Publish Xperience-10M minimal and neural derived artifacts

PROJECT_README.md

@@ -14,6 +14,9 @@ into:
 - 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,
+- a next TODO track for Qwen3-Omni fine-tuning and sensor-bridge evaluation,
 - metrics, predictions, model weights, manifests, charts, and a static website,
 - a clear explanation of what a single episode can and cannot prove.
 
@@ -33,11 +36,11 @@ This repo keeps that distinction explicit:
 
 Start with the visual dashboard:
 
-**https://chaoyue0307.github.io/ropedia-episode-task-suite/**
+**[chaoyue0307.github.io/ropedia-episode-task-suite](https://chaoyue0307.github.io/ropedia-episode-task-suite/)**
 
 Hugging Face Space app:
 
-**https://cy0307-ropedia-episode-task-suite.static.hf.space/**
+**[cy0307-ropedia-episode-task-suite.static.hf.space](https://cy0307-ropedia-episode-task-suite.static.hf.space/)**
 
 ## Read This Project In Three Layers
 
@@ -45,28 +48,29 @@ Hugging Face Space app:
 | --- | --- | --- |
 | Data contract | `windows.csv`, `feature_manifest.json`, modality manifests | Confirms what each sample window contains before modeling |
 | Minimal heads | softmax, ridge projection/regression, multi-label logistic heads | Keeps every input/output contract visible and debuggable |
+| 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 claims reviewable without rerunning first |
 
 ## Links
 
 | Resource | Link |
 | --- | --- |
-| This GitHub repo | https://github.com/ChaoYue0307/ropedia-episode-task-suite |
-| This project website | https://chaoyue0307.github.io/ropedia-episode-task-suite/ |
-| This Hugging Face Space | https://huggingface.co/spaces/cy0307/ropedia-episode-task-suite |
-| Live Hugging Face static app | https://cy0307-ropedia-episode-task-suite.static.hf.space/ |
-| Derived artifacts on Hugging Face | https://huggingface.co/datasets/cy0307/ropedia-episode-task-suite-artifacts |
-| Minimal baseline models on Hugging Face | https://huggingface.co/cy0307/ropedia-minimal-task-baselines |
-| Hugging Face collection | https://huggingface.co/collections/cy0307/ropedia-episode-task-suite |
-| Xperience-10M dataset website | https://ropedia.com/dataset |
-| Xperience-10M release page | https://ropedia.com/blog/20260316_xperience_10m |
-| Ropedia GitHub organization | https://github.com/Ropedia |
-| HOMIE Toolkit | https://github.com/Ropedia/HOMIE-toolkit |
-| Xperience-10M Hugging Face dataset | https://huggingface.co/datasets/ropedia-ai/xperience-10m |
-| Xperience-10M sample on Hugging Face | https://huggingface.co/datasets/ropedia-ai/xperience-10m-sample |
-| Ropedia Hugging Face organization | https://huggingface.co/ropedia-ai |
-
-![ChatGPT-image-backed Xperience-10M 12-task infographic](docs/assets/task_suite_infographic.png?v=xperience10m)
+| This GitHub repo | [github.com/ChaoYue0307/ropedia-episode-task-suite](https://github.com/ChaoYue0307/ropedia-episode-task-suite) |
+| This project website | [chaoyue0307.github.io/ropedia-episode-task-suite](https://chaoyue0307.github.io/ropedia-episode-task-suite/) |
+| This Hugging Face Space | [huggingface.co/spaces/cy0307/ropedia-episode-task-suite](https://huggingface.co/spaces/cy0307/ropedia-episode-task-suite) |
+| Live Hugging Face static app | [cy0307-ropedia-episode-task-suite.static.hf.space](https://cy0307-ropedia-episode-task-suite.static.hf.space/) |
+| Derived artifacts on Hugging Face | [huggingface.co/datasets/cy0307/ropedia-episode-task-suite-artifacts](https://huggingface.co/datasets/cy0307/ropedia-episode-task-suite-artifacts) |
+| Minimal baseline models on Hugging Face | [huggingface.co/cy0307/ropedia-minimal-task-baselines](https://huggingface.co/cy0307/ropedia-minimal-task-baselines) |
+| Hugging Face collection | [huggingface.co/collections/cy0307/ropedia-episode-task-suite](https://huggingface.co/collections/cy0307/ropedia-episode-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) |
+
+![ChatGPT-image-backed Xperience-10M 12-task infographic](docs/assets/task_suite_infographic.png?v=xperience10m-nn)
 
 The infographic uses a ChatGPT-image-generated text-free research background and
 low-resolution modality thumbnails extracted from the public sample episode. The
@@ -76,9 +80,9 @@ with [`scripts/render_task_suite_infographic.py`](scripts/render_task_suite_info
 so the published PNG is a presentation graphic with verified labels and metrics,
 not a hallucinated metric sheet.
 
-![Verified Pipeline](docs/assets/pipeline_diagram.png?v=xperience10m)
+![Verified Pipeline](docs/assets/pipeline_diagram.png?v=xperience10m-nn)
 
-![Minimal 12-task model architectures](docs/assets/task_architectures.png?v=xperience10m)
+![Minimal and neural 12-task model architectures](docs/assets/task_architectures.png?v=xperience10m-nn)
 
 The pipeline and architecture figures use the same pattern: ChatGPT-image
 provides text-free visual backgrounds, while
@@ -99,6 +103,8 @@ 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
   generate_visualizations.py        # refreshes SVG charts + summary JSON
   render_task_suite_infographic.py  # renders the ChatGPT-image-backed PNG
   render_overview_figures.py        # renders polished pipeline/architecture PNGs
@@ -114,11 +120,14 @@ results/
   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
   omni_exploration/                 # H20/ModelScope smoke-test artifacts
 
 docs/
   index.html                        # GitHub Pages dashboard
   data/summary_metrics.json         # website-readable metrics bundle
+  data/research_directions.json     # four-track website data bundle
   assets/task_suite_infographic.png # 12-task presentation graphic
   assets/pipeline_diagram.png       # verified episode pipeline graphic
   assets/task_architectures.png     # verified 12-task minimal architecture map
@@ -202,6 +211,15 @@ cd ropedia-episode-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
@@ -300,6 +318,7 @@ python scripts/omni/plan_finetune_sample_budget.py \
 Refresh charts and the website data bundle:
 
 ```bash
+python scripts/research_direction_taxonomy.py
 python scripts/generate_visualizations.py
 python scripts/render_overview_figures.py
 python scripts/render_task_suite_infographic.py
@@ -362,6 +381,37 @@ python3 scripts/omni/upload_qwen3_omni_lora_to_hf.py \
 This script requires a valid Hugging Face token via `HF_TOKEN` or `--token`.
 Network availability to `huggingface.co` is required.
 
+## 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 8,378-d window feature vector. 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_forecast` and `contact_prediction` are direct; `timeline_action` and `object_relevance` are proxies. Neural MLP improves hand forecasting from `0.8223` to `0.1116` 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`, `modality_reconstruction`, and `misalignment_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. | Single-episode chronological split limits generalization; audio and stronger video-language backbones still need to be added. |
+| D. Scene Reconstruction & World Modeling | Early proxy tasks | Subtask/next-action, object relevance, retrieval, reconstruction, temporal order, and misalignment 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.
+
 ## Minimal 12-Task Architectures
 
 These are deliberately minimal baselines. They are useful because every
@@ -385,6 +435,13 @@ There are four reusable head families:
 | Ridge + cosine ranking | `caption_grounding`, `cross_modal_retrieval` | project one modality into another feature space, then rank candidates by cosine |
 | Multi-label logistic regression | `object_relevance` | z-score non-caption features, sigmoid object heads, threshold at 0.5 |
 
+The optional neural run keeps the same feature vectors, 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 |
@@ -413,6 +470,31 @@ The task-specific heads are:
 | Cross-modal retrieval | 0.3764 top-5 | n/a | Motion/IMU/camera retrieves matching depth/video |
 | Transition detection | 0.6552 macro-F1 | 0.9253 | Boundary F1 is 0.2143 |
 | Hand trajectory forecast | 0.8223 MPJPE | n/a | Predicts future hand-joint trajectory |
+| Neural MLP hand forecast | 0.1116 MPJPE | n/a | Same features/split, nonlinear regression head |
+| Neural MLP temporal order | 0.8718 F1 | 0.8707 | Strong improvement on adjacent-window ordering |
+| Neural MLP misalignment | 0.7335 F1 | 0.7312 | Detects shifted motion/visual pairs better than the linear head |
+
+## 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 8,378-d handcrafted window features.
+
+| Task | Neural metric | Minimal metric | Readout |
+| --- | ---: | ---: | --- |
+| `timeline_action` | 0.0263 macro-F1 | 0.0500 macro-F1 | Still blocked by unseen future classes |
+| `timeline_subtask` | 0.0175 macro-F1 | 0.0495 macro-F1 | Same single-episode split limitation |
+| `transition_detection` | 0.6485 macro-F1 | 0.6552 macro-F1 | Similar to the linear baseline |
+| `next_action` | 0.0235 macro-F1 | 0.0593 macro-F1 | Same unseen-label issue |
+| `hand_trajectory_forecast` | 0.1116 MPJPE | 0.8223 MPJPE | Neural regression improves this target |
+| `contact_prediction` | 1.0000 macro-F1 | 1.0000 macro-F1 | Degenerate one-class sample |
+| `object_relevance` | 0.1798 micro-F1 | 0.1839 micro-F1 | Similar weak object signal |
+| `caption_grounding` | 0.0178 MRR | 0.0172 MRR | Similar ranking behavior |
+| `cross_modal_retrieval` | 0.1530 MRR | 0.2634 MRR | Linear ridge remains stronger here |
+| `modality_reconstruction` | -0.0102 R2 | -0.0160 R2 | Small improvement but still weak |
+| `temporal_order` | 0.8718 F1 | 0.5487 F1 | Neural head captures local temporal structure |
+| `misalignment_detection` | 0.7335 F1 | 0.4866 F1 | Neural head improves alignment detection |
 
 The strongest single-episode self-supervised signal is cross-modal retrieval:
 motion/IMU/camera features retrieve matching depth/video windows substantially

README.md

@@ -48,6 +48,8 @@ This is the reviewable half of the project. You can inspect the task outputs, co
 - `docs/assets/*.svg` and `docs/assets/*.png`: generated diagrams, charts, and ChatGPT-image-backed overview figures
 - `docs/assets/task_suite_infographic.png`: ChatGPT-image-backed infographic with low-resolution public-sample modality thumbnails, including audio waveform context, and verified metric overlays
 - `docs/data/summary_metrics.json`: dashboard-readable summary bundle
+- `docs/data/research_directions.json`: generated four-track taxonomy for the website
+- `results/episode_task_suite/research_directions/`: JSON, CSV, and Markdown task-to-research-track mapping
 - `scripts/*.py`: reproduction scripts
 - `notes/*.md`: interpretation and reproducibility notes
 
@@ -108,6 +110,22 @@ Primary NN artifact path:
 
 `results/episode_task_suite/neural_mlp/<task>/`
 
+## Four Research Directions
+
+The current 12 tasks are organized into the four Ropedia research directions
+with two baselines per task: minimal interpretable heads and neural MLP heads.
+
+| Direction | Current status | Evidence |
+| --- | --- | --- |
+| A. Human Modeling & Motion Understanding | partially implemented | hand trajectory and contact are direct; action/object tasks are proxies |
+| B. 3D/4D Reconstruction & Neural Rendering | proxy tasks only | retrieval, reconstruction, and misalignment diagnose prerequisites |
+| C. Egocentric Vision & Interaction | strongest implemented track | 6 direct tasks plus order/alignment diagnostics |
+| D. Scene Reconstruction & World Modeling | early proxy tasks | state, object, retrieval, reconstruction, and temporal probes |
+
+Primary taxonomy artifact:
+
+`results/episode_task_suite/research_directions/research_direction_taxonomy.json`
+
 ## Pending 32-Episode Pilot
 
 | Item | Value |

docs/data/research_directions.json

@@ -0,0 +1,384 @@
+{
+  "source": "results/episode_task_suite/summary_report.json",
+  "dataset_scope": {
+    "sample_episode_count": 1,
+    "num_frames": 5821,
+    "num_windows": 1161,
+    "feature_dim": 8378,
+    "warning": "Single public sample episode; this supports pipeline/task evidence, not cross-episode generalization claims."
+  },
+  "baselines": {
+    "minimal": "Interpretable softmax, logistic, ridge, and retrieval heads over the 8,378-d window feature vector.",
+    "neural_mlp": "Small PyTorch MLP classifiers/regressors using the same features, splits, and task contracts."
+  },
+  "directions": {
+    "A": {
+      "id": "human_motion",
+      "name": "Human Modeling & Motion Understanding",
+      "focus": "Human/hand/body motion, deformation priors, human-object interaction, affordance modeling.",
+      "preferred_background": "Human pose/shape estimation, SMPL-style models, motion capture, or motion generation.",
+      "current_status": "partially implemented",
+      "current_readout": "The sample supports hand trajectory forecasting and contact/object probes, but it does not yet include a full body/shape model or multi-person priors.",
+      "next_steps": [
+        "Add SMPL/SMPL-X or MANO-style body/hand parameter targets where available.",
+        "Train sequence models over multi-episode motion trajectories instead of isolated windows.",
+        "Evaluate affordance prediction on held-out objects and held-out episodes."
+      ],
+      "tasks": [
+        "timeline_action",
+        "hand_trajectory_forecast",
+        "contact_prediction",
+        "object_relevance"
+      ],
+      "counts": {
+        "direct": 2,
+        "proxy": 2,
+        "diagnostic": 0,
+        "total_links": 4
+      }
+    },
+    "B": {
+      "id": "reconstruction_rendering",
+      "name": "3D/4D Reconstruction & Neural Rendering",
+      "focus": "Multi-view dynamic scene reconstruction, NeRF/Gaussian Splatting, novel-view synthesis.",
+      "preferred_background": "3D reconstruction, neural rendering, camera calibration, and bundle adjustment.",
+      "current_status": "proxy tasks only",
+      "current_readout": "The current suite checks cross-modal alignment and depth/video reconstruction proxies; it does not yet train a renderer or reconstruct geometry.",
+      "next_steps": [
+        "Use calibrated multi-view video plus SLAM pose to build per-episode camera trajectories.",
+        "Add depth-supervised point clouds, TSDF, Gaussian Splatting, or NeRF baselines.",
+        "Evaluate novel-view synthesis and temporal consistency across held-out views/time."
+      ],
+      "tasks": [
+        "cross_modal_retrieval",
+        "modality_reconstruction",
+        "misalignment_detection"
+      ],
+      "counts": {
+        "direct": 0,
+        "proxy": 2,
+        "diagnostic": 1,
+        "total_links": 3
+      }
+    },
+    "C": {
+      "id": "egocentric_interaction",
+      "name": "Egocentric Vision & Interaction",
+      "focus": "Egocentric action and intention understanding, hand-object interaction, gaze/attention modeling, task structure modeling.",
+      "preferred_background": "Video understanding, action recognition, or egocentric vision.",
+      "current_status": "strongest implemented track",
+      "current_readout": "Most of the 12 tasks directly target egocentric action, task state, interaction, grounding, and alignment.",
+      "next_steps": [
+        "Move from single-episode chronological splits to held-out-episode splits.",
+        "Add audio features and stronger multimodal backbones for action, intent, and grounding.",
+        "Evaluate long-horizon task success prediction and action-conditioned generation."
+      ],
+      "tasks": [
+        "timeline_action",
+        "timeline_subtask",
+        "transition_detection",
+        "next_action",
+        "hand_trajectory_forecast",
+        "contact_prediction",
+        "object_relevance",
+        "caption_grounding",
+        "cross_modal_retrieval",
+        "temporal_order",
+        "misalignment_detection"
+      ],
+      "counts": {
+        "direct": 6,
+        "proxy": 2,
+        "diagnostic": 3,
+        "total_links": 11
+      }
+    },
+    "D": {
+      "id": "world_modeling",
+      "name": "Scene Reconstruction & World Modeling",
+      "focus": "Long-term consistent 3D/4D scene mapping, scene graphs, object- and space-centric representations, spatial reasoning.",
+      "preferred_background": "Large-scale mapping, semantic reconstruction, or agent world models.",
+      "current_status": "early proxy tasks",
+      "current_readout": "The current tasks probe temporal structure, object relevance, cross-modal retrieval, and modality prediction, but they do not yet build persistent maps or scene graphs.",
+      "next_steps": [
+        "Convert windows into persistent object/scene-state nodes with timestamps and camera poses.",
+        "Add map consistency, object permanence, and spatial relation prediction tasks.",
+        "Train held-out-episode world models that predict future observations and task state."
+      ],
+      "tasks": [
+        "timeline_subtask",
+        "transition_detection",
+        "next_action",
+        "object_relevance",
+        "caption_grounding",
+        "cross_modal_retrieval",
+        "modality_reconstruction",
+        "temporal_order",
+        "misalignment_detection"
+      ],
+      "counts": {
+        "direct": 0,
+        "proxy": 6,
+        "diagnostic": 3,
+        "total_links": 9
+      }
+    }
+  },
+  "tasks": {
+    "timeline_action": {
+      "name": "Timeline action recognition",
+      "family": "supervised",
+      "input": "all featurized modalities",
+      "output": "current action label",
+      "primary_direction": "C",
+      "direction_roles": {
+        "C": "direct",
+        "A": "proxy"
+      },
+      "why": "Reads egocentric sensor state as the current human action; also provides a weak human-motion readout.",
+      "current_limit": "Chronological single-episode split creates unseen future action classes.",
+      "metric": {
+        "key": "macro_f1",
+        "name": "macro-F1",
+        "direction": "higher",
+        "minimal": 0.05,
+        "neural_mlp": 0.02631578947368421,
+        "better_baseline": "minimal"
+      }
+    },
+    "timeline_subtask": {
+      "name": "Timeline subtask recognition",
+      "family": "supervised",
+      "input": "all featurized modalities",
+      "output": "current subtask label",
+      "primary_direction": "C",
+      "direction_roles": {
+        "C": "direct",
+        "D": "proxy"
+      },
+      "why": "Segments egocentric task state and provides a first proxy for symbolic world/task state.",
+      "current_limit": "Single-episode ordering makes future subtasks hard to generalize.",
+      "metric": {
+        "key": "macro_f1",
+        "name": "macro-F1",
+        "direction": "higher",
+        "minimal": 0.04954121121178666,
+        "neural_mlp": 0.017518248175182476,
+        "better_baseline": "minimal"
+      }
+    },
+    "transition_detection": {
+      "name": "Action transition detection",
+      "family": "diagnostic",
+      "input": "all featurized modalities",
+      "output": "boundary vs steady state",
+      "primary_direction": "C",
+      "direction_roles": {
+        "C": "direct",
+        "D": "diagnostic"
+      },
+      "why": "Localizes egocentric task boundaries and diagnoses temporal state changes.",
+      "current_limit": "Boundary class is sparse, so accuracy alone is misleading.",
+      "metric": {
+        "key": "macro_f1",
+        "name": "macro-F1",
+        "direction": "higher",
+        "minimal": 0.6551829268292684,
+        "neural_mlp": 0.6484848484848484,
+        "better_baseline": "minimal"
+      }
+    },
+    "next_action": {
+      "name": "Short-horizon next action",
+      "family": "supervised",
+      "input": "current multimodal window",
+      "output": "action 20 frames later",
+      "primary_direction": "C",
+      "direction_roles": {
+        "C": "direct",
+        "D": "proxy"
+      },
+      "why": "Tests action intention/task-flow prediction from egocentric context.",
+      "current_limit": "Unseen future labels dominate the single-episode chronological test.",
+      "metric": {
+        "key": "macro_f1",
+        "name": "macro-F1",
+        "direction": "higher",
+        "minimal": 0.05925925925925927,
+        "neural_mlp": 0.023529411764705882,
+        "better_baseline": "minimal"
+      }
+    },
+    "hand_trajectory_forecast": {
+      "name": "Hand trajectory forecasting",
+      "family": "forecast",
+      "input": "current multimodal window",
+      "output": "future left/right hand 3D joints",
+      "primary_direction": "A",
+      "direction_roles": {
+        "A": "direct",
+        "C": "proxy"
+      },
+      "why": "Directly predicts human hand motion and supports hand-object interaction modeling.",
+      "current_limit": "Forecasting is window-level and not yet a full sequence or policy model.",
+      "metric": {
+        "key": "mpjpe",
+        "name": "MPJPE",
+        "direction": "lower",
+        "minimal": 0.8222644925117493,
+        "neural_mlp": 0.11163123697042465,
+        "better_baseline": "neural_mlp"
+      }
+    },
+    "contact_prediction": {
+      "name": "Body/object contact prediction",
+      "family": "supervised",
+      "input": "non-contact/non-caption features",
+      "output": "binary contact label",
+      "primary_direction": "A",
+      "direction_roles": {
+        "A": "direct",
+        "C": "proxy"
+      },
+      "why": "Targets physical interaction state, a core affordance and manipulation signal.",
+      "current_limit": "The public sample is degenerate for this target because one class dominates.",
+      "metric": {
+        "key": "macro_f1",
+        "name": "macro-F1",
+        "direction": "higher",
+        "minimal": 1.0,
+        "neural_mlp": 1.0,
+        "better_baseline": "tie"
+      }
+    },
+    "object_relevance": {
+      "name": "Relevant object set prediction",
+      "family": "supervised",
+      "input": "non-caption feature blocks",
+      "output": "multi-label object set",
+      "primary_direction": "C",
+      "direction_roles": {
+        "C": "direct",
+        "A": "proxy",
+        "D": "proxy"
+      },
+      "why": "Connects egocentric activity to manipulated objects and early object-centric state.",
+      "current_limit": "Object labels are language-derived and sparse in one episode.",
+      "metric": {
+        "key": "micro_f1",
+        "name": "micro-F1",
+        "direction": "higher",
+        "minimal": 0.18393030009680542,
+        "neural_mlp": 0.1797583081570997,
+        "better_baseline": "minimal"
+      }
+    },
+    "caption_grounding": {
+      "name": "Caption-to-window grounding",
+      "family": "retrieval",
+      "input": "caption objects/interaction query and candidate sensor windows",
+      "output": "matching time window",
+      "primary_direction": "C",
+      "direction_roles": {
+        "C": "direct",
+        "D": "proxy"
+      },
+      "why": "Grounds language annotation into egocentric sensor time and task state.",
+      "current_limit": "Bag-of-objects language features are too weak for rich grounding.",
+      "metric": {
+        "key": "mrr",
+        "name": "MRR",
+        "direction": "higher",
+        "minimal": 0.017183946083791223,
+        "neural_mlp": 0.01781111161035397,
+        "better_baseline": "neural_mlp"
+      }
+    },
+    "cross_modal_retrieval": {
+      "name": "Cross-modal retrieval",
+      "family": "retrieval",
+      "input": "motion/IMU/camera query",
+      "output": "matching depth/video window",
+      "primary_direction": "C",
+      "direction_roles": {
+        "C": "diagnostic",
+        "B": "proxy",
+        "D": "proxy"
+      },
+      "why": "Tests whether synchronized modalities identify the same 4D moment, a prerequisite for reconstruction and world modeling.",
+      "current_limit": "Retrieval proves alignment signal, not geometric reconstruction.",
+      "metric": {
+        "key": "mrr",
+        "name": "MRR",
+        "direction": "higher",
+        "minimal": 0.26335984006618296,
+        "neural_mlp": 0.1530070022204131,
+        "better_baseline": "minimal"
+      }
+    },
+    "modality_reconstruction": {
+      "name": "Modality reconstruction",
+      "family": "forecast",
+      "input": "motion/IMU/camera",
+      "output": "depth/video feature vector",
+      "primary_direction": "B",
+      "direction_roles": {
+        "B": "proxy",
+        "D": "proxy"
+      },
+      "why": "Predicts visual/depth state from non-target sensors as a weak reconstruction/world-model objective.",
+      "current_limit": "Feature-vector reconstruction is not pixel, depth-map, mesh, NeRF, or Gaussian reconstruction.",
+      "metric": {
+        "key": "r2",
+        "name": "R2",
+        "direction": "higher",
+        "minimal": -0.016022846771134747,
+        "neural_mlp": -0.010198171891414143,
+        "better_baseline": "neural_mlp"
+      }
+    },
+    "temporal_order": {
+      "name": "Temporal order verification",
+      "family": "diagnostic",
+      "input": "two adjacent windows",
+      "output": "correct vs reversed order",
+      "primary_direction": "C",
+      "direction_roles": {
+        "C": "diagnostic",
+        "D": "diagnostic"
+      },
+      "why": "Checks whether features encode local time direction and task progression.",
+      "current_limit": "Only local adjacent ordering, not long-horizon causal modeling.",
+      "metric": {
+        "key": "f1",
+        "name": "F1",
+        "direction": "higher",
+        "minimal": 0.5487364620938628,
+        "neural_mlp": 0.8717948717948718,
+        "better_baseline": "neural_mlp"
+      }
+    },
+    "misalignment_detection": {
+      "name": "Cross-modal misalignment detection",
+      "family": "diagnostic",
+      "input": "motion plus visual/depth pair",
+      "output": "aligned vs shifted",
+      "primary_direction": "C",
+      "direction_roles": {
+        "C": "diagnostic",
+        "B": "diagnostic",
+        "D": "diagnostic"
+      },
+      "why": "Detects temporal desynchronization, a key data-quality gate for multimodal reconstruction and world models.",
+      "current_limit": "Synthetic shifts diagnose alignment but do not solve calibration or mapping.",
+      "metric": {
+        "key": "f1",
+        "name": "F1",
+        "direction": "higher",
+        "minimal": 0.4865671641791045,
+        "neural_mlp": 0.7335243553008597,
+        "better_baseline": "neural_mlp"
+      }
+    }
+  }
+}

docs/index.html

@@ -419,6 +419,52 @@
       gap: 18px;
       align-items: start;
     }
+    .direction-grid {
+      display: grid;
+      grid-template-columns: repeat(4, minmax(0, 1fr));
+      gap: 14px;
+      margin-bottom: 22px;
+    }
+    .direction-card {
+      border: 1px solid var(--line);
+      border-radius: var(--radius);
+      background: var(--surface);
+      padding: 18px;
+      min-height: 230px;
+      display: grid;
+      gap: 12px;
+      align-content: start;
+    }
+    .direction-card h3 { margin: 0; font-size: 16px; line-height: 1.25; }
+    .direction-card p { margin: 0; color: var(--muted); font-size: 13px; line-height: 1.55; }
+    .status-pill {
+      width: fit-content;
+      border-radius: 999px;
+      padding: 4px 9px;
+      background: #edf2f7;
+      color: #334155;
+      font-size: 11px;
+      font-weight: 740;
+    }
+    .direction-counts {
+      display: grid;
+      grid-template-columns: repeat(3, minmax(0, 1fr));
+      gap: 8px;
+      font-size: 12px;
+      color: #4e514c;
+    }
+    .direction-counts strong {
+      display: block;
+      font-family: var(--font-mono);
+      font-size: 18px;
+      color: var(--ink);
+    }
+    .baseline-strip {
+      display: grid;
+      grid-template-columns: repeat(2, minmax(0, 1fr));
+      gap: 14px;
+      margin-top: 18px;
+    }
     .artifact {
       border: 1px solid var(--line);
       border-radius: var(--radius);
@@ -460,14 +506,14 @@
     @media (max-width: 960px) {
       .hero-inner, .two-col { grid-template-columns: 1fr; }
       .hero-inner { min-height: 0; }
-      .hero-stats, .models, .task-grid, .artifact-grid, .chart-grid, .callout-row { grid-template-columns: repeat(2, minmax(0, 1fr)); }
+      .hero-stats, .models, .task-grid, .artifact-grid, .chart-grid, .callout-row, .direction-grid, .baseline-strip { grid-template-columns: repeat(2, minmax(0, 1fr)); }
       .section-head { display: block; }
       .section-head p { margin-top: 14px; }
       .nav-links { display: none; }
     }
     @media (max-width: 640px) {
       .wrap { width: min(100% - 28px, var(--max)); }
-      .hero-stats, .models, .task-grid, .artifact-grid, .chart-grid, .callout-row { grid-template-columns: 1fr; }
+      .hero-stats, .models, .task-grid, .artifact-grid, .chart-grid, .callout-row, .direction-grid, .baseline-strip { grid-template-columns: 1fr; }
       .hero-inner, section { padding: 46px 0; }
       .signal { grid-template-columns: 1fr; }
       .signal strong { text-align: left; }
@@ -496,6 +542,7 @@
         <a href="#pipeline">Pipeline</a>
         <a href="#models">Models</a>
         <a href="#neural">Neural</a>
+        <a href="#directions">Directions</a>
         <a href="#architectures">Arch</a>
         <a href="#tasks">Tasks</a>
         <a href="#features">Signals</a>
@@ -611,6 +658,52 @@
       </div>
     </section>
 
+    <section id="directions">
+      <div class="wrap">
+        <div class="section-head">
+          <h2>The 12 tasks organized into four research directions.</h2>
+          <p>Each task is mapped as direct, proxy, or diagnostic evidence for the Ropedia research tracks. The mapping uses two current baselines: minimal interpretable heads and neural MLP heads over the same feature contract.</p>
+        </div>
+        <div class="direction-grid">
+          <article class="direction-card">
+            <span class="status-pill">partially implemented</span>
+            <h3>A. Human Modeling & Motion Understanding</h3>
+            <p>Direct evidence comes from hand trajectory forecasting and contact prediction; action and object relevance are supporting proxies.</p>
+            <div class="direction-counts"><span><strong>2</strong>direct</span><span><strong>2</strong>proxy</span><span><strong>0</strong>diagnostic</span></div>
+          </article>
+          <article class="direction-card">
+            <span class="status-pill">proxy tasks only</span>
+            <h3>B. 3D/4D Reconstruction & Neural Rendering</h3>
+            <p>Cross-modal retrieval, modality reconstruction, and misalignment detection check reconstruction prerequisites, not full geometry.</p>
+            <div class="direction-counts"><span><strong>0</strong>direct</span><span><strong>2</strong>proxy</span><span><strong>1</strong>diagnostic</span></div>
+          </article>
+          <article class="direction-card">
+            <span class="status-pill">strongest implemented</span>
+            <h3>C. Egocentric Vision & Interaction</h3>
+            <p>Action, subtask, transition, next-action, object, caption, order, and alignment tasks directly stress egocentric understanding.</p>
+            <div class="direction-counts"><span><strong>6</strong>direct</span><span><strong>2</strong>proxy</span><span><strong>3</strong>diagnostic</span></div>
+          </article>
+          <article class="direction-card">
+            <span class="status-pill">early proxy tasks</span>
+            <h3>D. Scene Reconstruction & World Modeling</h3>
+            <p>Current probes cover task state, object relevance, retrieval, reconstruction, temporal order, and alignment but no persistent map yet.</p>
+            <div class="direction-counts"><span><strong>0</strong>direct</span><span><strong>6</strong>proxy</span><span><strong>3</strong>diagnostic</span></div>
+          </article>
+        </div>
+        <img class="chart" src="assets/charts/research_direction_coverage.svg" alt="Coverage of the 12 Xperience-10M tasks across four research directions">
+        <div class="baseline-strip">
+          <div class="callout">
+            <h3>Baseline 1: minimal heads</h3>
+            <p>Softmax, logistic, ridge, and retrieval heads keep every input/output contract readable. They are the first sanity check for whether a task is well-posed.</p>
+          </div>
+          <div class="callout">
+            <h3>Baseline 2: neural MLP heads</h3>
+            <p>Small PyTorch MLP classifiers/regressors reuse the same features and splits. They test nonlinear gains before heavier Omni fine-tuning.</p>
+          </div>
+        </div>
+      </div>
+    </section>
+
     <section id="architectures">
       <div class="wrap">
         <div class="section-head">
@@ -685,6 +778,7 @@
         <div class="artifact-grid">
           <article class="artifact"><h3>Task-suite report</h3><p>One JSON file with every task metric and split detail.</p><a href="https://github.com/ChaoYue0307/ropedia-episode-task-suite/blob/main/results/episode_task_suite/summary_report.json">summary_report.json</a></article>
           <article class="artifact"><h3>Neural MLP task results</h3><p>Per-task PyTorch MLP metrics, predictions, histories, and checkpoints for the same 12 task contracts.</p><a href="https://github.com/ChaoYue0307/ropedia-episode-task-suite/tree/main/results/episode_task_suite/neural_mlp">neural_mlp/</a></article>
+          <article class="artifact"><h3>Four-direction taxonomy</h3><p>Generated JSON, CSV, Markdown, and website data mapping all 12 tasks to the four research tracks with two baselines.</p><a href="https://github.com/ChaoYue0307/ropedia-episode-task-suite/tree/main/results/episode_task_suite/research_directions">research_directions/</a></article>
           <article class="artifact"><h3>Feature manifest</h3><p>Start/end index and dimension for every current feature block.</p><a href="https://github.com/ChaoYue0307/ropedia-episode-task-suite/blob/main/results/episode_task_suite/feature_manifest.json">feature_manifest.json</a></article>
           <article class="artifact"><h3>Cross-modal retrieval</h3><p>The strongest self-supervised signal from the single episode.</p><a href="https://github.com/ChaoYue0307/ropedia-episode-task-suite/blob/main/results/episode_task_suite/cross_modal_retrieval/metrics.json">metrics.json</a></article>
           <article class="artifact"><h3>Current all-feature action model</h3><p>Classifier metrics, predictions, confusion matrix, and model weights.</p><a href="https://github.com/ChaoYue0307/ropedia-episode-task-suite/blob/main/results/min_all_modalities_action_model/metrics.json">metrics.json</a></article>

results/episode_task_suite/research_directions/research_direction_summary.md

@@ -0,0 +1,71 @@
+# Four-Direction Task Taxonomy
+
+This file is generated by `scripts/research_direction_taxonomy.py` from the committed 12-task metrics.
+It maps the current Xperience-10M sample tasks to the four Ropedia research directions without claiming that a single episode solves any full direction.
+
+## Baseline Families
+
+| Baseline | Meaning |
+| --- | --- |
+| Minimal | Interpretable softmax, logistic, ridge, and retrieval heads over the 8,378-d window feature vector. |
+| Neural MLP | Small PyTorch MLP classifiers/regressors using the same features, splits, and task contracts. |
+
+## Direction Coverage
+
+| Direction | Current status | Direct | Proxy | Diagnostic | Current readout |
+| --- | --- | ---: | ---: | ---: | --- |
+| A. Human Modeling & Motion Understanding | partially implemented | 2 | 2 | 0 | The sample supports hand trajectory forecasting and contact/object probes, but it does not yet include a full body/shape model or multi-person priors. |
+| B. 3D/4D Reconstruction & Neural Rendering | proxy tasks only | 0 | 2 | 1 | The current suite checks cross-modal alignment and depth/video reconstruction proxies; it does not yet train a renderer or reconstruct geometry. |
+| C. Egocentric Vision & Interaction | strongest implemented track | 6 | 2 | 3 | Most of the 12 tasks directly target egocentric action, task state, interaction, grounding, and alignment. |
+| D. Scene Reconstruction & World Modeling | early proxy tasks | 0 | 6 | 3 | The current tasks probe temporal structure, object relevance, cross-modal retrieval, and modality prediction, but they do not yet build persistent maps or scene graphs. |
+
+## Task Mapping With Two Baselines
+
+| Task | Primary direction | Related directions | Minimal | Neural MLP | Readout |
+| --- | --- | --- | ---: | ---: | --- |
+| `timeline_action` | C | C:direct, A:proxy | 0.0500 macro-F1 | 0.0263 macro-F1 | Minimal baseline is stronger. Chronological single-episode split creates unseen future action classes. |
+| `timeline_subtask` | C | C:direct, D:proxy | 0.0495 macro-F1 | 0.0175 macro-F1 | Minimal baseline is stronger. Single-episode ordering makes future subtasks hard to generalize. |
+| `transition_detection` | C | C:direct, D:diagnostic | 0.6552 macro-F1 | 0.6485 macro-F1 | Minimal baseline is stronger. Boundary class is sparse, so accuracy alone is misleading. |
+| `next_action` | C | C:direct, D:proxy | 0.0593 macro-F1 | 0.0235 macro-F1 | Minimal baseline is stronger. Unseen future labels dominate the single-episode chronological test. |
+| `hand_trajectory_forecast` | A | A:direct, C:proxy | 0.8223 MPJPE | 0.1116 MPJPE | Neural MLP is stronger. Forecasting is window-level and not yet a full sequence or policy model. |
+| `contact_prediction` | A | A:direct, C:proxy | 1.0000 macro-F1 | 1.0000 macro-F1 | Both baselines are tied. The public sample is degenerate for this target because one class dominates. |
+| `object_relevance` | C | C:direct, A:proxy, D:proxy | 0.1839 micro-F1 | 0.1798 micro-F1 | Minimal baseline is stronger. Object labels are language-derived and sparse in one episode. |
+| `caption_grounding` | C | C:direct, D:proxy | 0.0172 MRR | 0.0178 MRR | Neural MLP is stronger. Bag-of-objects language features are too weak for rich grounding. |
+| `cross_modal_retrieval` | C | C:diagnostic, B:proxy, D:proxy | 0.2634 MRR | 0.1530 MRR | Minimal baseline is stronger. Retrieval proves alignment signal, not geometric reconstruction. |
+| `modality_reconstruction` | B | B:proxy, D:proxy | -0.0160 R2 | -0.0102 R2 | Neural MLP is stronger. Feature-vector reconstruction is not pixel, depth-map, mesh, NeRF, or Gaussian reconstruction. |
+| `temporal_order` | C | C:diagnostic, D:diagnostic | 0.5487 F1 | 0.8718 F1 | Neural MLP is stronger. Only local adjacent ordering, not long-horizon causal modeling. |
+| `misalignment_detection` | C | C:diagnostic, B:diagnostic, D:diagnostic | 0.4866 F1 | 0.7335 F1 | Neural MLP is stronger. Synthetic shifts diagnose alignment but do not solve calibration or mapping. |
+
+## Next-Step Interpretation
+
+### A. Human Modeling & Motion Understanding
+
+The sample supports hand trajectory forecasting and contact/object probes, but it does not yet include a full body/shape model or multi-person priors.
+
+- Add SMPL/SMPL-X or MANO-style body/hand parameter targets where available.
+- Train sequence models over multi-episode motion trajectories instead of isolated windows.
+- Evaluate affordance prediction on held-out objects and held-out episodes.
+
+### B. 3D/4D Reconstruction & Neural Rendering
+
+The current suite checks cross-modal alignment and depth/video reconstruction proxies; it does not yet train a renderer or reconstruct geometry.
+
+- Use calibrated multi-view video plus SLAM pose to build per-episode camera trajectories.
+- Add depth-supervised point clouds, TSDF, Gaussian Splatting, or NeRF baselines.
+- Evaluate novel-view synthesis and temporal consistency across held-out views/time.
+
+### C. Egocentric Vision & Interaction
+
+Most of the 12 tasks directly target egocentric action, task state, interaction, grounding, and alignment.
+
+- Move from single-episode chronological splits to held-out-episode splits.
+- Add audio features and stronger multimodal backbones for action, intent, and grounding.
+- Evaluate long-horizon task success prediction and action-conditioned generation.
+
+### D. Scene Reconstruction & World Modeling
+
+The current tasks probe temporal structure, object relevance, cross-modal retrieval, and modality prediction, but they do not yet build persistent maps or scene graphs.
+
+- Convert windows into persistent object/scene-state nodes with timestamps and camera poses.
+- Add map consistency, object permanence, and spatial relation prediction tasks.
+- Train held-out-episode world models that predict future observations and task state.

results/episode_task_suite/research_directions/research_direction_task_map.csv

@@ -0,0 +1,28 @@
+direction,direction_name,task,task_name,family,relationship,primary_direction,metric_name,minimal_metric,neural_mlp_metric,better_baseline,why,current_limit
+C,Egocentric Vision & Interaction,timeline_action,Timeline action recognition,supervised,direct,C,macro-F1,0.05,0.0263157894737,minimal,Reads egocentric sensor state as the current human action; also provides a weak human-motion readout.,Chronological single-episode split creates unseen future action classes.
+A,Human Modeling & Motion Understanding,timeline_action,Timeline action recognition,supervised,proxy,C,macro-F1,0.05,0.0263157894737,minimal,Reads egocentric sensor state as the current human action; also provides a weak human-motion readout.,Chronological single-episode split creates unseen future action classes.
+C,Egocentric Vision & Interaction,timeline_subtask,Timeline subtask recognition,supervised,direct,C,macro-F1,0.0495412112118,0.0175182481752,minimal,Segments egocentric task state and provides a first proxy for symbolic world/task state.,Single-episode ordering makes future subtasks hard to generalize.
+D,Scene Reconstruction & World Modeling,timeline_subtask,Timeline subtask recognition,supervised,proxy,C,macro-F1,0.0495412112118,0.0175182481752,minimal,Segments egocentric task state and provides a first proxy for symbolic world/task state.,Single-episode ordering makes future subtasks hard to generalize.
+C,Egocentric Vision & Interaction,transition_detection,Action transition detection,diagnostic,direct,C,macro-F1,0.655182926829,0.648484848485,minimal,Localizes egocentric task boundaries and diagnoses temporal state changes.,"Boundary class is sparse, so accuracy alone is misleading."
+D,Scene Reconstruction & World Modeling,transition_detection,Action transition detection,diagnostic,diagnostic,C,macro-F1,0.655182926829,0.648484848485,minimal,Localizes egocentric task boundaries and diagnoses temporal state changes.,"Boundary class is sparse, so accuracy alone is misleading."
+C,Egocentric Vision & Interaction,next_action,Short-horizon next action,supervised,direct,C,macro-F1,0.0592592592593,0.0235294117647,minimal,Tests action intention/task-flow prediction from egocentric context.,Unseen future labels dominate the single-episode chronological test.
+D,Scene Reconstruction & World Modeling,next_action,Short-horizon next action,supervised,proxy,C,macro-F1,0.0592592592593,0.0235294117647,minimal,Tests action intention/task-flow prediction from egocentric context.,Unseen future labels dominate the single-episode chronological test.
+A,Human Modeling & Motion Understanding,hand_trajectory_forecast,Hand trajectory forecasting,forecast,direct,A,MPJPE,0.822264492512,0.11163123697,neural_mlp,Directly predicts human hand motion and supports hand-object interaction modeling.,Forecasting is window-level and not yet a full sequence or policy model.
+C,Egocentric Vision & Interaction,hand_trajectory_forecast,Hand trajectory forecasting,forecast,proxy,A,MPJPE,0.822264492512,0.11163123697,neural_mlp,Directly predicts human hand motion and supports hand-object interaction modeling.,Forecasting is window-level and not yet a full sequence or policy model.
+A,Human Modeling & Motion Understanding,contact_prediction,Body/object contact prediction,supervised,direct,A,macro-F1,1,1,tie,"Targets physical interaction state, a core affordance and manipulation signal.",The public sample is degenerate for this target because one class dominates.
+C,Egocentric Vision & Interaction,contact_prediction,Body/object contact prediction,supervised,proxy,A,macro-F1,1,1,tie,"Targets physical interaction state, a core affordance and manipulation signal.",The public sample is degenerate for this target because one class dominates.
+C,Egocentric Vision & Interaction,object_relevance,Relevant object set prediction,supervised,direct,C,micro-F1,0.183930300097,0.179758308157,minimal,Connects egocentric activity to manipulated objects and early object-centric state.,Object labels are language-derived and sparse in one episode.
+A,Human Modeling & Motion Understanding,object_relevance,Relevant object set prediction,supervised,proxy,C,micro-F1,0.183930300097,0.179758308157,minimal,Connects egocentric activity to manipulated objects and early object-centric state.,Object labels are language-derived and sparse in one episode.
+D,Scene Reconstruction & World Modeling,object_relevance,Relevant object set prediction,supervised,proxy,C,micro-F1,0.183930300097,0.179758308157,minimal,Connects egocentric activity to manipulated objects and early object-centric state.,Object labels are language-derived and sparse in one episode.
+C,Egocentric Vision & Interaction,caption_grounding,Caption-to-window grounding,retrieval,direct,C,MRR,0.0171839460838,0.0178111116104,neural_mlp,Grounds language annotation into egocentric sensor time and task state.,Bag-of-objects language features are too weak for rich grounding.
+D,Scene Reconstruction & World Modeling,caption_grounding,Caption-to-window grounding,retrieval,proxy,C,MRR,0.0171839460838,0.0178111116104,neural_mlp,Grounds language annotation into egocentric sensor time and task state.,Bag-of-objects language features are too weak for rich grounding.
+C,Egocentric Vision & Interaction,cross_modal_retrieval,Cross-modal retrieval,retrieval,diagnostic,C,MRR,0.263359840066,0.15300700222,minimal,"Tests whether synchronized modalities identify the same 4D moment, a prerequisite for reconstruction and world modeling.","Retrieval proves alignment signal, not geometric reconstruction."
+B,3D/4D Reconstruction & Neural Rendering,cross_modal_retrieval,Cross-modal retrieval,retrieval,proxy,C,MRR,0.263359840066,0.15300700222,minimal,"Tests whether synchronized modalities identify the same 4D moment, a prerequisite for reconstruction and world modeling.","Retrieval proves alignment signal, not geometric reconstruction."
+D,Scene Reconstruction & World Modeling,cross_modal_retrieval,Cross-modal retrieval,retrieval,proxy,C,MRR,0.263359840066,0.15300700222,minimal,"Tests whether synchronized modalities identify the same 4D moment, a prerequisite for reconstruction and world modeling.","Retrieval proves alignment signal, not geometric reconstruction."
+B,3D/4D Reconstruction & Neural Rendering,modality_reconstruction,Modality reconstruction,forecast,proxy,B,R2,-0.0160228467711,-0.0101981718914,neural_mlp,Predicts visual/depth state from non-target sensors as a weak reconstruction/world-model objective.,"Feature-vector reconstruction is not pixel, depth-map, mesh, NeRF, or Gaussian reconstruction."
+D,Scene Reconstruction & World Modeling,modality_reconstruction,Modality reconstruction,forecast,proxy,B,R2,-0.0160228467711,-0.0101981718914,neural_mlp,Predicts visual/depth state from non-target sensors as a weak reconstruction/world-model objective.,"Feature-vector reconstruction is not pixel, depth-map, mesh, NeRF, or Gaussian reconstruction."
+C,Egocentric Vision & Interaction,temporal_order,Temporal order verification,diagnostic,diagnostic,C,F1,0.548736462094,0.871794871795,neural_mlp,Checks whether features encode local time direction and task progression.,"Only local adjacent ordering, not long-horizon causal modeling."
+D,Scene Reconstruction & World Modeling,temporal_order,Temporal order verification,diagnostic,diagnostic,C,F1,0.548736462094,0.871794871795,neural_mlp,Checks whether features encode local time direction and task progression.,"Only local adjacent ordering, not long-horizon causal modeling."
+C,Egocentric Vision & Interaction,misalignment_detection,Cross-modal misalignment detection,diagnostic,diagnostic,C,F1,0.486567164179,0.733524355301,neural_mlp,"Detects temporal desynchronization, a key data-quality gate for multimodal reconstruction and world models.",Synthetic shifts diagnose alignment but do not solve calibration or mapping.
+B,3D/4D Reconstruction & Neural Rendering,misalignment_detection,Cross-modal misalignment detection,diagnostic,diagnostic,C,F1,0.486567164179,0.733524355301,neural_mlp,"Detects temporal desynchronization, a key data-quality gate for multimodal reconstruction and world models.",Synthetic shifts diagnose alignment but do not solve calibration or mapping.
+D,Scene Reconstruction & World Modeling,misalignment_detection,Cross-modal misalignment detection,diagnostic,diagnostic,C,F1,0.486567164179,0.733524355301,neural_mlp,"Detects temporal desynchronization, a key data-quality gate for multimodal reconstruction and world models.",Synthetic shifts diagnose alignment but do not solve calibration or mapping.

results/episode_task_suite/research_directions/research_direction_taxonomy.json

@@ -0,0 +1,384 @@
+{
+  "source": "results/episode_task_suite/summary_report.json",
+  "dataset_scope": {
+    "sample_episode_count": 1,
+    "num_frames": 5821,
+    "num_windows": 1161,
+    "feature_dim": 8378,
+    "warning": "Single public sample episode; this supports pipeline/task evidence, not cross-episode generalization claims."
+  },
+  "baselines": {
+    "minimal": "Interpretable softmax, logistic, ridge, and retrieval heads over the 8,378-d window feature vector.",
+    "neural_mlp": "Small PyTorch MLP classifiers/regressors using the same features, splits, and task contracts."
+  },
+  "directions": {
+    "A": {
+      "id": "human_motion",
+      "name": "Human Modeling & Motion Understanding",
+      "focus": "Human/hand/body motion, deformation priors, human-object interaction, affordance modeling.",
+      "preferred_background": "Human pose/shape estimation, SMPL-style models, motion capture, or motion generation.",
+      "current_status": "partially implemented",
+      "current_readout": "The sample supports hand trajectory forecasting and contact/object probes, but it does not yet include a full body/shape model or multi-person priors.",
+      "next_steps": [
+        "Add SMPL/SMPL-X or MANO-style body/hand parameter targets where available.",
+        "Train sequence models over multi-episode motion trajectories instead of isolated windows.",
+        "Evaluate affordance prediction on held-out objects and held-out episodes."
+      ],
+      "tasks": [
+        "timeline_action",
+        "hand_trajectory_forecast",
+        "contact_prediction",
+        "object_relevance"
+      ],
+      "counts": {
+        "direct": 2,
+        "proxy": 2,
+        "diagnostic": 0,
+        "total_links": 4
+      }
+    },
+    "B": {
+      "id": "reconstruction_rendering",
+      "name": "3D/4D Reconstruction & Neural Rendering",
+      "focus": "Multi-view dynamic scene reconstruction, NeRF/Gaussian Splatting, novel-view synthesis.",
+      "preferred_background": "3D reconstruction, neural rendering, camera calibration, and bundle adjustment.",
+      "current_status": "proxy tasks only",
+      "current_readout": "The current suite checks cross-modal alignment and depth/video reconstruction proxies; it does not yet train a renderer or reconstruct geometry.",
+      "next_steps": [
+        "Use calibrated multi-view video plus SLAM pose to build per-episode camera trajectories.",
+        "Add depth-supervised point clouds, TSDF, Gaussian Splatting, or NeRF baselines.",
+        "Evaluate novel-view synthesis and temporal consistency across held-out views/time."
+      ],
+      "tasks": [
+        "cross_modal_retrieval",
+        "modality_reconstruction",
+        "misalignment_detection"
+      ],
+      "counts": {
+        "direct": 0,
+        "proxy": 2,
+        "diagnostic": 1,
+        "total_links": 3
+      }
+    },
+    "C": {
+      "id": "egocentric_interaction",
+      "name": "Egocentric Vision & Interaction",
+      "focus": "Egocentric action and intention understanding, hand-object interaction, gaze/attention modeling, task structure modeling.",
+      "preferred_background": "Video understanding, action recognition, or egocentric vision.",
+      "current_status": "strongest implemented track",
+      "current_readout": "Most of the 12 tasks directly target egocentric action, task state, interaction, grounding, and alignment.",
+      "next_steps": [
+        "Move from single-episode chronological splits to held-out-episode splits.",
+        "Add audio features and stronger multimodal backbones for action, intent, and grounding.",
+        "Evaluate long-horizon task success prediction and action-conditioned generation."
+      ],
+      "tasks": [
+        "timeline_action",
+        "timeline_subtask",
+        "transition_detection",
+        "next_action",
+        "hand_trajectory_forecast",
+        "contact_prediction",
+        "object_relevance",
+        "caption_grounding",
+        "cross_modal_retrieval",
+        "temporal_order",
+        "misalignment_detection"
+      ],
+      "counts": {
+        "direct": 6,
+        "proxy": 2,
+        "diagnostic": 3,
+        "total_links": 11
+      }
+    },
+    "D": {
+      "id": "world_modeling",
+      "name": "Scene Reconstruction & World Modeling",
+      "focus": "Long-term consistent 3D/4D scene mapping, scene graphs, object- and space-centric representations, spatial reasoning.",
+      "preferred_background": "Large-scale mapping, semantic reconstruction, or agent world models.",
+      "current_status": "early proxy tasks",
+      "current_readout": "The current tasks probe temporal structure, object relevance, cross-modal retrieval, and modality prediction, but they do not yet build persistent maps or scene graphs.",
+      "next_steps": [
+        "Convert windows into persistent object/scene-state nodes with timestamps and camera poses.",
+        "Add map consistency, object permanence, and spatial relation prediction tasks.",
+        "Train held-out-episode world models that predict future observations and task state."
+      ],
+      "tasks": [
+        "timeline_subtask",
+        "transition_detection",
+        "next_action",
+        "object_relevance",
+        "caption_grounding",
+        "cross_modal_retrieval",
+        "modality_reconstruction",
+        "temporal_order",
+        "misalignment_detection"
+      ],
+      "counts": {
+        "direct": 0,
+        "proxy": 6,
+        "diagnostic": 3,
+        "total_links": 9
+      }
+    }
+  },
+  "tasks": {
+    "timeline_action": {
+      "name": "Timeline action recognition",
+      "family": "supervised",
+      "input": "all featurized modalities",
+      "output": "current action label",
+      "primary_direction": "C",
+      "direction_roles": {
+        "C": "direct",
+        "A": "proxy"
+      },
+      "why": "Reads egocentric sensor state as the current human action; also provides a weak human-motion readout.",
+      "current_limit": "Chronological single-episode split creates unseen future action classes.",
+      "metric": {
+        "key": "macro_f1",
+        "name": "macro-F1",
+        "direction": "higher",
+        "minimal": 0.05,
+        "neural_mlp": 0.02631578947368421,
+        "better_baseline": "minimal"
+      }
+    },
+    "timeline_subtask": {
+      "name": "Timeline subtask recognition",
+      "family": "supervised",
+      "input": "all featurized modalities",
+      "output": "current subtask label",
+      "primary_direction": "C",
+      "direction_roles": {
+        "C": "direct",
+        "D": "proxy"
+      },
+      "why": "Segments egocentric task state and provides a first proxy for symbolic world/task state.",
+      "current_limit": "Single-episode ordering makes future subtasks hard to generalize.",
+      "metric": {
+        "key": "macro_f1",
+        "name": "macro-F1",
+        "direction": "higher",
+        "minimal": 0.04954121121178666,
+        "neural_mlp": 0.017518248175182476,
+        "better_baseline": "minimal"
+      }
+    },
+    "transition_detection": {
+      "name": "Action transition detection",
+      "family": "diagnostic",
+      "input": "all featurized modalities",
+      "output": "boundary vs steady state",
+      "primary_direction": "C",
+      "direction_roles": {
+        "C": "direct",
+        "D": "diagnostic"
+      },
+      "why": "Localizes egocentric task boundaries and diagnoses temporal state changes.",
+      "current_limit": "Boundary class is sparse, so accuracy alone is misleading.",
+      "metric": {
+        "key": "macro_f1",
+        "name": "macro-F1",
+        "direction": "higher",
+        "minimal": 0.6551829268292684,
+        "neural_mlp": 0.6484848484848484,
+        "better_baseline": "minimal"
+      }
+    },
+    "next_action": {
+      "name": "Short-horizon next action",
+      "family": "supervised",
+      "input": "current multimodal window",
+      "output": "action 20 frames later",
+      "primary_direction": "C",
+      "direction_roles": {
+        "C": "direct",
+        "D": "proxy"
+      },
+      "why": "Tests action intention/task-flow prediction from egocentric context.",
+      "current_limit": "Unseen future labels dominate the single-episode chronological test.",
+      "metric": {
+        "key": "macro_f1",
+        "name": "macro-F1",
+        "direction": "higher",
+        "minimal": 0.05925925925925927,
+        "neural_mlp": 0.023529411764705882,
+        "better_baseline": "minimal"
+      }
+    },
+    "hand_trajectory_forecast": {
+      "name": "Hand trajectory forecasting",
+      "family": "forecast",
+      "input": "current multimodal window",
+      "output": "future left/right hand 3D joints",
+      "primary_direction": "A",
+      "direction_roles": {
+        "A": "direct",
+        "C": "proxy"
+      },
+      "why": "Directly predicts human hand motion and supports hand-object interaction modeling.",
+      "current_limit": "Forecasting is window-level and not yet a full sequence or policy model.",
+      "metric": {
+        "key": "mpjpe",
+        "name": "MPJPE",
+        "direction": "lower",
+        "minimal": 0.8222644925117493,
+        "neural_mlp": 0.11163123697042465,
+        "better_baseline": "neural_mlp"
+      }
+    },
+    "contact_prediction": {
+      "name": "Body/object contact prediction",
+      "family": "supervised",
+      "input": "non-contact/non-caption features",
+      "output": "binary contact label",
+      "primary_direction": "A",
+      "direction_roles": {
+        "A": "direct",
+        "C": "proxy"
+      },
+      "why": "Targets physical interaction state, a core affordance and manipulation signal.",
+      "current_limit": "The public sample is degenerate for this target because one class dominates.",
+      "metric": {
+        "key": "macro_f1",
+        "name": "macro-F1",
+        "direction": "higher",
+        "minimal": 1.0,
+        "neural_mlp": 1.0,
+        "better_baseline": "tie"
+      }
+    },
+    "object_relevance": {
+      "name": "Relevant object set prediction",
+      "family": "supervised",
+      "input": "non-caption feature blocks",
+      "output": "multi-label object set",
+      "primary_direction": "C",
+      "direction_roles": {
+        "C": "direct",
+        "A": "proxy",
+        "D": "proxy"
+      },
+      "why": "Connects egocentric activity to manipulated objects and early object-centric state.",
+      "current_limit": "Object labels are language-derived and sparse in one episode.",
+      "metric": {
+        "key": "micro_f1",
+        "name": "micro-F1",
+        "direction": "higher",
+        "minimal": 0.18393030009680542,
+        "neural_mlp": 0.1797583081570997,
+        "better_baseline": "minimal"
+      }
+    },
+    "caption_grounding": {
+      "name": "Caption-to-window grounding",
+      "family": "retrieval",
+      "input": "caption objects/interaction query and candidate sensor windows",
+      "output": "matching time window",
+      "primary_direction": "C",
+      "direction_roles": {
+        "C": "direct",
+        "D": "proxy"
+      },
+      "why": "Grounds language annotation into egocentric sensor time and task state.",
+      "current_limit": "Bag-of-objects language features are too weak for rich grounding.",
+      "metric": {
+        "key": "mrr",
+        "name": "MRR",
+        "direction": "higher",
+        "minimal": 0.017183946083791223,
+        "neural_mlp": 0.01781111161035397,
+        "better_baseline": "neural_mlp"
+      }
+    },
+    "cross_modal_retrieval": {
+      "name": "Cross-modal retrieval",
+      "family": "retrieval",
+      "input": "motion/IMU/camera query",
+      "output": "matching depth/video window",
+      "primary_direction": "C",
+      "direction_roles": {
+        "C": "diagnostic",
+        "B": "proxy",
+        "D": "proxy"
+      },
+      "why": "Tests whether synchronized modalities identify the same 4D moment, a prerequisite for reconstruction and world modeling.",
+      "current_limit": "Retrieval proves alignment signal, not geometric reconstruction.",
+      "metric": {
+        "key": "mrr",
+        "name": "MRR",
+        "direction": "higher",
+        "minimal": 0.26335984006618296,
+        "neural_mlp": 0.1530070022204131,
+        "better_baseline": "minimal"
+      }
+    },
+    "modality_reconstruction": {
+      "name": "Modality reconstruction",
+      "family": "forecast",
+      "input": "motion/IMU/camera",
+      "output": "depth/video feature vector",
+      "primary_direction": "B",
+      "direction_roles": {
+        "B": "proxy",
+        "D": "proxy"
+      },
+      "why": "Predicts visual/depth state from non-target sensors as a weak reconstruction/world-model objective.",
+      "current_limit": "Feature-vector reconstruction is not pixel, depth-map, mesh, NeRF, or Gaussian reconstruction.",
+      "metric": {
+        "key": "r2",
+        "name": "R2",
+        "direction": "higher",
+        "minimal": -0.016022846771134747,
+        "neural_mlp": -0.010198171891414143,
+        "better_baseline": "neural_mlp"
+      }
+    },
+    "temporal_order": {
+      "name": "Temporal order verification",
+      "family": "diagnostic",
+      "input": "two adjacent windows",
+      "output": "correct vs reversed order",
+      "primary_direction": "C",
+      "direction_roles": {
+        "C": "diagnostic",
+        "D": "diagnostic"
+      },
+      "why": "Checks whether features encode local time direction and task progression.",
+      "current_limit": "Only local adjacent ordering, not long-horizon causal modeling.",
+      "metric": {
+        "key": "f1",
+        "name": "F1",
+        "direction": "higher",
+        "minimal": 0.5487364620938628,
+        "neural_mlp": 0.8717948717948718,
+        "better_baseline": "neural_mlp"
+      }
+    },
+    "misalignment_detection": {
+      "name": "Cross-modal misalignment detection",
+      "family": "diagnostic",
+      "input": "motion plus visual/depth pair",
+      "output": "aligned vs shifted",
+      "primary_direction": "C",
+      "direction_roles": {
+        "C": "diagnostic",
+        "B": "diagnostic",
+        "D": "diagnostic"
+      },
+      "why": "Detects temporal desynchronization, a key data-quality gate for multimodal reconstruction and world models.",
+      "current_limit": "Synthetic shifts diagnose alignment but do not solve calibration or mapping.",
+      "metric": {
+        "key": "f1",
+        "name": "F1",
+        "direction": "higher",
+        "minimal": 0.4865671641791045,
+        "neural_mlp": 0.7335243553008597,
+        "better_baseline": "neural_mlp"
+      }
+    }
+  }
+}

scripts/research_direction_taxonomy.py

@@ -0,0 +1,589 @@
+#!/usr/bin/env python3
+"""Organize the 12 Xperience-10M tasks into the four Ropedia research tracks.
+
+The script is intentionally deterministic: it reads the committed task metrics,
+adds a hand-audited taxonomy, and writes machine-readable artifacts used by the
+README, website, and Hugging Face pages.
+"""
+
+from __future__ import annotations
+
+import csv
+import html
+import json
+from collections import OrderedDict
+from pathlib import Path
+from typing import Any
+
+
+ROOT = Path(__file__).resolve().parents[1]
+RESULTS = ROOT / "results" / "episode_task_suite"
+OUT_DIR = RESULTS / "research_directions"
+DOCS_DATA = ROOT / "docs" / "data"
+CHARTS = ROOT / "docs" / "assets" / "charts"
+
+SUMMARY_REPORT = RESULTS / "summary_report.json"
+
+
+DIRECTIONS: OrderedDict[str, dict[str, Any]] = OrderedDict(
+    [
+        (
+            "A",
+            {
+                "id": "human_motion",
+                "name": "Human Modeling & Motion Understanding",
+                "focus": "Human/hand/body motion, deformation priors, human-object interaction, affordance modeling.",
+                "preferred_background": "Human pose/shape estimation, SMPL-style models, motion capture, or motion generation.",
+                "current_status": "partially implemented",
+                "current_readout": "The sample supports hand trajectory forecasting and contact/object probes, but it does not yet include a full body/shape model or multi-person priors.",
+                "next_steps": [
+                    "Add SMPL/SMPL-X or MANO-style body/hand parameter targets where available.",
+                    "Train sequence models over multi-episode motion trajectories instead of isolated windows.",
+                    "Evaluate affordance prediction on held-out objects and held-out episodes.",
+                ],
+            },
+        ),
+        (
+            "B",
+            {
+                "id": "reconstruction_rendering",
+                "name": "3D/4D Reconstruction & Neural Rendering",
+                "focus": "Multi-view dynamic scene reconstruction, NeRF/Gaussian Splatting, novel-view synthesis.",
+                "preferred_background": "3D reconstruction, neural rendering, camera calibration, and bundle adjustment.",
+                "current_status": "proxy tasks only",
+                "current_readout": "The current suite checks cross-modal alignment and depth/video reconstruction proxies; it does not yet train a renderer or reconstruct geometry.",
+                "next_steps": [
+                    "Use calibrated multi-view video plus SLAM pose to build per-episode camera trajectories.",
+                    "Add depth-supervised point clouds, TSDF, Gaussian Splatting, or NeRF baselines.",
+                    "Evaluate novel-view synthesis and temporal consistency across held-out views/time.",
+                ],
+            },
+        ),
+        (
+            "C",
+            {
+                "id": "egocentric_interaction",
+                "name": "Egocentric Vision & Interaction",
+                "focus": "Egocentric action and intention understanding, hand-object interaction, gaze/attention modeling, task structure modeling.",
+                "preferred_background": "Video understanding, action recognition, or egocentric vision.",
+                "current_status": "strongest implemented track",
+                "current_readout": "Most of the 12 tasks directly target egocentric action, task state, interaction, grounding, and alignment.",
+                "next_steps": [
+                    "Move from single-episode chronological splits to held-out-episode splits.",
+                    "Add audio features and stronger multimodal backbones for action, intent, and grounding.",
+                    "Evaluate long-horizon task success prediction and action-conditioned generation.",
+                ],
+            },
+        ),
+        (
+            "D",
+            {
+                "id": "world_modeling",
+                "name": "Scene Reconstruction & World Modeling",
+                "focus": "Long-term consistent 3D/4D scene mapping, scene graphs, object- and space-centric representations, spatial reasoning.",
+                "preferred_background": "Large-scale mapping, semantic reconstruction, or agent world models.",
+                "current_status": "early proxy tasks",
+                "current_readout": "The current tasks probe temporal structure, object relevance, cross-modal retrieval, and modality prediction, but they do not yet build persistent maps or scene graphs.",
+                "next_steps": [
+                    "Convert windows into persistent object/scene-state nodes with timestamps and camera poses.",
+                    "Add map consistency, object permanence, and spatial relation prediction tasks.",
+                    "Train held-out-episode world models that predict future observations and task state.",
+                ],
+            },
+        ),
+    ]
+)
+
+
+TASK_TAXONOMY: OrderedDict[str, dict[str, Any]] = OrderedDict(
+    [
+        (
+            "timeline_action",
+            {
+                "name": "Timeline action recognition",
+                "family": "supervised",
+                "input": "all featurized modalities",
+                "output": "current action label",
+                "primary_direction": "C",
+                "direction_roles": {"C": "direct", "A": "proxy"},
+                "why": "Reads egocentric sensor state as the current human action; also provides a weak human-motion readout.",
+                "current_limit": "Chronological single-episode split creates unseen future action classes.",
+            },
+        ),
+        (
+            "timeline_subtask",
+            {
+                "name": "Timeline subtask recognition",
+                "family": "supervised",
+                "input": "all featurized modalities",
+                "output": "current subtask label",
+                "primary_direction": "C",
+                "direction_roles": {"C": "direct", "D": "proxy"},
+                "why": "Segments egocentric task state and provides a first proxy for symbolic world/task state.",
+                "current_limit": "Single-episode ordering makes future subtasks hard to generalize.",
+            },
+        ),
+        (
+            "transition_detection",
+            {
+                "name": "Action transition detection",
+                "family": "diagnostic",
+                "input": "all featurized modalities",
+                "output": "boundary vs steady state",
+                "primary_direction": "C",
+                "direction_roles": {"C": "direct", "D": "diagnostic"},
+                "why": "Localizes egocentric task boundaries and diagnoses temporal state changes.",
+                "current_limit": "Boundary class is sparse, so accuracy alone is misleading.",
+            },
+        ),
+        (
+            "next_action",
+            {
+                "name": "Short-horizon next action",
+                "family": "supervised",
+                "input": "current multimodal window",
+                "output": "action 20 frames later",
+                "primary_direction": "C",
+                "direction_roles": {"C": "direct", "D": "proxy"},
+                "why": "Tests action intention/task-flow prediction from egocentric context.",
+                "current_limit": "Unseen future labels dominate the single-episode chronological test.",
+            },
+        ),
+        (
+            "hand_trajectory_forecast",
+            {
+                "name": "Hand trajectory forecasting",
+                "family": "forecast",
+                "input": "current multimodal window",
+                "output": "future left/right hand 3D joints",
+                "primary_direction": "A",
+                "direction_roles": {"A": "direct", "C": "proxy"},
+                "why": "Directly predicts human hand motion and supports hand-object interaction modeling.",
+                "current_limit": "Forecasting is window-level and not yet a full sequence or policy model.",
+            },
+        ),
+        (
+            "contact_prediction",
+            {
+                "name": "Body/object contact prediction",
+                "family": "supervised",
+                "input": "non-contact/non-caption features",
+                "output": "binary contact label",
+                "primary_direction": "A",
+                "direction_roles": {"A": "direct", "C": "proxy"},
+                "why": "Targets physical interaction state, a core affordance and manipulation signal.",
+                "current_limit": "The public sample is degenerate for this target because one class dominates.",
+            },
+        ),
+        (
+            "object_relevance",
+            {
+                "name": "Relevant object set prediction",
+                "family": "supervised",
+                "input": "non-caption feature blocks",
+                "output": "multi-label object set",
+                "primary_direction": "C",
+                "direction_roles": {"C": "direct", "A": "proxy", "D": "proxy"},
+                "why": "Connects egocentric activity to manipulated objects and early object-centric state.",
+                "current_limit": "Object labels are language-derived and sparse in one episode.",
+            },
+        ),
+        (
+            "caption_grounding",
+            {
+                "name": "Caption-to-window grounding",
+                "family": "retrieval",
+                "input": "caption objects/interaction query and candidate sensor windows",
+                "output": "matching time window",
+                "primary_direction": "C",
+                "direction_roles": {"C": "direct", "D": "proxy"},
+                "why": "Grounds language annotation into egocentric sensor time and task state.",
+                "current_limit": "Bag-of-objects language features are too weak for rich grounding.",
+            },
+        ),
+        (
+            "cross_modal_retrieval",
+            {
+                "name": "Cross-modal retrieval",
+                "family": "retrieval",
+                "input": "motion/IMU/camera query",
+                "output": "matching depth/video window",
+                "primary_direction": "C",
+                "direction_roles": {"C": "diagnostic", "B": "proxy", "D": "proxy"},
+                "why": "Tests whether synchronized modalities identify the same 4D moment, a prerequisite for reconstruction and world modeling.",
+                "current_limit": "Retrieval proves alignment signal, not geometric reconstruction.",
+            },
+        ),
+        (
+            "modality_reconstruction",
+            {
+                "name": "Modality reconstruction",
+                "family": "forecast",
+                "input": "motion/IMU/camera",
+                "output": "depth/video feature vector",
+                "primary_direction": "B",
+                "direction_roles": {"B": "proxy", "D": "proxy"},
+                "why": "Predicts visual/depth state from non-target sensors as a weak reconstruction/world-model objective.",
+                "current_limit": "Feature-vector reconstruction is not pixel, depth-map, mesh, NeRF, or Gaussian reconstruction.",
+            },
+        ),
+        (
+            "temporal_order",
+            {
+                "name": "Temporal order verification",
+                "family": "diagnostic",
+                "input": "two adjacent windows",
+                "output": "correct vs reversed order",
+                "primary_direction": "C",
+                "direction_roles": {"C": "diagnostic", "D": "diagnostic"},
+                "why": "Checks whether features encode local time direction and task progression.",
+                "current_limit": "Only local adjacent ordering, not long-horizon causal modeling.",
+            },
+        ),
+        (
+            "misalignment_detection",
+            {
+                "name": "Cross-modal misalignment detection",
+                "family": "diagnostic",
+                "input": "motion plus visual/depth pair",
+                "output": "aligned vs shifted",
+                "primary_direction": "C",
+                "direction_roles": {"C": "diagnostic", "B": "diagnostic", "D": "diagnostic"},
+                "why": "Detects temporal desynchronization, a key data-quality gate for multimodal reconstruction and world models.",
+                "current_limit": "Synthetic shifts diagnose alignment but do not solve calibration or mapping.",
+            },
+        ),
+    ]
+)
+
+
+METRIC_SPECS = {
+    "timeline_action": ("macro_f1", "macro-F1", "higher"),
+    "timeline_subtask": ("macro_f1", "macro-F1", "higher"),
+    "transition_detection": ("macro_f1", "macro-F1", "higher"),
+    "next_action": ("macro_f1", "macro-F1", "higher"),
+    "hand_trajectory_forecast": ("mpjpe", "MPJPE", "lower"),
+    "contact_prediction": ("macro_f1", "macro-F1", "higher"),
+    "object_relevance": ("micro_f1", "micro-F1", "higher"),
+    "caption_grounding": ("mrr", "MRR", "higher"),
+    "cross_modal_retrieval": ("mrr", "MRR", "higher"),
+    "modality_reconstruction": ("r2", "R2", "higher"),
+    "temporal_order": ("f1", "F1", "higher"),
+    "misalignment_detection": ("f1", "F1", "higher"),
+}
+
+
+def load_summary() -> dict[str, Any]:
+    return json.loads(SUMMARY_REPORT.read_text(encoding="utf-8"))
+
+
+def metric_value(metrics: dict[str, Any] | None, task: str) -> float | None:
+    if not metrics:
+        return None
+    key = METRIC_SPECS[task][0]
+    value = metrics.get(key)
+    return float(value) if value is not None else None
+
+
+def choose_better(task: str, minimal: float | None, neural: float | None) -> str:
+    if minimal is None or neural is None:
+        return "unavailable"
+    _, _, direction = METRIC_SPECS[task]
+    delta = neural - minimal
+    if abs(delta) < 1e-9:
+        return "tie"
+    if direction == "lower":
+        return "neural_mlp" if delta < 0 else "minimal"
+    return "neural_mlp" if delta > 0 else "minimal"
+
+
+def fmt_metric(value: float | None) -> str:
+    if value is None:
+        return "n/a"
+    if abs(value) >= 10:
+        return f"{value:.3f}"
+    return f"{value:.4f}"
+
+
+def baseline_readout(label: str) -> str:
+    if label == "tie":
+        return "Both baselines are tied"
+    if label == "minimal":
+        return "Minimal baseline is stronger"
+    if label == "neural_mlp":
+        return "Neural MLP is stronger"
+    return "Baseline comparison is unavailable"
+
+
+def build_taxonomy(summary: dict[str, Any]) -> dict[str, Any]:
+    minimal_tasks = summary["tasks"]
+    neural_tasks = summary.get("neural_tasks", {})
+
+    task_records: OrderedDict[str, dict[str, Any]] = OrderedDict()
+    direction_counts = {
+        code: {"direct": 0, "proxy": 0, "diagnostic": 0, "total_links": 0}
+        for code in DIRECTIONS
+    }
+
+    for task, spec in TASK_TAXONOMY.items():
+        metric_key, metric_name, metric_direction = METRIC_SPECS[task]
+        minimal_metric = metric_value(minimal_tasks.get(task), task)
+        neural_metric = metric_value(neural_tasks.get(task), task)
+        better = choose_better(task, minimal_metric, neural_metric)
+
+        roles = spec["direction_roles"]
+        for direction_code, role in roles.items():
+            direction_counts[direction_code][role] += 1
+            direction_counts[direction_code]["total_links"] += 1
+
+        task_records[task] = {
+            **spec,
+            "metric": {
+                "key": metric_key,
+                "name": metric_name,
+                "direction": metric_direction,
+                "minimal": minimal_metric,
+                "neural_mlp": neural_metric,
+                "better_baseline": better,
+            },
+        }
+
+    direction_records = OrderedDict()
+    for code, info in DIRECTIONS.items():
+        linked_tasks = [
+            task
+            for task, spec in task_records.items()
+            if code in spec["direction_roles"]
+        ]
+        direction_records[code] = {
+            **info,
+            "tasks": linked_tasks,
+            "counts": direction_counts[code],
+        }
+
+    return {
+        "source": "results/episode_task_suite/summary_report.json",
+        "dataset_scope": {
+            "sample_episode_count": 1,
+            "num_frames": summary.get("num_frames"),
+            "num_windows": summary.get("num_windows"),
+            "feature_dim": summary.get("feature_dim"),
+            "warning": "Single public sample episode; this supports pipeline/task evidence, not cross-episode generalization claims.",
+        },
+        "baselines": {
+            "minimal": "Interpretable softmax, logistic, ridge, and retrieval heads over the 8,378-d window feature vector.",
+            "neural_mlp": "Small PyTorch MLP classifiers/regressors using the same features, splits, and task contracts.",
+        },
+        "directions": direction_records,
+        "tasks": task_records,
+    }
+
+
+def write_csv(taxonomy: dict[str, Any]) -> None:
+    path = OUT_DIR / "research_direction_task_map.csv"
+    with path.open("w", newline="", encoding="utf-8") as handle:
+        writer = csv.writer(handle, lineterminator="\n")
+        writer.writerow(
+            [
+                "direction",
+                "direction_name",
+                "task",
+                "task_name",
+                "family",
+                "relationship",
+                "primary_direction",
+                "metric_name",
+                "minimal_metric",
+                "neural_mlp_metric",
+                "better_baseline",
+                "why",
+                "current_limit",
+            ]
+        )
+        for task, spec in taxonomy["tasks"].items():
+            metric = spec["metric"]
+            for direction_code, relationship in spec["direction_roles"].items():
+                writer.writerow(
+                    [
+                        direction_code,
+                        taxonomy["directions"][direction_code]["name"],
+                        task,
+                        spec["name"],
+                        spec["family"],
+                        relationship,
+                        spec["primary_direction"],
+                        metric["name"],
+                        "" if metric["minimal"] is None else f"{metric['minimal']:.12g}",
+                        "" if metric["neural_mlp"] is None else f"{metric['neural_mlp']:.12g}",
+                        metric["better_baseline"],
+                        spec["why"],
+                        spec["current_limit"],
+                    ]
+                )
+
+
+def write_markdown(taxonomy: dict[str, Any]) -> None:
+    lines = [
+        "# Four-Direction Task Taxonomy",
+        "",
+        "This file is generated by `scripts/research_direction_taxonomy.py` from the committed 12-task metrics.",
+        "It maps the current Xperience-10M sample tasks to the four Ropedia research directions without claiming that a single episode solves any full direction.",
+        "",
+        "## Baseline Families",
+        "",
+        "| Baseline | Meaning |",
+        "| --- | --- |",
+        f"| Minimal | {taxonomy['baselines']['minimal']} |",
+        f"| Neural MLP | {taxonomy['baselines']['neural_mlp']} |",
+        "",
+        "## Direction Coverage",
+        "",
+        "| Direction | Current status | Direct | Proxy | Diagnostic | Current readout |",
+        "| --- | --- | ---: | ---: | ---: | --- |",
+    ]
+    for code, info in taxonomy["directions"].items():
+        counts = info["counts"]
+        lines.append(
+            f"| {code}. {info['name']} | {info['current_status']} | {counts['direct']} | {counts['proxy']} | {counts['diagnostic']} | {info['current_readout']} |"
+        )
+
+    lines.extend(
+        [
+            "",
+            "## Task Mapping With Two Baselines",
+            "",
+            "| Task | Primary direction | Related directions | Minimal | Neural MLP | Readout |",
+            "| --- | --- | --- | ---: | ---: | --- |",
+        ]
+    )
+    for task, spec in taxonomy["tasks"].items():
+        metric = spec["metric"]
+        related = ", ".join(
+            f"{code}:{role}" for code, role in spec["direction_roles"].items()
+        )
+        minimal = f"{fmt_metric(metric['minimal'])} {metric['name']}"
+        neural = f"{fmt_metric(metric['neural_mlp'])} {metric['name']}"
+        readout = f"{baseline_readout(metric['better_baseline'])}. {spec['current_limit']}"
+        lines.append(
+            f"| `{task}` | {spec['primary_direction']} | {related} | {minimal} | {neural} | {readout} |"
+        )
+
+    lines.extend(["", "## Next-Step Interpretation", ""])
+    for code, info in taxonomy["directions"].items():
+        lines.append(f"### {code}. {info['name']}")
+        lines.append("")
+        lines.append(info["current_readout"])
+        lines.append("")
+        for step in info["next_steps"]:
+            lines.append(f"- {step}")
+        lines.append("")
+
+    (OUT_DIR / "research_direction_summary.md").write_text(
+        "\n".join(lines).rstrip() + "\n", encoding="utf-8"
+    )
+
+
+def svg_text(x: int, y: int, text: str, size: int = 16, weight: int = 500, color: str = "#16213a") -> str:
+    return (
+        f'<text x="{x}" y="{y}" font-size="{size}" font-weight="{weight}" '
+        f'fill="{color}">{html.escape(text)}</text>'
+    )
+
+
+def write_svg(taxonomy: dict[str, Any]) -> None:
+    width = 1180
+    height = 700
+    margin = 58
+    card_w = 515
+    card_h = 220
+    colors = {"direct": "#1f6c9f", "proxy": "#2e7775", "diagnostic": "#956400"}
+    cards = []
+
+    for idx, (code, info) in enumerate(taxonomy["directions"].items()):
+        row = idx // 2
+        col = idx % 2
+        x = margin + col * (card_w + 34)
+        y = 130 + row * (card_h + 34)
+        counts = info["counts"]
+        total = max(1, counts["direct"] + counts["proxy"] + counts["diagnostic"])
+        bar_x = x + 24
+        bar_y = y + 132
+        bar_w = card_w - 48
+        cursor = bar_x
+        segments = []
+        for key in ("direct", "proxy", "diagnostic"):
+            seg_w = round(bar_w * counts[key] / total)
+            if counts[key] > 0:
+                segments.append(
+                    f'<rect x="{cursor}" y="{bar_y}" width="{seg_w}" height="16" rx="8" fill="{colors[key]}"/>'
+                )
+            cursor += seg_w
+
+        task_labels = ", ".join(info["tasks"][:5])
+        if len(info["tasks"]) > 5:
+            task_labels += f", +{len(info['tasks']) - 5}"
+
+        cards.append(
+            "\n".join(
+                [
+                    f'<rect x="{x}" y="{y}" width="{card_w}" height="{card_h}" rx="8" fill="#ffffff" stroke="#d9e1ea"/>',
+                    svg_text(x + 24, y + 42, f"{code}. {info['name']}", 21, 700),
+                    svg_text(x + 24, y + 75, info["current_status"], 15, 700, "#566273"),
+                    svg_text(x + 24, y + 108, f"Tasks: {task_labels}", 14, 500, "#30394a"),
+                    *segments,
+                    svg_text(x + 24, y + 174, f"Direct {counts['direct']}", 14, 700, colors["direct"]),
+                    svg_text(x + 150, y + 174, f"Proxy {counts['proxy']}", 14, 700, colors["proxy"]),
+                    svg_text(x + 270, y + 174, f"Diagnostic {counts['diagnostic']}", 14, 700, colors["diagnostic"]),
+                ]
+            )
+        )
+
+    legend = []
+    lx = margin
+    for key, label in (
+        ("direct", "Direct task"),
+        ("proxy", "Proxy / prerequisite"),
+        ("diagnostic", "Diagnostic probe"),
+    ):
+        legend.extend(
+            [
+                f'<rect x="{lx}" y="622" width="16" height="16" rx="4" fill="{colors[key]}"/>',
+                svg_text(lx + 24, 636, label, 14, 600, "#30394a"),
+            ]
+        )
+        lx += 200
+
+    svg = f"""<svg xmlns="http://www.w3.org/2000/svg" width="{width}" height="{height}" viewBox="0 0 {width} {height}" role="img" aria-label="Xperience-10M task coverage across four research directions">
+  <rect width="100%" height="100%" fill="#f7f9fb"/>
+  {svg_text(margin, 64, "Xperience-10M 12-Task Suite: Four Research Directions", 30, 800)}
+  {svg_text(margin, 96, "One public sample episode, two baseline families, explicit direct/proxy/diagnostic coverage.", 16, 500, "#566273")}
+  {"".join(cards)}
+  {"".join(legend)}
+  {svg_text(margin, 670, "Generated from results/episode_task_suite/summary_report.json and scripts/research_direction_taxonomy.py", 13, 500, "#6d7787")}
+</svg>
+"""
+    (CHARTS / "research_direction_coverage.svg").write_text(svg, encoding="utf-8")
+
+
+def main() -> None:
+    OUT_DIR.mkdir(parents=True, exist_ok=True)
+    DOCS_DATA.mkdir(parents=True, exist_ok=True)
+    CHARTS.mkdir(parents=True, exist_ok=True)
+
+    taxonomy = build_taxonomy(load_summary())
+    json_text = json.dumps(taxonomy, indent=2, ensure_ascii=False)
+    (OUT_DIR / "research_direction_taxonomy.json").write_text(json_text + "\n", encoding="utf-8")
+    (DOCS_DATA / "research_directions.json").write_text(json_text + "\n", encoding="utf-8")
+    write_csv(taxonomy)
+    write_markdown(taxonomy)
+    write_svg(taxonomy)
+
+    print(f"Wrote {OUT_DIR / 'research_direction_taxonomy.json'}")
+    print(f"Wrote {OUT_DIR / 'research_direction_task_map.csv'}")
+    print(f"Wrote {OUT_DIR / 'research_direction_summary.md'}")
+    print(f"Wrote {DOCS_DATA / 'research_directions.json'}")
+    print(f"Wrote {CHARTS / 'research_direction_coverage.svg'}")
+
+
+if __name__ == "__main__":
+    main()