Publish Ropedia Xperience-10M derived artifacts

PROJECT_README.md

@@ -37,6 +37,14 @@ The central research questions are:
 | Scale-up path | Data-gated Qwen3-Omni LoRA pilot plan for 32 held-out episodes; not claimed as completed until data and evaluation are present |
 | Public surfaces | GitHub repo, GitHub Pages dashboard, HF Space, HF artifact dataset, HF baseline-model repo, and HF collection |
 
+For the fastest interpretation of the current metrics, start with
+[`RESEARCH_TAKEAWAYS.md`](RESEARCH_TAKEAWAYS.md) and
+[`docs/data/research_takeaways.json`](docs/data/research_takeaways.json).
+They summarize what the public sample results actually show: class shift under
+chronological splits, neural gains on dynamics/order/alignment, harder
+retrieval/reconstruction probes, and why the next model-quality step needs
+held-out episodes.
+
 Current contributions:
 
 - manifested sliding-window features over the currently extracted modalities,
@@ -65,6 +73,7 @@ multi-episode model claims:
 | Data windows | `results/episode_task_suite/windows.csv`, `shared_windows.npz`, `summary_report.json` | one public sample episode |
 | Feature contract | `results/episode_task_suite/feature_manifest.json`, `available_modalities.json` | 8,378 current features; audio documented but not featurized |
 | Evaluation protocol | `EVALUATION_PROTOCOL.md`, `docs/data/evaluation_protocol.json`, `scripts/build_evaluation_protocol.py` | defines windowing, chronological split, leakage controls, per-task metrics, and unsupported interpretations |
+| Research takeaways | `RESEARCH_TAKEAWAYS.md`, `docs/data/research_takeaways.json`, `scripts/build_research_takeaways.py` | summarizes result interpretation from committed metrics without broad model-quality overclaims |
 | 12-task suite | `scripts/episode_task_suite.py`, per-task `metrics.json`, predictions | chronological single-episode split |
 | Neural heads | `scripts/neural_task_models.py`, `results/episode_task_suite/neural_mlp/` | compact MLP heads, not a foundation model |
 | Research directions | `research_direction_taxonomy.json`, extension probe results | direct/proxy/diagnostic evidence, not full solutions |
@@ -102,6 +111,9 @@ The public-surface QA report is at
 The generated evaluation protocol is at
 [`EVALUATION_PROTOCOL.md`](EVALUATION_PROTOCOL.md) and
 [`docs/data/evaluation_protocol.json`](docs/data/evaluation_protocol.json).
+The generated research takeaways are at
+[`RESEARCH_TAKEAWAYS.md`](RESEARCH_TAKEAWAYS.md) and
+[`docs/data/research_takeaways.json`](docs/data/research_takeaways.json).
 The source-of-truth artifact index is at
 [`docs/data/artifact_index.json`](docs/data/artifact_index.json).
 For a human-readable artifact map, use
@@ -156,11 +168,12 @@ If you are reading the project cold, open these in order:
 | 2 | What is the official upstream dataset? | [`XPERIENCE10M_DATASET_CARD_ALIGNMENT.md`](XPERIENCE10M_DATASET_CARD_ALIGNMENT.md), [`docs/data/xperience10m_dataset_card_alignment.json`](docs/data/xperience10m_dataset_card_alignment.json), [official HF dataset](https://huggingface.co/datasets/ropedia-ai/xperience-10m) | The full dataset is described as a gated large-scale 4D multimodal egocentric source; this repo validates only one public sample episode. |
 | 3 | Are source facts consistently presented? | [`SOURCE_ALIGNMENT_AUDIT.md`](SOURCE_ALIGNMENT_AUDIT.md), [`docs/data/source_alignment_audit.json`](docs/data/source_alignment_audit.json), [`scripts/validate_source_alignment.py`](scripts/validate_source_alignment.py) | Repo, website, and HF cards preserve full-dataset, sample-card, API-listing, and project-boundary markers. |
 | 4 | How exactly are tasks evaluated? | [`EVALUATION_PROTOCOL.md`](EVALUATION_PROTOCOL.md), [`docs/data/evaluation_protocol.json`](docs/data/evaluation_protocol.json), [`scripts/build_evaluation_protocol.py`](scripts/build_evaluation_protocol.py) | The window unit, chronological split, leakage controls, task metrics, and unsupported interpretations are explicit. |
-| 5 | How do I reproduce it? | [`REPRODUCIBILITY.md`](REPRODUCIBILITY.md), [`docs/data/reproducibility_matrix.json`](docs/data/reproducibility_matrix.json), [`notes/reproducibility_audit.md`](notes/reproducibility_audit.md) | Public commands, expected outputs, and exact-match reproduction evidence are explicit. |
-| 6 | What is one model input? | [`windows.csv`](results/episode_task_suite/windows.csv), [`feature_manifest.json`](results/episode_task_suite/feature_manifest.json), [`available_modalities.json`](results/episode_task_suite/available_modalities.json) | The input is an aligned 8,378-d window vector with explicit feature-block boundaries. |
-| 7 | Are the task results backed by files? | [`summary_report.json`](results/episode_task_suite/summary_report.json), [`neural_mlp/`](results/episode_task_suite/neural_mlp/), [`docs/data/summary_metrics.json`](docs/data/summary_metrics.json) | Each task has minimal and neural-head evidence over the same window contracts. |
-| 8 | Is the website internally coherent? | [`docs/data/website_integrity.json`](docs/data/website_integrity.json), [`scripts/validate_website_integrity.py`](scripts/validate_website_integrity.py) | Local links, anchors, tab routing, JSON data, and referenced images are checked before publishing. |
-| 9 | What is still pending? | [`DATA_BLOCKER_REPORT.md`](results/omni_finetune/DATA_BLOCKER_REPORT.md), [`MULTI_EPISODE_ACCESS_STATUS.md`](results/omni_finetune/MULTI_EPISODE_ACCESS_STATUS.md), [`scripts/omni/discover_xperience10m_sources.py`](scripts/omni/discover_xperience10m_sources.py) | The 32-episode Qwen3-Omni run is prepared but not yet a real model-quality claim. |
+| 5 | What do the current results mean? | [`RESEARCH_TAKEAWAYS.md`](RESEARCH_TAKEAWAYS.md), [`docs/data/research_takeaways.json`](docs/data/research_takeaways.json), [`docs/data/summary_metrics.json`](docs/data/summary_metrics.json) | The takeaways are generated from committed metrics and separate useful signals from unsupported model-quality claims. |
+| 6 | How do I reproduce it? | [`REPRODUCIBILITY.md`](REPRODUCIBILITY.md), [`docs/data/reproducibility_matrix.json`](docs/data/reproducibility_matrix.json), [`notes/reproducibility_audit.md`](notes/reproducibility_audit.md) | Public commands, expected outputs, and exact-match reproduction evidence are explicit. |
+| 7 | What is one model input? | [`windows.csv`](results/episode_task_suite/windows.csv), [`feature_manifest.json`](results/episode_task_suite/feature_manifest.json), [`available_modalities.json`](results/episode_task_suite/available_modalities.json) | The input is an aligned 8,378-d window vector with explicit feature-block boundaries. |
+| 8 | Are the task results backed by files? | [`summary_report.json`](results/episode_task_suite/summary_report.json), [`neural_mlp/`](results/episode_task_suite/neural_mlp/), [`docs/data/summary_metrics.json`](docs/data/summary_metrics.json) | Each task has minimal and neural-head evidence over the same window contracts. |
+| 9 | Is the website internally coherent? | [`docs/data/website_integrity.json`](docs/data/website_integrity.json), [`scripts/validate_website_integrity.py`](scripts/validate_website_integrity.py) | Local links, anchors, tab routing, JSON data, and referenced images are checked before publishing. |
+| 10 | What is still pending? | [`DATA_BLOCKER_REPORT.md`](results/omni_finetune/DATA_BLOCKER_REPORT.md), [`MULTI_EPISODE_ACCESS_STATUS.md`](results/omni_finetune/MULTI_EPISODE_ACCESS_STATUS.md), [`scripts/omni/discover_xperience10m_sources.py`](scripts/omni/discover_xperience10m_sources.py) | The 32-episode Qwen3-Omni run is prepared but not yet a real model-quality claim. |
 
 The machine-readable project packet is
 [`docs/data/project_packet.json`](docs/data/project_packet.json).

PROJECT_STATUS.md

@@ -10,6 +10,7 @@ the next development step.
 | Public-sample pipeline | Verified | `results/episode_task_suite/summary_report.json`, `results/episode_task_suite/windows.csv`, `results/episode_task_suite/feature_manifest.json` | One public Xperience-10M sample episode is converted into 5,821 frames, 1,161 aligned 20-frame windows, and an 8,378-dimensional current feature contract. |
 | Task suite | Verified | `scripts/episode_task_suite.py`, `results/episode_task_suite/`, `docs/data/summary_metrics.json` | All 12 task contracts have committed metrics, predictions, and minimal baseline outputs. |
 | Neural heads | Verified | `scripts/neural_task_models.py`, `results/episode_task_suite/neural_mlp/` | Each task also has a compact PyTorch MLP run over the same feature tensor and chronological split. |
+| Research takeaways | Verified | `RESEARCH_TAKEAWAYS.md`, `docs/data/research_takeaways.json`, `scripts/build_research_takeaways.py` | The main result interpretation is generated from committed metrics: chronological class shift, neural gains on dynamics/order/alignment, open retrieval/reconstruction problems, and the need for held-out episodes. |
 | Evaluation protocol | Verified | `EVALUATION_PROTOCOL.md`, `docs/data/evaluation_protocol.json`, `scripts/build_evaluation_protocol.py` | Windowing, chronological split, per-task metrics, leakage controls, and unsupported interpretations are generated from committed metric artifacts. |
 | Official dataset wording | Verified | `XPERIENCE10M_DATASET_CARD_ALIGNMENT.md`, `docs/data/xperience10m_dataset_card_alignment.json` | Public wording is aligned to the official gated Xperience-10M dataset card, public sample card, and HF API metadata, including modalities, scale, access boundary, sample license/tooling, and unsupported claims. |
 | Source alignment | Verified | `SOURCE_ALIGNMENT_AUDIT.md`, `docs/data/source_alignment_audit.json`, `scripts/validate_source_alignment.py` | Source facts and boundary markers are checked across repo docs, website, and HF cards. |
@@ -24,14 +25,16 @@ the next development step.
 1. Read this status file and `EVIDENCE_CONTRACT.md` to establish what is
    claimed.
 2. Open `docs/data/project_packet.json` for the machine-readable project path.
-3. Inspect `docs/data/summary_metrics.json` and
+3. Inspect `RESEARCH_TAKEAWAYS.md` and
+   `docs/data/research_takeaways.json` for the generated result interpretation.
+4. Inspect `docs/data/summary_metrics.json` and
    `results/episode_task_suite/neural_mlp/` to check the 12-task outputs.
-4. Inspect `EVALUATION_PROTOCOL.md` before judging task metrics or leakage
+5. Inspect `EVALUATION_PROTOCOL.md` before judging task metrics or leakage
    controls.
-5. Inspect `SOURCE_ALIGNMENT_AUDIT.md` and
+6. Inspect `SOURCE_ALIGNMENT_AUDIT.md` and
    `XPERIENCE10M_DATASET_CARD_ALIGNMENT.md` before judging dataset
    wording.
-6. Inspect `results/omni_finetune/DATA_BLOCKER_REPORT.md` before judging
+7. Inspect `results/omni_finetune/DATA_BLOCKER_REPORT.md` before judging
    Qwen3-Omni scale-up status.
 
 ## Do Not Infer

README.md

@@ -80,6 +80,11 @@ excluded raw data.
 `EVALUATION_PROTOCOL.md` and `docs/data/evaluation_protocol.json` define the
 window unit, chronological split, leakage controls, per-task metrics, and
 unsupported interpretations before a reader compares scores.
+`RESEARCH_TAKEAWAYS.md` and `docs/data/research_takeaways.json`, regenerated by
+`scripts/build_research_takeaways.py`, summarize what the committed metrics
+actually show: chronological class shift, neural gains on
+dynamics/order/alignment, harder retrieval/reconstruction probes, and the need
+for held-out episodes before model-quality claims.
 `FIGURE_INDEX.md` and `docs/data/figure_index.json` catalog the public figures,
 charts, modality thumbnails, dimensions, stable hashes, and source scripts.
 `docs/data/brand_assets.json` catalogs the generated logo variants used for the
@@ -105,13 +110,15 @@ This is the explorable artifact half of the project. You can inspect the task ou
 | --- | --- | --- |
 | 1 | What has been implemented? | `PROJECT_STATUS.md`, `docs/data/project_status.json`, `EVIDENCE_CONTRACT.md`, `ARTIFACT_GUIDE.md`, `QUALITY_GATES.md`, `PUBLIC_SURFACE_QA.md`, `FIGURE_INDEX.md`, `docs/data/evidence_contract.json`, `docs/data/artifact_index.json`, `docs/data/figure_index.json`, `docs/data/live_publication_status.json`, `docs/data/quality_gates.json`, `docs/data/mirror_parity.json`, `docs/data/public_surface_qa.json`, `docs/data/scope_claims_audit.json`, `docs/data/publication_audit.json`, `docs/data/task_surface_integrity.json`, `docs/data/website_integrity.json` |
 | 2 | Are source facts consistently presented? | `SOURCE_ALIGNMENT_AUDIT.md`, `docs/data/source_alignment_audit.json`, `scripts/validate_source_alignment.py` |
-| 3 | How do I reproduce it? | `REPRODUCIBILITY.md`, `docs/data/reproducibility_matrix.json`, `notes/reproducibility_audit.md` |
-| 4 | What is one model input? | `results/episode_task_suite/windows.csv`, `results/episode_task_suite/feature_manifest.json`, `results/episode_task_suite/available_modalities.json` |
-| 5 | Are the task results backed by files? | `results/episode_task_suite/summary_report.json`, `results/episode_task_suite/neural_mlp/`, `docs/data/summary_metrics.json` |
-| 6 | What is still pending? | `results/omni_finetune/DATA_BLOCKER_REPORT.md`, `results/omni_finetune/MULTI_EPISODE_ACCESS_STATUS.md`, `scripts/omni/discover_xperience10m_sources.py` |
+| 3 | What do the current results mean? | `RESEARCH_TAKEAWAYS.md`, `docs/data/research_takeaways.json`, `docs/data/summary_metrics.json` |
+| 4 | How do I reproduce it? | `REPRODUCIBILITY.md`, `docs/data/reproducibility_matrix.json`, `notes/reproducibility_audit.md` |
+| 5 | What is one model input? | `results/episode_task_suite/windows.csv`, `results/episode_task_suite/feature_manifest.json`, `results/episode_task_suite/available_modalities.json` |
+| 6 | Are the task results backed by files? | `results/episode_task_suite/summary_report.json`, `results/episode_task_suite/neural_mlp/`, `docs/data/summary_metrics.json` |
+| 7 | What is still pending? | `results/omni_finetune/DATA_BLOCKER_REPORT.md`, `results/omni_finetune/MULTI_EPISODE_ACCESS_STATUS.md`, `scripts/omni/discover_xperience10m_sources.py` |
 
 Human-readable artifact guide: `ARTIFACT_GUIDE.md`.
 Project status: `PROJECT_STATUS.md` and `docs/data/project_status.json`.
+Research Takeaways: `RESEARCH_TAKEAWAYS.md` and `docs/data/research_takeaways.json`.
 Official dataset-card alignment: `XPERIENCE10M_DATASET_CARD_ALIGNMENT.md` and `docs/data/xperience10m_dataset_card_alignment.json`.
 Source alignment: `SOURCE_ALIGNMENT_AUDIT.md` and `docs/data/source_alignment_audit.json`.
 Publication quality gates: `QUALITY_GATES.md` and `docs/data/quality_gates.json`.
@@ -130,6 +137,7 @@ Source-of-truth brand asset index: `docs/data/brand_assets.json`.
 | Data windows | `results/episode_task_suite/windows.csv`, `shared_windows.npz`, `summary_report.json` | one public sample episode |
 | Feature contract | `results/episode_task_suite/feature_manifest.json`, `available_modalities.json` | 8,378 current features; audio documented but not featurized |
 | Evaluation protocol | `EVALUATION_PROTOCOL.md`, `docs/data/evaluation_protocol.json` | windowing, chronological split, leakage controls, and task metrics |
+| Research Takeaways | `RESEARCH_TAKEAWAYS.md`, `docs/data/research_takeaways.json`, `scripts/build_research_takeaways.py` | generated interpretation of the committed metrics and scale-up boundary |
 | 12-task suite | per-task `metrics.json`, predictions, confusion matrices | chronological single-episode split |
 | Neural heads | `results/episode_task_suite/neural_mlp/` | compact MLP heads, not a foundation model |
 | Research directions | `research_direction_taxonomy.json`, extension probe results | direct/proxy/diagnostic evidence, not full solutions |
@@ -155,6 +163,7 @@ Source-of-truth brand asset index: `docs/data/brand_assets.json`.
 - `PROJECT_STATUS.md` and `docs/data/project_status.json`: compact current-state decision table
 - `REPRODUCIBILITY.md` and `docs/data/reproducibility_matrix.json`: public commands, expected outputs, exact-match reproduction evidence, and non-reproducible boundaries
 - `EVALUATION_PROTOCOL.md` and `docs/data/evaluation_protocol.json`: generated task protocol, split policy, leakage controls, and unsupported interpretations
+- `RESEARCH_TAKEAWAYS.md` and `docs/data/research_takeaways.json`: generated metric interpretation and scale-up readout
 - `results/**/*.json`: verified metrics and metadata for minimal and neural MLP runs
 - `results/**/*.csv`: predictions, confusion matrices, per-class metrics, windows, boundaries
 - `results/**/history.json`: neural MLP training traces
@@ -187,6 +196,7 @@ Source-of-truth brand asset index: `docs/data/brand_assets.json`.
 - `scripts/*.py`: reproduction scripts
 - `scripts/export_modality_atlas_assets.py`: regenerates the responsive modality-card thumbnails and manifest from the local public sample
 - `scripts/build_artifact_index.py`: source-of-truth artifact-index builder
+- `scripts/build_research_takeaways.py`: regenerates Research Takeaways from committed metric artifacts
 - `scripts/validate_mirror_parity.py`: prepared mirror parity validator
 - `scripts/validate_scope_claims.py`: validates the Qwen3-Omni readiness/result claim boundary
 - `scripts/validate_publication_package.py`: public bundle validator

RESEARCH_TAKEAWAYS.md

@@ -0,0 +1,101 @@
+# Research Takeaways
+
+This generated note summarizes what the current public Xperience-10M sample
+pipeline actually shows. It is built from committed metric artifacts, not
+from hand-entered benchmark claims.
+
+## Scope
+
+- validated episodes: 1
+- frames: 5,821
+- aligned windows: 1,161
+- current feature dimension: 8,378
+- raw Xperience-10M data is not redistributed
+- audio is documented and visualized, but not yet featurized
+
+## Takeaways
+
+### One episode can become a real benchmark contract
+
+The public sample is converted into 5,821 frames, 1,161 aligned 20-frame windows, and an 8,378-dimensional feature contract.
+
+| Metric | Value |
+| --- | ---: |
+| `frames` | 5,821 |
+| `windows` | 1,161 |
+| `feature_dim` | 8,378 |
+
+Source: `docs/data/summary_metrics.json`.
+
+Boundary: This is a task-development benchmark, not cross-episode generalization.
+
+### Chronological splits expose action-class shift
+
+Earlier all-feature action classifiers reach high macro-F1 on their local split, but the 12-task chronological action/subtask heads are much harder because later held-out windows include unseen labels.
+
+| Metric | Value |
+| --- | ---: |
+| `all_feature_action_macro_f1` | 0.9791 |
+| `suite_action_macro_f1` | 0.0500 |
+| `suite_subtask_macro_f1` | 0.0495 |
+| `unseen_action_test_classes` | 4 |
+
+Source: `results/episode_task_suite/summary_report.json`.
+
+Boundary: This is an important leakage/split lesson, not evidence that action recognition is solved.
+
+### Small neural heads help dynamic and temporal probes
+
+The MLP heads substantially improve hand trajectory forecasting, temporal-order verification, and motion/visual synchronization.
+
+| Metric | Value |
+| --- | ---: |
+| `hand_mpjpe_minimal` | 0.8223 |
+| `hand_mpjpe_neural` | 0.1116 |
+| `hand_mpjpe_relative_improvement` | 0.8642 |
+| `temporal_order_f1_minimal` | 0.5487 |
+| `temporal_order_f1_neural` | 0.8718 |
+| `misalignment_f1_minimal` | 0.4866 |
+| `misalignment_f1_neural` | 0.7335 |
+
+Source: `results/episode_task_suite/neural_mlp/*/metrics.json`.
+
+Boundary: These gains are within one episode and should be re-tested on held-out episodes.
+
+### Retrieval and reconstruction remain the harder multimodal problems
+
+Ridge/cosine retrieval remains stronger than the neural projection on this sample, and cross-modal reconstruction still has negative R2.
+
+| Metric | Value |
+| --- | ---: |
+| `retrieval_mrr_minimal` | 0.2634 |
+| `retrieval_mrr_neural` | 0.1530 |
+| `retrieval_top5_minimal` | 0.3764 |
+| `reconstruction_r2_minimal` | -0.0160 |
+| `reconstruction_r2_neural` | -0.0102 |
+
+Source: `results/episode_task_suite/cross_modal_retrieval/metrics.json`.
+
+Boundary: The current reconstruction task is feature-vector reconstruction, not depth, mesh, NeRF, or Gaussian splatting.
+
+### The next scientific unit is held-out episodes, not more adjacent windows
+
+The prepared Qwen3-Omni path targets 32 episodes from 32 sessions, but it remains data-gated until access and held-out evaluation complete.
+
+| Metric | Value |
+| --- | ---: |
+| `target_episodes` | 32 |
+| `selected_sessions` | 32 |
+| `valid_candidates` | 680 |
+
+Source: `results/omni_finetune/MULTI_EPISODE_ACCESS_STATUS.md`.
+
+Boundary: No real 32-episode fine-tune is claimed until gated data is available locally and held-out evaluation runs.
+
+## How To Read These Results
+
+- High single-episode scores are useful pipeline checks, not broad embodied-AI claims.
+- Low chronological action/subtask scores are informative because they expose later-label shift.
+- Neural gains on trajectory/order/alignment make those tasks good candidates for the next fine-tuning stage.
+- Retrieval and reconstruction remain the main multimodal representation challenges.
+- The next credible model-quality result needs held-out episodes.

docs/data/artifact_index.json

@@ -1,8 +1,8 @@
 {
   "title": "Ropedia Xperience-10M Task Suite Artifact Index",
-  "generated_at_utc": "2026-06-02T08:34:08+00:00",
+  "generated_at_utc": "2026-06-02T09:12:35+00:00",
   "status": "pass",
-  "artifact_count": 55,
+  "artifact_count": 58,
   "missing": [],
   "by_kind": {
     "project_path": 5,
@@ -10,6 +10,7 @@
     "source_alignment": 5,
     "publication_workflow": 1,
     "evaluation_protocol": 3,
+    "result_interpretation": 3,
     "visual_evidence": 6,
     "quality_gate": 9,
     "reproducibility": 2,
@@ -40,8 +41,8 @@
       "surface": "repo_hf",
       "proves": "Gives a compact verified/data-gated/not-redistributed decision table for first-pass readers.",
       "exists": true,
-      "bytes": 4377,
-      "sha256": "f1b07e5d177c171216f1d0c88ae59c37c510c9f53995c6a3ec2b674bdcdc32e7"
+      "bytes": 4847,
+      "sha256": "b7a9d8ab6fffb3757e7fa920a6b32c6ae94b3e9fdfabc69357cffec40c793664"
     },
     {
       "id": "project_status_json",
@@ -51,8 +52,8 @@
       "surface": "website_hf",
       "proves": "Machine-readable copy of the current project status for website and HF mirrors.",
       "exists": true,
-      "bytes": 6111,
-      "sha256": "ca7b72cd6256adbbc40e08478093f0f433d889f2c52717f50c7b68be1fda56ea"
+      "bytes": 6748,
+      "sha256": "b575469918d01ae76539d69096cc7ff5282b4014aa38796528372ee0d61a2d09"
     },
     {
       "id": "evidence_contract",
@@ -129,7 +130,7 @@
       "proves": "Machine-readable source-alignment pass/fail check for repo, website, and HF surfaces.",
       "exists": true,
       "bytes": 4425,
-      "sha256": "8f5b2f435b3e9ba4f8cee70336a5005413a842ef03e03788f0b7322fe2e291ae"
+      "sha256": "ffd3fa2b7b1f4cbca4acd569706b31b559cf4084cbb312da6f338393016af3ee"
     },
     {
       "id": "source_alignment_validator",
@@ -186,6 +187,39 @@
       "bytes": 16102,
       "sha256": "0781265b37af226432d93b25c18b6278484ba7d6d78e3c56991aaaf78bb0ba76"
     },
+    {
+      "id": "research_takeaways",
+      "title": "Research takeaways",
+      "path": "RESEARCH_TAKEAWAYS.md",
+      "kind": "result_interpretation",
+      "surface": "repo_hf",
+      "proves": "Summarizes the main research lessons from committed metrics without broad model-quality overclaims.",
+      "exists": true,
+      "bytes": 3742,
+      "sha256": "d8cd581e05879bed9f79e967af55d92a83154fc41eb20f80b923a19fef1bd95a"
+    },
+    {
+      "id": "research_takeaways_json",
+      "title": "Research takeaways JSON",
+      "path": "docs/data/research_takeaways.json",
+      "kind": "result_interpretation",
+      "surface": "website_hf",
+      "proves": "Machine-readable result interpretation for the website, HF cards, and mirror checks.",
+      "exists": true,
+      "bytes": 5251,
+      "sha256": "d7508f7b4dc18c509b27c755f84cd90c838d142bdb502706b1d3fb89df780883"
+    },
+    {
+      "id": "research_takeaways_builder",
+      "title": "Research takeaways builder",
+      "path": "scripts/build_research_takeaways.py",
+      "kind": "result_interpretation",
+      "surface": "repo_hf",
+      "proves": "Regenerates the research takeaways from committed summary metrics and task result artifacts.",
+      "exists": true,
+      "bytes": 10451,
+      "sha256": "4bfa8f679cba0f27b463a8e9b0a92c9d3661db80b9657a79b35beea88561da17"
+    },
     {
       "id": "figure_index",
       "title": "Figure index",
@@ -272,7 +306,7 @@
       "proves": "Machine-readable release-gate summary for validators, mirrors, and public project surfaces.",
       "exists": true,
       "bytes": 7480,
-      "sha256": "ae1c786adf80a93a45b1587fc7f378eb21f2d9fa351318ea491e844b24d9b707"
+      "sha256": "963418bfc6181b52cbce7a21c3fff8a08db8d230d58a6dc83e2bc1809c9f2f65"
     },
     {
       "id": "public_surface_qa",
@@ -351,8 +385,8 @@
       "surface": "repo",
       "proves": "Fetches the published GitHub/HF URLs and compares live hashes and public-card markers against the release assets.",
       "exists": true,
-      "bytes": 25837,
-      "sha256": "bf3f37956f97112432f8b57ed396f530af0a6aade629efb728b73a90fd4dccab"
+      "bytes": 26872,
+      "sha256": "a22a524aaf07968918a6fde0c8ab117cd49ccbfab0acc1e4f267be11ac654eb2"
     },
     {
       "id": "reproducibility_contract",
@@ -384,8 +418,8 @@
       "surface": "repo_hf",
       "proves": "Generates the selective proof-artifact catalog from local files.",
       "exists": true,
-      "bytes": 20448,
-      "sha256": "199323940ab056c50687a93b29591e903478f0c594507b01fcb12f5d29742e8b"
+      "bytes": 21444,
+      "sha256": "19f4feddf842a73e298066f6d681188253f427ff87f7f6895a0b1f42dcf38e69"
     },
     {
       "id": "publication_audit",
@@ -396,7 +430,7 @@
       "volatile": true,
       "proves": "Confirms public bundles pass raw-data, cache, archive, and token-string checks.",
       "exists": true,
-      "bytes": 6796,
+      "bytes": 6878,
       "hash_policy": "existence_and_size_only"
     },
     {
@@ -420,7 +454,7 @@
       "volatile": true,
       "proves": "Confirms prepared GitHub/HF Space/artifact/model mirrors share the same critical data, figure, website HTML, and validator files.",
       "exists": true,
-      "bytes": 69773,
+      "bytes": 72677,
       "hash_policy": "existence_and_size_only"
     },
     {
@@ -432,7 +466,7 @@
       "volatile": true,
       "proves": "Confirms local website links, anchors, JSON data files, and referenced images resolve.",
       "exists": true,
-      "bytes": 11168,
+      "bytes": 11279,
       "hash_policy": "existence_and_size_only"
     },
     {

docs/data/mirror_parity.json

@@ -1,9 +1,9 @@
 {
   "status": "pass",
-  "generated_at_utc": "2026-06-02T08:46:23+00:00",
+  "generated_at_utc": "2026-06-02T09:32:30+00:00",
   "hf_root": "hf_publish",
   "summary": {
-    "group_count": 65,
+    "group_count": 68,
     "failure_count": 0,
     "failures_by_surface": {}
   },
@@ -36,27 +36,27 @@
       "local": {
         "path": "repo:docs/data/artifact_index.json",
         "exists": true,
-        "bytes": 24778,
-        "sha256": "3603082c1ea1f6a80a6204ddb178c32f8f612b11cb183f02365c7ea3d0d39c25"
+        "bytes": 26149,
+        "sha256": "9bbaa487cd3eeb5ae58fa36bf972c8a3dba6a0bf451ef58291c9798b19351c83"
       },
       "mirrors": {
         "hf_space": {
           "path": "hf_space:data/artifact_index.json",
           "exists": true,
-          "bytes": 24778,
-          "sha256": "3603082c1ea1f6a80a6204ddb178c32f8f612b11cb183f02365c7ea3d0d39c25"
+          "bytes": 26149,
+          "sha256": "9bbaa487cd3eeb5ae58fa36bf972c8a3dba6a0bf451ef58291c9798b19351c83"
         },
         "hf_artifacts": {
           "path": "hf_artifacts:docs/data/artifact_index.json",
           "exists": true,
-          "bytes": 24778,
-          "sha256": "3603082c1ea1f6a80a6204ddb178c32f8f612b11cb183f02365c7ea3d0d39c25"
+          "bytes": 26149,
+          "sha256": "9bbaa487cd3eeb5ae58fa36bf972c8a3dba6a0bf451ef58291c9798b19351c83"
         },
         "hf_model": {
           "path": "hf_model:metrics/artifact_index.json",
           "exists": true,
-          "bytes": 24778,
-          "sha256": "3603082c1ea1f6a80a6204ddb178c32f8f612b11cb183f02365c7ea3d0d39c25"
+          "bytes": 26149,
+          "sha256": "9bbaa487cd3eeb5ae58fa36bf972c8a3dba6a0bf451ef58291c9798b19351c83"
         }
       },
       "failures": []
@@ -315,27 +315,27 @@
       "local": {
         "path": "repo:docs/data/project_status.json",
         "exists": true,
-        "bytes": 6111,
-        "sha256": "ca7b72cd6256adbbc40e08478093f0f433d889f2c52717f50c7b68be1fda56ea"
+        "bytes": 6748,
+        "sha256": "b575469918d01ae76539d69096cc7ff5282b4014aa38796528372ee0d61a2d09"
       },
       "mirrors": {
         "hf_space": {
           "path": "hf_space:data/project_status.json",
           "exists": true,
-          "bytes": 6111,
-          "sha256": "ca7b72cd6256adbbc40e08478093f0f433d889f2c52717f50c7b68be1fda56ea"
+          "bytes": 6748,
+          "sha256": "b575469918d01ae76539d69096cc7ff5282b4014aa38796528372ee0d61a2d09"
         },
         "hf_artifacts": {
           "path": "hf_artifacts:docs/data/project_status.json",
           "exists": true,
-          "bytes": 6111,
-          "sha256": "ca7b72cd6256adbbc40e08478093f0f433d889f2c52717f50c7b68be1fda56ea"
+          "bytes": 6748,
+          "sha256": "b575469918d01ae76539d69096cc7ff5282b4014aa38796528372ee0d61a2d09"
         },
         "hf_model": {
           "path": "hf_model:metrics/project_status.json",
           "exists": true,
-          "bytes": 6111,
-          "sha256": "ca7b72cd6256adbbc40e08478093f0f433d889f2c52717f50c7b68be1fda56ea"
+          "bytes": 6748,
+          "sha256": "b575469918d01ae76539d69096cc7ff5282b4014aa38796528372ee0d61a2d09"
         }
       },
       "failures": []
@@ -346,27 +346,27 @@
       "local": {
         "path": "repo:docs/data/publication_audit.json",
         "exists": true,
-        "bytes": 6796,
-        "sha256": "3173d2b5c5032a68f728a0a421b58f53f1008f5bc0dcf3a9586691226a524c3c"
+        "bytes": 6878,
+        "sha256": "ac845b4b601b1631e4f02595870897359d6afbe21f2e08c002a9f73703b516e4"
       },
       "mirrors": {
         "hf_space": {
           "path": "hf_space:data/publication_audit.json",
           "exists": true,
-          "bytes": 6796,
-          "sha256": "3173d2b5c5032a68f728a0a421b58f53f1008f5bc0dcf3a9586691226a524c3c"
+          "bytes": 6878,
+          "sha256": "ac845b4b601b1631e4f02595870897359d6afbe21f2e08c002a9f73703b516e4"
         },
         "hf_artifacts": {
           "path": "hf_artifacts:docs/data/publication_audit.json",
           "exists": true,
-          "bytes": 6796,
-          "sha256": "3173d2b5c5032a68f728a0a421b58f53f1008f5bc0dcf3a9586691226a524c3c"
+          "bytes": 6878,
+          "sha256": "ac845b4b601b1631e4f02595870897359d6afbe21f2e08c002a9f73703b516e4"
         },
         "hf_model": {
           "path": "hf_model:metrics/publication_audit.json",
           "exists": true,
-          "bytes": 6796,
-          "sha256": "3173d2b5c5032a68f728a0a421b58f53f1008f5bc0dcf3a9586691226a524c3c"
+          "bytes": 6878,
+          "sha256": "ac845b4b601b1631e4f02595870897359d6afbe21f2e08c002a9f73703b516e4"
         }
       },
       "failures": []
@@ -377,27 +377,27 @@
       "local": {
         "path": "repo:docs/data/public_surface_qa.json",
         "exists": true,
-        "bytes": 5288,
-        "sha256": "b1e6654f3f8d0ad77a29024dc886afbb1ad5401a5c54f732ec51aa5a259c2a40"
+        "bytes": 5290,
+        "sha256": "4d55f990e755ddea0cba4a88683b28f903ffaf570130de95b4e14a43059afedc"
       },
       "mirrors": {
         "hf_space": {
           "path": "hf_space:data/public_surface_qa.json",
           "exists": true,
-          "bytes": 5288,
-          "sha256": "b1e6654f3f8d0ad77a29024dc886afbb1ad5401a5c54f732ec51aa5a259c2a40"
+          "bytes": 5290,
+          "sha256": "4d55f990e755ddea0cba4a88683b28f903ffaf570130de95b4e14a43059afedc"
         },
         "hf_artifacts": {
           "path": "hf_artifacts:docs/data/public_surface_qa.json",
           "exists": true,
-          "bytes": 5288,
-          "sha256": "b1e6654f3f8d0ad77a29024dc886afbb1ad5401a5c54f732ec51aa5a259c2a40"
+          "bytes": 5290,
+          "sha256": "4d55f990e755ddea0cba4a88683b28f903ffaf570130de95b4e14a43059afedc"
         },
         "hf_model": {
           "path": "hf_model:metrics/public_surface_qa.json",
           "exists": true,
-          "bytes": 5288,
-          "sha256": "b1e6654f3f8d0ad77a29024dc886afbb1ad5401a5c54f732ec51aa5a259c2a40"
+          "bytes": 5290,
+          "sha256": "4d55f990e755ddea0cba4a88683b28f903ffaf570130de95b4e14a43059afedc"
         }
       },
       "failures": []
@@ -409,26 +409,26 @@
         "path": "repo:docs/data/quality_gates.json",
         "exists": true,
         "bytes": 7480,
-        "sha256": "ae1c786adf80a93a45b1587fc7f378eb21f2d9fa351318ea491e844b24d9b707"
+        "sha256": "fe7bd935c6317c31eb86004683f0deaeb47cff454b8fbf34fab4c38747fe8455"
       },
       "mirrors": {
         "hf_space": {
           "path": "hf_space:data/quality_gates.json",
           "exists": true,
           "bytes": 7480,
-          "sha256": "ae1c786adf80a93a45b1587fc7f378eb21f2d9fa351318ea491e844b24d9b707"
+          "sha256": "fe7bd935c6317c31eb86004683f0deaeb47cff454b8fbf34fab4c38747fe8455"
         },
         "hf_artifacts": {
           "path": "hf_artifacts:docs/data/quality_gates.json",
           "exists": true,
           "bytes": 7480,
-          "sha256": "ae1c786adf80a93a45b1587fc7f378eb21f2d9fa351318ea491e844b24d9b707"
+          "sha256": "fe7bd935c6317c31eb86004683f0deaeb47cff454b8fbf34fab4c38747fe8455"
         },
         "hf_model": {
           "path": "hf_model:metrics/quality_gates.json",
           "exists": true,
           "bytes": 7480,
-          "sha256": "ae1c786adf80a93a45b1587fc7f378eb21f2d9fa351318ea491e844b24d9b707"
+          "sha256": "fe7bd935c6317c31eb86004683f0deaeb47cff454b8fbf34fab4c38747fe8455"
         }
       },
       "failures": []
@@ -464,6 +464,37 @@
       },
       "failures": []
     },
+    {
+      "name": "data/research_takeaways.json",
+      "status": "pass",
+      "local": {
+        "path": "repo:docs/data/research_takeaways.json",
+        "exists": true,
+        "bytes": 5251,
+        "sha256": "d7508f7b4dc18c509b27c755f84cd90c838d142bdb502706b1d3fb89df780883"
+      },
+      "mirrors": {
+        "hf_space": {
+          "path": "hf_space:data/research_takeaways.json",
+          "exists": true,
+          "bytes": 5251,
+          "sha256": "d7508f7b4dc18c509b27c755f84cd90c838d142bdb502706b1d3fb89df780883"
+        },
+        "hf_artifacts": {
+          "path": "hf_artifacts:docs/data/research_takeaways.json",
+          "exists": true,
+          "bytes": 5251,
+          "sha256": "d7508f7b4dc18c509b27c755f84cd90c838d142bdb502706b1d3fb89df780883"
+        },
+        "hf_model": {
+          "path": "hf_model:metrics/research_takeaways.json",
+          "exists": true,
+          "bytes": 5251,
+          "sha256": "d7508f7b4dc18c509b27c755f84cd90c838d142bdb502706b1d3fb89df780883"
+        }
+      },
+      "failures": []
+    },
     {
       "name": "data/research_direction_extensions.json",
       "status": "pass",
@@ -564,26 +595,26 @@
         "path": "repo:docs/data/source_alignment_audit.json",
         "exists": true,
         "bytes": 4425,
-        "sha256": "8f5b2f435b3e9ba4f8cee70336a5005413a842ef03e03788f0b7322fe2e291ae"
+        "sha256": "6ef3d0e78355aa3c2fce3031e30a2516b5568935408884256142bc581877602b"
       },
       "mirrors": {
         "hf_space": {
           "path": "hf_space:data/source_alignment_audit.json",
           "exists": true,
           "bytes": 4425,
-          "sha256": "8f5b2f435b3e9ba4f8cee70336a5005413a842ef03e03788f0b7322fe2e291ae"
+          "sha256": "6ef3d0e78355aa3c2fce3031e30a2516b5568935408884256142bc581877602b"
         },
         "hf_artifacts": {
           "path": "hf_artifacts:docs/data/source_alignment_audit.json",
           "exists": true,
           "bytes": 4425,
-          "sha256": "8f5b2f435b3e9ba4f8cee70336a5005413a842ef03e03788f0b7322fe2e291ae"
+          "sha256": "6ef3d0e78355aa3c2fce3031e30a2516b5568935408884256142bc581877602b"
         },
         "hf_model": {
           "path": "hf_model:metrics/source_alignment_audit.json",
           "exists": true,
           "bytes": 4425,
-          "sha256": "8f5b2f435b3e9ba4f8cee70336a5005413a842ef03e03788f0b7322fe2e291ae"
+          "sha256": "6ef3d0e78355aa3c2fce3031e30a2516b5568935408884256142bc581877602b"
         }
       },
       "failures": []
@@ -687,27 +718,27 @@
       "local": {
         "path": "repo:docs/data/website_integrity.json",
         "exists": true,
-        "bytes": 11168,
-        "sha256": "f3e4b6903f798673988b4c9ce620ced6922c0a9fd89ecc56097478b8892f494d"
+        "bytes": 11339,
+        "sha256": "52a729ce4778f8231336550eb4f6aaa4b9601ecb06091fe0f40865570619f069"
       },
       "mirrors": {
         "hf_space": {
           "path": "hf_space:data/website_integrity.json",
           "exists": true,
-          "bytes": 11168,
-          "sha256": "f3e4b6903f798673988b4c9ce620ced6922c0a9fd89ecc56097478b8892f494d"
+          "bytes": 11339,
+          "sha256": "52a729ce4778f8231336550eb4f6aaa4b9601ecb06091fe0f40865570619f069"
         },
         "hf_artifacts": {
           "path": "hf_artifacts:docs/data/website_integrity.json",
           "exists": true,
-          "bytes": 11168,
-          "sha256": "f3e4b6903f798673988b4c9ce620ced6922c0a9fd89ecc56097478b8892f494d"
+          "bytes": 11339,
+          "sha256": "52a729ce4778f8231336550eb4f6aaa4b9601ecb06091fe0f40865570619f069"
         },
         "hf_model": {
           "path": "hf_model:metrics/website_integrity.json",
           "exists": true,
-          "bytes": 11168,
-          "sha256": "f3e4b6903f798673988b4c9ce620ced6922c0a9fd89ecc56097478b8892f494d"
+          "bytes": 11339,
+          "sha256": "52a729ce4778f8231336550eb4f6aaa4b9601ecb06091fe0f40865570619f069"
         }
       },
       "failures": []
@@ -1378,21 +1409,21 @@
       "local": {
         "path": "repo:scripts/build_artifact_index.py",
         "exists": true,
-        "bytes": 20448,
-        "sha256": "199323940ab056c50687a93b29591e903478f0c594507b01fcb12f5d29742e8b"
+        "bytes": 21444,
+        "sha256": "19f4feddf842a73e298066f6d681188253f427ff87f7f6895a0b1f42dcf38e69"
       },
       "mirrors": {
         "hf_artifacts": {
           "path": "hf_artifacts:scripts/build_artifact_index.py",
           "exists": true,
-          "bytes": 20448,
-          "sha256": "199323940ab056c50687a93b29591e903478f0c594507b01fcb12f5d29742e8b"
+          "bytes": 21444,
+          "sha256": "19f4feddf842a73e298066f6d681188253f427ff87f7f6895a0b1f42dcf38e69"
         },
         "hf_model": {
           "path": "hf_model:scripts/build_artifact_index.py",
           "exists": true,
-          "bytes": 20448,
-          "sha256": "199323940ab056c50687a93b29591e903478f0c594507b01fcb12f5d29742e8b"
+          "bytes": 21444,
+          "sha256": "19f4feddf842a73e298066f6d681188253f427ff87f7f6895a0b1f42dcf38e69"
         }
       },
       "failures": []
@@ -1503,21 +1534,46 @@
       "local": {
         "path": "repo:scripts/build_public_surface_qa.py",
         "exists": true,
-        "bytes": 10144,
-        "sha256": "447a76966d40834c7c2fbdd5977eafcc24a09f2354e7f090940a62e07ba89abb"
+        "bytes": 10365,
+        "sha256": "bc1633b52d8f7151ee328326b0a0dace1982ba103f9e1145b20b143b41f087f9"
       },
       "mirrors": {
         "hf_artifacts": {
           "path": "hf_artifacts:scripts/build_public_surface_qa.py",
           "exists": true,
-          "bytes": 10144,
-          "sha256": "447a76966d40834c7c2fbdd5977eafcc24a09f2354e7f090940a62e07ba89abb"
+          "bytes": 10365,
+          "sha256": "bc1633b52d8f7151ee328326b0a0dace1982ba103f9e1145b20b143b41f087f9"
         },
         "hf_model": {
           "path": "hf_model:scripts/build_public_surface_qa.py",
           "exists": true,
-          "bytes": 10144,
-          "sha256": "447a76966d40834c7c2fbdd5977eafcc24a09f2354e7f090940a62e07ba89abb"
+          "bytes": 10365,
+          "sha256": "bc1633b52d8f7151ee328326b0a0dace1982ba103f9e1145b20b143b41f087f9"
+        }
+      },
+      "failures": []
+    },
+    {
+      "name": "scripts/build_research_takeaways.py",
+      "status": "pass",
+      "local": {
+        "path": "repo:scripts/build_research_takeaways.py",
+        "exists": true,
+        "bytes": 10451,
+        "sha256": "4bfa8f679cba0f27b463a8e9b0a92c9d3661db80b9657a79b35beea88561da17"
+      },
+      "mirrors": {
+        "hf_artifacts": {
+          "path": "hf_artifacts:scripts/build_research_takeaways.py",
+          "exists": true,
+          "bytes": 10451,
+          "sha256": "4bfa8f679cba0f27b463a8e9b0a92c9d3661db80b9657a79b35beea88561da17"
+        },
+        "hf_model": {
+          "path": "hf_model:scripts/build_research_takeaways.py",
+          "exists": true,
+          "bytes": 10451,
+          "sha256": "4bfa8f679cba0f27b463a8e9b0a92c9d3661db80b9657a79b35beea88561da17"
         }
       },
       "failures": []
@@ -1528,21 +1584,21 @@
       "local": {
         "path": "repo:scripts/verify_live_publication.py",
         "exists": true,
-        "bytes": 25837,
-        "sha256": "bf3f37956f97112432f8b57ed396f530af0a6aade629efb728b73a90fd4dccab"
+        "bytes": 26872,
+        "sha256": "a22a524aaf07968918a6fde0c8ab117cd49ccbfab0acc1e4f267be11ac654eb2"
       },
       "mirrors": {
         "hf_artifacts": {
           "path": "hf_artifacts:scripts/verify_live_publication.py",
           "exists": true,
-          "bytes": 25837,
-          "sha256": "bf3f37956f97112432f8b57ed396f530af0a6aade629efb728b73a90fd4dccab"
+          "bytes": 26872,
+          "sha256": "a22a524aaf07968918a6fde0c8ab117cd49ccbfab0acc1e4f267be11ac654eb2"
         },
         "hf_model": {
           "path": "hf_model:scripts/verify_live_publication.py",
           "exists": true,
-          "bytes": 25837,
-          "sha256": "bf3f37956f97112432f8b57ed396f530af0a6aade629efb728b73a90fd4dccab"
+          "bytes": 26872,
+          "sha256": "a22a524aaf07968918a6fde0c8ab117cd49ccbfab0acc1e4f267be11ac654eb2"
         }
       },
       "failures": []
@@ -1553,21 +1609,21 @@
       "local": {
         "path": "repo:scripts/validate_mirror_parity.py",
         "exists": true,
-        "bytes": 10147,
-        "sha256": "9be726975691591b8d68aab6cc039feae56f9d8a133cc797a73ab6171d19c5a2"
+        "bytes": 10242,
+        "sha256": "672c42b82d1dc1aa8c75b25b53d5526357704769f74c3f7f1c2930c9f04abfff"
       },
       "mirrors": {
         "hf_artifacts": {
           "path": "hf_artifacts:scripts/validate_mirror_parity.py",
           "exists": true,
-          "bytes": 10147,
-          "sha256": "9be726975691591b8d68aab6cc039feae56f9d8a133cc797a73ab6171d19c5a2"
+          "bytes": 10242,
+          "sha256": "672c42b82d1dc1aa8c75b25b53d5526357704769f74c3f7f1c2930c9f04abfff"
         },
         "hf_model": {
           "path": "hf_model:scripts/validate_mirror_parity.py",
           "exists": true,
-          "bytes": 10147,
-          "sha256": "9be726975691591b8d68aab6cc039feae56f9d8a133cc797a73ab6171d19c5a2"
+          "bytes": 10242,
+          "sha256": "672c42b82d1dc1aa8c75b25b53d5526357704769f74c3f7f1c2930c9f04abfff"
         }
       },
       "failures": []
@@ -1578,21 +1634,21 @@
       "local": {
         "path": "repo:scripts/validate_publication_package.py",
         "exists": true,
-        "bytes": 18163,
-        "sha256": "d47f33506ca21fa340a0281cafcb6400effd8c8abf64c3c89b4ab6333856e02c"
+        "bytes": 18725,
+        "sha256": "bfc95f30548c97608b1c833085dee4b27bd4a5de050e0a9f7ef8114c7d4220fb"
       },
       "mirrors": {
         "hf_artifacts": {
           "path": "hf_artifacts:scripts/validate_publication_package.py",
           "exists": true,
-          "bytes": 18163,
-          "sha256": "d47f33506ca21fa340a0281cafcb6400effd8c8abf64c3c89b4ab6333856e02c"
+          "bytes": 18725,
+          "sha256": "bfc95f30548c97608b1c833085dee4b27bd4a5de050e0a9f7ef8114c7d4220fb"
         },
         "hf_model": {
           "path": "hf_model:scripts/validate_publication_package.py",
           "exists": true,
-          "bytes": 18163,
-          "sha256": "d47f33506ca21fa340a0281cafcb6400effd8c8abf64c3c89b4ab6333856e02c"
+          "bytes": 18725,
+          "sha256": "bfc95f30548c97608b1c833085dee4b27bd4a5de050e0a9f7ef8114c7d4220fb"
         }
       },
       "failures": []
@@ -1678,21 +1734,21 @@
       "local": {
         "path": "repo:scripts/validate_website_integrity.py",
         "exists": true,
-        "bytes": 19977,
-        "sha256": "cbdde3ef45f1c24e7505dedaf8d46f7ddce46b04d2d7307509fee782e39483d4"
+        "bytes": 20438,
+        "sha256": "d6d59f8de726593c43a9aacd04cda5adb1ad7194259f954ece7e546d6aa77bf9"
       },
       "mirrors": {
         "hf_artifacts": {
           "path": "hf_artifacts:scripts/validate_website_integrity.py",
           "exists": true,
-          "bytes": 19977,
-          "sha256": "cbdde3ef45f1c24e7505dedaf8d46f7ddce46b04d2d7307509fee782e39483d4"
+          "bytes": 20438,
+          "sha256": "d6d59f8de726593c43a9aacd04cda5adb1ad7194259f954ece7e546d6aa77bf9"
         },
         "hf_model": {
           "path": "hf_model:scripts/validate_website_integrity.py",
           "exists": true,
-          "bytes": 19977,
-          "sha256": "cbdde3ef45f1c24e7505dedaf8d46f7ddce46b04d2d7307509fee782e39483d4"
+          "bytes": 20438,
+          "sha256": "d6d59f8de726593c43a9aacd04cda5adb1ad7194259f954ece7e546d6aa77bf9"
         }
       },
       "failures": []
@@ -1778,21 +1834,21 @@
       "local": {
         "path": "repo:docs/index.html",
         "exists": true,
-        "bytes": 138368,
-        "sha256": "a330106a1c3c9b44215dcbb597593e00db60fe81d7f0b474206e47c51938491e"
+        "bytes": 146877,
+        "sha256": "3574b46588a181ce58efabcad3a0755de8d6ebd6a919f51ca476b05018cbd613"
       },
       "mirrors": {
         "hf_space": {
           "path": "hf_space:index.html",
           "exists": true,
-          "bytes": 138368,
-          "sha256": "a330106a1c3c9b44215dcbb597593e00db60fe81d7f0b474206e47c51938491e"
+          "bytes": 146877,
+          "sha256": "3574b46588a181ce58efabcad3a0755de8d6ebd6a919f51ca476b05018cbd613"
         },
         "hf_artifacts_docs": {
           "path": "hf_artifacts:docs/index.html",
           "exists": true,
-          "bytes": 138368,
-          "sha256": "a330106a1c3c9b44215dcbb597593e00db60fe81d7f0b474206e47c51938491e"
+          "bytes": 146877,
+          "sha256": "3574b46588a181ce58efabcad3a0755de8d6ebd6a919f51ca476b05018cbd613"
         }
       },
       "failures": []
@@ -1921,27 +1977,27 @@
       "local": {
         "path": "repo:PROJECT_STATUS.md",
         "exists": true,
-        "bytes": 4377,
-        "sha256": "f1b07e5d177c171216f1d0c88ae59c37c510c9f53995c6a3ec2b674bdcdc32e7"
+        "bytes": 4847,
+        "sha256": "b7a9d8ab6fffb3757e7fa920a6b32c6ae94b3e9fdfabc69357cffec40c793664"
       },
       "mirrors": {
         "hf_space": {
           "path": "hf_space:PROJECT_STATUS.md",
           "exists": true,
-          "bytes": 4377,
-          "sha256": "f1b07e5d177c171216f1d0c88ae59c37c510c9f53995c6a3ec2b674bdcdc32e7"
+          "bytes": 4847,
+          "sha256": "b7a9d8ab6fffb3757e7fa920a6b32c6ae94b3e9fdfabc69357cffec40c793664"
         },
         "hf_artifacts": {
           "path": "hf_artifacts:PROJECT_STATUS.md",
           "exists": true,
-          "bytes": 4377,
-          "sha256": "f1b07e5d177c171216f1d0c88ae59c37c510c9f53995c6a3ec2b674bdcdc32e7"
+          "bytes": 4847,
+          "sha256": "b7a9d8ab6fffb3757e7fa920a6b32c6ae94b3e9fdfabc69357cffec40c793664"
         },
         "hf_model": {
           "path": "hf_model:PROJECT_STATUS.md",
           "exists": true,
-          "bytes": 4377,
-          "sha256": "f1b07e5d177c171216f1d0c88ae59c37c510c9f53995c6a3ec2b674bdcdc32e7"
+          "bytes": 4847,
+          "sha256": "b7a9d8ab6fffb3757e7fa920a6b32c6ae94b3e9fdfabc69357cffec40c793664"
         }
       },
       "failures": []
@@ -1977,6 +2033,37 @@
       },
       "failures": []
     },
+    {
+      "name": "docs/RESEARCH_TAKEAWAYS.md",
+      "status": "pass",
+      "local": {
+        "path": "repo:RESEARCH_TAKEAWAYS.md",
+        "exists": true,
+        "bytes": 3742,
+        "sha256": "d8cd581e05879bed9f79e967af55d92a83154fc41eb20f80b923a19fef1bd95a"
+      },
+      "mirrors": {
+        "hf_space": {
+          "path": "hf_space:RESEARCH_TAKEAWAYS.md",
+          "exists": true,
+          "bytes": 3742,
+          "sha256": "d8cd581e05879bed9f79e967af55d92a83154fc41eb20f80b923a19fef1bd95a"
+        },
+        "hf_artifacts": {
+          "path": "hf_artifacts:RESEARCH_TAKEAWAYS.md",
+          "exists": true,
+          "bytes": 3742,
+          "sha256": "d8cd581e05879bed9f79e967af55d92a83154fc41eb20f80b923a19fef1bd95a"
+        },
+        "hf_model": {
+          "path": "hf_model:RESEARCH_TAKEAWAYS.md",
+          "exists": true,
+          "bytes": 3742,
+          "sha256": "d8cd581e05879bed9f79e967af55d92a83154fc41eb20f80b923a19fef1bd95a"
+        }
+      },
+      "failures": []
+    },
     {
       "name": "docs/SOURCE_ALIGNMENT_AUDIT.md",
       "status": "pass",

docs/data/project_status.json

@@ -54,6 +54,16 @@
             ],
             "readout": "Windowing, chronological split, per-task metrics, leakage controls, and unsupported interpretations are generated from committed metric artifacts."
         },
+        {
+            "area": "Research takeaways",
+            "status": "verified",
+            "evidence": [
+                "RESEARCH_TAKEAWAYS.md",
+                "docs/data/research_takeaways.json",
+                "scripts/build_research_takeaways.py"
+            ],
+            "readout": "The main result interpretation is generated from committed metrics: chronological class shift, neural gains on dynamics/order/alignment, open retrieval/reconstruction problems, and the need for held-out episodes."
+        },
         {
             "area": "Official dataset wording",
             "status": "verified",
@@ -127,6 +137,7 @@
         "Open docs/data/project_packet.json for the machine-readable project path.",
         "Inspect docs/data/summary_metrics.json and results/episode_task_suite/neural_mlp/ to check the 12-task outputs.",
         "Inspect EVALUATION_PROTOCOL.md before judging task metrics or leakage controls.",
+        "Inspect RESEARCH_TAKEAWAYS.md and docs/data/research_takeaways.json before interpreting model scores.",
         "Inspect SOURCE_ALIGNMENT_AUDIT.md before judging source-card consistency across public surfaces.",
         "Inspect XPERIENCE10M_DATASET_CARD_ALIGNMENT.md before judging dataset wording.",
     "Inspect results/omni_finetune/DATA_BLOCKER_REPORT.md before judging Qwen3-Omni scale-up status."

docs/data/public_surface_qa.json

@@ -1,7 +1,7 @@
 {
   "title": "Ropedia Xperience-10M Public Surface QA",
   "status": "pass",
-  "generated_at_utc": "2026-06-02T08:34:07+00:00",
+  "generated_at_utc": "2026-06-02T09:30:35+00:00",
   "scope": "Repo README, GitHub Pages HTML, Hugging Face Space card, artifact dataset card, and model card.",
   "checks": [
     {
@@ -18,7 +18,7 @@
         "website_integrity": {
           "exists": true,
           "status": "pass",
-          "generated_at_utc": "2026-06-02T08:19:22+00:00"
+          "generated_at_utc": "2026-06-02T09:14:19+00:00"
         },
         "task_surface_integrity": {
           "exists": true,
@@ -28,7 +28,7 @@
         "source_alignment": {
           "exists": true,
           "status": "pass",
-          "generated_at_utc": "2026-06-02T06:43:29+00:00"
+          "generated_at_utc": "2026-06-02T09:14:13+00:00"
         },
         "scope_claims": {
           "exists": true,
@@ -38,12 +38,12 @@
         "publication_hygiene": {
           "exists": true,
           "status": "pass",
-          "generated_at_utc": "2026-06-02T08:19:42+00:00"
+          "generated_at_utc": "2026-06-02T09:20:06+00:00"
         },
         "mirror_parity": {
           "exists": true,
           "status": "pass",
-          "generated_at_utc": "2026-06-02T08:20:43+00:00"
+          "generated_at_utc": "2026-06-02T09:21:39+00:00"
         },
         "live_publication": {
           "exists": true,
@@ -74,15 +74,15 @@
       "status": "pass",
       "reason": "The long research dashboard should be navigable as real tabs, including keyboard support.",
       "marker_counts": {
-        "role=\"tablist\"": 2,
-        "role=\"tab\"": 5,
-        "role=\"tabpanel\"": 19,
-        "aria-selected": 7,
-        "aria-controls": 6,
+        "role=\"tablist\"": 3,
+        "role=\"tab\"": 9,
+        "role=\"tabpanel\"": 24,
+        "aria-selected": 12,
+        "aria-controls": 10,
         "moveProjectTabFocus": 2,
-        "ArrowRight": 4,
-        "Home": 4,
-        "End": 4
+        "ArrowRight": 6,
+        "Home": 6,
+        "End": 6
       }
     },
     {
@@ -97,7 +97,7 @@
       "marker_counts": {
         "Ropedia Xperience-10M Task Suite": 15,
         "Xperience-10M": 95,
-        "12-task": 20,
+        "12-task": 21,
         "Qwen3-Omni": 35,
         "one public Xperience-10M sample episode": 2
       }
@@ -107,7 +107,7 @@
       "status": "pass",
       "reason": "Public cards should link the repo, Space, artifacts, model baselines, upstream dataset, and Ropedia dataset page.",
       "marker_counts": {
-        "https://github.com/ChaoYue0307/ropedia-xperience-10m-task-suite": 55,
+        "https://github.com/ChaoYue0307/ropedia-xperience-10m-task-suite": 58,
         "https://huggingface.co/spaces/cy0307/ropedia-xperience-10m-task-suite": 6,
         "https://huggingface.co/datasets/cy0307/ropedia-xperience-10m-task-suite-artifacts": 4,
         "https://huggingface.co/cy0307/ropedia-xperience-10m-task-baselines": 5,

docs/data/publication_audit.json

@@ -1,6 +1,6 @@
 {
   "status": "pass",
-  "generated_at_utc": "2026-06-02T08:45:24+00:00",
+  "generated_at_utc": "2026-06-02T09:30:39+00:00",
   "checks": [
     {
       "name": "required_publication_assets_present",
@@ -50,6 +50,7 @@
     "codemeta.json": true,
     "ARTIFACT_GUIDE.md": true,
     "PROJECT_STATUS.md": true,
+    "RESEARCH_TAKEAWAYS.md": true,
     "QUALITY_GATES.md": true,
     "PUBLIC_SURFACE_QA.md": true,
     "EVALUATION_PROTOCOL.md": true,
@@ -78,6 +79,7 @@
     "docs/data/project_manifest.json": true,
     "docs/data/project_packet.json": true,
     "docs/data/project_status.json": true,
+    "docs/data/research_takeaways.json": true,
     "docs/data/xperience10m_dataset_card_alignment.json": true,
     "docs/data/reproducibility_matrix.json": true,
     "docs/data/modality_atlas.json": true,
@@ -131,7 +133,7 @@
       "surface": "github_repo",
       "path": "README.md",
       "exists": true,
-      "required_marker_count": 16,
+      "required_marker_count": 19,
       "missing_markers": [],
       "status": "pass"
     },
@@ -139,7 +141,7 @@
       "surface": "hf_space_bundle",
       "path": "README.md",
       "exists": true,
-      "required_marker_count": 17,
+      "required_marker_count": 19,
       "missing_markers": [],
       "status": "pass"
     },
@@ -147,7 +149,7 @@
       "surface": "hf_artifact_bundle",
       "path": "README.md",
       "exists": true,
-      "required_marker_count": 16,
+      "required_marker_count": 18,
       "missing_markers": [],
       "status": "pass"
     },
@@ -155,7 +157,7 @@
       "surface": "hf_artifact_bundle",
       "path": "PROJECT_README.md",
       "exists": true,
-      "required_marker_count": 16,
+      "required_marker_count": 19,
       "missing_markers": [],
       "status": "pass"
     },
@@ -163,7 +165,7 @@
       "surface": "hf_model_bundle",
       "path": "README.md",
       "exists": true,
-      "required_marker_count": 17,
+      "required_marker_count": 19,
       "missing_markers": [],
       "status": "pass"
     }
@@ -172,8 +174,8 @@
     "github_repo": {
       "root": "repo",
       "exists": true,
-      "file_count": 322,
-      "text_file_count": 258,
+      "file_count": 325,
+      "text_file_count": 261,
       "largest_file": {
         "path": "results/episode_task_suite/modality_reconstruction/predictions.npz",
         "bytes": 52601010
@@ -183,8 +185,8 @@
     "hf_space_bundle": {
       "root": "hf_publish/space",
       "exists": true,
-      "file_count": 78,
-      "text_file_count": 56,
+      "file_count": 80,
+      "text_file_count": 58,
       "largest_file": {
         "path": "assets/task_suite_infographic.png",
         "bytes": 2600193
@@ -194,8 +196,8 @@
     "hf_artifact_bundle": {
       "root": "hf_publish/artifacts",
       "exists": true,
-      "file_count": 363,
-      "text_file_count": 277,
+      "file_count": 366,
+      "text_file_count": 280,
       "largest_file": {
         "path": "results/episode_task_suite/modality_reconstruction/predictions.npz",
         "bytes": 52601010
@@ -205,8 +207,8 @@
     "hf_model_bundle": {
       "root": "hf_publish/model",
       "exists": true,
-      "file_count": 231,
-      "text_file_count": 181,
+      "file_count": 234,
+      "text_file_count": 184,
       "largest_file": {
         "path": "artifacts/episode_task_suite/cross_modal_retrieval/model.npz",
         "bytes": 41310574

docs/data/quality_gates.json

@@ -1,7 +1,7 @@
 {
   "title": "Ropedia Xperience-10M Publication Quality Gates",
   "status": "pass",
-  "generated_at_utc": "2026-06-02T08:34:07+00:00",
+  "generated_at_utc": "2026-06-02T09:30:55+00:00",
   "rule": "Do not present a release as current unless every automated gate passes, then verify live GitHub/HF mirrors after publishing.",
   "automated_gates": [
     {

docs/data/research_takeaways.json

@@ -0,0 +1,154 @@
+{
+  "title": "Ropedia Xperience-10M Research Takeaways",
+  "status": "pass",
+  "generated_at_utc": "2026-06-02T09:02:43+00:00",
+  "source_files": [
+    "docs/data/summary_metrics.json",
+    "results/episode_task_suite/summary_report.json",
+    "results/episode_task_suite/neural_mlp/*/metrics.json",
+    "results/omni_finetune/MULTI_EPISODE_ACCESS_STATUS.md"
+  ],
+  "scope": {
+    "validated_episode_count": 1,
+    "num_frames": 5821,
+    "num_windows": 1161,
+    "feature_dim": 8378,
+    "audio_featurized": false,
+    "raw_data_redistributed": false
+  },
+  "takeaways": [
+    {
+      "id": "episode_to_benchmark",
+      "title": "One episode can become a real benchmark contract",
+      "claim": "The public sample is converted into 5,821 frames, 1,161 aligned 20-frame windows, and an 8,378-dimensional feature contract.",
+      "evidence": [
+        {
+          "label": "frames",
+          "value": 5821
+        },
+        {
+          "label": "windows",
+          "value": 1161
+        },
+        {
+          "label": "feature_dim",
+          "value": 8378
+        }
+      ],
+      "source": "docs/data/summary_metrics.json",
+      "boundary": "This is a task-development benchmark, not cross-episode generalization."
+    },
+    {
+      "id": "chronological_split_exposes_class_shift",
+      "title": "Chronological splits expose action-class shift",
+      "claim": "Earlier all-feature action classifiers reach high macro-F1 on their local split, but the 12-task chronological action/subtask heads are much harder because later held-out windows include unseen labels.",
+      "evidence": [
+        {
+          "label": "all_feature_action_macro_f1",
+          "value": 0.9791023658779895
+        },
+        {
+          "label": "suite_action_macro_f1",
+          "value": 0.05
+        },
+        {
+          "label": "suite_subtask_macro_f1",
+          "value": 0.04954121121178666
+        },
+        {
+          "label": "unseen_action_test_classes",
+          "value": 4
+        }
+      ],
+      "source": "results/episode_task_suite/summary_report.json",
+      "boundary": "This is an important leakage/split lesson, not evidence that action recognition is solved."
+    },
+    {
+      "id": "neural_heads_help_dynamics",
+      "title": "Small neural heads help dynamic and temporal probes",
+      "claim": "The MLP heads substantially improve hand trajectory forecasting, temporal-order verification, and motion/visual synchronization.",
+      "evidence": [
+        {
+          "label": "hand_mpjpe_minimal",
+          "value": 0.8222644925117493
+        },
+        {
+          "label": "hand_mpjpe_neural",
+          "value": 0.11163123697042465
+        },
+        {
+          "label": "hand_mpjpe_relative_improvement",
+          "value": 0.8642392587944207
+        },
+        {
+          "label": "temporal_order_f1_minimal",
+          "value": 0.5487364620938628
+        },
+        {
+          "label": "temporal_order_f1_neural",
+          "value": 0.8717948717948718
+        },
+        {
+          "label": "misalignment_f1_minimal",
+          "value": 0.4865671641791045
+        },
+        {
+          "label": "misalignment_f1_neural",
+          "value": 0.7335243553008597
+        }
+      ],
+      "source": "results/episode_task_suite/neural_mlp/*/metrics.json",
+      "boundary": "These gains are within one episode and should be re-tested on held-out episodes."
+    },
+    {
+      "id": "retrieval_and_reconstruction_remain_open",
+      "title": "Retrieval and reconstruction remain the harder multimodal problems",
+      "claim": "Ridge/cosine retrieval remains stronger than the neural projection on this sample, and cross-modal reconstruction still has negative R2.",
+      "evidence": [
+        {
+          "label": "retrieval_mrr_minimal",
+          "value": 0.26335984006618296
+        },
+        {
+          "label": "retrieval_mrr_neural",
+          "value": 0.1530070022204131
+        },
+        {
+          "label": "retrieval_top5_minimal",
+          "value": 0.3764367816091954
+        },
+        {
+          "label": "reconstruction_r2_minimal",
+          "value": -0.016022846771134747
+        },
+        {
+          "label": "reconstruction_r2_neural",
+          "value": -0.010198171891414143
+        }
+      ],
+      "source": "results/episode_task_suite/cross_modal_retrieval/metrics.json",
+      "boundary": "The current reconstruction task is feature-vector reconstruction, not depth, mesh, NeRF, or Gaussian splatting."
+    },
+    {
+      "id": "scale_requires_episodes",
+      "title": "The next scientific unit is held-out episodes, not more adjacent windows",
+      "claim": "The prepared Qwen3-Omni path targets 32 episodes from 32 sessions, but it remains data-gated until access and held-out evaluation complete.",
+      "evidence": [
+        {
+          "label": "target_episodes",
+          "value": 32
+        },
+        {
+          "label": "selected_sessions",
+          "value": 32
+        },
+        {
+          "label": "valid_candidates",
+          "value": 680
+        }
+      ],
+      "source": "results/omni_finetune/MULTI_EPISODE_ACCESS_STATUS.md",
+      "boundary": "No real 32-episode fine-tune is claimed until gated data is available locally and held-out evaluation runs."
+    }
+  ]
+}

docs/data/source_alignment_audit.json

@@ -1,7 +1,7 @@
 {
   "title": "Ropedia Xperience-10M Source Alignment Audit",
   "status": "pass",
-  "generated_at_utc": "2026-06-02T08:34:07+00:00",
+  "generated_at_utc": "2026-06-02T09:14:13+00:00",
   "alignment_json": "docs/data/xperience10m_dataset_card_alignment.json",
   "alignment_summary": {
     "full_dataset_repo": "ropedia-ai/xperience-10m",

docs/data/website_integrity.json

@@ -1,13 +1,13 @@
 {
   "status": "pass",
-  "generated_at_utc": "2026-06-02T08:34:10+00:00",
+  "generated_at_utc": "2026-06-02T09:30:36+00:00",
   "docs_root": "docs",
   "site_base": "/ropedia-xperience-10m-task-suite/",
   "summary": {
     "html_pages": 2,
-    "local_references": 90,
-    "external_reference_count": 72,
-    "json_files": 24,
+    "local_references": 92,
+    "external_reference_count": 75,
+    "json_files": 25,
     "image_assets_referenced": 19,
     "failure_count": 0
   },
@@ -30,7 +30,7 @@
       "name": "project_sections_are_assigned_to_tabs",
       "status": "pass",
       "reason": "Every major research section should be assigned to a tab group.",
-      "section_count": 19
+      "section_count": 20
     },
     {
       "name": "project_hash_router_preserves_deep_links",
@@ -43,21 +43,23 @@
       "name": "project_tabs_use_accessible_roles",
       "status": "pass",
       "reason": "The tabbed research dashboard should expose tablist/tab semantics.",
-      "tab_role_count": 5,
+      "tab_role_count": 9,
+      "project_tab_count": 5,
+      "nested_tab_count": 4,
       "has_tablist": true
     },
     {
       "name": "project_sections_are_labeled_tabpanels",
       "status": "pass",
       "reason": "Every tabbed research section should expose a labeled panel role.",
-      "panel_count": 19,
-      "labeled_panel_count": 19
+      "panel_count": 24,
+      "labeled_panel_count": 20
     },
     {
       "name": "project_tabs_update_selected_state",
       "status": "pass",
       "reason": "Tab activation should update selected state for assistive technology.",
-      "selected_count": 7,
+      "selected_count": 12,
       "updates_selected_state": true
     },
     {
@@ -72,8 +74,8 @@
       "name": "project_overview_precedes_progress_ledger",
       "status": "pass",
       "reason": "The project overview should appear before the deeper progress ledger.",
-      "overview_index": 52926,
-      "evidence_index": 59780
+      "overview_index": 55077,
+      "evidence_index": 61931
     },
     {
       "name": "project_status_links_json",
@@ -85,9 +87,9 @@
       "name": "evaluation_protocol_between_overview_and_progress",
       "status": "pass",
       "reason": "The evaluation protocol should appear before the deeper evidence ledger.",
-      "overview_index": 52926,
-      "protocol_index": 57288,
-      "evidence_index": 59780
+      "overview_index": 55077,
+      "protocol_index": 59439,
+      "evidence_index": 61931
     },
     {
       "name": "evaluation_protocol_links_json",
@@ -160,15 +162,15 @@
     },
     {
       "path": "index.html",
-      "id_count": 65,
-      "reference_count": 89,
+      "id_count": 74,
+      "reference_count": 91,
       "image_count": 22
     }
   ],
   "json_files": [
     {
       "path": "data/artifact_index.json",
-      "bytes": 24778,
+      "bytes": 26149,
       "top_level_type": "dict"
     },
     {
@@ -198,7 +200,7 @@
     },
     {
       "path": "data/mirror_parity.json",
-      "bytes": 69773,
+      "bytes": 72677,
       "top_level_type": "dict"
     },
     {
@@ -218,17 +220,17 @@
     },
     {
       "path": "data/project_status.json",
-      "bytes": 6111,
+      "bytes": 6748,
       "top_level_type": "dict"
     },
     {
       "path": "data/public_surface_qa.json",
-      "bytes": 5288,
+      "bytes": 5290,
       "top_level_type": "dict"
     },
     {
       "path": "data/publication_audit.json",
-      "bytes": 6796,
+      "bytes": 6878,
       "top_level_type": "dict"
     },
     {
@@ -251,6 +253,11 @@
       "bytes": 14390,
       "top_level_type": "dict"
     },
+    {
+      "path": "data/research_takeaways.json",
+      "bytes": 5251,
+      "top_level_type": "dict"
+    },
     {
       "path": "data/scope_claims_audit.json",
       "bytes": 20081,
@@ -278,7 +285,7 @@
     },
     {
       "path": "data/website_integrity.json",
-      "bytes": 11168,
+      "bytes": 11339,
       "top_level_type": "dict"
     },
     {

docs/index.html

@@ -309,6 +309,7 @@
       background: rgba(2, 5, 2, 0.72);
       scroll-margin-top: 132px;
     }
+    main > section:focus { outline: none; }
     main.tabbed > section[hidden] { display: none; }
     .project-tabs-shell {
       order: 0;
@@ -404,25 +405,79 @@
       background: rgba(164, 242, 127, 0.14);
       color: var(--green);
     }
+    .content-tabs {
+      display: flex;
+      gap: 8px;
+      overflow-x: auto;
+      padding: 2px 0 4px;
+      margin-bottom: 18px;
+      scrollbar-width: thin;
+      scrollbar-color: rgba(164, 242, 127, 0.42) rgba(7, 18, 7, 0.72);
+    }
+    .content-tab {
+      appearance: none;
+      flex: 0 0 auto;
+      min-width: 190px;
+      min-height: 56px;
+      border: 1px solid rgba(164, 242, 127, 0.18);
+      border-radius: 6px;
+      background:
+        linear-gradient(180deg, rgba(164, 242, 127, 0.05), rgba(7, 18, 7, 0.72)),
+        rgba(2, 5, 2, 0.54);
+      color: #dce8d6;
+      cursor: pointer;
+      padding: 10px 12px;
+      text-align: left;
+      font: inherit;
+      transition: transform 220ms cubic-bezier(0.16, 1, 0.3, 1), border-color 220ms cubic-bezier(0.16, 1, 0.3, 1), background 220ms cubic-bezier(0.16, 1, 0.3, 1), color 220ms cubic-bezier(0.16, 1, 0.3, 1);
+    }
+    .content-tab:hover {
+      transform: translateY(-1px);
+      border-color: rgba(164, 242, 127, 0.42);
+      color: var(--ink);
+    }
+    .content-tab strong {
+      display: block;
+      color: var(--ink);
+      font-family: var(--font-ui);
+      font-size: 14px;
+      line-height: 1.15;
+    }
+    .content-tab span {
+      display: block;
+      margin-top: 4px;
+      color: var(--muted);
+      font-size: 11px;
+      line-height: 1.25;
+    }
+    .content-tab.active {
+      border-color: rgba(164, 242, 127, 0.82);
+      background: rgba(164, 242, 127, 0.14);
+      color: var(--green);
+    }
+    .content-tab.active strong,
+    .content-tab.active span { color: var(--green); }
+    .tabbed-panel[hidden] { display: none; }
     #overview { order: 1; }
     #reading-path { order: 2; }
     #dataset-card { order: 3; }
     #suite { order: 4; }
     #pipeline { order: 5; }
     #protocol { order: 6; }
-    #models { order: 7; }
-    #neural { order: 8; }
-    #directions { order: 9; }
-    #extensions { order: 10; }
-    #architectures { order: 11; }
-    #walkthroughs { order: 12; }
-    #tasks { order: 13; }
-    #features { order: 14; }
-    #diagnostics { order: 15; }
-    #evidence { order: 16; }
-    #artifacts { order: 17; }
-    #omni-relay { order: 18; }
-    #run { order: 19; }
+    #takeaways { order: 7; }
+    #models { order: 8; }
+    #neural { order: 9; }
+    #directions { order: 10; }
+    #extensions { order: 11; }
+    #architectures { order: 12; }
+    #walkthroughs { order: 13; }
+    #tasks { order: 14; }
+    #features { order: 15; }
+    #diagnostics { order: 16; }
+    #evidence { order: 17; }
+    #artifacts { order: 18; }
+    #omni-relay { order: 19; }
+    #run { order: 20; }
     #suite { padding: 62px 0 76px; }
     #suite .wrap { width: min(1680px, calc(100% - 48px)); }
     #suite .section-head { max-width: var(--max); margin-inline: auto; }
@@ -1517,32 +1572,48 @@
       .wrap { width: min(100% - 28px, var(--max)); }
       .project-tabs-shell { top: 64px; padding: 10px 0; }
       .project-tabs {
-        grid-template-columns: repeat(2, minmax(0, 1fr));
+        display: flex;
+        grid-template-columns: none;
         gap: 8px;
+        overflow-x: auto;
+        padding-bottom: 4px;
+        scroll-snap-type: x proximity;
+        scrollbar-width: thin;
+        scrollbar-color: rgba(164, 242, 127, 0.42) rgba(7, 18, 7, 0.72);
       }
       .project-tab {
-        min-height: 54px;
-        padding: 10px;
+        flex: 0 0 min(42vw, 168px);
+        min-height: 50px;
+        padding: 10px 11px;
+        scroll-snap-align: start;
       }
       .project-tab strong { font-size: 14px; }
       .project-tab span {
-        font-size: 10px;
+        font-size: 10.5px;
         line-height: 1.25;
       }
       .section-tabs {
-        flex-wrap: wrap;
-        gap: 6px;
-        overflow-x: visible;
+        flex-wrap: nowrap;
+        gap: 8px;
+        overflow-x: auto;
         padding-top: 8px;
+        padding-bottom: 4px;
       }
       .section-tab {
-        flex: 1 1 calc(50% - 6px);
+        flex: 0 0 auto;
         min-height: 34px;
         padding: 7px 10px;
         font-size: 12px;
-        white-space: normal;
+        white-space: nowrap;
+      }
+      .content-tabs {
+        margin-bottom: 14px;
       }
-      main > section { scroll-margin-top: 246px; }
+      .content-tab {
+        min-width: min(76vw, 210px);
+        min-height: 52px;
+      }
+      main > section { scroll-margin-top: 184px; }
       .hero-stats, .models, .task-grid, .artifact-grid, .evidence-grid, .reading-grid, .snapshot-grid, .boundary-strip, .chart-grid, .callout-row, .direction-grid, .baseline-strip, .extension-grid, .walk-flow, .flow-steps, .storyboard-steps, .task-selector, .atlas-rows { grid-template-columns: 1fr; }
       .artifact-group { padding: 16px; }
       .modality-atlas-panel { padding: 14px; }
@@ -1659,7 +1730,7 @@
           <strong>Method</strong>
           <span>pipeline and model design</span>
         </button>
-        <button type="button" class="project-tab" id="tab-results" role="tab" data-tab-key="results" data-default-section="models" aria-selected="false" aria-pressed="false" aria-controls="models neural directions extensions diagnostics" tabindex="-1">
+        <button type="button" class="project-tab" id="tab-results" role="tab" data-tab-key="results" data-default-section="takeaways" aria-selected="false" aria-pressed="false" aria-controls="takeaways models neural directions extensions diagnostics" tabindex="-1">
           <strong>Results</strong>
           <span>baselines and research tracks</span>
         </button>
@@ -2054,6 +2125,44 @@
       </div>
     </section>
 
+    <section id="takeaways" data-project-tab="results" role="tabpanel" aria-labelledby="tab-results" tabindex="-1">
+      <div class="wrap">
+        <div class="section-head">
+          <h2>What the current results actually say.</h2>
+          <p>A generated takeaways layer reads the committed metrics and separates useful research signals from claims that still require held-out episodes.</p>
+        </div>
+        <div class="artifact-grid">
+          <article class="artifact primary-artifact">
+            <div>
+              <h3>One episode becomes a benchmark contract</h3>
+              <p>The public sample is converted into 5,821 frames, 1,161 aligned 20-frame windows, and an 8,378-dimensional feature contract.</p>
+            </div>
+            <a href="data/research_takeaways.json">research_takeaways.json</a>
+          </article>
+          <article class="artifact">
+            <h3>Chronological split exposes class shift</h3>
+            <p>All-feature action reaches 0.9791 macro-F1 on its local split, while the 12-task chronological action head is 0.0500 macro-F1 with four unseen later action labels.</p>
+            <a href="https://github.com/ChaoYue0307/ropedia-xperience-10m-task-suite/blob/main/RESEARCH_TAKEAWAYS.md">takeaways</a>
+          </article>
+          <article class="artifact">
+            <h3>Neural heads help dynamics</h3>
+            <p>Hand MPJPE improves from 0.8223 to 0.1116; temporal-order F1 rises from 0.5487 to 0.8718; misalignment F1 rises from 0.4866 to 0.7335.</p>
+            <a href="data/research_takeaways.json">metrics</a>
+          </article>
+          <article class="artifact">
+            <h3>Retrieval and reconstruction remain open</h3>
+            <p>Ridge/cosine retrieval remains stronger than the neural projection here, and cross-modal feature reconstruction still has negative R2.</p>
+            <a href="https://github.com/ChaoYue0307/ropedia-xperience-10m-task-suite/blob/main/results/episode_task_suite/cross_modal_retrieval/metrics.json">retrieval metrics</a>
+          </article>
+          <article class="artifact">
+            <h3>Scale means held-out episodes</h3>
+            <p>The next credible model-quality unit is a 32-episode held-out pilot across 32 sessions, not more adjacent windows from one sample.</p>
+            <a href="https://github.com/ChaoYue0307/ropedia-xperience-10m-task-suite/blob/main/results/omni_finetune/MULTI_EPISODE_ACCESS_STATUS.md">scale-up status</a>
+          </article>
+        </div>
+      </div>
+    </section>
+
     <section id="models" data-project-tab="results" role="tabpanel" aria-labelledby="tab-results" tabindex="-1">
       <div class="wrap">
         <div class="section-head">
@@ -2302,7 +2411,25 @@
           <p>Metrics, predictions, manifests, lightweight model weights, and derived window artifacts are organized so the project can be inspected, extended, and scaled before rerunning the full pipeline. Raw Xperience-10M data and Qwen weights are not redistributed.</p>
         </div>
         <div class="artifact-library">
-          <section class="artifact-group">
+          <div class="content-tabs" role="tablist" aria-label="Artifact categories">
+            <button type="button" class="content-tab active" id="artifact-tab-task-heads" role="tab" data-panel-target="artifact-panel-task-heads" aria-selected="true" aria-pressed="true" aria-controls="artifact-panel-task-heads">
+              <strong>Task Heads</strong>
+              <span>windows, features, metrics</span>
+            </button>
+            <button type="button" class="content-tab" id="artifact-tab-public-surfaces" role="tab" data-panel-target="artifact-panel-public-surfaces" aria-selected="false" aria-pressed="false" aria-controls="artifact-panel-public-surfaces" tabindex="-1">
+              <strong>Public Surfaces</strong>
+              <span>repo, HF, project map</span>
+            </button>
+            <button type="button" class="content-tab" id="artifact-tab-scale-up" role="tab" data-panel-target="artifact-panel-scale-up" aria-selected="false" aria-pressed="false" aria-controls="artifact-panel-scale-up" tabindex="-1">
+              <strong>Scale-Up</strong>
+              <span>data gate and Omni path</span>
+            </button>
+            <button type="button" class="content-tab" id="artifact-tab-checks" role="tab" data-panel-target="artifact-panel-checks" aria-selected="false" aria-pressed="false" aria-controls="artifact-panel-checks" tabindex="-1">
+              <strong>Checks</strong>
+              <span>validators and parity</span>
+            </button>
+          </div>
+          <section class="artifact-group tabbed-panel" id="artifact-panel-task-heads" role="tabpanel" aria-labelledby="artifact-tab-task-heads">
             <div class="artifact-group-head">
               <div><span>Research artifacts</span><h3>From one episode to task heads</h3></div>
               <p>Start with the files that define the sample windows, feature blocks, task contracts, metrics, walkthroughs, and research-direction mapping.</p>
@@ -2319,7 +2446,7 @@
             </div>
           </section>
 
-          <section class="artifact-group">
+          <section class="artifact-group tabbed-panel" id="artifact-panel-public-surfaces" role="tabpanel" aria-labelledby="artifact-tab-public-surfaces" hidden>
             <div class="artifact-group-head">
               <div><span>Public surfaces</span><h3>Project map, mirrors, and runnable code</h3></div>
               <p>Use these files to navigate the whole project, open the published mirrors, or reproduce the public-sample pipeline.</p>
@@ -2336,7 +2463,7 @@
             </div>
           </section>
 
-          <section class="artifact-group">
+          <section class="artifact-group tabbed-panel" id="artifact-panel-scale-up" role="tabpanel" aria-labelledby="artifact-tab-scale-up" hidden>
             <div class="artifact-group-head">
               <div><span>Scale-up path</span><h3>Prepared for multi-episode training</h3></div>
               <p>The multi-episode Qwen3-Omni path is documented and scripted, but no full-pilot metric is claimed until the data gate and held-out evaluation pass.</p>
@@ -2349,7 +2476,7 @@
             </div>
           </section>
 
-          <section class="artifact-group">
+          <section class="artifact-group tabbed-panel" id="artifact-panel-checks" role="tabpanel" aria-labelledby="artifact-tab-checks" hidden>
             <div class="artifact-group-head">
               <div><span>Consistency checks</span><h3>Release checks behind the research site</h3></div>
               <p>These validator outputs support the public research artifacts by keeping links, mirrors, figures, source wording, and package boundaries consistent.</p>
@@ -2475,6 +2602,7 @@ python scripts/validate_publication_package.py</code></pre>
       { id: "protocol", label: "Evaluation Protocol" },
       { id: "architectures", label: "Model Architectures" },
       { id: "features", label: "Feature Blocks" },
+      { id: "takeaways", label: "Research Takeaways" },
       { id: "models", label: "Minimal Baselines" },
       { id: "neural", label: "Neural Heads" },
       { id: "directions", label: "Four Directions" },
@@ -2630,6 +2758,44 @@ python scripts/validate_publication_package.py</code></pre>
     window.addEventListener("hashchange", () => activateTabForHash({ scroll: true }));
     activateTabForHash({ scroll: Boolean(window.location.hash) });
 
+    function initContentTabs() {
+      document.querySelectorAll(".content-tabs").forEach((tablist) => {
+        const buttons = Array.from(tablist.querySelectorAll("[data-panel-target]"));
+        if (!buttons.length) return;
+
+        const activatePanel = (activeButton, options = {}) => {
+          buttons.forEach((button) => {
+            const active = button === activeButton;
+            const panel = document.getElementById(button.dataset.panelTarget);
+            button.classList.toggle("active", active);
+            button.setAttribute("aria-selected", active ? "true" : "false");
+            button.setAttribute("aria-pressed", active ? "true" : "false");
+            button.tabIndex = active ? 0 : -1;
+            if (panel) panel.hidden = !active;
+          });
+          if (options.focus) activeButton.focus();
+        };
+
+        buttons.forEach((button, index) => {
+          button.addEventListener("click", () => activatePanel(button));
+          button.addEventListener("keydown", (event) => {
+            if (!["ArrowRight", "ArrowDown", "ArrowLeft", "ArrowUp", "Home", "End"].includes(event.key)) return;
+            event.preventDefault();
+            const lastIndex = buttons.length - 1;
+            let nextIndex = index;
+            if (event.key === "ArrowRight" || event.key === "ArrowDown") nextIndex = index === lastIndex ? 0 : index + 1;
+            if (event.key === "ArrowLeft" || event.key === "ArrowUp") nextIndex = index === 0 ? lastIndex : index - 1;
+            if (event.key === "Home") nextIndex = 0;
+            if (event.key === "End") nextIndex = lastIndex;
+            activatePanel(buttons[nextIndex], { focus: true });
+          });
+        });
+
+        activatePanel(buttons.find((button) => button.classList.contains("active")) || buttons[0]);
+      });
+    }
+    initContentTabs();
+
     const escapeHtml = (value) => String(value ?? "")
       .replaceAll("&", "&amp;")
       .replaceAll("<", "&lt;")

scripts/build_artifact_index.py

@@ -129,6 +129,30 @@ ARTIFACTS = [
         "surface": "repo_hf",
         "proves": "Regenerates the protocol from committed summary metrics and task artifacts.",
     },
+    {
+        "id": "research_takeaways",
+        "title": "Research takeaways",
+        "path": "RESEARCH_TAKEAWAYS.md",
+        "kind": "result_interpretation",
+        "surface": "repo_hf",
+        "proves": "Summarizes the main research lessons from committed metrics without broad model-quality overclaims.",
+    },
+    {
+        "id": "research_takeaways_json",
+        "title": "Research takeaways JSON",
+        "path": "docs/data/research_takeaways.json",
+        "kind": "result_interpretation",
+        "surface": "website_hf",
+        "proves": "Machine-readable result interpretation for the website, HF cards, and mirror checks.",
+    },
+    {
+        "id": "research_takeaways_builder",
+        "title": "Research takeaways builder",
+        "path": "scripts/build_research_takeaways.py",
+        "kind": "result_interpretation",
+        "surface": "repo_hf",
+        "proves": "Regenerates the research takeaways from committed summary metrics and task result artifacts.",
+    },
     {
         "id": "figure_index",
         "title": "Figure index",

scripts/build_public_surface_qa.py

@@ -165,9 +165,12 @@ def build_report() -> dict:
         check(
             "website_tabs_are_accessible_and_keyboardable",
             'role="tablist"' in website
-            and website.count('role="tab"') == 5
+            and website.count("data-tab-key=") == 5
+            and website.count("data-panel-target=") >= 4
+            and website.count('role="tab"') >= website.count("data-tab-key=") + website.count("data-panel-target=")
             and website.count('role="tabpanel"') >= 19
             and "moveProjectTabFocus" in website
+            and "initContentTabs" in website
             and "ArrowRight" in website
             and "Home" in website
             and "End" in website,

scripts/build_research_takeaways.py

@@ -0,0 +1,234 @@
+#!/usr/bin/env python3
+"""Build research takeaways from committed Xperience-10M metric artifacts."""
+
+from __future__ import annotations
+
+import json
+from datetime import datetime, timezone
+from pathlib import Path
+
+
+ROOT = Path(__file__).resolve().parents[1]
+SUMMARY_PATH = ROOT / "docs/data/summary_metrics.json"
+OUTPUT_JSON = ROOT / "docs/data/research_takeaways.json"
+OUTPUT_MD = ROOT / "RESEARCH_TAKEAWAYS.md"
+
+
+def pct_delta(new: float, old: float, higher_is_better: bool = True) -> float:
+    if old == 0:
+        return 0.0
+    if higher_is_better:
+        return (new - old) / abs(old)
+    return (old - new) / abs(old)
+
+
+def fmt(value: float | int | None, digits: int = 4) -> str:
+    if value is None:
+        return "n/a"
+    if isinstance(value, int):
+        return f"{value:,}"
+    return f"{value:.{digits}f}"
+
+
+def task_metric(tasks: dict, task: str, key: str) -> float:
+    return float(tasks[task][key])
+
+
+def build_payload() -> dict:
+    summary = json.loads(SUMMARY_PATH.read_text(encoding="utf-8"))
+    suite = summary["suite"]
+    tasks = suite["tasks"]
+    neural = suite.get("neural_tasks", {})
+    models = summary["models"]
+    omni = summary.get("omni_relay", {})
+
+    hand_min = task_metric(tasks, "hand_trajectory_forecast", "mpjpe")
+    hand_neural = task_metric(neural, "hand_trajectory_forecast", "mpjpe")
+    temporal_min = task_metric(tasks, "temporal_order", "f1")
+    temporal_neural = task_metric(neural, "temporal_order", "f1")
+    misalign_min = task_metric(tasks, "misalignment_detection", "f1")
+    misalign_neural = task_metric(neural, "misalignment_detection", "f1")
+    retrieval_min_mrr = task_metric(tasks, "cross_modal_retrieval", "mrr")
+    retrieval_neural_mrr = task_metric(neural, "cross_modal_retrieval", "mrr")
+    recon_min_r2 = task_metric(tasks, "modality_reconstruction", "r2")
+    recon_neural_r2 = task_metric(neural, "modality_reconstruction", "r2")
+    action_chrono = task_metric(tasks, "timeline_action", "macro_f1")
+    subtask_chrono = task_metric(tasks, "timeline_subtask", "macro_f1")
+
+    takeaways = [
+        {
+            "id": "episode_to_benchmark",
+            "title": "One episode can become a real benchmark contract",
+            "claim": (
+                "The public sample is converted into 5,821 frames, 1,161 aligned "
+                "20-frame windows, and an 8,378-dimensional feature contract."
+            ),
+            "evidence": [
+                {"label": "frames", "value": suite["num_frames"]},
+                {"label": "windows", "value": suite["num_windows"]},
+                {"label": "feature_dim", "value": suite["feature_dim"]},
+            ],
+            "source": "docs/data/summary_metrics.json",
+            "boundary": "This is a task-development benchmark, not cross-episode generalization.",
+        },
+        {
+            "id": "chronological_split_exposes_class_shift",
+            "title": "Chronological splits expose action-class shift",
+            "claim": (
+                "Earlier all-feature action classifiers reach high macro-F1 on their "
+                "local split, but the 12-task chronological action/subtask heads are "
+                "much harder because later held-out windows include unseen labels."
+            ),
+            "evidence": [
+                {"label": "all_feature_action_macro_f1", "value": models["all_modalities_action"]["macro_f1"]},
+                {"label": "suite_action_macro_f1", "value": action_chrono},
+                {"label": "suite_subtask_macro_f1", "value": subtask_chrono},
+                {"label": "unseen_action_test_classes", "value": len(tasks["timeline_action"].get("unseen_test_classes", []))},
+            ],
+            "source": "results/episode_task_suite/summary_report.json",
+            "boundary": "This is an important leakage/split lesson, not evidence that action recognition is solved.",
+        },
+        {
+            "id": "neural_heads_help_dynamics",
+            "title": "Small neural heads help dynamic and temporal probes",
+            "claim": (
+                "The MLP heads substantially improve hand trajectory forecasting, "
+                "temporal-order verification, and motion/visual synchronization."
+            ),
+            "evidence": [
+                {"label": "hand_mpjpe_minimal", "value": hand_min},
+                {"label": "hand_mpjpe_neural", "value": hand_neural},
+                {"label": "hand_mpjpe_relative_improvement", "value": pct_delta(hand_neural, hand_min, higher_is_better=False)},
+                {"label": "temporal_order_f1_minimal", "value": temporal_min},
+                {"label": "temporal_order_f1_neural", "value": temporal_neural},
+                {"label": "misalignment_f1_minimal", "value": misalign_min},
+                {"label": "misalignment_f1_neural", "value": misalign_neural},
+            ],
+            "source": "results/episode_task_suite/neural_mlp/*/metrics.json",
+            "boundary": "These gains are within one episode and should be re-tested on held-out episodes.",
+        },
+        {
+            "id": "retrieval_and_reconstruction_remain_open",
+            "title": "Retrieval and reconstruction remain the harder multimodal problems",
+            "claim": (
+                "Ridge/cosine retrieval remains stronger than the neural projection on "
+                "this sample, and cross-modal reconstruction still has negative R2."
+            ),
+            "evidence": [
+                {"label": "retrieval_mrr_minimal", "value": retrieval_min_mrr},
+                {"label": "retrieval_mrr_neural", "value": retrieval_neural_mrr},
+                {"label": "retrieval_top5_minimal", "value": tasks["cross_modal_retrieval"]["top5_accuracy"]},
+                {"label": "reconstruction_r2_minimal", "value": recon_min_r2},
+                {"label": "reconstruction_r2_neural", "value": recon_neural_r2},
+            ],
+            "source": "results/episode_task_suite/cross_modal_retrieval/metrics.json",
+            "boundary": "The current reconstruction task is feature-vector reconstruction, not depth, mesh, NeRF, or Gaussian splatting.",
+        },
+        {
+            "id": "scale_requires_episodes",
+            "title": "The next scientific unit is held-out episodes, not more adjacent windows",
+            "claim": (
+                "The prepared Qwen3-Omni path targets 32 episodes from 32 sessions, "
+                "but it remains data-gated until access and held-out evaluation complete."
+            ),
+            "evidence": [
+                {"label": "target_episodes", "value": omni.get("target_episodes")},
+                {"label": "selected_sessions", "value": omni.get("selected_sessions")},
+                {"label": "valid_candidates", "value": omni.get("valid_candidates")},
+            ],
+            "source": "results/omni_finetune/MULTI_EPISODE_ACCESS_STATUS.md",
+            "boundary": omni.get("claim_boundary", "No 32-episode fine-tune is claimed yet."),
+        },
+    ]
+
+    return {
+        "title": "Ropedia Xperience-10M Research Takeaways",
+        "status": "pass",
+        "generated_at_utc": datetime.now(timezone.utc).isoformat(timespec="seconds"),
+        "source_files": [
+            "docs/data/summary_metrics.json",
+            "results/episode_task_suite/summary_report.json",
+            "results/episode_task_suite/neural_mlp/*/metrics.json",
+            "results/omni_finetune/MULTI_EPISODE_ACCESS_STATUS.md",
+        ],
+        "scope": {
+            "validated_episode_count": 1,
+            "num_frames": suite["num_frames"],
+            "num_windows": suite["num_windows"],
+            "feature_dim": suite["feature_dim"],
+            "audio_featurized": False,
+            "raw_data_redistributed": False,
+        },
+        "takeaways": takeaways,
+    }
+
+
+def render_md(payload: dict) -> str:
+    lines = [
+        "# Research Takeaways",
+        "",
+        "This generated note summarizes what the current public Xperience-10M sample",
+        "pipeline actually shows. It is built from committed metric artifacts, not",
+        "from hand-entered benchmark claims.",
+        "",
+        "## Scope",
+        "",
+        f"- validated episodes: {payload['scope']['validated_episode_count']}",
+        f"- frames: {payload['scope']['num_frames']:,}",
+        f"- aligned windows: {payload['scope']['num_windows']:,}",
+        f"- current feature dimension: {payload['scope']['feature_dim']:,}",
+        "- raw Xperience-10M data is not redistributed",
+        "- audio is documented and visualized, but not yet featurized",
+        "",
+        "## Takeaways",
+        "",
+    ]
+    for item in payload["takeaways"]:
+        lines.extend(
+            [
+                f"### {item['title']}",
+                "",
+                item["claim"],
+                "",
+                "| Metric | Value |",
+                "| --- | ---: |",
+            ]
+        )
+        for evidence in item["evidence"]:
+            value = evidence["value"]
+            if isinstance(value, float):
+                value_text = fmt(value)
+            elif isinstance(value, int):
+                value_text = fmt(value)
+            elif value is None:
+                value_text = "n/a"
+            else:
+                value_text = str(value)
+            lines.append(f"| `{evidence['label']}` | {value_text} |")
+        lines.extend(["", f"Source: `{item['source']}`.", "", f"Boundary: {item['boundary']}", ""])
+    lines.extend(
+        [
+            "## How To Read These Results",
+            "",
+            "- High single-episode scores are useful pipeline checks, not broad embodied-AI claims.",
+            "- Low chronological action/subtask scores are informative because they expose later-label shift.",
+            "- Neural gains on trajectory/order/alignment make those tasks good candidates for the next fine-tuning stage.",
+            "- Retrieval and reconstruction remain the main multimodal representation challenges.",
+            "- The next credible model-quality result needs held-out episodes.",
+            "",
+        ]
+    )
+    return "\n".join(lines)
+
+
+def main() -> int:
+    payload = build_payload()
+    OUTPUT_JSON.write_text(json.dumps(payload, indent=2) + "\n", encoding="utf-8")
+    OUTPUT_MD.write_text(render_md(payload), encoding="utf-8")
+    print(f"PASS: wrote {OUTPUT_JSON}")
+    print(f"PASS: wrote {OUTPUT_MD}")
+    return 0
+
+
+if __name__ == "__main__":
+    raise SystemExit(main())

scripts/validate_mirror_parity.py

@@ -34,6 +34,7 @@ DATA_FILES = [
     "public_surface_qa.json",
     "quality_gates.json",
     "reproducibility_matrix.json",
+    "research_takeaways.json",
     "research_direction_extensions.json",
     "research_directions.json",
     "scope_claims_audit.json",
@@ -72,6 +73,7 @@ SCRIPT_FILES = [
     "build_figure_index.py",
     "build_quality_gates.py",
     "build_public_surface_qa.py",
+    "build_research_takeaways.py",
     "verify_live_publication.py",
     "validate_mirror_parity.py",
     "validate_publication_package.py",
@@ -95,6 +97,7 @@ DOC_FILES = [
     "FIGURE_INDEX.md",
     "PROJECT_STATUS.md",
     "PUBLIC_SURFACE_QA.md",
+    "RESEARCH_TAKEAWAYS.md",
     "SOURCE_ALIGNMENT_AUDIT.md",
     "XPERIENCE10M_DATASET_CARD_ALIGNMENT.md",
 ]

scripts/validate_publication_package.py

@@ -69,8 +69,11 @@ CARD_FRESHNESS_EXPECTATIONS = [
             "FIGURE_INDEX.md",
             "brand_assets.json",
             "PROJECT_STATUS.md",
+            "RESEARCH_TAKEAWAYS.md",
             "xperience10m-logo-social-card.png",
             "build_brand_assets.py",
+            "build_research_takeaways.py",
+            "research_takeaways.json",
             "cc-by-nc-4.0",
             "12,103 episode folders",
             "all 12 task families before the",
@@ -91,8 +94,10 @@ CARD_FRESHNESS_EXPECTATIONS = [
             "figure_index.json",
             "brand_assets.json",
             "project_status.json",
+            "research_takeaways.json",
             "xperience10m-logo-social-card.png",
             "build_brand_assets.py",
+            "build_research_takeaways.py",
             "cc-by-nc-4.0",
             "12,103 episode folders",
             "task-first 12-task infographic",
@@ -114,8 +119,10 @@ CARD_FRESHNESS_EXPECTATIONS = [
             "figure_index.json",
             "brand_assets.json",
             "project_status.json",
+            "research_takeaways.json",
             "xperience10m-logo-social-card.png",
             "build_brand_assets.py",
+            "build_research_takeaways.py",
             "cc-by-nc-4.0",
             "12,103 episode folders",
             "task-first 12-task map",
@@ -136,8 +143,11 @@ CARD_FRESHNESS_EXPECTATIONS = [
             "FIGURE_INDEX.md",
             "brand_assets.json",
             "PROJECT_STATUS.md",
+            "RESEARCH_TAKEAWAYS.md",
             "xperience10m-logo-social-card.png",
             "build_brand_assets.py",
+            "build_research_takeaways.py",
+            "research_takeaways.json",
             "cc-by-nc-4.0",
             "12,103 episode folders",
             "all 12 task families before the",
@@ -158,8 +168,10 @@ CARD_FRESHNESS_EXPECTATIONS = [
             "figure_index.json",
             "brand_assets.json",
             "project_status.json",
+            "research_takeaways.json",
             "xperience10m-logo-social-card.png",
             "build_brand_assets.py",
+            "build_research_takeaways.py",
             "cc-by-nc-4.0",
             "12,103 episode folders",
             "task-first 12-head",
@@ -276,6 +288,7 @@ def required_assets(root: Path) -> dict[str, bool]:
         "codemeta.json",
         "ARTIFACT_GUIDE.md",
         "PROJECT_STATUS.md",
+        "RESEARCH_TAKEAWAYS.md",
         "QUALITY_GATES.md",
         "PUBLIC_SURFACE_QA.md",
         "EVALUATION_PROTOCOL.md",
@@ -304,6 +317,7 @@ def required_assets(root: Path) -> dict[str, bool]:
         "docs/data/project_manifest.json",
         "docs/data/project_packet.json",
         "docs/data/project_status.json",
+        "docs/data/research_takeaways.json",
         "docs/data/xperience10m_dataset_card_alignment.json",
         "docs/data/reproducibility_matrix.json",
         "docs/data/modality_atlas.json",

scripts/validate_website_integrity.py

@@ -377,8 +377,20 @@ def validate(docs_root: Path, site_base: str) -> dict:
             detail = {"marker_count": marker_count, "has_section_tab_map": "sectionTabMap" in index_text}
         elif name == "project_tabs_use_accessible_roles":
             tab_role_count = index_text.count(marker)
-            passed = 'role="tablist"' in index_text and tab_role_count == 5
-            detail = {"tab_role_count": tab_role_count, "has_tablist": 'role="tablist"' in index_text}
+            project_tab_count = index_text.count("data-tab-key=")
+            nested_tab_count = index_text.count("data-panel-target=")
+            passed = (
+                'role="tablist"' in index_text
+                and project_tab_count == 5
+                and nested_tab_count >= 4
+                and tab_role_count >= project_tab_count + nested_tab_count
+            )
+            detail = {
+                "tab_role_count": tab_role_count,
+                "project_tab_count": project_tab_count,
+                "nested_tab_count": nested_tab_count,
+                "has_tablist": 'role="tablist"' in index_text,
+            }
         elif name == "project_sections_are_labeled_tabpanels":
             panel_count = index_text.count(marker)
             passed = panel_count >= 19 and index_text.count('aria-labelledby="tab-') >= 19

scripts/verify_live_publication.py

@@ -105,6 +105,17 @@ HASH_GROUPS = [
             "hf_model": "https://huggingface.co/cy0307/ropedia-xperience-10m-task-baselines/resolve/main/metrics/evaluation_protocol.json",
         },
     },
+    {
+        "id": "research_takeaways_json",
+        "title": "Research takeaways JSON",
+        "local_path": "docs/data/research_takeaways.json",
+        "urls": {
+            "github_pages": "https://chaoyue0307.github.io/ropedia-xperience-10m-task-suite/data/research_takeaways.json",
+            "hf_space": "https://huggingface.co/spaces/cy0307/ropedia-xperience-10m-task-suite/raw/main/data/research_takeaways.json",
+            "hf_artifacts": "https://huggingface.co/datasets/cy0307/ropedia-xperience-10m-task-suite-artifacts/resolve/main/docs/data/research_takeaways.json",
+            "hf_model": "https://huggingface.co/cy0307/ropedia-xperience-10m-task-baselines/resolve/main/metrics/research_takeaways.json",
+        },
+    },
     {
         "id": "figure_index_json",
         "title": "Figure index JSON",
@@ -185,6 +196,8 @@ MARKER_CHECKS = [
             "xperience10m_dataset_card_alignment.json",
             "source_alignment_audit.json",
             "evaluation_protocol.json",
+            "research_takeaways.json",
+            "Research Takeaways",
             "figure_index.json",
             "brand_assets.json",
             "xperience10m-logo-social-card.png",
@@ -216,6 +229,8 @@ MARKER_CHECKS = [
             "xperience10m_dataset_card_alignment.json",
             "source_alignment_audit.json",
             "evaluation_protocol.json",
+            "research_takeaways.json",
+            "Research Takeaways",
             "figure_index.json",
             "brand_assets.json",
             "xperience10m-logo-social-card.png",
@@ -247,6 +262,8 @@ MARKER_CHECKS = [
             "xperience10m_dataset_card_alignment.json",
             "source_alignment_audit.json",
             "evaluation_protocol.json",
+            "research_takeaways.json",
+            "Research Takeaways",
             "figure_index.json",
             "brand_assets.json",
             "xperience10m-logo-social-card.png",
@@ -270,6 +287,8 @@ MARKER_CHECKS = [
             "xperience10m_dataset_card_alignment.json",
             "source_alignment_audit.json",
             "evaluation_protocol.json",
+            "research_takeaways.json",
+            "Research Takeaways",
             "figure_index.json",
             "brand_assets.json",
             "xperience10m-logo-social-card.png",