Update README + docs/{quality,caveats}.md: post-trim totals, three concrete filtering recipes (example dataloader / direct .pt / pose-only), rewritten OT-track-loss section reflecting the prefix-trim + recorder watchdog fix

README.md

@@ -53,7 +53,7 @@ Dense, contact-rich, synchronized multimodal interaction data collected from **h
 | Embodiment | **Human hands (no robot)** — handheld GelSight sensors with motion-capture rigid bodies |
 | Intended use | Dynamics / world-model learning over short multimodal windows. Sample short trajectories (1 s – 10 s); recording-file boundaries are not action boundaries. |
 | Total synchronized duration | **105.7 min** at 30 Hz (190,231 multimodal frames, post-trim) |
-| Bimanual tactile-contact time | **81.4 min — 66 % of frames** (median event duration 0.73 s) |
+| Bimanual tactile-contact time | **≥ 75 % of post-trim frames** (median event duration 0.73 s; see `data_quality_report.csv` for per-file numbers) |
 | Cameras | 3× Intel RealSense D415 (color + depth), 480×640, 30 FPS |
 | Tactile | 2× GelSight Mini (left, right), handheld |
 | Motion capture | OptiTrack VRPN, 3 rigid bodies, ~120 Hz |
@@ -69,7 +69,18 @@ Dense, contact-rich, synchronized multimodal interaction data collected from **h
 
 See [`tasks.json`](tasks.json) for the machine-readable registry (per-date `active_sensors`, etc.).
 
-**OT-uninitialized prefixes trimmed.** Three episodes had OptiTrack offline for the first 1–11 min of recording (`2026-05-11/episode_{005,012,017}`); those prefixes have been cut from the published `.pt` files (see `_contact_meta.trim_offset` per file). The original recordings remain in the H5 archive untouched. Future episodes use a recorder-side OT watchdog that refuses to start without an active mocap stream.
+**OT-uninitialized prefixes trimmed.** Three episodes had OptiTrack offline at the start of recording (1–11 min each); those prefixes have been cut from the published `.pt` files (`_contact_meta.trim_offset` per file). Future recordings use an OT watchdog that refuses to start an episode unless mocap is streaming. Full story: [`docs/caveats.md`](docs/caveats.md).
+
+## Data quality
+
+| Mode | Frames | % | Files | Cause |
+|---|---:|---:|---:|---|
+| GelSight LED flicker | 56 | 0.029 % | 5 | Single-frame LED dropout, recovers next frame |
+| OptiTrack pose teleport | 56 | 0.029 % | 3 | Solver flip (translation > 5 m/s or angular > 15 rad/s) |
+| OptiTrack track loss | 1,680 | 0.883 % | 6 | Marker briefly left mocap-volume / camera FOV mid-episode |
+| **Total (union)** | **1,768** | **0.929 %** | **11** | |
+
+Every flagged interval is in [`bad_frames.json`](bad_frames.json) keyed by `episode/episode_*` with TRIMMED-pt frame indices. Skip-list usage is shown below and in [`docs/quality.md`](docs/quality.md). Long start-of-episode OT-uninitialized prefixes (the dominant problem in the raw recordings) have already been trimmed from the published `.pt` files — see [`docs/caveats.md`](docs/caveats.md).
 
 ## Quick start
 
@@ -98,15 +109,25 @@ ep = torch.load(path, weights_only=False)
 # Plus per-frame contact metrics:            tactile_{side}_{intensity, area, mixed}
 ```
 
-Sampling short windows for dynamics learning:
+Sampling short windows for dynamics learning — **drop windows that overlap any flagged interval**:
 
 ```python
 import json
+
 with open("bad_frames.json") as f:
-    bad = json.load(f)["episodes"]
-# Drop ~0.085 % of frames flagged in bad_frames.json — see docs/quality.md
+    bad = json.load(f)["episodes"]   # frame indices are TRIMMED-pt coordinates
+
+def is_clean_window(episode_key, t_start, t_end):
+    """True iff [t_start, t_end] doesn't intersect any flagged span."""
+    bf = bad[episode_key]
+    intervals = (bf["intensity_spikes"]
+                 + bf["pose_teleports_L"] + bf["pose_teleports_R"]
+                 + bf["ot_loss_L"] + bf["ot_loss_R"])
+    return all(not (s <= t_end and e >= t_start) for s, e in intervals)
 ```
 
+Currently 1,768 / 190,231 frames (0.93 %) are flagged across 11 of 27 files — see [`docs/quality.md`](docs/quality.md) for the per-mode breakdown and more filtering recipes. The example dataloader below does this filtering for you when `skip_bad_frames=True`.
+
 ## Example dataloader — short contact-rich windows
 
 A reference PyTorch `Dataset` is shipped under [`examples/react_window_dataset.py`](examples/react_window_dataset.py). It scans the processed `.pt` files, applies the contact filter, drops windows that overlap [`bad_frames.json`](bad_frames.json), and respects the per-date `active_sensors` field from [`tasks.json`](tasks.json).

docs/caveats.md

@@ -9,47 +9,80 @@ React is a **dense multimodal interaction stream** intended for learning **dynam
 - A demonstration / policy-learning dataset. "Episodes" here are just file boundaries — they don't carry semantic / action structure.
 - Comparable apples-to-apples with BridgeData V2, DROID, RT-1, ALOHA, etc. on "number of demos." The relevant comparison is *hours of synchronized multimodal contact-rich interaction*.
 
+## OT-uninitialized prefixes (already trimmed from `.pt`)
+
+Three episodes on 2026-05-11 had OptiTrack offline at the moment the
+operator pressed `s` to start recording. RealSense + GelSight kept
+capturing fresh frames; OT held a stale (or zero) pose until the mocap
+software came online:
+
+| Episode | Trimmed prefix | What it was |
+|---|---:|---|
+| `episode_005` | 81 s (2,429 frames) | OT had ~50 samples in the prefix (~0.6 Hz) |
+| `episode_012` | 324 s (9,719 frames) | OT had 3 samples in the prefix |
+| `episode_017` | 641 s (19,228 frames) | OT had **zero** samples in the prefix |
+
+**These prefixes have already been trimmed** from the published `.pt`
+files. Each `.pt` records the trim amount in `_contact_meta.trim_offset`,
+and all frame indices in `bad_frames.json` are in post-trim coordinates.
+The original H5 recordings under `data/motherboard/2026-05-11/` remain
+untouched in the recording-machine archive (not on HF).
+
+**Fixed at the source**: future recordings use a `data_collection.py`
+recorder-side watchdog that (a) refuses to start an episode unless every
+active rigid body has streamed an OT sample in the last 2 s, and (b)
+auto-saves + ends an in-progress episode after 10 s of OT silence on any
+active body. The bug that produced these prefixes can no longer happen.
+
+## OT track loss mid-episode (the remaining ~0.9 %)
+
+After trimming the prefixes, the rest of the `ot_loss` intervals in
+[`../bad_frames.json`](../bad_frames.json) are short (typically 1–3 s)
+mid-episode mocap dropouts. Per `freeze_diagnose`'s cross-modal motion
+check, the gel pad is still healthy and the cameras still capture — only
+the OptiTrack solver lost the rigid body, usually because the operator's
+hand moved near the edge of the mocap volume / camera FOV.
+
+The shipped `bad_frames.json` flags these as `ot_loss_L` / `ot_loss_R`
+and the example dataloader skips overlapping windows automatically when
+`skip_bad_frames=True`.
+
+**Mitigation for future recordings**: keep the workspace centred in the
+mocap volume; add a fourth or fifth marker to each GelSight rigid body
+so the solver tolerates more occlusion; add a fourth OT camera if budget
+allows. None of these affect the current dataset retroactively — they're
+just guidance for follow-up recording sessions.
+
 ## Other caveats
 
 - **Missing / dropped files** on `motherboard/2026-05-11`:
-  - `episode_000` and `episode_002` — short test recordings (8.8 s and 10.4 s) with no tactile contact on either sensor; intentionally excluded.
-  - `episode_001` — lost at recording time (HDF5 superblock never finalized when the writer was killed mid-write); intentionally absent.
-  - The remaining episode IDs are non-contiguous as a result. **Don't infer ordering from filename gaps.**
-- **Lossy resize**: the 128×128 `view` and tactile fields are downsampled from native 480×640. Native resolution is **not** preserved in this release.
-- **Single camera**: only `realsense/cam0/color` is included. The other two RealSense views and all depth streams are not in `processed/mode1_v1/`.
-- **OptiTrack alignment**: the per-step poses are nearest-neighbor matched to camera ticks. The full ~120 Hz pose streams are not preserved here.
-- **Mode is opinionated**: contact metrics depend on the chosen `tau` and the p01 reference strategy. If you want a different `tau`, re-deriving from raw is necessary.
+  - `episode_000` and `episode_002` — short test recordings (8.8 s and
+    10.4 s) with no tactile contact on either sensor; intentionally
+    excluded.
+  - `episode_001` — lost at recording time (HDF5 superblock never
+    finalized when the writer was killed mid-write); intentionally absent.
+  - The remaining episode IDs are non-contiguous as a result. **Don't
+    infer ordering from filename gaps.**
+- **Lossy resize**: the 128×128 `view` and tactile fields are downsampled
+  from native 480×640. Native resolution is **not** preserved in this
+  release.
+- **Single camera**: only `realsense/cam0/color` is included in the
+  `processed/.../*.pt` slice. The other two RealSense views (and depth
+  streams) are recorded in the raw H5 archive but not exposed in
+  `processed/mode1_v1/`.
+- **OptiTrack alignment**: the per-step poses in `.pt` are nearest-neighbour
+  matched to camera ticks. The full ~120 Hz pose streams are not
+  preserved here.
+- **Contact metrics are opinionated**: `tactile_{side}_{intensity, area, mixed}`
+  depend on the chosen `tau` and the `p01` reference strategy. If you
+  want a different `tau`, you'll need to re-derive them from raw GelSight
+  frames.
 
 ## Roadmap
 
 - **More tasks** — the registry in [`../tasks.json`](../tasks.json) will grow.
-- **LeRobot-format full-fidelity** variant (`lerobot/v1.0/`) is planned. It will include all three RealSense color and depth streams, GelSight at native resolution (FFV1 lossless), full-rate OptiTrack pose tracks for all three rigid bodies, and HF-native browser previews. The current `processed/mode1_v1/` slice will remain as a stable training-task view.
-
-## OptiTrack track loss (the dominant data-quality issue)
-
-The 47 `ot_loss` intervals in [`bad_frames.json`](../bad_frames.json) all
-come from the same root cause: **the GelSight rigid body moved near the
-edge of the OptiTrack mocap volume / camera FOV**, the solver lost the
-body, and the recorder kept emitting the last good pose. On re-acquire
-the solver sometimes flips orientation (per-sample angular velocity up to
-51,000 rad/s), which the cross-modal detector picks up as a teleport
-boundary.
-
-What's *not* broken during these intervals:
-
-- RealSense streams — all three cameras keep capturing fresh frames.
-- GelSight streams — the gel pad just isn't deforming (sensor moved in
-  free air or set down). Frame-duplicate rate is high but that's a
-  side-effect of low contact, not a pipeline stall.
-- The recorder itself — all writes land in the H5 on schedule.
-
-What is broken: `sensor_{side}_pose` is stale during the interval, so any
-training window overlapping it has an unreliable label. The shipped
-`bad_frames.json` flags these as `ot_loss_L` / `ot_loss_R` and the
-example dataloader skips overlapping windows automatically when
-`skip_bad_frames=True`.
-
-**Mitigation for future recordings**: keep the workspace centred in the
-mocap volume; add a fourth/fifth marker to each GelSight rigid body so
-the solver tolerates more occlusion; add a fourth OT camera if budget
-allows. None of these affect the current dataset retroactively.
+- **LeRobot-format full-fidelity** variant (`lerobot/v1.0/`) is planned.
+  It will include all three RealSense color and depth streams, GelSight
+  at native resolution (FFV1 lossless), full-rate OptiTrack pose tracks
+  for all three rigid bodies, and HF-native browser previews. The current
+  `processed/mode1_v1/` slice will remain as a stable training-task view.

docs/quality.md

@@ -1,39 +1,109 @@
 # Data quality
 
-A small fraction of frames contain known sensor artifacts. The repo ships a [`../bad_frames.json`](../bad_frames.json) index so downstream code can avoid sampling on top of these intervals — useful since this dataset is intended for **short-window dynamics / world-model learning**, where a glitch landing inside a training window can dominate the loss for that step.
+A small fraction of frames contain known sensor artifacts. The repo ships [`../bad_frames.json`](../bad_frames.json) as a skip-list so downstream code can avoid sampling on top of these intervals — important since this dataset is intended for **short-window dynamics / world-model learning**, where a glitch landing inside a training window can dominate the loss for that step.
+
+> **All frame indices in `bad_frames.json` are in TRIMMED `.pt` coordinates.** The original H5 recordings have not been edited; only the published `.pt` files have had their OT-uninitialized prefixes cut (see [`caveats.md`](caveats.md) §OT track loss). The per-file trim offset is stored in `_contact_meta.trim_offset` inside each `.pt`.
 
 ## Headline numbers
 
 | | |
 |---|---:|
-| Total synchronized frames | 190,231 (105.7 min @ 30 Hz, post-trim) |
+| Total synchronized frames (post-trim) | 190,231 (105.7 min @ 30 Hz) |
 | Recording files | 27 |
 | Frames flagged in `bad_frames.json` | **1,768 (0.929 %)** |
-| Recording files with ≥1 flagged frame | 11 |
+| Recording files with ≥ 1 flagged frame | 11 |
 
 ## Failure modes — quantitative breakdown
 
 | # | Mode | Frames | % of dataset | Files | Symptom | Example |
 |---|---|---:|---:|---:|---|---|
-| 1 | **GelSight LED flicker** | 66 | **0.030 %** | 5 / 27 | 1–2 frames of uniform pink/magenta wash across the gel surface; adjacent frames normal. 2026-05-11 / {003, 007, 008, 011, 017}. | [`intensity_spike_overview.png`](../figures/dataset_figures/intensity_spike_samples/intensity_spike_overview.png) |
-| 2 | **OptiTrack pose teleport (>5 m/s)** | 64 | **0.029 %** | 3 / 27 | Position jumps by >5 cm in 33 ms — mocap lost lock and reacquired. 2026-05-10/{001, 002}, 2026-05-11/015. | [`pose_teleport_samples/`](../figures/dataset_figures/pose_teleport_samples/) |
-| 3 | **OptiTrack track loss** | 33,056 | **14.92 %** | 8 / 27 | OptiTrack lost the rigid body — the recorder kept emitting the held pose. Boundary glitches (ang vel up to 51 k rad/s) appear on re-acquire. Worst single outage: 10.7 min (ep_017). Cause: marker near mocap-volume / camera-FOV boundary. GelSight + RealSense streams are *not* affected. | [`freeze_diagnose/`](../figures/dataset_figures/freeze_diagnose/) |
+| 1 | **GelSight LED flicker** | 56 | **0.029 %** | 5 / 27 | 1–2 frames of uniform pink/magenta wash across the gel surface; adjacent frames normal. | [`intensity_spike_overview.png`](../figures/dataset_figures/intensity_spike_samples/intensity_spike_overview.png) |
+| 2 | **OptiTrack pose teleport** | 56 | **0.029 %** | 3 / 27 | Translation velocity > 5 m/s **or** angular velocity > 15 rad/s in a single OT sample (solver flip — physically impossible for human-hand motion). | [`pose_teleport_samples/`](../figures/dataset_figures/pose_teleport_samples/) |
+| 3 | **OptiTrack track loss** | 1,680 | **0.883 %** | 6 / 27 | Sensor pose held at a stale value for ≥ 0.25 s; cross-modal motion check confirms hand was actually moving. Cause: marker briefly left the mocap volume / camera FOV. | [`freeze_diagnose/`](../figures/dataset_figures/freeze_diagnose/) |
+
+**Union of modes 1+2+3: 1,768 / 190,231 = 0.929 % of post-trim frames flagged.** Mode 3 is by far the largest — but it's already orders of magnitude smaller than what the *raw* recordings looked like (15 % of frames before the OT-uninitialized prefixes were trimmed). The remaining 0.88 % is genuine mid-episode mocap dropout that no amount of post-processing can recover.
+
+## How to use the data — three recipes
+
+### A. Easiest: the shipped example dataloader
+
+`ReactWindowDataset` does the right thing out of the box. Just turn `skip_bad_frames=True`:
+
+```python
+from examples.react_window_dataset import ReactWindowDataset
+from torch.utils.data import DataLoader
+
+ds = ReactWindowDataset(
+    data_root       = "processed/mode1_v1/motherboard",
+    bad_frames_path = "bad_frames.json",
+    tasks_json_path = "tasks.json",
+    window_length   = 16,
+    skip_bad_frames = True,    # ← drops windows touching ANY of modes 1/2/3
+    respect_active_sensors = True,
+)
+```
 
-**Modes 1+2+3 together: 33,144 / 221,621 = 14.96% of frames flagged.** Earlier note: 122 / 221621 = 0.055 % of frames are corrupted in any modality.** Mode 3 is tracked separately because it is healthy data, just non-informative for dynamics. `bad_frames.json` covers modes 1 and 2 only; if you also want to skip rest periods, intersect with the velocity track from `sensor_*_pose`.
+Any window whose [t_start, t_end] range overlaps an `intensity_spikes` /
+`pose_teleports_{L,R}` / `ot_loss_{L,R}` interval is silently dropped at
+window-enumeration time, so your DataLoader never sees them.
+
+### B. Loading a single `.pt` directly — DIY skip-list
+
+If you're rolling your own sampler or scanning a single file:
+
+```python
+import json, torch
+
+ep_key = "2026-05-11/episode_017"
+ep = torch.load(f"processed/mode1_v1/motherboard/{ep_key}.pt", weights_only=False)
+T = ep["view"].shape[0]
+trim_offset = ep["_contact_meta"].get("trim_offset", 0)   # already applied to ep
+bad = json.load(open("bad_frames.json"))["episodes"][ep_key]
+
+# Build a per-frame boolean mask (True = drop)
+import numpy as np
+mask = np.zeros(T, dtype=bool)
+for s_, e_ in (bad["intensity_spikes"]
+               + bad["pose_teleports_L"] + bad["pose_teleports_R"]
+               + bad["ot_loss_L"] + bad["ot_loss_R"]):
+    mask[s_:e_ + 1] = True
+print(f"{mask.sum()}/{T} flagged ({100 * mask.mean():.2f} %)")
+
+# Then sample only clean windows:
+def clean_window_start(t_start, win_len):
+    return not mask[t_start:t_start + win_len].any()
+```
+
+### C. Only care about action labels (pose)? Skip just ot_loss
+
+If your model treats vision + tactile as observations and sensor pose as the *action* label (UMI-style imitation learning), the LED flicker and translation teleports are observation-side noise that you can usually tolerate (single frames or short < 0.3 s spikes). The OT track losses are the one mode you *must* exclude, because the recorded "action" is stale:
+
+```python
+# Just the action-label cuts
+def clean_pose_window(t_start, win_len, bad_ep):
+    for s_, e_ in (bad_ep["ot_loss_L"] + bad_ep["ot_loss_R"]
+                   + bad_ep["pose_teleports_L"] + bad_ep["pose_teleports_R"]):
+        if s_ <= t_start + win_len - 1 and e_ >= t_start:
+            return False
+    return True
+```
+
+### What NOT to do
+
+- Don't ignore `bad_frames.json` — for `ot_loss` intervals the *recorded pose looks plausible* (it's just a stale held value), but the actual sensor was moving. A model trained on those will learn that contact-rich tactile signals are uncorrelated with motion.
+- Don't try to skip individual *frames* and stitch the rest — windows must be contiguous over time. Drop the **whole window**, not just the bad frames in it.
 
 ## Inspection figures
 
-**Mode 1 — GelSight LED flicker** (overview across all 5 affected files):
+**Mode 1 — GelSight LED flicker** (overview across all affected files):
 
 ![Intensity spike overview](../figures/dataset_figures/intensity_spike_samples/intensity_spike_overview.png)
 
-Per-file close-ups (reference frame, ±1 s neighbors, peak frame): [`figures/dataset_figures/intensity_spike_samples/`](../figures/dataset_figures/intensity_spike_samples/).
+Per-file close-ups: [`figures/dataset_figures/intensity_spike_samples/`](../figures/dataset_figures/intensity_spike_samples/).
 
-**Mode 2 — OptiTrack pose teleport.** GIFs (10 s playback at 2× speed) for all 3 affected files: [`pose_teleport_samples/`](../figures/dataset_figures/pose_teleport_samples/).
+**Mode 2 — OptiTrack pose teleport.** GIFs (10 s playback at 2× speed) for all affected files: [`pose_teleport_samples/`](../figures/dataset_figures/pose_teleport_samples/).
 
-**Modes 1+2+3 — top-6 worst-offenders chart** (tactile intensity + sensor velocity time-series):
-
-![Data quality report](../figures/dataset_figures/data_quality_report.png)
+**Mode 3 — OptiTrack track loss.** 15 sample clips (recording-viewer layout, 30 fps real time, red ring on the frozen-sensor projected dot): [`freeze_diagnose/ot_loss/`](../figures/dataset_figures/freeze_diagnose/ot_loss/).
 
 ## `bad_frames.json` schema
 
@@ -43,21 +113,26 @@ Frame indices are inclusive on both ends and pre-padded by `buffer_frames` (3) o
 {
   "tau_intensity": 30.0,
   "tau_velocity_mps": 5.0,
+  "tau_angular_rad_per_s": 15.0,
+  "tau_opt_gap_s": 0.10,
+  "freeze_threshold_s": 0.25,
   "buffer_frames": 3,
   "summary": {
     "n_episodes": 27,
-    "total_frames": 221621,
-    "total_bad_frames": 122,
-    "bad_fraction_overall": 0.00055,
-    "n_episodes_with_bad_frames": 8
+    "total_frames": 190231,
+    "total_bad_frames": 1768,
+    "bad_fraction_overall": 0.009294,
+    "n_episodes_with_bad_frames": 11
   },
   "episodes": {
     "2026-05-11/episode_003": {
       "n_frames": 10032,
-      "duration_s": 338.2,
+      "duration_s": 334.4,
       "intensity_spikes": [[260, 268], "..."],
       "pose_teleports_L": [],
       "pose_teleports_R": [],
+      "ot_loss_L": [],
+      "ot_loss_R": [],
       "total_bad_frames": 19,
       "bad_fraction": 0.0019
     }
@@ -67,37 +142,6 @@ Frame indices are inclusive on both ends and pre-padded by `buffer_frames` (3) o
 
 A per-mode aggregate is also available: [`figures/dataset_figures/data_quality_breakdown.json`](../figures/dataset_figures/data_quality_breakdown.json).
 
-## Use in a dataloader
-
-```python
-import json
-from pathlib import Path
-
-with open("bad_frames.json") as f:
-    bad = json.load(f)["episodes"]
-with open("tasks.json") as f:
-    sessions = json.load(f)["tasks"]["motherboard"]["per_date_notes"]
-
-def is_clean_window(ep_name, t_start, t_end):
-    """Return True if [t_start, t_end] does not overlap any flagged interval."""
-    intervals = (bad[ep_name]["intensity_spikes"]
-                 + bad[ep_name]["pose_teleports_L"]
-                 + bad[ep_name]["pose_teleports_R"])
-    for s, e in intervals:
-        if s <= t_end and e >= t_start:
-            return False
-    return True
-
-def active_sensors(ep_name):
-    """Returns e.g. ['right'] or ['left', 'right']."""
-    date = ep_name.split("/")[0]
-    return sessions[date]["active_sensors"]
-
-# In your sampler:
-#   if not is_clean_window(...): resample
-#   sides = active_sensors(...)   # mask out inactive tactile + pose modalities
-```
-
 ## Full report (per-file CSV)
 
 Every file's individual stats (frames, duration, contact %, max intensity left/right, drift, max pose velocity, etc.) are in [`figures/dataset_figures/data_quality_report.csv`](../figures/dataset_figures/data_quality_report.csv).