
+Place snack in box,0,0,0.0,0.0,0.0
+Place snack on shelf,0,0,0.0,0.0,0.0
+Place snack package in box,0,0,0.0,0.0,0.0
+Place snack package on shelf,0,2,0.0,0.0,0.0
+Place snack packages on shelf,0,0,0.0,0.0,0.0
+Place snack pouch in container,0,0,0.0,0.0,0.0
+Place snack pouch on shelf,0,0,0.0,0.0,0.0
+Place spice jar in container,0,0,0.0,0.0,0.0
+Place star,0,0,0.0,0.0,0.0
+Place star in row,0,0,0.0,0.0,0.0
+Place star on table,0,0,0.0,0.0,0.0
+Place stars in container,0,0,0.0,0.0,0.0
+Place stool on floor,0,0,0.0,0.0,0.0
+Place storage container on floor,0,0,0.0,0.0,0.0
+Place strip on table,0,0,0.0,0.0,0.0
+Place supplement bottle in container,0,0,0.0,0.0,0.0
+Place tool on table,0,0,0.0,0.0,0.0
+Place towel,1,0,0.0,0.0,0.0
+Place water bottle on table,0,0,0.0,0.0,0.0
+Place white box on table,0,0,0.0,0.0,0.0
+Placing labeled cardboard square,0,0,0.0,0.0,0.0
+Placing labeled square,0,0,0.0,0.0,0.0
+Placing paper strip,4,0,0.0,0.0,0.0
+Placing pen on table,0,0,0.0,0.0,0.0
+Placing phone down,0,0,0.0,0.0,0.0
+Placing piece on stack,0,0,0.0,0.0,0.0
+Placing stock on shelf,0,0,0.0,0.0,0.0
+Plug cable into portable charger,0,0,0.0,0.0,0.0
+Position cardboard for cutting,0,0,0.0,0.0,0.0
+Position cardboard piece,0,0,0.0,0.0,0.0
+Position cardboard strip,0,0,0.0,0.0,0.0
+Position cardboard tray,0,0,0.0,0.0,0.0
+Position cardboard tube,0,0,0.0,0.0,0.0
+Position container near shelf,0,0,0.0,0.0,0.0
+Position container on shelf,0,0,0.0,0.0,0.0
+Position hands for work,0,0,0.0,0.0,0.0
+Position ribbon piece,0,0,0.0,0.0,0.0
+Position ruler and mark cardboard,0,0,0.0,0.0,0.0
+Position ruler on cardboard,0,0,0.0,0.0,0.0
+Position scissors,0,0,0.0,0.0,0.0
+Position scissors for next cut,0,0,0.0,0.0,0.0
+Position scissors to cut cardboard,0,0,0.0,0.0,0.0
+Position shelving divider,0,0,0.0,0.0,0.0
+Position the ruler,0,0,0.0,0.0,0.0
+Position tray,0,0,0.0,0.0,0.0
+Position utility knife,0,0,0.0,0.0,0.0
+Position utility knife on cardboard,0,0,0.0,0.0,0.0
+Positioning cardboard on workspace,0,0,0.0,0.0,0.0
+Positioning paper strip,0,0,0.0,0.0,0.0
+Positioning puzzle piece,0,0,0.0,0.0,0.0
+Positioning ruler on cardboard,0,0,0.0,0.0,0.0
+Prepare paper strip,0,0,0.0,0.0,0.0
+Prepare to cut cardboard,0,0,0.0,0.0,0.0
+Prepare to draw lines,0,0,0.0,0.0,0.0
+Prepare to pick up item,0,0,0.0,0.0,0.0
+Prepare to place bottle on shelf,0,0,0.0,0.0,0.0
+Prepare to place cardboard,0,0,0.0,0.0,0.0
+Prepare to place item in bag,0,0,0.0,0.0,0.0
+Prepare to place product,0,0,0.0,0.0,0.0
+Prepare to resume cutting,0,0,0.0,0.0,0.0
+Prepare to sort beads,0,0,0.0,0.0,0.0
+Preparing to craft,2,0,0.0,0.0,0.0
+Press fold,0,0,0.0,0.0,0.0
+Pull back hand,0,0,0.0,0.0,0.0
+Pull paper strip,0,0,0.0,0.0,0.0
+Push vacuum cleaner,0,0,0.0,0.0,0.0
+Put down phone,0,0,0.0,0.0,0.0
+Put down scissors,0,0,0.0,0.0,0.0
+Put down smartphone,3,0,0.0,0.0,0.0
+Put down utility knife,0,0,0.0,0.0,0.0
+Put down water bottle,0,0,0.0,0.0,0.0
+Putting away smartphone,0,0,0.0,0.0,0.0
+Reach and sort buttons,0,0,0.0,0.0,0.0
+Reach for Mahjong tiles,0,0,0.0,0.0,0.0
+Reach for additional items,0,0,0.0,0.0,0.0
+Reach for and examine canned goods,0,0,0.0,0.0,0.0
+Reach for and pick up smartphone,0,3,0.0,0.0,0.0
+Reach for another container,0,0,0.0,0.0,0.0
+Reach for another item,1,0,0.0,0.0,0.0
+Reach for beads,0,0,0.0,0.0,0.0
+Reach for black button,0,0,0.0,0.0,0.0
+Reach for button,0,0,0.0,0.0,0.0
+Reach for can,0,0,0.0,0.0,0.0
+Reach for canned food,0,0,0.0,0.0,0.0
+Reach for canned goods,0,0,0.0,0.0,0.0
+Reach for cardboard box,0,0,0.0,0.0,0.0
+Reach for cardboard piece,0,3,0.0,0.0,0.0
+Reach for cleaning supplies,1,0,0.0,0.0,0.0
+Reach for container,0,0,0.0,0.0,0.0
+Reach for craft items,3,0,0.0,0.0,0.0
+Reach for empty shelf space,0,0,0.0,0.0,0.0
+Reach for item,0,0,0.0,0.0,0.0
+Reach for item in box,0,0,0.0,0.0,0.0
+Reach for item on shelf,0,0,0.0,0.0,0.0
+Reach for items,0,0,0.0,0.0,0.0
+Reach for items in box,0,0,0.0,0.0,0.0
+Reach for more pieces,0,0,0.0,0.0,0.0
+Reach for next can,1,0,0.0,0.0,0.0
+Reach for next canned food,1,0,0.0,0.0,0.0
+Reach for next canned food item,0,0,0.0,0.0,0.0
+Reach for next canned product,0,0,0.0,0.0,0.0
+Reach for next item,1,0,0.0,0.0,0.0
+Reach for next piece,0,0,0.0,0.0,0.0
+Reach for next product,0,0,0.0,0.0,0.0
+Reach for object,0,0,0.0,0.0,0.0
+Reach for paper strip,0,6,0.0,0.0,0.0
+Reach for paper strips,0,0,0.0,0.0,0.0
+Reach for phone,0,0,0.0,0.0,0.0
+Reach for product,0,0,0.0,0.0,0.0
+Reach for product labels,0,0,0.0,0.0,0.0
+Reach for product on shelf,0,0,0.0,0.0,0.0
+Reach for puzzle piece,2,0,0.0,0.0,0.0
+Reach for retail item,0,0,0.0,0.0,0.0
+Reach for shelf,0,0,0.0,0.0,0.0
+Reach for shelving divider,0,0,0.0,0.0,0.0
+Reach for snack package,0,1,0.0,0.0,0.0
+Reach for snack pouch,0,0,0.0,0.0,0.0
+Reach for star,0,0,0.0,0.0,0.0
+Reach for stars,0,0,0.0,0.0,0.0
+Reach for utility knife,0,0,0.0,0.0,0.0
+Reach for water bottle,0,0,0.0,0.0,0.0
+Reach for wire hangers,1,0,0.0,0.0,0.0
+Reach into bag,0,0,0.0,0.0,0.0
+Reach into box,1,0,0.0,0.0,0.0
+Reach towards shelf,0,0,0.0,0.0,0.0
+Reaching for beads,0,0,0.0,0.0,0.0
+Realign Mahjong tiles,0,0,0.0,0.0,0.0
+Rearrange Mahjong tile,0,0,0.0,0.0,0.0
+Rearrange Mahjong tiles,0,0,0.0,0.0,0.0
+Rearrange shelf item,0,0,0.0,0.0,0.0
+Record count,2,0,0.0,0.0,0.0
+Record count on notepad,0,0,0.0,0.0,0.0
+Record star count,0,0,0.0,0.0,0.0
+Record star count on paper,0,0,0.0,0.0,0.0
+Release and prepare new strip,0,0,0.0,0.0,0.0
+Release bottle,0,0,0.0,0.0,0.0
+Release cardboard,0,0,0.0,0.0,0.0
+Release cardboard piece,0,0,0.0,0.0,0.0
+Release cardboard piece and gesture,2,0,0.0,0.0,0.0
+Release cardboard shape,0,0,0.0,0.0,0.0
+Release container,0,0,0.0,0.0,0.0
+Release folded paper,0,0,0.0,0.0,0.0
+Release food item,0,0,0.0,0.0,0.0
+Release hook,1,0,0.0,0.0,0.0
+Release label,0,0,0.0,0.0,0.0
+Release lantern,1,0,0.0,0.0,0.0
+Release paper,0,0,0.0,0.0,0.0
+Release paper coil,0,0,0.0,0.0,0.0
+Release paper star,0,0,0.0,0.0,0.0
+Release paper strip,4,0,0.0,0.0,0.0
+Release pickle jar,0,0,0.0,0.0,0.0
+Release product on shelf,0,0,0.0,0.0,0.0
+Release puzzle piece,2,0,0.0,0.0,0.0
+Release quilling strip,0,0,0.0,0.0,0.0
+Release scissors,3,0,0.0,0.0,0.0
+Release smartphone,3,0,0.0,0.0,0.0
+Remove cardboard flap,0,0,0.0,0.0,0.0
+Remove cardboard pattern,0,0,0.0,0.0,0.0
+Remove cardboard pattern piece,0,0,0.0,0.0,0.0
+Remove cleaning bottle,1,0,0.0,0.0,0.0
+Remove item from bag,0,0,0.0,0.0,0.0
+Remove item from shelf,0,0,0.0,0.0,0.0
+Remove lid from container,0,0,0.0,0.0,0.0
+Remove paper lantern part from packaging,1,0,0.0,0.0,0.0
+Remove plastic container from shelf,0,0,0.0,0.0,0.0
+Remove plastic container from storage box,0,0,0.0,0.0,0.0
+Remove plastic packaging,1,0,0.0,0.0,0.0
+Remove ruler,0,0,0.0,0.0,0.0
+Remove ruler and marker,0,0,0.0,0.0,0.0
+Remove shelf label,0,0,0.0,0.0,0.0
+Remove storage bin from shelf,0,0,0.0,0.0,0.0
+Reorganize bin contents,0,0,0.0,0.0,0.0
+Reposition and cut,0,0,0.0,0.0,0.0
+Reposition cardboard for cutting,0,0,0.0,0.0,0.0
+Reposition hand,3,0,0.0,0.0,0.0
+Reposition hands,0,0,0.0,0.0,0.0
+Reposition hands and ruler,0,0,0.0,0.0,0.0
+Reposition marker,0,0,0.0,0.0,0.0
+Reposition newspaper,0,0,0.0,0.0,0.0
+Reposition pen and prepare for next line,0,0,0.0,0.0,0.0
+Reposition ruler,0,0,0.0,0.0,0.0
+Reposition ruler and pen,0,0,0.0,0.0,0.0
+Reposition scissors,0,0,0.0,0.0,0.0
+Reposition sign and organize beads,0,0,0.0,0.0,0.0
+Reposition tools,0,0,0.0,0.0,0.0
+Reposition utility knife,0,0,0.0,0.0,0.0
+Repositioning ruler,0,0,0.0,0.0,0.0
+Repositioning ruler and cardboard,0,0,0.0,0.0,0.0
+Resume counting stars,0,0,0.0,0.0,0.0
+Resume observation,2,0,0.0,0.0,0.0
+Resume sorting blue beads,0,0,0.0,0.0,0.0
+Resume writing on paper,0,0,0.0,0.0,0.0
+Retract camera/reposition view,0,0,0.0,0.0,0.0
+Retract hand,0,0,0.0,0.0,0.0
+Retract hand from bag,0,0,0.0,0.0,0.0
+Retrieve another container,0,0,0.0,0.0,0.0
+Retrieve canned food from box,1,0,0.0,0.0,0.0
+Retrieve hand to table,0,0,0.0,0.0,0.0
+Retrieve items from bag,0,0,0.0,0.0,0.0
+Retrieve next canned food item,1,0,0.0,0.0,0.0
+Retrieve paper strip,0,1,0.0,0.0,0.0
+Retrieve paper strips,0,0,0.0,0.0,0.0
+Retrieve snack from container,0,0,0.0,0.0,0.0
+Retrieve star,0,0,0.0,0.0,0.0
+Retrieving more beads,2,0,0.0,0.0,0.0
+Return to sorting,0,0,0.0,0.0,0.0
+Reviewing count record,0,0,0.0,0.0,0.0
+Rinse cloth in sink,1,0,0.0,0.0,0.0
+Roll quilling paper,0,0,0.0,0.0,0.0
+Rolling paper strip,0,0,0.0,0.0,0.0
+Rub hands together,0,0,0.0,0.0,0.0
+Scan for next piece,0,0,0.0,0.0,0.0
+Scan supermarket shelves,0,0,0.0,0.0,0.0
+Score cardboard,0,0,0.0,0.0,0.0
+Scroll on smartphone,0,0,0.0,0.0,0.0
+Scroll smartphone screen,3,0,0.0,0.0,0.0
+Scroll through photo gallery,0,0,0.0,0.0,0.0
+Scrolling and viewing content on phone,0,0,0.0,0.0,0.0
+Scrolling or navigating on phone,0,0,0.0,0.0,0.0
+Search for puzzle piece,2,0,0.0,0.0,0.0
+Secure paper edges with adhesive,4,0,0.0,0.0,0.0
+Secure ribbon with needle,0,0,0.0,0.0,0.0
+Securing paper structure,4,0,0.0,0.0,0.0
+Select a bottle,0,0,0.0,0.0,0.0
+Select and pick up a canned item,0,0,0.0,0.0,0.0
+Select another item,0,0,0.0,0.0,0.0
+Select paper strip,0,0,0.0,0.0,0.0
+Select product from box,0,0,0.0,0.0,0.0
+Selecting new paper strip,0,0,0.0,0.0,0.0
+Separate cardboard piece,0,0,0.0,0.0,0.0
+Set down scissors and pick up power bank,0,0,0.0,0.0,0.0
+Set down utility knife,0,0,0.0,0.0,0.0
+Slide utility knife along ruler,0,0,0.0,0.0,0.0
+Sort Mahjong tiles,0,0,0.0,0.0,0.0
+Sort and adjust button line,3,0,0.0,0.0,0.0
+Sort and arrange buttons,3,0,0.0,0.0,0.0
+Sort and arrange cardboard pieces,0,0,0.0,0.0,0.0
+Sort and count beads,2,0,0.0,0.0,0.0
+Sort and place buttons,3,0,0.0,0.0,0.0
+Sort and place paper star,0,0,0.0,0.0,0.0
+Sort and stack cardboard pieces,0,0,0.0,0.0,0.0
+Sort beads,2,0,0.0,0.0,0.0
+Sort beads and write count,1,0,0.0,0.0,0.0
+Sort beads by color,0,39,0.0,0.0,0.0
+Sort beads by hand,0,0,0.0,0.0,0.0
+Sort beads on table,0,0,0.0,0.0,0.0
+Sort beads on the table,0,6,0.0,0.0,0.0
+Sort blue beads,0,0,0.0,0.0,0.0
+Sort blue star-shaped pieces,0,0,0.0,0.0,0.0
+Sort button,3,0,0.0,0.0,0.0
+Sort button by color,0,0,0.0,0.0,0.0
+Sort buttons,3,0,0.0,0.0,0.0
+Sort buttons by color,0,0,0.0,0.0,0.0
+Sort canned goods in tray,0,0,0.0,0.0,0.0
+Sort colored tiles,0,0,0.0,0.0,0.0
+Sort colorful pieces,0,0,0.0,0.0,0.0
+Sort craft items,3,0,0.0,0.0,0.0
+Sort cut cardboard,0,0,0.0,0.0,0.0
+Sort light blue origami stars,0,0,0.0,0.0,0.0
+Sort orange button,0,0,0.0,0.0,0.0
+Sort orange buttons,0,0,0.0,0.0,0.0
+Sort origami stars,0,0,0.0,0.0,0.0
+Sort origami stars by color,0,0,0.0,0.0,0.0
+Sort paper star,0,0,0.0,0.0,0.0
+Sort paper stars,0,0,0.0,0.0,0.0
+Sort plastic pieces,0,0,0.0,0.0,0.0
+Sort purple beads,0,0,0.0,0.0,0.0
+Sort purple star-shaped objects,0,0,0.0,0.0,0.0
+Sort puzzle pieces,2,0,0.0,0.0,0.0
+Sort quilled paper pieces,0,0,0.0,0.0,0.0
+Sort small colorful pieces,0,0,0.0,0.0,0.0
+Sort small craft pieces,3,0,0.0,0.0,0.0
+Sort small objects,0,0,0.0,0.0,0.0
+Sort small plastic pieces,0,0,0.0,0.0,0.0
+Sort star-shaped beads,1,0,0.0,0.0,0.0
+Sort star-shaped objects,0,0,0.0,0.0,0.0
+Sort star-shaped objects by color,0,0,0.0,0.0,0.0
+Sort tiles,0,0,0.0,0.0,0.0
+Sort tiles by color,0,0,0.0,0.0,0.0
+Sort yellow star-shaped objects,0,1,0.0,0.0,0.0
+Sorting buttons,0,0,0.0,0.0,0.0
+Sorting colorful paper pieces,0,0,0.0,0.0,0.0
+Sorting paper stars,0,0,0.0,0.0,0.0
+Stabilize cardboard,0,0,0.0,0.0,0.0
+Stabilize ruler,0,0,0.0,0.0,0.0
+Stack cardboard pieces,0,0,0.0,0.0,0.0
+Stack cardboard square,0,0,0.0,0.0,0.0
+Stack cardboard squares,0,1,0.0,0.0,0.0
+Stacking cardboard pieces,0,0,0.0,0.0,0.0
+Stacking cardboard square,0,0,0.0,0.0,0.0
+Stacking cardboard squares,0,0,0.0,0.0,0.0
+Stand up and walk away,0,0,0.0,0.0,0.0
+Start cutting,2,0,0.0,0.0,0.0
+Start folding paper strip,0,0,0.0,0.0,0.0
+Starting to label next square,0,0,0.0,0.0,0.0
+Stir contents,1,0,0.0,0.0,0.0
+Stop measuring and put down tools,0,0,0.0,0.0,0.0
+Stop sorting stars,0,0,0.0,0.0,0.0
+Sweep debris,0,0,0.0,0.0,0.0
+Sweep floor debris,0,0,0.0,0.0,0.0
+Switch to scissors,0,0,0.0,0.0,0.0
+Switching marker,0,0,0.0,0.0,0.0
+Tap smartphone screen,0,0,0.0,0.0,0.0
+Tapping on smartphone screen,0,0,0.0,0.0,0.0
+Tapping smartphone screen,0,0,0.0,0.0,0.0
+Tear newspaper,0,0,0.0,0.0,0.0
+Tear off cardboard segment,0,0,0.0,0.0,0.0
+Touch canned goods,0,0,0.0,0.0,0.0
+Touch pieces in box,0,0,0.0,0.0,0.0
+Touch shelf edge,0,0,0.0,0.0,0.0
+Trace pattern on cardboard,0,0,0.0,0.0,0.0
+Transition to cutting,0,0,0.0,0.0,0.0
+Transition to standing position,0,0,0.0,0.0,0.0
+Trim cardboard,0,0,0.0,0.0,0.0
+Trim cardboard piece,0,0,0.0,0.0,0.0
+Type on smartphone,0,0,0.0,0.0,0.0
+Typing message on smartphone,0,0,0.0,0.0,0.0
+Typing on phone,0,0,0.0,0.0,0.0
+Typing on smartphone,0,0,0.0,0.0,0.0
+Update paper record,0,0,0.0,0.0,0.0
+Use phone,2,0,0.0,0.0,0.0
+Use phone to check instructions,0,0,0.0,0.0,0.0
+Use phone to check stock,0,0,0.0,0.0,0.0
+Use phone while crafting,1,0,0.0,0.0,0.0
+Use smartphone,6,0,0.0,0.0,0.0
+Vacuum edge of carpet,0,0,0.0,0.0,0.0
+Vacuum the carpet,0,0,0.0,0.0,0.0
+Vacuuming along the wall edge,0,0,0.0,0.0,0.0
+Vacuuming carpet corner,0,0,0.0,0.0,0.0
+Vacuuming carpet edge,0,0,0.0,0.0,0.0
+Vacuuming the carpet edge,0,0,0.0,0.0,0.0
+View content on smartphone,0,0,0.0,0.0,0.0
+View phone screen,0,0,0.0,0.0,0.0
+Viewing phone screen,0,0,0.0,0.0,0.0
+Walk across office,0,0,0.0,0.0,0.0
+Walk across room,0,0,0.0,0.0,0.0
+Walk across the room,0,0,0.0,0.0,0.0
+Walk away,0,0,0.0,0.0,0.0
+Walk in hallway,0,0,0.0,0.0,0.0
+Walk through corridor,0,0,0.0,0.0,0.0
+Walk through doorway,0,0,0.0,0.0,0.0
+Walk through hallway,0,0,0.0,0.0,0.0
+Walk through office,0,0,0.0,0.0,0.0
+Walk through store,0,0,0.0,0.0,0.0
+Walk through workspace,0,17,0.0,0.0,0.0
+Walk towards aisle,0,0,0.0,0.0,0.0
+Walk towards desk,0,0,0.0,0.0,0.0
+Walk towards next aisle,0,0,0.0,0.0,0.0
+Walk towards other aisles,1,0,0.0,0.0,0.0
+Walk towards room,0,0,0.0,0.0,0.0
+Walk towards shelf,0,0,0.0,0.0,0.0
+Walk towards shelves,1,0,0.0,0.0,0.0
+Walk towards storage area,0,0,0.0,0.0,0.0
+Walk towards table,0,0,0.0,0.0,0.0
+Walk towards workspace,0,0,0.0,0.0,0.0
+Walk with cardboard,0,0,0.0,0.0,0.0
+Walk with cardboard cutout,0,0,0.0,0.0,0.0
+Walk with marker,0,0,0.0,0.0,0.0
+Walk with shopping bag,0,0,0.0,0.0,0.0
+Walking across the room,1,0,0.0,0.0,0.0
+Walking along the aisle,0,0,0.0,0.0,0.0
+Walking in the hallway,3,0,0.0,0.0,0.0
+Walking in the workspace,0,0,0.0,0.0,0.0
+Walking through classroom,0,0,0.0,0.0,0.0
+Walking through office hallway,0,0,0.0,0.0,0.0
+Walking through the office,0,0,0.0,0.0,0.0
+Walking to sink,0,0,0.0,0.0,0.0
+Walking towards door,1,0,0.0,0.0,0.0
+Walking towards workstation,0,0,0.0,0.0,0.0
+Washing hands,0,0,0.0,0.0,0.0
+Washing hands in sink,1,0,0.0,0.0,0.0
+Wipe down shelf,0,0,0.0,0.0,0.0
+Wipe electronic item,0,0,0.0,0.0,0.0
+Wipe food product,0,0,0.0,0.0,0.0
+Wipe grocery shelf,0,0,0.0,0.0,0.0
+Wipe item,0,0,0.0,0.0,0.0
+Wipe jar,0,0,0.0,0.0,0.0
+Wipe ketchup bottle,0,0,0.0,0.0,0.0
+Wipe kitchen counter,1,0,0.0,0.0,0.0
+Wipe product,0,0,0.0,0.0,0.0
+Wipe retail item,0,0,0.0,0.0,0.0
+Wipe shelf,0,0,0.0,0.0,0.0
+Wipe shelf surface,0,0,0.0,0.0,0.0
+Wipe the plastic jar,0,0,0.0,0.0,0.0
+Wipe the product jar,0,0,0.0,0.0,0.0
+Wipe the shelf,0,0,0.0,0.0,0.0
+Wiping countertop,1,0,0.0,0.0,0.0
+Withdraw hand,0,0,0.0,0.0,0.0
+Write count on paper,1,0,0.0,0.0,0.0
+Write on notepad,0,0,0.0,0.0,0.0
+Write on paper,1,24,0.041666666666666664,1.0,0.07999999999999999
+Write on paper record,0,0,0.0,0.0,0.0
+Writing on notepad,0,0,0.0,0.0,0.0
+fold purple ribbon,0,0,0.0,0.0,0.0
+sort craft materials,3,0,0.0,0.0,0.0
+"{""action"": ""Pour liquid into bowl"", ""contact"": ""yes"", ""evidence_window"": {""end_frame"": 19, ""start_frame"": 0}, ""next_action"": ""Pour liquid into bowl"", ""objects"": [""bottle"", ""bowl"", ""kitchen counter"", ""kitchen sink"", ""kitchen tap"", ""kitchen cabinet"", ""kitchen shelf"", ""kitchen window"", ""kitchen utensils"", ""kitchen appliances"", ""kitchen utensil rack"", ""kitchen towel"", ""kitchen cabinet door"", ""kitchen cabinet drawer"", ""kitchen cabinet shelf"", ""kitchen cabinet handle"", ""kitchen cabinet knob"", ""kitchen cabinet latch"", ""kitchen cabinet hinge"", ""kitchen cabinet panel"", ""kitchen cabinet frame"", ""kitchen cabinet base"", ""kitchen cabinet top"", ""kitchen cabinet side"", ""kitchen cabinet back"", ""kitchen cabinet door panel"", ""kitchen cabinet door frame"", ""kitchen cabinet door handle"", ""kitchen cabinet door knob"", ""kitchen cabinet door latch"", ""kitchen cabinet door hinge"", ""kitchen cabinet door panel frame"", ""kitchen cabinet door panel handle"", ""kitchen cabinet door panel knob"", ""kitchen cabinet door panel latch"", ""kitchen cabinet door panel hinge"", ""kitchen cabinet door panel frame handle"", ""kitchen cabinet door panel frame knob"", ""kitchen cabinet door panel frame latch"", ""kitchen cabinet door panel frame hinge"", ""kitchen cabinet door panel frame panel"", ""kitchen cabinet door panel frame panel handle"", ""kitchen cabinet door panel frame panel knob"", ""kitchen cabinet door panel frame panel latch"", ""kitchen cabinet door panel frame panel hinge"", ""kitchen cabinet door panel frame panel frame"", ""kitchen cabinet door panel frame panel frame handle"", ""kitchen cabinet door panel frame panel frame knob"", ""kitchen cabinet door panel frame panel frame latch"", ""kitchen cabinet door panel frame panel frame hinge"", ""kitchen cabinet door panel frame panel frame panel"", ""kitchen cabinet door panel frame panel frame panel handle"", ""kitchen cabinet door panel frame panel frame panel knob"", ""kitchen cabinet door panel frame panel frame panel latch"", ""kitchen cabinet door panel frame panel frame panel hinge"", ""kitchen cabinet door panel frame panel frame panel frame"", ""kitchen cabinet door panel frame panel frame panel frame handle"", ""kitchen cabinet door panel frame panel frame panel frame knob"", ""kitchen cabinet door panel frame panel frame panel frame latch"", ""kitchen cabinet door panel frame panel frame panel frame hinge"", ""kitchen cabinet door panel frame panel frame panel frame panel"", ""kitchen cabinet door panel frame panel frame panel frame panel handle"", ""kitchen cabinet door panel frame panel frame panel frame panel knob"", ""kitchen cabinet door panel frame panel frame panel frame panel latch"", ""kitchen cabinet door panel frame panel frame panel frame panel hinge"", ""kitchen cabinet door panel frame panel frame panel frame panel frame"", ""kitchen cabinet door panel frame panel frame panel frame panel frame handle"", ""kitchen cabinet door panel frame panel frame panel frame panel frame knob"", ""kitchen cabinet door panel frame panel frame panel frame panel frame latch"", ""kitchen cabinet door panel frame panel frame panel frame panel frame hinge"", ""kitchen cabinet door panel frame panel frame panel frame panel frame panel"", ""kitchen cabinet door panel frame panel frame panel frame panel frame panel handle"", ""kitchen cabinet door panel frame panel frame panel frame panel frame panel knob"", ""kitchen cabinet door panel frame panel frame panel frame panel frame panel latch"", ""kitchen cabinet door panel frame panel frame panel frame panel frame panel hinge"", ""kitchen cabinet door panel frame panel frame panel frame panel frame panel frame"", ""kitchen cabinet door panel frame panel frame panel frame panel frame panel frame handle"", ""kitchen cabinet door panel frame panel frame panel frame panel frame panel frame knob"", ""kitchen cabinet door panel frame panel frame panel frame panel frame panel frame latch"", ""kitchen cabinet door panel frame panel frame panel frame panel frame panel frame hinge"", ""kitchen cabinet door panel frame panel frame panel frame panel frame panel frame panel"", ""kitchen cabinet door panel frame panel frame panel frame panel frame panel frame panel handle"", ""kitchen cabinet door panel frame panel frame panel frame panel frame panel frame panel knob"", ""kitchen cabinet door panel frame panel frame panel frame panel frame panel frame panel latch"", ""kitchen cabinet door panel frame panel frame panel frame panel frame panel frame panel hinge"", ""kitchen cabinet door panel frame panel frame panel frame panel frame panel frame panel frame"", ""kitchen cabinet door panel frame panel frame panel frame panel frame panel frame panel frame handle"", ""kitchen cabinet door panel frame panel frame panel frame panel frame panel frame panel frame knob"", ""kitchen cabinet door panel frame panel frame panel frame panel frame panel frame panel frame latch"", ""kitchen cabinet door panel frame panel frame panel frame panel frame panel frame panel frame hinge"", ""kitchen cabinet",0,1,0.0,0.0,0.0
+Close the door,0,2,0.0,0.0,0.0
+"{""action"": ""Close door"", ""contact"": ""yes"", ""evidence_window"": {""end_frame"": 2539, ""start_frame"": 2520}, ""next_action"": ""Close door"", ""objects"": [""door"", ""door handle"", ""plastic bag"", ""blue bucket"", ""red bucket"", ""blue container"", ""red container"", ""white container"", ""white box"", ""white bag"", ""white cloth"", ""white plastic bag"", ""white plastic container"", ""white cardboard box"", ""white paper"", ""white paper bag"", ""white paper sheet"", ""white paper strip"", ""white plastic sheet"", ""white plastic bag with red label"", ""white plastic container with red label"", ""white cardboard tray"", ""white cardboard piece"", ""white cardboard square"", ""white cardboard shape"", ""white paper lantern"", ""white paper lantern component"", ""white paper star"", ""white paper star shape"", ""white paper lantern piece"", ""white paper lantern part"", ""white paper decoration"", ""white paper cutout"", ""white paper shape"", ""white paper segment"", ""white paper coil"", ""white paper strip with red label"", ""white paper strip with blue label"", ""white paper strip with yellow label"", ""white paper strip with green label"", ""white paper strip with purple label"", ""white paper strip with orange label"", ""white paper strip with black label"", ""white paper strip with pink label"", ""white paper strip with brown label"", ""white paper strip with grey label"", ""white paper strip with turquoise label"", ""white paper strip with maroon label"", ""white paper strip with lavender label"", ""white paper strip with olive green label"", ""white paper strip with coral label"", ""white paper strip with navy blue label"", ""white paper strip with light blue label"", ""white paper strip with dark blue label"", ""white paper strip with light green label"", ""white paper strip with dark green label"", ""white paper strip with yellow-green label"", ""white paper strip with lime green label"", ""white paper strip with forest green label"", ""white paper strip with emerald green label"", ""white paper strip with teal label"", ""white paper strip with turquoise blue label"", ""white paper strip with sky blue label"", ""white paper strip with baby blue label"", ""white paper strip with periwinkle label"", ""white paper strip with lavender blue label"", ""white paper strip with violet label"", ""white paper strip with purple blue label"", ""white paper strip with indigo label"", ""white paper strip with dark purple label"", ""white paper strip with magenta label"", ""white paper strip with pink purple label"", ""white paper strip with rose pink label"", ""white paper strip with coral pink label"", ""white paper strip with salmon pink label"", ""white paper strip with peach label"", ""white paper strip with apricot label"", ""white paper strip with orange yellow label"", ""white paper strip with golden yellow label"", ""white paper strip with lemon yellow label"", ""white paper strip with yellow green label"", ""white paper strip with chartreuse label"", ""white paper strip with olive label"", ""white paper strip with khaki label"", ""white paper strip with tan label"", ""white paper strip with beige label"", ""white paper strip with cream label"", ""white paper strip with off white label"", ""white paper strip with light grey label"", ""white paper strip with grey label"", ""white paper strip with dark grey label"", ""white paper strip with black label"", ""white paper strip with brown label"", ""white paper strip with dark brown label"", ""white paper strip with chocolate label"", ""white paper strip with coffee label"", ""white paper strip with taupe label"", ""white paper strip with maroon label"", ""white paper strip with burgundy label"", ""white paper strip with red label"", ""white paper strip with dark red label"", ""white paper strip with crimson label"", ""white paper strip with cherry red label"", ""white paper strip with pink label"", ""white paper strip with light pink label"", ""white paper strip with pale pink label"", ""white paper strip with rose label"", ""white paper strip with blush label"", ""white paper strip with coral label"", ""white paper strip with salmon label"", ""white paper strip with peach label"", ""white paper strip with apricot label"", ""white paper strip with orange label"", ""white paper strip with light orange label"", ""white paper strip with pale orange label"", ""white paper strip with tangerine label"", ""white paper strip with mango label"", ""white paper strip with yellow label"", ""white paper strip with light yellow label"", ""white paper strip with pale yellow label"", ""white paper strip with lemon label"", ""white paper strip with lime label"", ""white paper strip with green label"", ""white paper strip with light green label"", ""white paper strip with pale green label"", ""white paper strip with mint label"", ""white paper strip with sage label"", ""white paper strip with olive label"", ""white paper strip with khaki",0,1,0.0,0.0,0.0
+"{""action"": ""Close the door"", ""contact"": ""yes"", ""evidence_window"": {""end_frame"": 2559, ""start_frame"": 2540}, ""next_action"": ""Close the door"", ""objects"": [""door"", ""door handle"", ""door knob"", ""door frame"", ""door hinge"", ""door latch"", ""door lock"", ""door panel"", ""door post"", ""door stop"", ""door weather stripping"", ""doorway"", ""garage door"", ""garage door opener"", ""garage door track"", ""garage door wheel"", ""garage door window"", ""garage door opener button"", ""garage door opener remote"", ""garage door opener receiver"", ""garage door opener motor"", ""garage door opener chain"", ""garage door opener belt"", ""garage door opener rail"", ""garage door opener sensor"", ""garage door opener control panel"", ""garage door opener light"", ""garage door opener emergency release"", ""garage door opener remote control"", ""garage door opener wall switch"", ""garage door opener keypad"", ""garage door opener smartphone app"", ""garage door opener voice control"", ""garage door opener timer"", ""garage door opener remote control"", ""garage door opener wall switch"", ""garage door opener keypad"", ""garage door opener smartphone app"", ""garage door opener voice control"", ""garage door opener timer"", ""garage door opener emergency release"", ""garage door opener sensor"", ""garage door opener control panel"", ""garage door opener light"", ""garage door opener motor"", ""garage door opener chain"", ""garage door opener belt"", ""garage door opener rail"", ""garage door opener remote"", ""garage door opener receiver"", ""garage door opener button"", ""garage door window"", ""garage door track"", ""garage door wheel"", ""garage door"", ""garage door opener"", ""garage door opener button"", ""garage door opener remote"", ""garage door opener receiver"", ""garage door opener motor"", ""garage door opener chain"", ""garage door opener belt"", ""garage door opener rail"", ""garage door opener sensor"", ""garage door opener control panel"", ""garage door opener light"", ""garage door opener emergency release"", ""garage door opener remote control"", ""garage door opener wall switch"", ""garage door opener keypad"", ""garage door opener smartphone app"", ""garage door opener voice control"", ""garage door opener timer"", ""garage door opener remote control"", ""garage door opener wall switch"", ""garage door opener keypad"", ""garage door opener smartphone app"", ""garage door opener voice control"", ""garage door opener timer"", ""garage door opener emergency release"", ""garage door opener sensor"", ""garage door opener control panel"", ""garage door opener light"", ""garage door opener motor"", ""garage door opener chain"", ""garage door opener belt"", ""garage door opener rail"", ""garage door opener remote"", ""garage door opener receiver"", ""garage door opener button"", ""garage door window"", ""garage door track"", ""garage door wheel"", ""garage door"", ""garage door opener"", ""garage door opener button"", ""garage door opener remote"", ""garage door opener receiver"", ""garage door opener motor"", ""garage door opener chain"", ""garage door opener belt"", ""garage door opener rail"", ""garage door opener sensor"", ""garage door opener control panel"", ""garage door opener light"", ""garage door opener emergency release"", ""garage door opener remote control"", ""garage door opener wall switch"", ""garage door opener keypad"", ""garage door opener smartphone app"", ""garage door opener voice control"", ""garage door opener timer"", ""garage door opener remote control"", ""garage door opener wall switch"", ""garage door opener keypad"", ""garage door opener smartphone app"", ""garage door opener voice control"", ""garage door opener timer"", ""garage door opener emergency release"", ""garage door opener sensor"", ""garage door opener control panel"", ""garage door opener light"", ""garage door opener motor"", ""garage door opener chain"", ""garage door opener belt"", ""garage door opener rail"", ""garage door opener remote"", ""garage door opener receiver"", ""garage door opener button"", ""garage door window"", ""garage door track"", ""garage door wheel"", ""garage door"", ""garage door opener"", ""garage door opener button"", ""garage door opener remote"", ""garage door opener receiver"", ""garage door opener motor"", ""garage door opener chain"", ""garage door opener belt"", ""garage door opener rail"", ""garage door opener sensor"", ""garage",0,1,0.0,0.0,0.0
+"{""action"": ""Close cabinet door"", ""contact"": ""yes"", ""evidence_window"": {""end_frame"": 3419, ""start_frame"": 3400}, ""next_action"": ""Close cabinet door"", ""objects"": [""cabinet"", ""cabinet door"", ""cabinet handle"", ""cabinet shelf"", ""cabinet drawer"", ""cabinet drawer handle"", ""cabinet drawer knob"", ""cabinet drawer pull"", ""cabinet drawer slide"", ""cabinet drawer slide rail"", ""cabinet drawer slide track"", ""cabinet drawer slide roller"", ""cabinet drawer slide bracket"", ""cabinet drawer slide support"", ""cabinet drawer slide guide"", ""cabinet drawer slide stop"", ""cabinet drawer slide clip"", ""cabinet drawer slide clip holder"", ""cabinet drawer slide clip bracket"", ""cabinet drawer slide clip support"", ""cabinet drawer slide clip guide"", ""cabinet drawer slide clip stop"", ""cabinet drawer slide clip holder bracket"", ""cabinet drawer slide clip holder support"", ""cabinet drawer slide clip holder guide"", ""cabinet drawer slide clip holder stop"", ""cabinet drawer slide clip holder clip"", ""cabinet drawer slide clip holder clip holder"", ""cabinet drawer slide clip holder clip bracket"", ""cabinet drawer slide clip holder clip support"", ""cabinet drawer slide clip holder clip guide"", ""cabinet drawer slide clip holder clip stop"", ""cabinet drawer slide clip holder clip holder bracket"", ""cabinet drawer slide clip holder clip holder support"", ""cabinet drawer slide clip holder clip holder guide"", ""cabinet drawer slide clip holder clip holder stop"", ""cabinet drawer slide clip holder clip holder clip"", ""cabinet drawer slide clip holder clip holder clip holder"", ""cabinet drawer slide clip holder clip holder clip bracket"", ""cabinet drawer slide clip holder clip holder clip support"", ""cabinet drawer slide clip holder clip holder clip guide"", ""cabinet drawer slide clip holder clip holder clip stop"", ""cabinet drawer slide clip holder clip holder clip holder bracket"", ""cabinet drawer slide clip holder clip holder clip holder support"", ""cabinet drawer slide clip holder clip holder clip holder guide"", ""cabinet drawer slide clip holder clip holder clip holder stop"", ""cabinet drawer slide clip holder clip holder clip holder clip"", ""cabinet drawer slide clip holder clip holder clip holder clip holder"", ""cabinet drawer slide clip holder clip holder clip holder clip bracket"", ""cabinet drawer slide clip holder clip holder clip holder clip support"", ""cabinet drawer slide clip holder clip holder clip holder clip guide"", ""cabinet drawer slide clip holder clip holder clip holder clip stop"", ""cabinet drawer slide clip holder clip holder clip holder clip holder bracket"", ""cabinet drawer slide clip holder clip holder clip holder clip holder support"", ""cabinet drawer slide clip holder clip holder clip holder clip holder guide"", ""cabinet drawer slide clip holder clip holder clip holder clip holder stop"", ""cabinet drawer slide clip holder clip holder clip holder clip holder clip"", ""cabinet drawer slide clip holder clip holder clip holder clip holder clip holder"", ""cabinet drawer slide clip holder clip holder clip holder clip holder clip bracket"", ""cabinet drawer slide clip holder clip holder clip holder clip holder clip support"", ""cabinet drawer slide clip holder clip holder clip holder clip holder clip guide"", ""cabinet drawer slide clip holder clip holder clip holder clip holder clip stop"", ""cabinet drawer slide clip holder clip holder clip holder clip holder clip holder bracket"", ""cabinet drawer slide clip holder clip holder clip holder clip holder clip holder support"", ""cabinet drawer slide clip holder clip holder clip holder clip holder clip holder guide"", ""cabinet drawer slide clip holder clip holder clip holder clip holder clip holder stop"", ""cabinet drawer slide clip holder clip holder clip holder clip holder clip holder clip"", ""cabinet drawer slide clip holder clip holder clip holder clip holder clip holder clip holder"", ""cabinet drawer slide clip holder clip holder clip holder clip holder clip holder clip bracket"", ""cabinet drawer slide clip holder clip holder clip holder clip holder clip holder clip support"", ""cabinet drawer slide clip holder clip holder clip holder clip holder clip holder clip guide"", ""cabinet drawer slide clip holder clip holder clip holder clip holder clip holder clip stop"", ""cabinet drawer slide clip holder clip holder clip holder clip holder clip holder clip holder bracket"", ""cabinet drawer slide clip holder clip holder clip holder clip holder clip holder clip holder support"", ""cabinet drawer slide clip holder clip holder clip holder clip holder clip holder clip holder guide"", ""cabinet drawer slide clip holder clip holder clip holder clip holder clip holder clip holder stop"", ""cabinet drawer slide clip holder clip holder clip holder clip holder clip holder clip holder clip"", ""cabinet drawer slide clip holder clip holder clip holder clip holder clip holder clip holder clip holder"", ""cabinet drawer slide clip holder clip holder clip holder clip holder clip holder clip holder clip bracket"", ""cabinet drawer slide clip holder clip holder clip holder clip holder clip holder clip holder clip support",0,1,0.0,0.0,0.0
+Close cabinet,0,1,0.0,0.0,0.0
+Close cabinet door,0,1,0.0,0.0,0.0
+Wipe hands,0,1,0.0,0.0,0.0
+Hold paper star,0,8,0.0,0.0,0.0
+Place pot on the counter,0,1,0.0,0.0,0.0
+Rinse pot,0,1,0.0,0.0,0.0
+Place marker on table,0,2,0.0,0.0,0.0
+Carry snack package,0,1,0.0,0.0,0.0
+Place paper strip on table,0,12,0.0,0.0,0.0
+Reach for jar,0,4,0.0,0.0,0.0
+Hold jar,0,1,0.0,0.0,0.0
+Place beads on table,0,10,0.0,0.0,0.0
+"{""action"": ""Hold paper star"", ""contact"": ""yes"", ""evidence_window"": {""end_frame"": 1959, ""start_frame"": 1940}, ""next_action"": ""Hold paper star"", ""objects"": [""paper star"", ""table"", ""scissors"", ""phone"", ""mouse"", ""beads"", ""pen"", ""light blue paper"", ""yellow paper"", ""purple paper"", ""cardboard"", ""cardboard pieces"", ""cardboard shapes"", ""cardboard square"", ""cardboard tray"", ""cardboard tube"", ""cardboard strip"", ""utility knife"", ""ruler"", ""marker"", ""jar"", ""canned food"", ""tin can"", ""plastic container"", ""plastic packaging"", ""box"", ""shopping bag"", ""gift box"", ""product"", ""product box"", ""product package"", ""supplement bottle"", ""bottle"", ""sauce bottle"", ""pickle jar"", ""water bottle"", ""pink water bottle"", ""cleaning bottle"", ""earbud case"", ""charging case"", ""portable charger"", ""power bank"", ""charger"", ""charging cable"", ""power adapter"", ""smartphone"", ""smartphone box"", ""computer mouse"", ""pen and paper"", ""paper"", ""paper pieces"", ""paper segments"", ""paper lantern"", ""paper lantern component"", ""blue product box"", ""blue strip"", ""light blue strip"", ""yellow item"", ""yellow paper strip"", ""purple paper strip"", ""adhesive tape"", ""ribbon"", ""ribbon piece"", ""plastic strip"", ""foam piece"", ""foam strips"", ""cardboard pattern"", ""cardboard pattern piece"", ""stapler"", ""cleaning cloth"", ""dustpan"", ""box of beads"", ""container"", ""container of canned food"", ""small cardboard piece"", ""small piece of material"", ""small object"", ""small product bag"", ""small white box"", ""bean"", ""bean pile"", ""button"", ""black button"", ""orange button"", ""red button"", ""small button"", ""button cluster"", ""button row"", ""star"", ""star bead"", ""star-shaped bead"", ""purple bead"", ""light blue origami star"", ""yellow star-shaped object"", ""purple star-shaped object"", ""colored tile"", ""colored pieces"", ""jewel"", ""jewel piece"", ""Mahjong tile"", ""Mahjong tiles"", ""puzzle piece"", ""quilled paper"", ""quilled paper piece"", ""quilled paper shape"", ""quilling paper"", ""quilling paper strip"", ""paper quilling piece"", ""paper quilling strip"", ""paper star"", ""lucky star"", ""folded paper"", ""folded paper star"", ""paper strip"", ""paper decoration"", ""colored paper"", ""colorful pieces"", ""cardboard sheet"", ""cardboard square"", ""cardboard pieces"", ""cardboard shapes"", ""cardboard cutout"", ""cardboard tray"", ""cardboard tube"", ""cardboard strip"", ""newspaper"", ""newspaper piece"", ""plastic container"", ""plastic storage bin"", ""tin can"", ""jar"", ""canned good"", ""canned product"", ""packaged paper lantern component"", ""snack package"", ""snack pouch"", ""container of snack"", ""gift box"", ""product package"", ""supplement bottle"", ""bottle"", ""ketchup bottle"", ""sauce bottle"", ""pickle jar"", ""water bottle"", ""pink water bottle"", ""cleaning bottle"", ""earbud case"", ""charging case"", ""portable charger"", ""power bank"", ""charger"", ""charging cable"", ""power adapter"", ""smartphone"", ""smartphone box"", ""computer mouse"", ""pen"", ""notepad"", ""paper"", ""paper pieces"", ""paper segments"", ""pieces"", ""lantern"", ""lantern component"", ""blue product box"", ""blue strip"", ""light blue strip"", ""yellow item"", ""yellow paper strip"", ""purple paper strip"", ""adhesive strip"", ""ribbon knot"", ""small paper segment"", ""foam piece"", ""foam strips"", ""cardboard pattern"", ""cardboard pattern piece"", ""stapler"", ""cleaning cloth"", ""dustpan"", ""box of beads"", ""container"", ""small cardboard piece"", ""small piece of material"", ""small object"", ""small product bag"", ""small white box"", ""bean"", ""bean pile"", ""button"", ""black button"", ""orange button"", ""red button"", ""small button"", ""buttons"", ""button cluster"", ""button row"", ""star"", ""star bead"", ""star-shaped bead"", ""purple bead"", ""light blue origami star"", ""yellow star-shaped object"", ""purple star-shaped object"", ""colored tile"", ""colored pieces"", ""jewel"", ""jewel piece"", ""Mahjong tile"", ""Mahjong tiles"", ""puzzle piece"", ""quilled paper"", ""qu",0,1,0.0,0.0,0.0
+Place earbud case on table,0,2,0.0,0.0,0.0
+Hold red button,0,1,0.0,0.0,0.0
+Adjust beads by color,0,5,0.0,0.0,0.0
+Approach workspace,0,3,0.0,0.0,0.0
+"{""action"": ""Approach workspace"", ""contact"": ""no"", ""evidence_window"": {""end_frame"": 39, ""start_frame"": 20}, ""next_action"": ""Approach workspace"", ""objects"": [""cardboard"", ""cardboard pieces"", ""cardboard stack"", ""cardboard tray"", ""cardboard tube"", ""cardboard pieces"", ""cardboard shapes"", ""cardboard squares"", ""cardboard sheet"", ""cardboard cutouts"", ""cardboard shapes"", ""cardboard pieces"", ""cardboard shapes"", ""cardboard pieces"", ""cardboard shapes"", ""cardboard pieces"", ""cardboard shapes"", ""cardboard pieces"", ""cardboard shapes"", ""cardboard pieces"", ""cardboard shapes"", ""cardboard pieces"", ""cardboard shapes"", ""cardboard pieces"", ""cardboard shapes"", ""cardboard pieces"", ""cardboard shapes"", ""cardboard pieces"", ""cardboard shapes"", ""cardboard pieces"", ""cardboard shapes"", ""cardboard pieces"", ""cardboard shapes"", ""cardboard pieces"", ""cardboard shapes"", ""cardboard pieces"", ""cardboard shapes"", ""cardboard pieces"", ""cardboard shapes"", ""cardboard pieces"", ""cardboard shapes"", ""cardboard pieces"", ""cardboard shapes"", ""cardboard pieces"", ""cardboard shapes"", ""cardboard pieces"", ""cardboard shapes"", ""cardboard pieces"", ""cardboard shapes"", ""cardboard pieces"", ""cardboard shapes"", ""cardboard pieces"", ""cardboard shapes"", ""cardboard pieces"", ""cardboard shapes"", ""cardboard pieces"", ""cardboard shapes"", ""cardboard pieces"", ""cardboard shapes"", ""cardboard pieces"", ""cardboard shapes"", ""cardboard pieces"", ""cardboard shapes"", ""cardboard pieces"", ""cardboard shapes"", ""cardboard pieces"", ""cardboard shapes"", ""cardboard pieces"", ""cardboard shapes"", ""cardboard pieces"", ""cardboard shapes"", ""cardboard pieces"", ""cardboard shapes"", ""cardboard pieces"", ""cardboard shapes"", ""cardboard pieces"", ""cardboard shapes"", ""cardboard pieces"", ""cardboard shapes"", ""cardboard pieces"", ""cardboard shapes"", ""cardboard pieces"", ""cardboard shapes"", ""cardboard pieces"", ""cardboard shapes"", ""cardboard pieces"", ""cardboard shapes"", ""cardboard pieces"", ""cardboard shapes"", ""cardboard pieces"", ""cardboard shapes"", ""cardboard pieces"", ""cardboard shapes"", ""cardboard pieces"", ""cardboard shapes"", ""cardboard pieces"", ""cardboard shapes"", ""cardboard pieces"", ""cardboard shapes"", ""cardboard pieces"", ""cardboard shapes"", ""cardboard pieces"", ""cardboard shapes"", ""cardboard pieces"", ""cardboard shapes"", ""cardboard pieces"", ""cardboard shapes"", ""cardboard pieces"", ""cardboard shapes"", ""cardboard pieces"", ""cardboard shapes"", ""cardboard pieces"", ""cardboard shapes"", ""cardboard pieces"", ""cardboard shapes"", ""cardboard pieces"", ""cardboard shapes"", ""cardboard pieces"", ""cardboard shapes"", ""cardboard pieces"", ""cardboard shapes"", ""cardboard pieces"", ""cardboard shapes"", ""cardboard pieces"", ""cardboard shapes"", ""cardboard pieces"", ""cardboard shapes"", ""cardboard pieces"", ""cardboard shapes"", ""cardboard pieces"", ""cardboard shapes"", ""cardboard pieces"", ""cardboard shapes"", ""cardboard pieces"", ""cardboard shapes"", ""cardboard pieces"", ""cardboard shapes"", ""cardboard pieces"", ""cardboard shapes"", ""cardboard pieces"", ""cardboard shapes"", ""cardboard pieces"", ""cardboard shapes"", ""cardboard pieces"", ""cardboard shapes"", ""cardboard pieces"", ""cardboard shapes"", ""cardboard pieces"", ""cardboard shapes"", ""cardboard pieces"", ""cardboard shapes"", ""cardboard pieces"", ""cardboard shapes"", ""cardboard pieces"", ""cardboard shapes"", ""cardboard pieces"", ""cardboard shapes"", ""cardboard pieces"", ""cardboard shapes"", ""cardboard pieces"", ""cardboard shapes"", ""cardboard pieces"", ""cardboard shapes"", ""cardboard pieces"", ""cardboard shapes"", ""cardboard pieces"", ""cardboard shapes"", ""cardboard pieces"", ""cardboard shapes"", ""cardboard pieces"", ""cardboard shapes"", ""cardboard pieces"", ""cardboard shapes"", ""cardboard pieces"", ""cardboard shapes"", ""cardboard pieces"", ""cardboard shapes"", ""cardboard pieces"", ""cardboard shapes"", ""cardboard pieces"", ""cardboard shapes"", ""cardboard pieces"", ""cardboard shapes"", ""cardboard pieces"", ""cardboard shapes"", ""cardboard pieces"", ""cardboard shapes"", ""cardboard pieces"", ""cardboard shapes"", ""cardboard pieces"", ""cardboard shapes"", ""cardboard pieces"", ""cardboard shapes"", ""cardboard pieces"", ""cardboard shapes"", ""card",0,1,0.0,0.0,0.0

results/omni_finetune/verified_public/xperience10m_qwen3_omni_128ep_96train_16val_16test_valmon_20260605_eval/package_audit.json

@@ -0,0 +1,24 @@
+{
+  "status": "pass",
+  "package_dir": "results/omni_finetune/verified_public/xperience10m_qwen3_omni_128ep_96train_16val_16test_valmon_20260605_eval",
+  "backbone": "qwen3_omni_lora",
+  "required_eval_files": [
+    "metrics.json",
+    "predictions.jsonl",
+    "predictions.csv",
+    "per_class_metrics.csv",
+    "confusion_matrix.csv",
+    "RUN_REPORT.md"
+  ],
+  "primary_metrics": [
+    "action_macro_f1",
+    "contact_accuracy",
+    "held_out_episode_count",
+    "json_validity_rate",
+    "next_action_accuracy",
+    "object_micro_f1",
+    "subtask_accuracy",
+    "transition_accuracy"
+  ],
+  "issues": []
+}

results/omni_finetune/verified_public/xperience10m_qwen3_omni_128ep_96train_16val_16test_valmon_20260605_eval/training/progress.jsonl

@@ -0,0 +1,47 @@
+{"event": "setup_done", "run_id": "xperience10m_qwen3_omni_128ep_96train_16val_16test_valmon_20260605_lora", "dataset_jsonl": "<project>/results/omni_finetune/xperience10m_qwen3_omni_128ep_96train_16val_16test_valmon_20260605_dataset/dataset.jsonl", "num_processes": 8, "num_train_samples": 2848, "num_val_samples": 512, "rank0_samples_per_epoch": 356, "backbone_id": "qwen3_omni_lora", "dataset_contract": "xperience10m_episode_json_qa_v1", "training_objective": "structured_episode_understanding_json_qa", "loss_mode": "answer_token_ce", "loss_logit_tail_only": true, "timestamp": 1780646466.9641912}
+{"event": "model_load_start", "run_id": "xperience10m_qwen3_omni_128ep_96train_16val_16test_valmon_20260605_lora", "model_id": "<workspace-parent>/modelscope_models/Qwen__Qwen3-Omni-30B-A3B-Instruct", "backbone_id": "qwen3_omni_lora", "device_map": "none", "dtype": "bfloat16", "timestamp": 1780646466.9644935}
+{"event": "model_load_done", "run_id": "xperience10m_qwen3_omni_128ep_96train_16val_16test_valmon_20260605_lora", "timestamp": 1780646529.778516}
+{"event": "accelerator_prepare_start", "run_id": "xperience10m_qwen3_omni_128ep_96train_16val_16test_valmon_20260605_lora", "timestamp": 1780646529.7835758}
+{"event": "accelerator_prepare_done", "run_id": "xperience10m_qwen3_omni_128ep_96train_16val_16test_valmon_20260605_lora", "timestamp": 1780647258.9977164}
+{"event": "train_loop_start", "run_id": "xperience10m_qwen3_omni_128ep_96train_16val_16test_valmon_20260605_lora", "model_id": "<workspace-parent>/modelscope_models/Qwen__Qwen3-Omni-30B-A3B-Instruct", "dataset_jsonl": "<project>/results/omni_finetune/xperience10m_qwen3_omni_128ep_96train_16val_16test_valmon_20260605_dataset/dataset.jsonl", "num_processes": 8, "num_train_samples": 2848, "num_val_samples": 512, "rank_samples_per_epoch": 356, "epochs": 1, "timestamp": 1780647259.0075266}
+{"event": "train_step", "epoch": 1, "global_step": 10, "rank0_seen": 10, "rank0_samples_per_epoch": 356, "rank0_batch_loss": 1.230179786682129, "timestamp": 1780647366.8140962}
+{"event": "train_step", "epoch": 1, "global_step": 20, "rank0_seen": 20, "rank0_samples_per_epoch": 356, "rank0_batch_loss": 1.2016743421554565, "timestamp": 1780647469.5297396}
+{"event": "train_step", "epoch": 1, "global_step": 30, "rank0_seen": 30, "rank0_samples_per_epoch": 356, "rank0_batch_loss": 1.1469742059707642, "timestamp": 1780647572.59419}
+{"event": "train_step", "epoch": 1, "global_step": 40, "rank0_seen": 40, "rank0_samples_per_epoch": 356, "rank0_batch_loss": 1.0939034223556519, "timestamp": 1780647677.915944}
+{"event": "train_step", "epoch": 1, "global_step": 50, "rank0_seen": 50, "rank0_samples_per_epoch": 356, "rank0_batch_loss": 1.0159685611724854, "timestamp": 1780647782.0573242}
+{"event": "train_step", "epoch": 1, "global_step": 60, "rank0_seen": 60, "rank0_samples_per_epoch": 356, "rank0_batch_loss": 0.9657138586044312, "timestamp": 1780647884.5396533}
+{"event": "train_step", "epoch": 1, "global_step": 70, "rank0_seen": 70, "rank0_samples_per_epoch": 356, "rank0_batch_loss": 0.9127762913703918, "timestamp": 1780647988.1080294}
+{"event": "train_step", "epoch": 1, "global_step": 80, "rank0_seen": 80, "rank0_samples_per_epoch": 356, "rank0_batch_loss": 0.8944218754768372, "timestamp": 1780648093.5427563}
+{"event": "train_step", "epoch": 1, "global_step": 90, "rank0_seen": 90, "rank0_samples_per_epoch": 356, "rank0_batch_loss": 0.8086783289909363, "timestamp": 1780648199.2200294}
+{"event": "train_step", "epoch": 1, "global_step": 100, "rank0_seen": 100, "rank0_samples_per_epoch": 356, "rank0_batch_loss": 0.7389833331108093, "timestamp": 1780648302.853592}
+{"event": "train_step", "epoch": 1, "global_step": 110, "rank0_seen": 110, "rank0_samples_per_epoch": 356, "rank0_batch_loss": 0.6932628750801086, "timestamp": 1780648406.7458248}
+{"event": "train_step", "epoch": 1, "global_step": 120, "rank0_seen": 120, "rank0_samples_per_epoch": 356, "rank0_batch_loss": 0.6040449738502502, "timestamp": 1780648513.4280822}
+{"event": "train_step", "epoch": 1, "global_step": 130, "rank0_seen": 130, "rank0_samples_per_epoch": 356, "rank0_batch_loss": 0.4854508638381958, "timestamp": 1780648618.7809823}
+{"event": "train_step", "epoch": 1, "global_step": 140, "rank0_seen": 140, "rank0_samples_per_epoch": 356, "rank0_batch_loss": 0.44700533151626587, "timestamp": 1780648722.3871136}
+{"event": "train_step", "epoch": 1, "global_step": 150, "rank0_seen": 150, "rank0_samples_per_epoch": 356, "rank0_batch_loss": 0.3768823444843292, "timestamp": 1780648826.5495484}
+{"event": "train_step", "epoch": 1, "global_step": 160, "rank0_seen": 160, "rank0_samples_per_epoch": 356, "rank0_batch_loss": 0.3188228905200958, "timestamp": 1780648933.4630425}
+{"event": "train_step", "epoch": 1, "global_step": 170, "rank0_seen": 170, "rank0_samples_per_epoch": 356, "rank0_batch_loss": 0.21572333574295044, "timestamp": 1780649040.920719}
+{"event": "train_step", "epoch": 1, "global_step": 180, "rank0_seen": 180, "rank0_samples_per_epoch": 356, "rank0_batch_loss": 0.15064814686775208, "timestamp": 1780649145.727296}
+{"event": "train_step", "epoch": 1, "global_step": 190, "rank0_seen": 190, "rank0_samples_per_epoch": 356, "rank0_batch_loss": 0.10380999743938446, "timestamp": 1780649250.9975677}
+{"event": "train_step", "epoch": 1, "global_step": 200, "rank0_seen": 200, "rank0_samples_per_epoch": 356, "rank0_batch_loss": 0.08768776804208755, "timestamp": 1780649358.2915566}
+{"event": "train_step", "epoch": 1, "global_step": 210, "rank0_seen": 210, "rank0_samples_per_epoch": 356, "rank0_batch_loss": 0.04660193994641304, "timestamp": 1780649465.2272158}
+{"event": "train_step", "epoch": 1, "global_step": 220, "rank0_seen": 220, "rank0_samples_per_epoch": 356, "rank0_batch_loss": 0.04696264863014221, "timestamp": 1780649570.7967262}
+{"event": "train_step", "epoch": 1, "global_step": 230, "rank0_seen": 230, "rank0_samples_per_epoch": 356, "rank0_batch_loss": 0.06465131789445877, "timestamp": 1780649675.9557557}
+{"event": "train_step", "epoch": 1, "global_step": 240, "rank0_seen": 240, "rank0_samples_per_epoch": 356, "rank0_batch_loss": 0.030367165803909302, "timestamp": 1780649782.9315662}
+{"event": "train_step", "epoch": 1, "global_step": 250, "rank0_seen": 250, "rank0_samples_per_epoch": 356, "rank0_batch_loss": 0.044979315251111984, "timestamp": 1780649890.433016}
+{"event": "train_step", "epoch": 1, "global_step": 260, "rank0_seen": 260, "rank0_samples_per_epoch": 356, "rank0_batch_loss": 0.06045611947774887, "timestamp": 1780649995.4567258}
+{"event": "train_step", "epoch": 1, "global_step": 270, "rank0_seen": 270, "rank0_samples_per_epoch": 356, "rank0_batch_loss": 0.04012652114033699, "timestamp": 1780650099.2143335}
+{"event": "train_step", "epoch": 1, "global_step": 280, "rank0_seen": 280, "rank0_samples_per_epoch": 356, "rank0_batch_loss": 0.05115583539009094, "timestamp": 1780650206.8328776}
+{"event": "train_step", "epoch": 1, "global_step": 290, "rank0_seen": 290, "rank0_samples_per_epoch": 356, "rank0_batch_loss": 0.011530512943863869, "timestamp": 1780650313.893128}
+{"event": "train_step", "epoch": 1, "global_step": 300, "rank0_seen": 300, "rank0_samples_per_epoch": 356, "rank0_batch_loss": 0.055682338774204254, "timestamp": 1780650418.009996}
+{"event": "train_step", "epoch": 1, "global_step": 310, "rank0_seen": 310, "rank0_samples_per_epoch": 356, "rank0_batch_loss": 0.03179597854614258, "timestamp": 1780650521.7131195}
+{"event": "train_step", "epoch": 1, "global_step": 320, "rank0_seen": 320, "rank0_samples_per_epoch": 356, "rank0_batch_loss": 0.018152443692088127, "timestamp": 1780650628.7296107}
+{"event": "train_step", "epoch": 1, "global_step": 330, "rank0_seen": 330, "rank0_samples_per_epoch": 356, "rank0_batch_loss": 0.05016171559691429, "timestamp": 1780650734.1827128}
+{"event": "train_step", "epoch": 1, "global_step": 340, "rank0_seen": 340, "rank0_samples_per_epoch": 356, "rank0_batch_loss": 0.014182890765368938, "timestamp": 1780650837.4421883}
+{"event": "train_step", "epoch": 1, "global_step": 350, "rank0_seen": 350, "rank0_samples_per_epoch": 356, "rank0_batch_loss": 0.024329334497451782, "timestamp": 1780650940.6010504}
+{"event": "train_step", "epoch": 1, "global_step": 356, "rank0_seen": 356, "rank0_samples_per_epoch": 356, "rank0_batch_loss": 0.035586968064308167, "timestamp": 1780651004.4774256}
+{"event": "epoch_end", "epoch": 1, "train_loss": 0.41304643672440994, "val_loss": 0.0330660454928875, "global_step": 356, "timestamp": 1780651173.976799}
+{"event": "save_start", "checkpoint_dir": "<project>/checkpoints/xperience10m_qwen3_omni_128ep_96train_16val_16test_valmon_20260605_lora/adapter_lora", "save_mode": "trainable_lora_state_dict", "timestamp": 1780651173.9771187}
+{"event": "save_state_dict_built", "checkpoint_dir": "<project>/checkpoints/xperience10m_qwen3_omni_128ep_96train_16val_16test_valmon_20260605_lora/adapter_lora", "state_source": "accelerator_full_state_dict", "trainable_tensors": 576, "trainable_bytes": 69206016, "shape_summary": {"adapter_tensors": 576, "adapter_bytes": 69206016, "prefixes": {"audio_tower": {"tensors": 192, "numel": 3932160}, "model": {"tensors": 384, "numel": 13369344}}}, "timestamp": 1780651307.4472642}
+{"event": "save_done", "checkpoint_dir": "<project>/checkpoints/xperience10m_qwen3_omni_128ep_96train_16val_16test_valmon_20260605_lora/adapter_lora", "timestamp": 1780651307.977997}
+{"event": "complete", "checkpoint_dir": "<project>/checkpoints/xperience10m_qwen3_omni_128ep_96train_16val_16test_valmon_20260605_lora/adapter_lora", "timestamp": 1780651307.9785495}

results/omni_finetune/verified_public/xperience10m_qwen3_omni_128ep_96train_16val_16test_valmon_20260605_eval/training/training_metadata.json

@@ -0,0 +1,79 @@
+{
+  "run_id": "xperience10m_qwen3_omni_128ep_96train_16val_16test_valmon_20260605_lora",
+  "model_id": "<workspace-parent>/modelscope_models/Qwen__Qwen3-Omni-30B-A3B-Instruct",
+  "backbone": {
+    "id": "qwen3_omni_lora",
+    "display_name": "Qwen3-Omni LoRA",
+    "status": "implemented",
+    "model_family": "Qwen3-Omni",
+    "dataset_contract": "xperience10m_episode_json_qa_v1",
+    "training_objective": "structured_episode_understanding_json_qa",
+    "split_policy": {
+      "unit": "episode",
+      "default_counts": {
+        "train": 96,
+        "val": 16,
+        "test": 16
+      },
+      "leakage_guard": "episode_id and path-qualified session split checks"
+    },
+    "modalities": {
+      "direct_inputs": [
+        "multi-camera rendered mosaic video",
+        "AAC audio extracted from fisheye_cam0 MP4",
+        "language prompt and label options"
+      ],
+      "bridged_inputs": [
+        "depth and confidence",
+        "pose and SLAM trajectory",
+        "body and hand mocap",
+        "IMU acceleration and gyro",
+        "object/contact summary features"
+      ],
+      "excluded_inputs": [
+        "visualization.rrd"
+      ]
+    },
+    "primary_metrics": [
+      "json_validity_rate",
+      "action_macro_f1",
+      "subtask_accuracy",
+      "transition_accuracy",
+      "next_action_accuracy",
+      "contact_accuracy",
+      "object_micro_f1",
+      "held_out_episode_count"
+    ],
+    "config_path": "<project>/configs/omni_backbones/qwen3_omni_lora.json"
+  },
+  "dataset_jsonl": "<project>/results/omni_finetune/xperience10m_qwen3_omni_128ep_96train_16val_16test_valmon_20260605_dataset/dataset.jsonl",
+  "checkpoint_dir": "<project>/checkpoints/xperience10m_qwen3_omni_128ep_96train_16val_16test_valmon_20260605_lora/adapter_lora",
+  "num_processes": 8,
+  "num_train_samples": 2848,
+  "num_val_samples": 512,
+  "history": [
+    {
+      "epoch": 1,
+      "train_loss": 0.41304643672440994,
+      "val_loss": 0.0330660454928875,
+      "global_step": 356
+    }
+  ],
+  "lora": {
+    "r": 16,
+    "alpha": 32,
+    "dropout": 0.05,
+    "target_modules": [
+      "q_proj",
+      "k_proj",
+      "v_proj",
+      "o_proj",
+      "gate_proj",
+      "up_proj",
+      "down_proj"
+    ]
+  },
+  "use_audio_in_video": false,
+  "loss_mode": "answer_token_ce",
+  "loss_logit_tail_only": true
+}

results/omni_finetune/verified_public/xperience10m_qwen3_omni_128ep_96train_16val_16test_valmon_20260605_eval/validation/eval.json

@@ -0,0 +1,81 @@
+{
+  "status": "pass",
+  "summary": {
+    "run_id": "xperience10m_qwen3_omni_128ep_96train_16val_16test_valmon_20260605",
+    "dataset_run_id": "xperience10m_qwen3_omni_128ep_96train_16val_16test_valmon_20260605",
+    "train_run_id": "xperience10m_qwen3_omni_128ep_96train_16val_16test_valmon_20260605",
+    "eval_run_id": "xperience10m_qwen3_omni_128ep_96train_16val_16test_valmon_20260605",
+    "backbone": "qwen3_omni_lora",
+    "backbone_status": "implemented",
+    "checkpoint_gate": "lora_safetensors_shape_check",
+    "required_stage": "eval",
+    "workspace": "<project>",
+    "manifest": {
+      "path": "<project>/results/omni_finetune/xperience10m_qwen3_omni_128ep_96train_16val_16test_valmon_20260605/episode_manifest.json",
+      "episode_count": 128,
+      "split_counts": {
+        "test": 16,
+        "train": 96,
+        "val": 16
+      },
+      "session_leakage": []
+    },
+    "dataset": {
+      "dataset_dir": "<project>/results/omni_finetune/xperience10m_qwen3_omni_128ep_96train_16val_16test_valmon_20260605_dataset",
+      "manifest_path": "<project>/results/omni_finetune/xperience10m_qwen3_omni_128ep_96train_16val_16test_valmon_20260605_dataset/dataset_manifest.json",
+      "dataset_path": "<project>/results/omni_finetune/xperience10m_qwen3_omni_128ep_96train_16val_16test_valmon_20260605_dataset/dataset.jsonl",
+      "manifest_num_samples": 3808,
+      "row_count": 3808,
+      "sample_split_counts": {
+        "train": 2848,
+        "val": 512,
+        "test": 448
+      },
+      "episode_split_counts": {
+        "train": 89,
+        "val": 16,
+        "test": 14
+      },
+      "skipped_episodes": 9
+    },
+    "training": {
+      "train_dir": "<project>/results/omni_finetune/xperience10m_qwen3_omni_128ep_96train_16val_16test_valmon_20260605_lora",
+      "checkpoint_candidates": [
+        "<project>/checkpoints/xperience10m_qwen3_omni_128ep_96train_16val_16test_valmon_20260605/adapter_lora",
+        "<project>/checkpoints/xperience10m_qwen3_omni_128ep_96train_16val_16test_valmon_20260605_lora/adapter_lora"
+      ],
+      "checkpoint_gate": "lora_safetensors_shape_check",
+      "required_training_files": [
+        "training_metadata.json",
+        "progress.jsonl",
+        "adapter_config.json",
+        "adapter_model.safetensors"
+      ],
+      "checkpoint_dir": "<project>/checkpoints/xperience10m_qwen3_omni_128ep_96train_16val_16test_valmon_20260605_lora/adapter_lora",
+      "num_processes": 8,
+      "num_train_samples": 2848,
+      "num_val_samples": 512,
+      "history_len": 1,
+      "checkpoint_dir_recorded": "<project>/checkpoints/xperience10m_qwen3_omni_128ep_96train_16val_16test_valmon_20260605_lora/adapter_lora"
+    },
+    "eval": {
+      "eval_dir": "<project>/results/omni_finetune/xperience10m_qwen3_omni_128ep_96train_16val_16test_valmon_20260605_eval",
+      "required_eval_files": [
+        "metrics.json",
+        "predictions.jsonl",
+        "predictions.csv",
+        "per_class_metrics.csv",
+        "confusion_matrix.csv",
+        "RUN_REPORT.md"
+      ],
+      "eval_split": "test",
+      "num_eval_episodes": 14,
+      "held_out_episode_count": 14,
+      "json_validity_rate": 0.875,
+      "action_macro_f1": 0.0026621494447581404,
+      "prediction_file": "predictions.jsonl",
+      "prediction_rows": 448
+    }
+  },
+  "issues": []
+}

results/omni_finetune/verified_public/xperience10m_qwen3_omni_128ep_96train_16val_16test_valmon_20260605_eval/verified_result_summary.json

@@ -0,0 +1,168 @@
+{
+  "status": "verified",
+  "backbone": "qwen3_omni_lora",
+  "backbone_display_name": "Qwen3-Omni LoRA",
+  "dataset_contract": "xperience10m_episode_json_qa_v1",
+  "training_objective": "structured_episode_understanding_json_qa",
+  "dataset_run_id": "xperience10m_qwen3_omni_128ep_96train_16val_16test_valmon_20260605",
+  "train_run_id": "xperience10m_qwen3_omni_128ep_96train_16val_16test_valmon_20260605_lora",
+  "eval_run_id": "xperience10m_qwen3_omni_128ep_96train_16val_16test_valmon_20260605_eval",
+  "dataset": {
+    "num_samples": 3808,
+    "num_episodes": 119,
+    "split_counts": {
+      "train": 2848,
+      "val": 512,
+      "test": 448
+    },
+    "skipped_episodes": 9
+  },
+  "training": {
+    "num_processes": 8,
+    "num_train_samples": 2848,
+    "num_val_samples": 512,
+    "history": [
+      {
+        "epoch": 1,
+        "train_loss": 0.41304643672440994,
+        "val_loss": 0.0330660454928875,
+        "global_step": 356
+      }
+    ]
+  },
+  "eval": {
+    "eval_split": "test",
+    "num_samples": 448,
+    "prediction_file": "predictions.jsonl",
+    "prediction_rows": 448,
+    "num_eval_episodes": 14,
+    "held_out_episode_count": 14,
+    "primary_metrics": {
+      "json_validity_rate": 0.875,
+      "action_macro_f1": 0.0026621494447581404,
+      "subtask_accuracy": 0.006696428571428571,
+      "transition_accuracy": 0.8504464285714286,
+      "next_action_accuracy": 0.024553571428571428,
+      "contact_accuracy": 0.6450892857142857,
+      "object_micro_f1": 0.22299431459254582,
+      "held_out_episode_count": 14
+    }
+  },
+  "validation_summary": {
+    "run_id": "xperience10m_qwen3_omni_128ep_96train_16val_16test_valmon_20260605",
+    "dataset_run_id": "xperience10m_qwen3_omni_128ep_96train_16val_16test_valmon_20260605",
+    "train_run_id": "xperience10m_qwen3_omni_128ep_96train_16val_16test_valmon_20260605",
+    "eval_run_id": "xperience10m_qwen3_omni_128ep_96train_16val_16test_valmon_20260605",
+    "backbone": "qwen3_omni_lora",
+    "backbone_status": "implemented",
+    "checkpoint_gate": "lora_safetensors_shape_check",
+    "required_stage": "eval",
+    "workspace": "<project>",
+    "manifest": {
+      "path": "<project>/results/omni_finetune/xperience10m_qwen3_omni_128ep_96train_16val_16test_valmon_20260605/episode_manifest.json",
+      "episode_count": 128,
+      "split_counts": {
+        "test": 16,
+        "train": 96,
+        "val": 16
+      },
+      "session_leakage": []
+    },
+    "dataset": {
+      "dataset_dir": "<project>/results/omni_finetune/xperience10m_qwen3_omni_128ep_96train_16val_16test_valmon_20260605_dataset",
+      "manifest_path": "<project>/results/omni_finetune/xperience10m_qwen3_omni_128ep_96train_16val_16test_valmon_20260605_dataset/dataset_manifest.json",
+      "dataset_path": "<project>/results/omni_finetune/xperience10m_qwen3_omni_128ep_96train_16val_16test_valmon_20260605_dataset/dataset.jsonl",
+      "manifest_num_samples": 3808,
+      "row_count": 3808,
+      "sample_split_counts": {
+        "train": 2848,
+        "val": 512,
+        "test": 448
+      },
+      "episode_split_counts": {
+        "train": 89,
+        "val": 16,
+        "test": 14
+      },
+      "skipped_episodes": 9
+    },
+    "training": {
+      "train_dir": "<project>/results/omni_finetune/xperience10m_qwen3_omni_128ep_96train_16val_16test_valmon_20260605_lora",
+      "checkpoint_candidates": [
+        "<project>/checkpoints/xperience10m_qwen3_omni_128ep_96train_16val_16test_valmon_20260605/adapter_lora",
+        "<project>/checkpoints/xperience10m_qwen3_omni_128ep_96train_16val_16test_valmon_20260605_lora/adapter_lora"
+      ],
+      "checkpoint_gate": "lora_safetensors_shape_check",
+      "required_training_files": [
+        "training_metadata.json",
+        "progress.jsonl",
+        "adapter_config.json",
+        "adapter_model.safetensors"
+      ],
+      "checkpoint_dir": "<project>/checkpoints/xperience10m_qwen3_omni_128ep_96train_16val_16test_valmon_20260605_lora/adapter_lora",
+      "num_processes": 8,
+      "num_train_samples": 2848,
+      "num_val_samples": 512,
+      "history_len": 1,
+      "checkpoint_dir_recorded": "<project>/checkpoints/xperience10m_qwen3_omni_128ep_96train_16val_16test_valmon_20260605_lora/adapter_lora"
+    },
+    "eval": {
+      "eval_dir": "<project>/results/omni_finetune/xperience10m_qwen3_omni_128ep_96train_16val_16test_valmon_20260605_eval",
+      "required_eval_files": [
+        "metrics.json",
+        "predictions.jsonl",
+        "predictions.csv",
+        "per_class_metrics.csv",
+        "confusion_matrix.csv",
+        "RUN_REPORT.md"
+      ],
+      "eval_split": "test",
+      "num_eval_episodes": 14,
+      "held_out_episode_count": 14,
+      "json_validity_rate": 0.875,
+      "action_macro_f1": 0.0026621494447581404,
+      "prediction_file": "predictions.jsonl",
+      "prediction_rows": 448
+    }
+  },
+  "included_files": [
+    "dataset/dataset_manifest.json",
+    "dataset/episode_manifest.json",
+    "eval/RUN_REPORT.md",
+    "eval/confusion_matrix.csv",
+    "eval/metrics.json",
+    "eval/per_class_metrics.csv",
+    "eval/predictions.csv",
+    "eval/predictions.jsonl",
+    "training/progress.jsonl",
+    "training/training_metadata.json",
+    "validation/eval.json"
+  ],
+  "required_eval_files": [
+    "metrics.json",
+    "predictions.jsonl",
+    "predictions.csv",
+    "per_class_metrics.csv",
+    "confusion_matrix.csv",
+    "RUN_REPORT.md"
+  ],
+  "public_package_allowed": [
+    "metrics",
+    "predictions",
+    "confusion matrices",
+    "run reports",
+    "episode and dataset manifests",
+    "training metadata",
+    "validation summaries"
+  ],
+  "public_package_forbidden": [
+    "raw MP4",
+    "annotation HDF5",
+    "Rerun RRD",
+    "base-model weights",
+    "LoRA adapter weights",
+    "full checkpoints",
+    "large archives"
+  ],
+  "excluded_policy": "Raw Xperience-10M files, base-model weights, adapter or checkpoint weights, full checkpoints, and large archives are not included."
+}