ARTIFACT_GUIDE.md

Artifact Guide

This guide is the human-readable map for the public Ropedia Xperience-10M task suite artifacts. It is organized around what a reader usually wants to do: understand the project, inspect the sample episode, compare baselines, read the task results, and follow the Qwen3-Omni scale-up path.

Start Here

Artifact	Why to open it first
PROJECT_STATUS.md	Gives the fastest current-state table: implemented, in staging, and outside current scope.
RESEARCH_ROADMAP.md	Shows the roadmap from public-sample task development to multi-episode data preparation, Qwen3-Omni LoRA, robustness runs, and larger omni-model extensions.
FOUNDATION_MODEL_PLAN.md	Explains which foundation backbones fit which Xperience-10M objective: Qwen3-Omni first, Cosmos 3 for world modeling, and VLA/policy models after action-target conversion.
EVALUATION_PROTOCOL.md	Defines the task unit, chronological split, metrics, leakage controls, and current limitations.
REPRODUCIBILITY.md	Defines public reproduction commands, expected outputs, and unreproducible boundaries.
results/audio_ablation/AUDIO_ABLATION_SUMMARY.md	Shows measured current-audio and raw log-mel replacement deltas across the 12 task contracts.
docs/single_episode_explorer.html	Gives a static window-level explorer for the public sample episode.
XPERIENCE10M_DATASET_CARD_ALIGNMENT.md	Optional detail for readers who need official dataset and access-term context.

Dataset Context

Artifact	What it shows
XPERIENCE10M_DATASET_CARD_ALIGNMENT.md	Human-readable summary of the official gated Xperience-10M dataset, public sample, modalities, access terms, intended uses, and limitations.
docs/data/xperience10m_dataset_card_alignment.json	Machine-readable dataset-context bundle for the website and Hub pages.
SOURCE_ALIGNMENT_AUDIT.md	Supporting provenance note for maintainers who want to inspect how public dataset descriptions were checked.
docs/data/source_alignment_audit.json	Machine-readable provenance record for generated project pages.
scripts/validate_source_alignment.py	Maintenance script for refreshing the dataset-context note.

Evaluation Protocol

Artifact	What it shows
EVALUATION_PROTOCOL.md	Human-readable task protocol: window unit, chronological split, input/target contracts, primary metrics, leakage controls, and current limitations.
docs/data/evaluation_protocol.json	Machine-readable protocol generated from committed task metrics.
scripts/build_evaluation_protocol.py	Regenerates the protocol from docs/data/summary_metrics.json and source task artifacts.

Visual Evidence

Artifact	What it shows
FIGURE_INDEX.md	Human-readable catalog of public visual assets, dimensions, hashes, roles, and source scripts.
docs/data/figure_index.json	Machine-readable visual asset index mirrored to the website, artifact dataset, and model repo.
scripts/build_figure_index.py	Regenerates visual-asset hashes, dimensions, and source-script provenance.
docs/data/brand_assets.json	Machine-readable logo/brand manifest for the website, README, Hugging Face cards, favicon, app icon, and social preview.
docs/assets/brand/xperience10m-logo-social-card.png	Project logo card used by README and Hugging Face cards.
scripts/build_brand_assets.py	Regenerates deterministic logo derivatives, favicon variants, app icons, and the social card from the generated logo mark.
docs/assets/task_suite_infographic.png	Primary 12-task suite map with sample modality thumbnails.
docs/assets/pipeline_diagram.png	Episode-to-task pipeline overview.
docs/assets/task_architectures.png	Minimal and neural task-head architecture map.

Data Contract

Artifact	What it shows
results/episode_task_suite/windows.csv	The sample episode is converted into 1,161 aligned 20-frame windows.
results/episode_task_suite/feature_manifest.json	The current input vector has 8,546 dimensions with explicit modality-group boundaries, including a 168-d audio group.
results/episode_task_suite/available_modalities.json	The sample modality coverage is recorded, including the current audio-featurization status.
results/audio_ablation/raw_logmel_fisheye_cam0_sr16000_mels64_fft512_hop160.npz	Derived 588-d raw log-mel window features decoded from the local public-sample MP4 audio stream; raw audio itself is not redistributed.
docs/data/modality_atlas.json	The responsive website modality cards and derived thumbnail assets are documented without redistributing raw data.
docs/assets/modalities/	Small public-sample thumbnails used by the readable modality atlas.

Task Evidence

Artifact	What it shows
results/episode_task_suite/summary_report.json	The 12 task contracts, chronological split, and minimal/neural metrics.
results/episode_task_suite/neural_mlp/	Matching PyTorch MLP heads for the same task contracts and feature windows.
results/episode_task_suite/research_directions/	Mapping from the 12 tasks to the four Ropedia research directions.
results/episode_task_suite/research_direction_extensions/	Four additional coded probes, one per research direction.
results/episode_task_suite/task_walkthroughs/	Human-readable research names and case studies explaining input, process modules, output, metric, limitation, and the website task-player data.
results/audio_ablation/audio_ablation_metrics.csv	All 72 measured audio rows: 12 tasks times six variants, including no-audio, audio-only, alternate-audio-only, representation replacement, and all-input variants.
results/audio_ablation/audio_delta_summary.csv	Compact per-task audio delta table for quick manual inspection.
scripts/audio_ablation_and_raw_upgrade.py	Regenerates audio contribution results from real task-suite artifacts plus the local public-sample MP4.
scripts/validate_task_surface.py	Fails publication if public task cards drift back to raw artifact ids or lose their thumbnail/player wiring.

Reproducibility

Artifact	What it shows
REPRODUCIBILITY.md	Public commands, expected outputs, and non-reproducible boundaries are explicit.
docs/data/reproducibility_matrix.json	Machine-readable command matrix for the website and Hub pages.
notes/reproducibility_audit.md	The last exact metric rebuild reproduced the public-sample metrics and matched committed artifacts.

Public Pages

Surface	Purpose
GitHub Pages dashboard	Primary public website and visual research flow.
Hugging Face Space	Static app mirror for HF users.
HF artifact dataset	Derived CSV/JSON/Markdown/figure artifacts without raw Xperience-10M data.
HF baseline model repo	Lightweight minimal and neural task-head model files.
HF collection	One grouped landing page for the Space, artifact dataset, and baseline model repo.

The public pages are meant to be the normal reader path. Supporting maintenance checks remain in the repo, but they are not required for understanding the research project.

Scale-Up Readiness

Artifact	Current status
results/omni_finetune/DATA_ACCESS_STATUS.md	Summarizes the staging requirement before the held-out Qwen3-Omni pilot can report metrics.
results/omni_finetune/MULTI_EPISODE_ACCESS_STATUS.md	Documents the public multi-episode access path, selected 128-episode pilot plan, and data requirements.
scripts/omni/discover_xperience10m_sources.py	Discovery gate for valid multi-episode Xperience-10M sources.
scripts/omni/train_qwen3_omni_lora.py	Training entrypoint for the Qwen3-Omni LoRA pilot after the data gate passes.
FOUNDATION_MODEL_PLAN.md	Adds the post-data-gate backbone selection plan: Qwen3-Omni first, Cosmos 3 for world modeling, and OpenVLA/openpi/GR00T for policy/action branches.
docs/data/foundation_model_plan.json	Machine-readable model-family registry with source links, entry conditions, and evaluation additions.

What Is Not Included

The public repo and Hugging Face mirrors do not redistribute raw Xperience-10M videos, raw annotation.hdf5, gated private dataset files, full Qwen weights, or large full checkpoints. Dataset use remains governed by the official Ropedia/Xperience-10M terms.