1.3.0 documentation update: camera-ready benchmark text, runnable quick start, LICENSE.md, Croissant fixes

CHANGELOG.md

@@ -2,6 +2,20 @@
 
 Finding identifiers in the 1.2 entries are (audit dimension, id) pairs of the ForceBody v1.2 audit; the same id can occur in two dimensions, so both are given.
 
+## 1.3.0 documentation update (2026-11-01)
+
+Documentation only. No data file changed: the trial files, models, manifest, split, persons, subjects, schema and map files are byte-identical to the 1.3.0 freeze (2026-10-29).
+
+- License decided: the data are released under CC BY-NC 4.0, SKEL-derived fields also follow the SKEL model license, and the code on GitHub is released under the MIT license. This replaces the open license decision recorded in the 1.3.0 README entry below.
+- README.md is now a short Hugging Face dataset card (overview, quick start, main fields, files, license, citation). The full documentation moved to DETAILS.md. The README sections named in earlier entries (Known issues, SKEL fit versions and fix rounds) are now in DETAILS.md.
+- Dataset card: the quick start runs as written (install of `huggingface_hub`, data path `../../ForceBody` from the `code/` folder of the GitHub repository, mask `mc_frame_ok` combined with `torque_label_usable`, joint-torque columns 6-36), step 0 states that the joint positions are stored and the SKEL model is needed only for meshes, the contact channels, frame-mask shapes and root-residual columns are defined, the files table is complete and links croissant.json and the AddBiomechanics site.
+- DETAILS.md "Paper numbers" and "Force benchmark": the camera-ready paper reports the baselines retrained on 1.2.1 (benchmark data identical in 1.3.0) as mean and standard deviation over three seeds, with checkpoints selected on the 12 validation persons and the 19 test persons scored once. The reviewed build 1.0.0 gave the numbers of the submitted paper.
+- DETAILS.md "How to cite": the ForceBody entry (`kwon2026forcebody`) replaces the placeholder, the two AddBiomechanics entries have their full author lists, and a reference list of the 10 source studies was added.
+- DETAILS.md, other corrections: SCHEMA_v1.3.csv named in "Quality masks", `dof_names` / `dof_names_opensim` described as `<OpenSim joint>_<coordinate index>` names with their OpenSim coordinates, the native-rate trials counted (24 Han2023 and 1 Carter2023 trial at 250 Hz, 111 Tiziana2019 trials at 201 Hz), the code location given as the `code/` folder of the GitHub repository, the requirements of `visualize.py` (public SKEL package and SKEL model files, the addb2skel package is no longer needed), and the layout tree rooted at `ForceBody/` with DETAILS.md, MD5SUMS.txt and croissant.json.
+- croissant.json added: Croissant metadata with Responsible AI fields.
+- LICENSE.md added: license and attribution terms (CC BY-NC 4.0 data, SKEL model license, AddBiomechanics CC BY 4.0, source studies, MIT code).
+- MD5SUMS.txt: the lines of README.md, DETAILS.md and CHANGELOG.md updated to the new files.
+
 ## 1.3.0 (2026-10-29, frozen)
 
 Final release audit of 1.2.1. Pose, force, CoP, torque, MC, mask and QC arrays are unchanged; the benchmark inputs, labels, masks and splits are identical to 1.2.1. MD5 of the manifest and split files at the freeze:
@@ -22,7 +36,7 @@ Changes:
 - New manifest column `dataset_version` ("1.3.0"). The trial-file field `release_version` stays "v1.2" (trial file format); only 265 trial files were rewritten (40 Hamner2013 + 225 Tiziana2019), the other files are byte-identical to 1.2.1.
 - SCHEMA_v1.2.csv renamed to SCHEMA_v1.3.csv: rows for `trial_uid_v12` and `dataset_version`; descriptions of `sex_resolved`, `trial_uid`, `study`, `release_version`, `trial_uid_m11`, `trial_uid_v11`, `trial_uid_v12_proposed` updated.
 - benchmark/BENCHMARK.md: split table filled from the cache meta files (train 109 persons / 8,374 trials, val 12 / 1,098, test 19 / 914) and the baselines retrained on 1.2.1 (T-tau: MLP, Conv1D, LSTM, GRU, Mamba2; T-GRF: MLP); Transformer, Transformer + UW and GLN are marked pending. spec_v12.json: `dataset_version` and decision D25 text; `spec_version` stays `forcebody-bench-1.2.0`, so 1.2.1 result files stay `official_table_eligible`.
-- README: Hugging Face dataset card header with the license marked TODO (decision open); dataset totals (9,689,808 frames, 26.89 h, 75 male / 55 female / 10 unknown persons); sigma statistics set of the paper (`has_uncertainty` and `grf_valid_frac` >= 0.5, 8,732 trials); `name_map` `split` column is the AddBiomechanics source split; Hugging Face splits; version table.
+- README: Hugging Face dataset card header with the license marked TODO (decision open at the freeze, decided on 2026-11-01, see "1.3.0 documentation update" above); dataset totals (9,689,808 frames, 26.89 h, 75 male / 55 female / 10 unknown persons); sigma statistics set of the paper (`has_uncertainty` and `grf_valid_frac` >= 0.5, 8,732 trials); `name_map` `split` column is the AddBiomechanics source split; Hugging Face splits; version table.
 - `__pycache__` folders removed from the release.
 
 Release audit of 1.2.1 (all 10,386 trial files, 24 workers): every file has the 547 core fields of the schema with the schema dtype and shape; no NaN or inf in poses, trans, joints, q_kin, q_gt, or in tau_selfconsistent / grf / grf_aligned / mu_tau / sigma_tau inside their frame masks (`markers` / `markers_aligned` carry NaN for missing markers, as documented); all frame masks have length T; `eval_*_frame_ok` definitions hold; `grf_foot_ok.all(1)` equals `grf_frame_strict_ok`; betas identical within each of the 140 persons; no duplicate `trial_uid`; 457 scalar manifest columns agree with the trial files; no person in two folds of `split_addb`, `split_addb_val` or `split_t2m`; MD5SUMS.txt of 1.2.1 matched all files. The same audit passes on 1.3.0.
@@ -117,5 +131,5 @@ Internal build after review (dynamics regenerated on the SKEL-fit timebase, corr
 
 ## 1.0.0-reviewed (2026-05-07)
 
-The build reviewed with the paper; the paper numbers refer to it.
+The build reviewed with the paper; the numbers of the submitted paper refer to it (the camera-ready numbers use 1.2.1, see "1.3.0 documentation update").
 

DETAILS.md

@@ -1,7 +1,7 @@
 
 # ForceBody: dataset details
 
-Release 1.3.0 (2026-10-29).
+Release 1.3.0 (2026-10-29). The documentation was updated on 2026-11-01 (CHANGELOG.md). No data file changed.
 
 ForceBody pairs the SKEL body model with measured ground reaction forces (GRF), inverse-dynamics joint torques and Monte Carlo torque uncertainty for 10,386 motion trials (9,689,808 frames, 26.89 h; duration = `num_frames` / `fps`) of 140 persons (75 male, 55 female, 10 sex not reported) from 10 studies (AddBiomechanics). Every trial is one NumPy `.npz` file; no nimblephysics, OpenSim or AddBiomechanics tooling is needed to read it.
 
@@ -27,8 +27,10 @@ Study ids follow the first author and year of the source study (`study_citation`
 
 This Hugging Face repository holds the data only. The loader (`forcebody_io.py`), the renderer (`visualize.py`, `VISUALIZATION.md`), the benchmark card, cache builder and scorer (`benchmark/`) and the baseline code are in the GitHub repository https://github.com/DominoAI-Lab/ForceBody-NeurIPS-ED-2026. The v1.1 provenance files and the Monte Carlo torque percentiles are not part of this upload.
 
+Downloaded as in the Quick start of the dataset card (`--local-dir ForceBody`), the folder holds:
+
 ```
-ForceBody_release_v1.3/
+ForceBody/
   trials/<study>/S<NN>/<trial_uid>.npz        core arrays, 547 fields (SCHEMA_v1.3.csv)
   manifest.csv, manifest_train.csv, manifest_test.csv   one row per trial (every scalar field + metadata)
   uid_map_v12_to_v13.csv                     the 40 trial keys and paths renamed in 1.3.0 (Hammer2013 -> Hamner2013)
@@ -37,10 +39,17 @@ ForceBody_release_v1.3/
   subjects.csv                               per-person metadata table (v1.1 layout, v1.2 values)
   gait_events.csv                            GRF heel strikes and toe-offs (walking / running)
   models/                                    AddB pass-0 and pass-2 scaled OpenSim models per b3d file (.osim.gz),
+                                              models_index.csv (model file -> source b3d file and processing pass),
                                               body_segments.csv (segment masses, local COM)
   marker_vertex_map.json                     marker name -> SKEL skin vertex used by the fit
   SCHEMA_v1.3.csv                            field, dtype, shape, unit, frame, version_added, description
-  README.md, CHANGELOG.md
+  README.md                                  dataset card
+  DETAILS.md                                 this file
+  CHANGELOG.md                               version history
+  MD5SUMS.txt                                md5 of the data files and of README.md, DETAILS.md, CHANGELOG.md
+  croissant.json                             Croissant metadata with Responsible AI fields
+  LICENSE.md                                 license and attribution terms
+  assets/                                    images of the dataset card
 ```
 
 ## Choosing a pose source
@@ -108,7 +117,7 @@ Rule (an allowlist): a model that predicts or scores forces or torques may take
 
 ## Force benchmark
 
-The force benchmark (joint torques T-tau and per-foot GRF T-GRF from motion) is defined in `code/benchmark/BENCHMARK.md` of the GitHub repository https://github.com/DominoAI-Lab/ForceBody-NeurIPS-ED-2026 with the machine-readable `benchmark/spec_v12.json`, a cache builder (`benchmark/build_bench_cache_v12.py`) and the reference scorer (`benchmark/score_v12.py`). The scripts import `forcebody_io.py` from the release root. BENCHMARK.md lists the split sizes and the baselines retrained on 1.2.1 (5 torque models and the GRF MLP; 3 torque models pending). The benchmark data of 1.3.0 are identical to 1.2.1 except the 40 renamed Hamner2013 keys.
+The force benchmark (joint torques T-tau and per-foot GRF T-GRF from motion) is defined in `code/benchmark/BENCHMARK.md` of the GitHub repository https://github.com/DominoAI-Lab/ForceBody-NeurIPS-ED-2026 with the machine-readable `benchmark/spec_v12.json`, a cache builder (`benchmark/build_bench_cache_v12.py`) and the reference scorer (`benchmark/score_v12.py`). The scripts import `forcebody_io.py` from the `code/` folder of the GitHub repository. The baselines retrained on 1.2.1 (three seeds, the numbers of the camera-ready paper, see Paper numbers) are in the Benchmark section of the GitHub README. The benchmark data of 1.3.0 are identical to 1.2.1 except the 40 renamed Hamner2013 keys.
 
 ## Conventions
 
@@ -119,11 +128,24 @@ The force benchmark (joint torques T-tau and per-foot GRF T-GRF from motion) is
 | contacts | index 0 = right foot (calcn_r), 1 = left foot (calcn_l), 2 = pelvis (Falisse2017 chair) |
 | trans | SKEL root translation (`trans_frame == "skel_world"`). The pelvis joint is `trans + j0(betas)`, not `trans`; j0 is about (0.0095, -0.236, 0.103) m for a male rest pose, so rotating `trans` alone misplaces the body by 0.12-0.16 m |
 | joints | `joints[T,24,3]` in `joint_names` order (SKEL / OpenSim: 0 pelvis, 1 femur_r, ...) |
-| DOF names | `skel_pose_names` (46 SKEL pose DOFs; `skel_dof_valid` False = pinned to 0 in the fit), `dof_names_opensim` (37 AddB DOFs, OpenSim Rajagopal coordinate names and sign conventions) |
+| DOF names | `skel_pose_names` (46 SKEL pose DOFs; `skel_dof_valid` False = pinned to 0 in the fit), `dof_names` = `dof_names_opensim` (37 AddB DOFs, `<OpenSim joint>_<coordinate index>` names, see below) |
 | time | `t_i = i / fps_true_grid`. `fps` is the nominal rate; the true grid differs slightly (61 Hz: 64, 99 Hz: 470, 100 Hz: 9,640, 101 Hz: 76, 201 Hz: 111, 250 Hz: 25 trials by rounded rate) |
 | heading | use `heading_canonical_deg` (SKEL frame) only |
 | mirroring | swapping left/right SKEL DOFs with a sign table is not an exact mirror (MPJPE 32 mm, feet 45-77 mm); mirror the mesh or the joint positions instead |
 
+AddB DOF order. `dof_names` and `dof_names_opensim` hold the same 37 names. They label the columns of `q_*`, `tau_*`, `mu_tau` and `sigma_tau`. The names are `<OpenSim joint>_<coordinate index>` (for example `hip_r_0`, `walker_knee_r_0`), not OpenSim coordinate names. Index k is the k-th coordinate of that joint in the AddB OpenSim model (`models/`, the Rajagopal coordinate set):
+
+| columns | names | OpenSim coordinates |
+|---|---|---|
+| 0-5 | `ground_pelvis_0`-`5` | pelvis_tilt, pelvis_list, pelvis_rotation, pelvis_tx, pelvis_ty, pelvis_tz |
+| 6-12 | `hip_r_0`-`2`, `walker_knee_r_0`, `ankle_r_0`, `subtalar_r_0`, `mtp_r_0` | hip_flexion_r, hip_adduction_r, hip_rotation_r, knee_angle_r, ankle_angle_r, subtalar_angle_r, mtp_angle_r |
+| 13-19 | same for the left leg (`_l`) | same coordinates, `_l` |
+| 20-22 | `back_0`-`2` | lumbar_extension, lumbar_bending, lumbar_rotation |
+| 23-29 | `acromial_r_0`-`2`, `elbow_r_0`, `radioulnar_r_0`, `radius_hand_r_0`-`1` | arm_flex_r, arm_add_r, arm_rot_r, elbow_flex_r, pro_sup_r, wrist_flex_r, wrist_dev_r |
+| 30-36 | same for the left arm (`_l`) | same coordinates, `_l` |
+
+In `tau_selfconsistent`, `tau_gt_det`, `mu_tau` and `sigma_tau`, columns 0-2 are root residual moments (N m) and columns 3-5 are root residual forces (N). Columns 6-36 are joint torques (N m).
+
 ## Body mass fields
 
 | field | meaning | use |
@@ -164,7 +186,7 @@ Trial flags:
 
 Evaluation flags. `eval_grf_trial_ok` and `eval_torque_trial_ok` require usable force labels and exclude single-channel force plates (`qc_grf_single_channel`: one plate carries both feet, so per-foot GRF and the foot assignment of the torques are wrong; also excluded from `torque_label_usable`), byte-identical duplicates (`qc_duplicate_drop`), the Han2023 trimmed copies of a re-exported capture (`qc_overlap_pair_drop` = `eval_exclude_duplicate` and not `qc_duplicate_drop`), static calibration, transition and pause segments, and trials whose dynamics mass disagrees with the GRF-implied mass by more than 5% (`qc_mass_dyn_vs_grf_flag`); `eval_torque_trial_ok` also requires a root residual force of at most 0.1 BW. Since 1.2.1 both also require at least one loaded evaluation frame (`qc_grf_loaded_eval_frames` > 0: a foot vertical GRF above 20 N on an `eval_grf_frame_ok` frame; for torques also `qc_tau_loaded_eval_frames` > 0); 83 trials with usable force labels have none (most have `grf_valid_mask` False on every frame). `eval_motion_trial_ok` applies the same duplicate / segment exclusions and in addition drops one member of each overlapping Tiziana2019 capture pair (`qc_overlap_motion_drop`: the member with fewer GRF-valid frames). The Tiziana2019 pairs share no GRF-valid frame, so both members stay in force evaluation. Always combine a trial flag with its frame mask.
 
-`force_label_usable` is False where force and motion capture are out of sync and could not be re-timed (14 trials); every force, torque, MC, contact and CoP frame mask is then False. `qc_skel_grf_newton_mismatch` marks trials whose SKEL root path does not follow the GRF without any time-lag evidence; their force labels are kept. Pathological persons are never excluded; `eval_stratum` reports them separately. Every QC field is described in SCHEMA_v1.2.csv.
+`force_label_usable` is False where force and motion capture are out of sync and could not be re-timed (14 trials); every force, torque, MC, contact and CoP frame mask is then False. `qc_skel_grf_newton_mismatch` marks trials whose SKEL root path does not follow the GRF without any time-lag evidence; their force labels are kept. Pathological persons are never excluded; `eval_stratum` reports them separately. Every QC field is described in SCHEMA_v1.3.csv.
 
 Falisse2017 chair channel (contact index 2, `contact_bodies` = `pelvis`). `chair_contact` [T] marks the frames where the subject sits on the instrumented chair (the `grf_contact_label` hysteresis on the chair vertical force, not restricted to `grf_frame_ok`; most seated frames are outside `grf_valid_mask`), `qc_chair_channel_active` marks trials with any such frame, and `qc_extra_channel_zero` marks trials whose chair channel carries no force at all (walk, stairs). On those trials `grf_contact_ok[:, 2]` is False on every frame and `grf_frame_ok` ignores the channel. The trial-level manifest column `chair_contact` (protocol metadata) is a different field from the per-frame npz array of the same name.
 
@@ -212,7 +234,7 @@ Added in 1.2.1 (T2M audit; flags and masks only, no pose refit):
 
 ## Uncertainty
 
-`mu_tau` and `sigma_tau` ship in the core file for all 10,386 trials, the torque `percentiles` (with `percentile_values`) are kept in the internal v1.1 provenance archive and are not part of this upload; `has_uncertainty` is True for 10,382 (False: the trial is shorter than one period of the 6 Hz noise band, sigma is 0). `has_uncertainty_v11` (manifest) marks the 8,652 trials that had MC in the paper version. sigma_tau is the kinematic sensitivity of the torque to 1.6 deg band-limited joint-angle noise on `q_gt` (pass 2; `q_kin` on re-timed trials) with the measured GRF held fixed. It is not the uncertainty of `poses`, and not a label-reliability score. Normalize it by `mass_kg_dynamics` only, and use it only where `mc_frame_ok` and `torque_label_usable` hold. Per-trial ratios to the paper-version sigma have p5-p95 0.85-1.20. The sigma statistics of the paper use the trials with `has_uncertainty` and `grf_valid_frac` >= 0.5 (8,732 trials).
+`mu_tau` and `sigma_tau` ship in the core file for all 10,386 trials, the torque `percentiles` (with `percentile_values`) are kept in the internal v1.1 provenance archive and are not part of this upload; `has_uncertainty` is True for 10,382 (False: the trial is shorter than one period of the 6 Hz noise band, sigma is 0). `has_uncertainty_v11` (manifest) marks the 8,652 trials that had MC in the version of the submitted paper. sigma_tau is the kinematic sensitivity of the torque to 1.6 deg band-limited joint-angle noise on `q_gt` (pass 2; `q_kin` on re-timed trials) with the measured GRF held fixed. It is not the uncertainty of `poses`, and not a label-reliability score. Normalize it by `mass_kg_dynamics` only, and use it only where `mc_frame_ok` and `torque_label_usable` hold. Per-trial ratios to the sigma of the submitted-paper version have p5-p95 0.85-1.20. The sigma statistics of the paper use the trials with `has_uncertainty` and `grf_valid_frac` >= 0.5 (8,732 trials).
 
 ## Time alignment
 
@@ -285,7 +307,7 @@ Intended: research on motion-to-force and force-aware motion models, text-to-mot
 
 ## Paper numbers
 
-The paper tables (Tables 3, 4, 5, 14 and 15) were computed on the reviewed build with the v1 labels and are not reproduced by v1.1, 1.2 or 1.3; benchmark baselines retrained on 1.2.1 are in benchmark/BENCHMARK.md. The Table 3 AddB row used dynamics-pass (pass 2) kinematics as input (see Leakage policy). The torque labels themselves moved between versions: scoring the v1.1 `tau_selfconsistent` against the paper labels gives RMSE / MAE / PCC 0.287 / 0.087 / 0.721, the same order as the model errors, so numbers from different versions must not be mixed in one table.
+The benchmark numbers of the camera-ready paper (the ForceBody rows of Table 3, and Tables 4, 5, 14 and 15) are the mean and standard deviation over three seeds of the baselines retrained on release 1.2.1. The benchmark inputs, labels, masks and splits of 1.2.1 are identical in 1.3.0. Checkpoints were selected on the 12 validation persons (`split_addb_val == "val"`, 1,098 trials), and the 19 test persons (914 trials, healthy and pathological strata) were scored once. The Benchmark section of the GitHub README gives Tables 3, 4 and 5. The AddB rows of Table 3 are copied from the reviewed build and serve as a reference, not as a matched comparison. They used dynamics-pass (pass 2) kinematics as input (see Leakage policy). The submitted version of the paper used the reviewed build 1.0.0 with the v1 labels. The torque labels moved between versions. Scoring the v1.1 `tau_selfconsistent` against the reviewed-build labels gives RMSE / MAE / PCC 0.287 / 0.087 / 0.721, the same order as the model errors, so numbers from different versions must not be mixed in one table.
 
 ## Known issues
 
@@ -304,8 +326,8 @@ The paper tables (Tables 3, 4, 5, 14 and 15) were computed on the reviewed build
 - Some dyn-pass arm events listed by the 2026-10-07 diagnosis refer to the dyn fit before the 2026-10-08/09 target repair; they are kept in `arm_event_ok_dyn` (conservative), together with the events of the re-scan of the final fits.
 - Carter2023 running: in some trials AddB marks the flight frames as missing GRF (`grf_valid_mask` False), so force labels and the T-GRF benchmark cover stance frames only there (most trials keep the flight frames valid with zero force).
 - `floor_height_m` is the loaded step height on stairs / step activities; no floor normal is shipped.
-- `visualize.py` needs the addb2skel package (SKEL rendering utilities), which is not distributed with ForceBody; reading the data needs only numpy.
-- Release verification left three small deviations: in one trial the fit joint error recomputed from the stored float32 `joints` / `addb_joints_kin` differs from the stored `qc_skel_fit_err_mm` by 0.054 mm (tolerance 0.05 mm, so one of 10,386 trials fails the independent reproduction check); the median lag between GRF and kinematic toe-off events is -74.5 ms for Uhlrich2023 (all other studies -50 to 0 ms), just outside the -70 to +10 ms window used as a sanity range.
+- `visualize.py` (GitHub repository, `code/`) needs the public SKEL package, the SKEL model files and an EGL-capable renderer (see `code/VISUALIZATION.md`). Reading the data needs only numpy.
+- Release verification left small deviations: in one trial the fit joint error recomputed from the stored float32 `joints` / `addb_joints_kin` differs from the stored `qc_skel_fit_err_mm` by 0.054 mm (tolerance 0.05 mm, so one of 10,386 trials fails the independent reproduction check); the median lag between GRF and kinematic toe-off events is -74.5 ms for Uhlrich2023 (all other studies -50 to 0 ms), just outside the -70 to +10 ms window used as a sanity range.
 - No acceleration filter was applied to the kinematics-pass target: `poses` keep a small frame-to-frame wobble on trials whose AddB pass-0 inverse kinematics is noisy. The dynamics-pass target is smoother (AddB dynamics pass; repaired targets use the AddB acceleration-minimizing filter).
 - Name clash: `chair_contact` in manifest.csv is the trial-level protocol flag (a chair is used in the trial), while `chair_contact` inside the trial file is the frame-level [T] seated mask of the Falisse2017 chair channel (all False without a chair channel). Use `qc_chair_channel_active` for the trial-level summary of the frame mask.
 - CoP placement at low load: frames where a foot channel is loaded (vertical force above 5% BW) and its CoP lies more than 0.15 m from both feet are not all flagged. A scan of the release found 3,227 GRF-valid such frames in 664 trials with `cop_frame_ok` True and `cop_off_foot` False; mask CoP losses by load if this matters.
@@ -315,50 +337,86 @@ The paper tables (Tables 3, 4, 5, 14 and 15) were computed on the reviewed build
 - Stairs (22 trials in activity class stairs): no per-step floor is shipped. `floor_height_m` is one loaded-step height per trial, and one fitted floor plane per trial (as in the T2M encoders) treats a staircase as a ramp, so heights above that floor are not step-relative (T2M audit R15).
 - Moore2015 standing (29 trials): in the wide-stance phases the feet sit 2-4 cm below the trial median floor level on about 40% of frames; whether this is real penetration or a floor estimate set too high is not resolved (T2M audit R18).
 - Pathological persons (Fregly2012 TKA 78 trials, Li2021 stroke 65 trials): the `caption_*` phrases never mention the pathology, and the Li2021_S05 fits have about 1.8 times the healthy fit error (standing 26.6 mm vs 14.7 mm) (T2M audit R21). Use them as an out-of-distribution stratum (`eval_stratum`).
-- Trials above 100 Hz (Han2023 and some Carter2023 at 250 Hz, Tiziana2019 at 201 Hz) ship on their native grid. Resampling them to 100 Hz without a low-pass filter (as the T2M encoders do) aliases force-plate noise (Han2023: 7-30 N RMS below 50 Hz; peaks are kept) (T2M audit R22).
+- Trials above 100 Hz (24 Han2023 trials and 1 Carter2023 trial at 250 Hz, 111 Tiziana2019 trials at 201 Hz, rounded `fps_true_grid`) ship on their native grid. Resampling them to 100 Hz without a low-pass filter (as the T2M encoders do) aliases force-plate noise (Han2023: 7-30 N RMS below 50 Hz; peaks are kept) (T2M audit R22).
 - vanderZee2022 walking: the median peak hip flexion torque is 1.95 N m/kg, against 1.0 (Carter2023) and 1.36 (Moore2015); the cause is not identified (possibly the plate cross-talk above) (T2M audit R23).
 - `q_kin` beyond 2 pi on a shoulder DOF in a few trials (`qc_q_kin_wrap`, T2M audit R12): an AddB Euler branch of the pass-0 target, not unwrapped; the arm frames are masked in `arm_event_ok_kin` (1.2.1). The mask covers only the frames beyond 2 pi (+-0.1 s); in some of these trials the wrapped stretch is wider (an arm DOF of `q_kin` more than pi/2 from `q_gt` outside the mask), so skip arm losses on the whole trial if in doubt.
 - Constant-labelled treadmill belts that change speed (`qc_belt_speed_change`, `_suspected`; T2M audit R16) and segments whose label does not match the motion (`label_conflict_codes` SEGMENT_MOTION_MISMATCH; R17) are flagged in 1.2.1; trials and labels are unchanged.
 
 ## License and attribution
 
-The ForceBody data are released under the Creative Commons Attribution-NonCommercial 4.0 International license (CC BY-NC 4.0). Commercial use is not permitted.
+The ForceBody data are released under the Creative Commons Attribution-NonCommercial 4.0 International license (CC BY-NC 4.0). Commercial use is not permitted. LICENSE.md holds the full terms.
 
 - Fields derived from AddBiomechanics (forces, AddB kinematics, torques, uncertainty, metadata) build on the AddBiomechanics dataset (CC BY 4.0). Cite AddBiomechanics and every source study you use.
 - SKEL-derived fields (`poses*`, `trans*`, `joints*`, `betas`, `skel_gender`) are also subject to the SKEL model license, which allows non-commercial scientific research only. Building meshes from these fields requires the SKEL model, which you obtain from its authors under their license.
-- The code in this release (`forcebody_io.py`, `visualize.py`, `benchmark/`) and in the GitHub repository is released under the MIT license.
+- The code in the GitHub repository (`forcebody_io.py`, `visualize.py`, `benchmark/`, `baselines/` in `code/`) is released under the MIT license. This Hugging Face repository holds no code.
 
 ## How to cite
 
 ```bibtex
-@inproceedings{forcebody,
-  title = {ForceBody},
-  note = {citation to be completed at publication}
+@inproceedings{kwon2026forcebody,
+  title     = {ForceBody: Force-Paired Parametric Body Motion with Torque Uncertainty},
+  author    = {Kwon, Joonwoo and Zhang, Yufei and Liu, Xiaoming and Cui, Zijun},
+  booktitle = {Advances in Neural Information Processing Systems (NeurIPS), Evaluations and Datasets Track},
+  year      = {2026}
 }
 @inproceedings{werling2024addbiomechanics,
-  title = {AddBiomechanics Dataset: Capturing the Physics of Human Motion at Scale},
-  booktitle = {ECCV},
-  year = {2024},
-  doi = {10.1007/978-3-031-73223-2_27}
+  title     = {AddBiomechanics Dataset: Capturing the Physics of Human Motion at Scale},
+  author    = {Werling, Keenon and Kaneda, Janelle and Tan, Tian and Agarwal, Rishi and Skov, Six and Van Wouwe, Tom and Uhlrich, Scott and Bianco, Nicholas and Ong, Carmichael and Falisse, Antoine and Sapkota, Shardul and Chandra, Aidan and Carter, Joshua and Preatoni, Ezio and Fregly, Benjamin and Hicks, Jennifer and Delp, Scott and Liu, C. Karen},
+  booktitle = {European Conference on Computer Vision (ECCV 2024)},
+  series    = {Lecture Notes in Computer Science},
+  pages     = {490--508},
+  publisher = {Springer},
+  year      = {2024},
+  doi       = {10.1007/978-3-031-73223-2_27}
 }
 @article{werling2023addbiomechanics,
-  title = {AddBiomechanics: Automating model scaling, inverse kinematics, and inverse dynamics from human motion data through sequential optimization},
+  title   = {AddBiomechanics: Automating model scaling, inverse kinematics, and inverse dynamics from human motion data through sequential optimization},
+  author  = {Werling, Keenon and Bianco, Nicholas A. and Raitor, Michael and Stingel, Jon and Hicks, Jennifer L. and Collins, Steven H. and Delp, Scott L. and Liu, C. Karen},
   journal = {PLOS ONE},
-  year = {2023},
-  doi = {10.1371/journal.pone.0295152}
+  volume  = {18},
+  number  = {11},
+  pages   = {e0295152},
+  year    = {2023},
+  doi     = {10.1371/journal.pone.0295152}
+}
+
+@article{keller2023skin,
+  title   = {From Skin to Skeleton: Towards Biomechanically Accurate {3D} Digital Humans},
+  author  = {Keller, Marilyn and Werling, Keenon and Shin, Soyong and Delp, Scott and Pujades, Sergi and Liu, C. Karen and Black, Michael J.},
+  journal = {ACM Transactions on Graphics},
+  volume  = {42},
+  number  = {6},
+  pages   = {1--12},
+  year    = {2023},
+  doi     = {10.1145/3618381}
 }
 ```
 
+Source studies. Cite every study whose trials you use (manifest columns `study` and `study_citation`). AddBiomechanics data: https://addbiomechanics.org/.
+
+| `study` | `study_citation` | reference |
+|---|---|---|
+| Carter2023 | Carter2023 | Carter, J., Chen, X., Cazzola, D., Trewartha, G., Preatoni, E. Full body kinematics and ground reaction forces of fifty heterogeneous runners completing treadmill running at various speeds and gradients. University of Bath Research Data Archive, 2024. https://doi.org/10.15125/BATH-01341 |
+| Falisse2017 | Falisse2017 | Falisse, A., Van Rossom, S., Jonkers, I., De Groote, F. EMG-driven optimal estimation of subject-specific Hill model muscle-tendon parameters of the knee joint actuators. IEEE Transactions on Biomedical Engineering 64(9), 2253-2262, 2017. https://doi.org/10.1109/TBME.2016.2630009 |
+| Fregly2012 | Fregly2012 | Fregly, B.J., Besier, T.F., Lloyd, D.G., Delp, S.L., Banks, S.A., Pandy, M.G., D'Lima, D.D. Grand challenge competition to predict in vivo knee loads. Journal of Orthopaedic Research 30(4), 503-513, 2012. https://doi.org/10.1002/jor.22023 |
+| Hamner2013 | Hamner2013 | Hamner, S.R., Delp, S.L. Muscle contributions to fore-aft and vertical body mass center accelerations over a range of running speeds. Journal of Biomechanics 46(4), 780-787, 2013. https://doi.org/10.1016/j.jbiomech.2012.11.024 |
+| Han2023 | Han2023 | Han, X., Senderling, B., To, S., Kumar, D., Whiting, E., Saito, J. GroundLink: A dataset unifying human body movement and ground reaction dynamics. SIGGRAPH Asia 2023 Conference Papers, 1-10, 2023. https://doi.org/10.1145/3610548.3618247 |
+| Li2021 | Li2021 | Li, G., Shourijeh, M.S., Ao, D., Patten, C., Fregly, B.J. How well do commonly used co-contraction indices approximate lower limb joint stiffness trends during gait for individuals post-stroke? Frontiers in Bioengineering and Biotechnology 8, 588908, 2021. https://doi.org/10.3389/fbioe.2020.588908 |
+| Moore2015 | Moore2015 | Moore, J.K., Hnat, S.K., van den Bogert, A.J. An elaborate data set on human gait and the effect of mechanical perturbations. PeerJ 3, e918, 2015. https://doi.org/10.7717/peerj.918 |
+| Tiziana2019 | Lencioni2019 | Lencioni, T., Carpinella, I., Rabuffetti, M., Marzegan, A., Ferrarin, M. Human kinematic, kinetic and EMG data during different walking and stair ascending and descending tasks. Scientific Data 6, 309, 2019. https://doi.org/10.1038/s41597-019-0323-z |
+| Uhlrich2023 | Uhlrich2023 | Uhlrich, S.D., Falisse, A., Kidziński, Ł., Muccini, J., Ko, M., Chaudhari, A.S., Hicks, J.L., Delp, S.L. OpenCap: Human movement dynamics from smartphone videos. PLOS Computational Biology 19(10), e1011462, 2023. https://doi.org/10.1371/journal.pcbi.1011462 |
+| vanderZee2022 | vanderZee2022 | van der Zee, T.J., Mundinger, E.M., Kuo, A.D. A biomechanics dataset of healthy human walking at various speeds, step lengths and step widths. Scientific Data 9, 704, 2022. https://doi.org/10.1038/s41597-022-01817-1 |
+
 ## Versions
 
 | version | date | note |
 |---|---|---|
-| 1.0.0-reviewed | 2026-05-07 | reviewed build (paper numbers) |
+| 1.0.0-reviewed | 2026-05-07 | reviewed build (numbers of the submitted paper) |
 | internal v1 | 2026-08-31 | internal build |
 | internal 1.1 | 2026-09-27 | internal build (frame and timebase fixes) |
 | 1.2.0 | 2026-10-11 | first frozen 1.2 build (first build 2026-10-09); superseded by 1.2.1 (QC mask erratum) |
 | 1.2.1 | 2026-10-12 | 1.2.0 with corrected QC masks and flags (pose, force and torque arrays unchanged) |
-| 1.3.0 | 2026-10-29 | this release: Hamner2013 study id (40 trial keys renamed, map shipped), `sex_resolved` corrected, benchmark results, documentation; arrays otherwise identical to 1.2.1, see CHANGELOG.md |
+| 1.3.0 | 2026-10-29 | this release: Hamner2013 study id (40 trial keys renamed, map shipped), `sex_resolved` corrected, benchmark results, documentation; arrays otherwise identical to 1.2.1, see CHANGELOG.md. Documentation updated on 2026-11-01, no data file changed |
 
 ## Maintenance
 

LICENSE.md

@@ -0,0 +1,24 @@
+# License and attribution
+
+## ForceBody data
+
+The ForceBody data in this repository are released under the Creative Commons Attribution-NonCommercial 4.0 International license (CC BY-NC 4.0).
+License text: https://creativecommons.org/licenses/by-nc/4.0/legalcode
+
+You may share and adapt the data for non-commercial purposes. You must give credit as described below. Commercial use is not permitted.
+
+## SKEL-derived fields
+
+The fields `poses*`, `trans*`, `joints*`, `betas` and `skel_gender` are parameters of the SKEL body model. They are also subject to the SKEL model license, which allows non-commercial scientific research only. Building meshes from these fields requires the SKEL model, which you obtain from its authors: https://skel.is.tue.mpg.de/
+
+## Upstream data
+
+ForceBody is derived from the AddBiomechanics dataset (https://addbiomechanics.org/), which is released under the Creative Commons Attribution 4.0 International license (CC BY 4.0, https://creativecommons.org/licenses/by/4.0/legalcode). The motion and force records come from 10 source studies: Carter2023, Falisse2017, Fregly2012, Hamner2013, Han2023, Li2021, Moore2015, Tiziana2019 (Lencioni et al. 2019), Uhlrich2023 and vanderZee2022. DETAILS.md ("How to cite") gives the full reference of each study and the AddBiomechanics BibTeX entries.
+
+## Attribution
+
+If you use ForceBody, cite the ForceBody paper, AddBiomechanics and every source study whose trials you use. The `study` and `study_citation` columns of manifest.csv identify the source study of each trial.
+
+## Code
+
+The loader, benchmark scorer, baselines and renderer are in https://github.com/DominoAI-Lab/ForceBody-NeurIPS-ED-2026 under the MIT license. This repository holds no code.

MD5SUMS.txt

@@ -1,5 +1,5 @@
-fe0153904dd9cf3ec5c174dc0c91919b  ./CHANGELOG.md
-29b1ac48ac11a03e445ffd7dababea85  ./README.md
+63ebfb3705948c9c2cf5bb2439051aa4  ./CHANGELOG.md
+876b68a25ae6360d4a25fa34e40fc171  ./README.md
 1989b413f574403d1aa72b52bdf394af  ./SCHEMA_v1.3.csv
 16725af040dd20ce2d7522a34aa8e5dc  ./gait_events.csv
 557efe2ff4a7c31d52e3eed707a59253  ./manifest.csv
@@ -11155,4 +11155,4 @@ f41b5d9a18a82419c1eb029911b82f7f  ./trials/vanderZee2022/S10/vanderZee2022_S10_w
 e2b12456ebf056da7cd9f5a451ac2afc  ./trials/vanderZee2022/S10/vanderZee2022_S10_walk_112.npz
 989be65c15591198dc4242e90046b8f5  ./trials/vanderZee2022/S10/vanderZee2022_S10_walk_113.npz
 288f681179e42c09619d77e007bb7530  ./uid_map_v12_to_v13.csv
-8bf3b934ca02812b4596b2b38ff1356d  ./DETAILS.md
+a2e37e5d78b81ea83cab5fe618abe955  ./DETAILS.md

README.md

@@ -31,7 +31,7 @@ configs:
 
 Joonwoo Kwon<sup>1</sup>, Yufei Zhang<sup>2</sup>, Xiaoming Liu<sup>3</sup>, Zijun Cui<sup>1</sup>
 
-<sup>1</sup>Michigan State University &nbsp; <sup>2</sup>Independent Researcher &nbsp; <sup>3</sup>University of North Carolina at Chapel Hill
+<sup>1</sup>Michigan State University, <sup>2</sup>Independent Researcher, <sup>3</sup>University of North Carolina at Chapel Hill
 
 <img src="assets/divider.png" width="100%" height="4">
 
@@ -41,7 +41,7 @@ Joonwoo Kwon<sup>1</sup>, Yufei Zhang<sup>2</sup>, Xiaoming Liu<sup>3</sup>, Zij
 
 <p align="center"><img src="assets/teaser.png" width="100%"></p>
 
-ForceBody pairs SKEL body motion with measured ground reaction forces (GRF), inverse-dynamics joint torques, and a per-frame, per-joint torque uncertainty σ<sub>τ</sub>. This page holds the data. Code, loader and benchmark are on GitHub.
+ForceBody pairs SKEL body motion with measured ground reaction forces (GRF), inverse-dynamics joint torques, and a per-frame, per-joint torque uncertainty σ<sub>τ</sub>. This page holds the data. Code, loader and benchmark are on GitHub. The source data come from [AddBiomechanics](https://addbiomechanics.org/). Croissant metadata with Responsible AI fields: [croissant.json](https://huggingface.co/datasets/kwonjoon/ForceBody/resolve/main/croissant.json).
 
 | Trials | Subjects | Hours | Frames | Sources | Release |
 |:---:|:---:|:---:|:---:|:---:|:---:|
@@ -67,11 +67,12 @@ ForceBody pairs SKEL body motion with measured ground reaction forces (GRF), inv
 
 ### 0. Download the SKEL model (required)
 
-The trial files store SKEL parameters (`poses`, `betas`, `trans`), not meshes. Register and download the SKEL model from the [SKEL website](https://skel.is.tue.mpg.de/) under its non-commercial license. ForceBody does not redistribute it. You need it to build meshes or joints. The force, torque and uncertainty arrays can be read without it.
+The trial files store SKEL parameters (`poses`, `betas`, `trans`), not meshes. Register and download the SKEL model from the [SKEL website](https://skel.is.tue.mpg.de/) under its non-commercial license. ForceBody does not redistribute it. You need it to build meshes. The joint positions (`joints`), force, torque and uncertainty arrays can be read without it.
 
 ### 1. Download the data
 
 ```bash
+pip install -U huggingface_hub
 hf download kwonjoon/ForceBody --repo-type dataset --local-dir ForceBody
 ```
 
@@ -79,6 +80,8 @@ About 46 GB.
 
 ### 2. Install the loader
 
+Run these commands in the folder of step 1.
+
 ```bash
 git clone https://github.com/DominoAI-Lab/ForceBody-NeurIPS-ED-2026.git
 cd ForceBody-NeurIPS-ED-2026/code
@@ -87,17 +90,20 @@ pip install numpy
 
 ### 3. Load a trial
 
+Run Python in `ForceBody-NeurIPS-ED-2026/code`. The data folder of step 1 is then `../../ForceBody`.
+
 ```python
 import forcebody_io as fio
 
-d = fio.load_trial("ForceBody/trials/Moore2015/S10/Moore2015_S10_walk_014.npz", pose_source="pass0")
+d = fio.load_trial("../../ForceBody/trials/Moore2015/S10/Moore2015_S10_walk_014.npz", pose_source="pass0")
 m = float(d["mass_kg_dynamics"])
-ok = d["mc_frame_ok"]                       # always apply the frame masks
-tau   = d["tau_selfconsistent"][ok] / m     # joint torque (N m / kg)
-sigma = d["sigma_tau"][ok] / m              # torque uncertainty (N m / kg)
+ok = d["mc_frame_ok"] & bool(d["torque_label_usable"])   # frame mask and trial flag
+tau   = d["tau_selfconsistent"][ok, 6:] / m   # joint torques (N m / kg), columns 0-5 are root residuals
+sigma = d["sigma_tau"][ok, 6:] / m            # torque uncertainty (N m / kg)
+print(tau.shape, sigma.shape)
 ```
 
-For force or torque prediction, use `pose_source="pass0"` as model input.
+For force or torque prediction, use `pose_source="pass0"` as model input. Always combine a frame mask with its trial flag.
 
 ### 4. Train and evaluate
 
@@ -107,13 +113,17 @@ See the GitHub repository (`code/baselines/`) for cache building, training, unce
 
 ## What is in a trial
 
+T = frames. C = contact channels: 2 (right foot, left foot), or 3 in Falisse2017, where channel 2 is the pelvis on an instrumented chair.
+
 | Field | Shape | Content |
 |---|---|---|
 | `poses`, `trans`, `betas` | `[T,46]`, `[T,3]`, `[10]` | SKEL pose, root translation, body shape |
-| `grf_aligned`, `cop_aligned_foot` | `[T,C,3]` | Measured GRF and centre of pressure per foot |
-| `tau_selfconsistent` | `[T,37]` | Inverse-dynamics joint torque |
-| `sigma_tau`, `mu_tau` | `[T,37]` | Monte Carlo torque uncertainty and mean |
-| `*_frame_ok` | `[T]` | Frame masks |
+| `grf_aligned`, `cop_aligned_foot` | `[T,C,3]` | Measured GRF (N) and centre of pressure (m) per contact channel |
+| `tau_selfconsistent` | `[T,37]` | Inverse dynamics in AddBiomechanics DOF order (`dof_names`). Columns 0-5 are pelvis root residuals (0-2 moments in N m, 3-5 forces in N). Columns 6-36 are joint torques (N m) |
+| `sigma_tau`, `mu_tau` | `[T,37]` | Monte Carlo torque uncertainty and mean, same columns |
+| `grf_frame_ok`, `tau_frame_ok`, `mc_frame_ok` | `[T]` | Frame masks for GRF, torque and σ<sub>τ</sub> |
+| `cop_frame_ok`, `grf_foot_ok` | `[T,2]` | Per-foot frame masks (right, left) |
+| `force_label_usable`, `torque_label_usable` | scalar | Trial flags |
 
 <img src="assets/divider.png" width="100%" height="4">
 
@@ -122,17 +132,21 @@ See the GitHub repository (`code/baselines/`) for cache building, training, unce
 | Path | Content |
 |---|---|
 | `trials/<study>/S<NN>/<trial_uid>.npz` | One file per trial |
-| `models/` | AddBiomechanics OpenSim models per subject (used for the torques) |
+| `models/` | AddBiomechanics OpenSim models (`.osim.gz`) per source b3d file, processing passes 0 and 2 (the torques use the pass-2 model) |
 | `manifest.csv` (+ `_train`, `_test`) | One row per trial: metadata, split, quality flags |
-| `persons.csv`, `subjects.csv` | One row per subject: SKEL shape, sex, height, mass |
+| `persons.csv` | One row per subject: SKEL shape (`beta_0` to `beta_9`), sex, height, mass, body measurements, splits |
+| `subjects.csv` | One row per subject: age, health status, footwear, mass sources, trial counts, activities |
 | `SCHEMA_v1.3.csv` | Shape, unit and frame of every field |
 | `DETAILS.md` | Full documentation (masks, conventions, known issues) |
+| `CHANGELOG.md`, `MD5SUMS.txt` | Version history, md5 checksums |
+| [`croissant.json`](https://huggingface.co/datasets/kwonjoon/ForceBody/resolve/main/croissant.json) | Croissant metadata with Responsible AI fields |
+| Other files | Gait events, key maps, marker map: see DETAILS.md, "Layout" |
 
 <img src="assets/divider.png" width="100%" height="4">
 
 ## License
 
-Data under **CC BY-NC 4.0** (non-commercial). SKEL-derived fields also follow the [SKEL license](https://skel.is.tue.mpg.de/). Code on GitHub under MIT. Please cite AddBiomechanics and the source studies you use.
+Data under **CC BY-NC 4.0** (non-commercial). SKEL-derived fields also follow the [SKEL license](https://skel.is.tue.mpg.de/). Code on GitHub under MIT. Please cite [AddBiomechanics](https://addbiomechanics.org/) and the source studies you use. DETAILS.md ("How to cite") gives the BibTeX entries and the source-study references. LICENSE.md gives the full license and attribution terms.
 
 <img src="assets/divider.png" width="100%" height="4">
 

croissant.json

@@ -50,7 +50,7 @@
   },
   "@type": "sc:Dataset",
   "name": "ForceBody",
-  "description": "ForceBody pairs the SKEL parametric body model with measured ground reaction forces and inverse-dynamics joint torques across 10,386 motion trials (9,689,808 frames, 26.9 hours) from 140 subjects in 10 source studies aggregated by AddBiomechanics. Each torque label comes with a per-frame, per-joint Monte Carlo uncertainty sigma_tau. Release 1.3.0.",
+  "description": "ForceBody pairs the SKEL parametric body model with measured ground reaction forces and inverse-dynamics joint torques across 10,386 motion trials (9,689,808 frames, 26.9 hours) from 140 subjects in 10 source studies aggregated by AddBiomechanics. Each torque label comes with a per-frame, per-DOF Monte Carlo uncertainty sigma_tau, computed for 10,382 of the 10,386 trials. Release 1.3.0.",
   "conformsTo": [
     "http://mlcommons.org/croissant/1.1",
     "http://mlcommons.org/croissant/RAI/1.0"
@@ -112,36 +112,36 @@
     "name": "Michigan State University"
   },
   "citeAs": "@inproceedings{kwon2026forcebody, title={ForceBody: Force-Paired Parametric Body Motion with Torque Uncertainty}, author={Kwon, Joonwoo and Zhang, Yufei and Liu, Xiaoming and Cui, Zijun}, booktitle={Advances in Neural Information Processing Systems (NeurIPS), Evaluations and Datasets Track}, year={2026}}",
-  "rai:dataLimitations": "ForceBody is limited to the 140 subjects and 10 source studies it takes from AddBiomechanics. Walking and running make up 93.5% of trials (running 5,547, walking 4,167). Rare classes have few trials (standing 272, mixed 208, jumping 75, sit-to-stand 71, squats 24, stairs 22). Carter2023 contributes 7,006 trials (67.5%). SKEL parameters are numerical fits with mean per-joint error 21.6 mm against the AddBiomechanics joint centers. sigma_tau models joint-angle noise only and does not cover GRF and CoP error, segment inertia, or the axis mismatch between SKEL and the Rajagopal model. Torques inherit any residual error of the upstream AddBiomechanics inverse dynamics.",
-  "rai:dataBiases": "Subjects: 75 male, 55 female, 10 unreported (vanderZee2022). Height 1.73 +/- 0.09 m, mass 68.8 +/- 11.5 kg, BMI 22.8 +/- 2.8. Mostly healthy adults. Non-healthy subjects are 1 post-stroke subject (Li2021, 65 trials) and 4 total knee arthroplasty subjects (Fregly2012, 78 trials). Activity and source imbalance as in dataLimitations. The 10 unreported-sex subjects are fitted with a SKEL sex model chosen by height (female below 1.683 m). sigma_q = 1.6 deg is applied isotropically to all 20 direct DOFs because AddBiomechanics reports only an aggregate angle error.",
-  "rai:personalSensitiveInformation": "No personally identifying information. Subjects are generic labels inherited from AddBiomechanics. Anthropometrics (height, mass, BMI, sex, age) are de-identified.",
-  "rai:dataUseCases": "Intended for non-commercial scientific research. Primary use cases include (i) supervised joint torque prediction from kinematics, (ii) uncertainty-aware regression where the released sigma_tau is used as a per-sample loss weight or label-noise estimate, (iii) ground reaction force prediction from kinematics, (iv) marker-based pose fitting and SKEL-fit benchmarks, and (v) physics-aware human motion modeling that requires force grounding rather than purely kinematic supervision. Not intended for clinical decision making, identification, or commercial product development or commercial use.",
-  "rai:dataSocialImpact": "ForceBody is intended for physically grounded human motion modeling, with applications in biomechanical assessment, exoskeleton and prosthetic design, sports analytics, and physics-aware animation. The released per-sample uncertainty exposes the inherent label noise of inverse-dynamics torques, which downstream models can account for during training. Misuse risk is modest given the de-identified nature of the data and the non-commercial license restriction inherited from the SKEL body model.",
+  "rai:dataLimitations": "ForceBody is limited to the 140 subjects and 10 source studies it takes from AddBiomechanics. Walking and running make up 93.5% of trials (running 5,547, walking 4,167). Rare classes have few trials (standing 272, mixed 208, jumping 75, sit-to-stand 71, squats 24, stairs 22). Carter2023 contributes 7,006 trials (67.5%). SKEL parameters are numerical fits with mean per-joint error 21.6 mm against the AddBiomechanics joint centers. sigma_tau models joint-angle noise only and does not cover GRF and CoP error, segment inertia, or the axis mismatch between SKEL and the Rajagopal model. Torques inherit any error of the AddBiomechanics kinematics, skeleton and force data they are computed from.",
+  "rai:dataBiases": "Subjects are 75 male, 55 female and 10 unreported (vanderZee2022). Height 1.73 +/- 0.09 m, mass 68.8 +/- 11.5 kg, BMI 22.8 +/- 2.8. Mostly healthy adults. Non-healthy subjects are 1 post-stroke subject (Li2021, 65 trials) and 4 total knee arthroplasty subjects (Fregly2012, 78 trials). Activity and source imbalance as in dataLimitations. The 10 unreported-sex subjects are fitted with a SKEL sex model chosen by height (female below 1.683 m). sigma_q = 1.6 deg is applied isotropically to all 20 direct DOFs because AddBiomechanics reports only an aggregate angle error.",
+  "rai:personalSensitiveInformation": "No full names or contact details. Subjects are generic labels inherited from AddBiomechanics. The data include sex, age, height, mass and BMI from the source studies, all de-identified. The data include health information. 1 subject after stroke and 4 subjects after total knee arthroplasty are labelled in the fields health_status and health_detail. Gait and body-size data can help to re-identify a person. Do not attempt to re-identify subjects.",
+  "rai:dataUseCases": "Intended for non-commercial scientific research. Use cases are (i) supervised joint torque prediction from kinematics, (ii) uncertainty-aware regression that uses sigma_tau as a per-sample loss weight, (iii) ground reaction force prediction from kinematics, (iv) marker-based pose fitting and SKEL-fit benchmarks, and (v) physics-aware human motion modeling that requires force grounding rather than purely kinematic supervision. Not intended for clinical decision making, identification of persons, or commercial use.",
+  "rai:dataSocialImpact": "ForceBody is intended for physically grounded human motion modeling, with applications in biomechanical assessment, exoskeleton and prosthetic design, sports analytics, and physics-aware animation. sigma_tau gives the sensitivity of each torque label to joint-angle noise, which downstream models can account for during training. The data are de-identified and released under CC BY-NC 4.0. SKEL-derived fields also follow the SKEL non-commercial license. The data include health labels for 5 subjects, and gait and body-size data can help to re-identify a person. Re-identification of subjects and clinical decisions are out of scope.",
   "rai:hasSyntheticData": false,
-  "rai:dataCollection": "ForceBody does not collect new motion-capture data. Motion, force and torque records come from the public AddBiomechanics dataset (CC BY 4.0), which aggregates optical motion capture with synchronized force-plate ground reaction forces from 10 source studies. Each source study recruited its subjects under its own ethics approval. Subject identities are replaced by generic labels.",
+  "rai:dataCollection": "ForceBody does not collect new motion-capture data. Motion, force and torque records come from the public AddBiomechanics dataset (CC BY 4.0), which aggregates optical motion capture with synchronized force-plate ground reaction forces from 10 source studies. The source studies are responsible for the ethics approval of their data collection. Subject identities are replaced by generic labels.",
   "rai:dataCollectionType": [
     "Secondary Data Analysis",
     "Software Collection"
   ],
-  "rai:dataCollectionRawData": "Source records are AddBiomechanics .b3d files. Each file holds marker trajectories, inverse-kinematics joint angles, force-plate GRF and CoP, and inverse-dynamics torques on the OpenSim Rajagopal full-body model. The 10 source studies are Carter2023, Moore2015, Hamner2013, Falisse2017, Fregly2012, Han2023, Li2021, Tiziana2019, Uhlrich2023 and vanderZee2022.",
-  "rai:dataCollectionMissingData": "Of the 10,386 released trials, 10,382 carry Monte Carlo sigma_tau. The paper statistics use the 8,732 trials with sigma_tau and GRF valid on at least 50% of frames. Missing markers are NaN in markers fields. Frames with invalid force, CoP, torque or kinematics are flagged by the per-frame masks (grf_frame_ok, tau_frame_ok, mc_frame_ok, kin_frame_ok) and are not imputed.",
+  "rai:dataCollectionRawData": "Source records are 378 AddBiomechanics .b3d files. Each file holds marker trajectories, inverse-kinematics joint angles, force-plate GRF and CoP, and inverse-dynamics torques on the OpenSim Rajagopal full-body model. The 10 source studies are Carter2023, Moore2015, Hamner2013, Falisse2017, Fregly2012, Han2023, Li2021, Tiziana2019, Uhlrich2023 and vanderZee2022.",
+  "rai:dataCollectionMissingData": "sigma_tau is computed for 10,382 of the 10,386 released trials. The other 4 trials are shorter than one period of the 6 Hz noise band, so their sigma_tau is 0 and has_uncertainty is False (mc_missing_reason). The sigma_tau statistics of the paper use the 8,732 trials with has_uncertainty True and GRF valid on at least 50% of frames (grf_valid_frac >= 0.5). Missing markers are NaN in markers fields. Frames with invalid force, CoP, torque or kinematics are flagged by the per-frame masks (grf_frame_ok, cop_frame_ok, tau_frame_ok, mc_frame_ok, kin_frame_ok) and are not imputed. In 14 trials force and motion capture are out of sync and could not be re-timed, so force_label_usable is False.",
   "rai:dataPreprocessingProtocol": [
-    "Kinematics are taken from the AddBiomechanics inverse-kinematics pass only (pass 0, field poses) to keep force labels out of kinematic inputs. A second pose set fitted to the dynamics pass (poses_dyn) is also released.",
-    "Kinematics are filtered with a 2nd-order 30 Hz zero-lag Butterworth filter, as in AddBiomechanics. Trials keep their native grid: about 98% are at 99 to 101 Hz, the rest at 61, 201 or 250 Hz (field fps).",
-    "Trials are quality-filtered at the 80th percentile of the unfiltered fitting error (10,386 of 12,986 trials kept).",
-    "Ground reaction force, CoP and free moment are cleaned and expressed in the SKEL frame. Joint torques tau_selfconsistent are recomputed with the cleaned forces."
+    "The default SKEL pose set (poses) is fitted to the AddBiomechanics inverse-kinematics pass (pass 0) to keep force labels out of kinematic inputs. A second pose set fitted to the dynamics pass (poses_dyn) is also released. The torque labels and sigma_tau are computed on the dynamics-pass joint angles (q_gt, or q_kin on the 27 re-timed trials).",
+    "The kinematics-pass target q_kin is AddBiomechanics pass 0 filtered with a 2nd-order 30 Hz zero-lag Butterworth filter, which equals AddBiomechanics pass 1. 10,186 of the 10,386 trials (98.1%) are on a grid of 99 to 101 Hz. 8,848 of them were resampled from a capture rate of 120 to 251 Hz (field fps_native_capture), and 1,338 were captured at 100 or 101 Hz. The other 200 trials keep their native grid of 61 Hz (64 trials), 201 Hz (111 trials) or 250 Hz (25 trials). fps is the nominal rate. Time stamps use fps_true_grid.",
+    "Trials are quality-filtered at the 80th percentile of the unfiltered fitting error (10,389 of 12,986 trials kept, 10,386 of them released).",
+    "Ground reaction force, CoP and free moment are cleaned (contact rule, plate re-zero, crossover, free-moment bound, raw-plate CoP, frame tilt and tail blanking). contact_status gives the result per frame and contact. The cleaned forces are released in the AddBiomechanics lab frame (grf, cop, contact_torque) and in the SKEL world frame (grf_aligned, cop_aligned, contact_torque_aligned). Joint torques tau_selfconsistent are recomputed with the cleaned forces."
   ],
-  "rai:dataManipulationProtocol": "Two steps. (i) AddBiomechanics motion is re-expressed in SKEL by the fitting described in dataAnnotationProtocol. (ii) GRF, CoP and free moment are cleaned and expressed in the SKEL frame, and the torque label tau_selfconsistent is recomputed from the released kinematics and the cleaned forces. The original AddBiomechanics angles (q_gt) and torques (tau_gt_det) are also released unchanged for reference.",
-  "rai:dataImputationProtocol": "No values are imputed. Missing markers are NaN. Frames with missing or invalid GRF, CoP, torque or kinematics are flagged by per-frame masks (grf_valid_mask, grf_frame_ok, tau_frame_ok, mc_frame_ok, kin_frame_ok). Users should apply these masks.",
-  "rai:dataAnnotationProtocol": "Two automated annotation layers. (i) SKEL fit. Stage 1 estimates one 10-dimensional shape per subject by matching eight bilateral bone lengths (pelvis to hip, thigh, shank, forearm), with terms for the recorded height and mass. Stage 2 fixes the shape and fits per-frame pose (46 DOFs) and translation with a joint-position loss on the 15 directly matched joints, a marker loss, anatomical regularizers on the spine, scapula and humerus, and temporal smoothness. (ii) Torque uncertainty. The AddBiomechanics joint angles at the 20 direct DOFs of 14 joints are perturbed with band-limited Gaussian noise (sigma_q = 1.6 deg, 6 Hz), each sample is passed through inverse dynamics with measured GRF held fixed, and the per-DOF mean and standard deviation over N = 1000 samples are stored. sigma_q = 1.6 deg is the joint-angle RMSE that AddBiomechanics reports for its inverse kinematics on synthetic walking data. No human annotators are involved.",
-  "rai:personalSensitiveInformationDescription": "Subjects appear only as generic labels (for example Carter2023_S45). No names, contact details or institution identifiers appear in any field. Height, mass, BMI, sex and age come from the source studies. AddBiomechanics requires de-identified uploads.",
+  "rai:dataManipulationProtocol": "Each manipulation below is recorded in the named field. (i) SKEL fit. AddBiomechanics motion is re-expressed in SKEL by the fitting described in dataAnnotationProtocol, once for each pose set (poses, poses_dyn). (ii) Resampling. 8,848 trials captured at 120 to 251 Hz are released on a resampled grid of 99 to 101 Hz, with kinematics, forces and markers on one grid per trial (fps_native_capture, fps_true_grid). (iii) Force cleaning. GRF, CoP and free moment are cleaned as described in dataPreprocessingProtocol (contact_status). In 270 trials the raw plate CoP is unusable and the CoP is a near-fixed lab point (qc_cop_fixed_point). (iv) Force re-timing. In 27 trials the force stream and grf_valid_mask are shifted in time to match the motion capture (skel_fit_target_dyn == kin_retimed, per-frame shift in grf_retimed_ms, median 277 ms, range -85 to 320 ms). In 14 trials that could not be re-timed, force_label_usable is False, and 1 trial has force invalid from force_invalid_from_frame. (v) Dynamics mass. In 491 trials (14 source files) the AddBiomechanics link masses are rescaled to agree with the measured GRF (mass_dyn_scale, already applied in mass_kg_dynamics). (vi) Torque label. tau_selfconsistent is recomputed from the released kinematics (q_gt, or q_kin on the 27 re-timed trials, field tau_sc_kinematics) and the cleaned forces, on the AddBiomechanics pass-2 skeleton, with an inverse-dynamics routine built on nimblephysics that re-implements the AddBiomechanics inverse dynamics (mc_engine). (vii) Arm re-solve. In 875 trials the AddBiomechanics pass-0 arm that parks on a joint limit or hops is re-solved by inverse kinematics to the markers before the 30 Hz filter (q_kin_source == addb_pass0_armfix_lp30). (viii) Pass-2 target repair. In 1,420 trials the poses_dyn fit target is a repaired pass-2 target (skel_fit_target_dyn addb_pass2_armsplice 9, addb_pass2_repaired_v3a 791, addb_pass2_repaired_v4s 620). 1,392 of them have modified frames, marked per frame and DOF in dyn_target_source (dyn_target_modified_frac). q_gt itself is unchanged. (ix) Head post-fit. In 1,948 trials with at least 2 head markers the head DOFs are fitted after the main fit (head_postfit). In the other 8,438 trials the head DOFs stay 0 (qc_head_untracked). (x) Study rename. Release 1.3.0 renames Hammer2013 to Hamner2013 in 40 trial keys (trial_uid_v12, uid_map_v12_to_v13.csv). The AddBiomechanics dynamics-pass angles (q_gt) and torques (tau_gt_det) are also released for reference, on the release grid. They are not usable on the 27 re-timed trials (q_gt_usable, tau_gt_det_usable).",
+  "rai:dataImputationProtocol": "Masked frames are not imputed and not set to NaN. Outside a mask the arrays keep their raw values, and a 0 there is not a measurement. Missing markers are NaN. During force cleaning some CoP values were filled, and contact_status 2 marks them. Frames with missing or invalid GRF, CoP, torque or kinematics are flagged by per-frame masks (grf_valid_mask, grf_frame_ok, cop_frame_ok, tau_frame_ok, mc_frame_ok, kin_frame_ok). Users should apply these masks.",
+  "rai:dataAnnotationProtocol": "Two automated annotation layers. (i) SKEL fit. Stage 1 estimates one 10-dimensional shape per subject by matching eight bilateral bone lengths (pelvis to hip, thigh, shank, forearm), with terms for the recorded height and a subject mass target. Stage 2 fixes the shape and fits per-frame pose (46 DOFs) and translation with a joint-position loss on the 15 directly matched joints, a marker loss, anatomical regularizers on the spine, scapula and humerus, and temporal smoothness. Stage 2 runs once on the kinematics-pass target (poses) and once on the dynamics-pass target (poses_dyn). (ii) Torque uncertainty. The AddBiomechanics dynamics-pass joint angles (q_gt, or q_kin on the 27 re-timed trials) at the 20 direct DOFs of 14 joints are perturbed with band-limited Gaussian noise (sigma_q = 1.6 deg, 6 Hz), each sample is passed through the inverse-dynamics routine that computes tau_selfconsistent with the cleaned measured GRF held fixed, and the per-DOF mean and standard deviation over N = 1000 samples are stored. sigma_q = 1.6 deg is the joint-angle RMSE that AddBiomechanics reports for its inverse kinematics on synthetic walking data. No human annotators are involved.",
+  "rai:personalSensitiveInformationDescription": "Subjects appear only as generic labels (for example Carter2023_S45). No full names or contact details appear in any field. trial_name keeps the AddBiomechanics source trial names. All 240 Han2023 names contain a capture date, and the 78 Fregly2012 names contain subject initials. Height, mass, BMI, sex and age (18 to 86 years, unknown in 200 trials) come from the source studies. health_status is healthy (135 subjects, 10,243 trials), TKA (total knee arthroplasty, 4 subjects, 78 trials) or stroke (1 subject, 65 trials). health_detail gives TKA left knee (40 trials), TKA right knee (38 trials) or stroke, right-sided hemiparesis (65 trials). AddBiomechanics requires de-identified uploads.",
   "rai:dataAnnotationPlatform": "No annotation platform or labeling service. SKEL fitting and Monte Carlo sampling run as Python scripts on research servers.",
-  "rai:dataAnnotationAnalysis": "Each trial stores mu_tau and sigma_tau per frame and DOF, and seven percentiles (1, 5, 25, 50, 75, 95, 99) of the Monte Carlo distribution. The standard error of sigma_tau is 7.1% at N = 100, 3.2% at N = 500 and 2.2% at N = 1000.",
-  "rai:annotationsPerItem": "Each trial carries one SKEL fit (shape per subject, pose and translation per frame), the inherited GRF and CoP, one torque label tau_selfconsistent per frame and DOF, and the Monte Carlo mean mu_tau and standard deviation sigma_tau from N = 1000 samples per frame.",
+  "rai:dataAnnotationAnalysis": "Each trial stores mu_tau and sigma_tau per frame and DOF. The Monte Carlo percentiles are not part of the public release. The standard error of sigma_tau is 7.1% at N = 100, 3.2% at N = 500 and 2.2% at N = 1000.",
+  "rai:annotationsPerItem": "Each trial carries one SKEL shape per subject and two SKEL pose sets with root translation per frame (poses and trans fitted to the kinematics pass, poses_dyn and trans_dyn fitted to the dynamics pass), the cleaned AddBiomechanics GRF, CoP and free moment, one torque label tau_selfconsistent per frame and DOF, and the Monte Carlo mean mu_tau and standard deviation sigma_tau from N = 1000 samples per frame.",
   "rai:annotatorDemographics": "No human annotators. All annotations were produced by the automated SKEL fitting and Monte Carlo pipeline.",
   "rai:machineAnnotationTools": [
     "SKEL body model and a Python fitting pipeline (Stage 1 shape, Stage 2 pose and translation).",
-    "A re-implementation of the AddBiomechanics inverse-dynamics routine, used for tau_selfconsistent and the Monte Carlo samples.",
+    "An inverse-dynamics routine built on nimblephysics that re-implements the AddBiomechanics inverse dynamics, used for tau_selfconsistent and the Monte Carlo samples.",
     "scipy.signal Butterworth filters."
   ],
   "rai:dataReleaseMaintenancePlan": "Released as version 1.3.0 on Hugging Face (kwonjoon/ForceBody), with code at https://github.com/DominoAI-Lab/ForceBody-NeurIPS-ED-2026. Fixes are issued as semantic-versioned releases on the same repository with a CHANGELOG.md entry and MD5SUMS.txt. Data under CC BY-NC 4.0. SKEL-derived fields also follow the SKEL non-commercial license. Code under MIT.",
@@ -154,16 +154,16 @@
       "license": "https://spdx.org/licenses/CC-BY-4.0.html"
     }
   ],
-  "prov:wasGeneratedBy": "AddBiomechanics motion in the Rajagopal skeleton was re-expressed in SKEL by a two-stage fit: one shape per subject from bone lengths, height and mass, then per-frame pose and translation on the 15 directly matched joints with marker, anatomical and temporal terms. sigma_tau comes from 1000-sample Monte Carlo perturbation of the AddBiomechanics joint angles (band-limited Gaussian, sigma_q = 1.6 deg, 6 Hz, 20 direct DOFs of 14 joints) through the AddBiomechanics inverse-dynamics solver with measured GRF held fixed.",
+  "prov:wasGeneratedBy": "AddBiomechanics motion in the Rajagopal skeleton was re-expressed in SKEL by a two-stage fit. Stage 1 fits one shape per subject from bone lengths, height and mass. Stage 2 fits per-frame pose and translation on the 15 directly matched joints with marker, anatomical and temporal terms, once to the kinematics pass (poses) and once to the dynamics pass (poses_dyn). GRF, CoP and free moment were cleaned. The torque label tau_selfconsistent was recomputed from the released kinematics and the cleaned forces with an inverse-dynamics routine built on nimblephysics that re-implements the AddBiomechanics inverse dynamics. sigma_tau comes from 1000-sample Monte Carlo perturbation of the dynamics-pass joint angles (band-limited Gaussian, sigma_q = 1.6 deg, 6 Hz, 20 direct DOFs of 14 joints) through the same routine with the cleaned measured GRF held fixed.",
   "distribution": [
     {
       "@type": "cr:FileObject",
       "@id": "huggingface-repo",
       "name": "huggingface-repo",
-      "description": "Hugging Face dataset repository kwonjoon/ForceBody.",
+      "description": "Hugging Face dataset repository kwonjoon/ForceBody. The repository as a whole has no file checksum. The sha256 value points to Croissant issue 80, as in the Croissant reference examples for git repositories. MD5SUMS.txt lists the md5 of every data file.",
       "contentUrl": "https://huggingface.co/datasets/kwonjoon/ForceBody",
       "encodingFormat": "git+https",
-      "sha256": "main"
+      "sha256": "https://github.com/mlcommons/croissant/issues/80"
     },
     {
       "@type": "cr:FileObject",
@@ -178,7 +178,7 @@
       "@type": "cr:FileObject",
       "@id": "manifest_train.csv",
       "name": "manifest_train.csv",
-      "description": "Train split of manifest.csv (9,472 rows, 121 subjects).",
+      "description": "Train split of manifest.csv (9,472 rows, 121 subjects). It holds the 12 validation subjects (1,098 trials, split_addb_val == val).",
       "contentUrl": "https://huggingface.co/datasets/kwonjoon/ForceBody/resolve/main/manifest_train.csv",
       "encodingFormat": "text/csv",
       "sha256": "f5dec7406223c71cf9af84b656aa207de7098c3439c7ba3de3e49b17cc961d71"
@@ -205,7 +205,7 @@
       "@type": "cr:FileObject",
       "@id": "subjects.csv",
       "name": "subjects.csv",
-      "description": "Per-subject metadata summary (140 rows).",
+      "description": "Per-subject metadata summary (140 rows, v1.1 layout with v1.2 values). Its n_with_uncertainty column is the v1.1 count (8,652 trials in total). For release 1.3.0 use has_uncertainty in manifest.csv (10,382 trials).",
       "contentUrl": "https://huggingface.co/datasets/kwonjoon/ForceBody/resolve/main/subjects.csv",
       "encodingFormat": "text/csv",
       "sha256": "f8072afaf863474a4b151575d5e6b28d50c74ebdd2004ada67ca0582658a4c84"
@@ -219,6 +219,69 @@
       "encodingFormat": "text/csv",
       "sha256": "ab5a523100feaf16241f6082e63b1055f09a1f42024d92492f3e7213f086bfbd"
     },
+    {
+      "@type": "cr:FileObject",
+      "@id": "gait_events.csv",
+      "name": "gait_events.csv",
+      "description": "GRF heel strikes and toe-offs of each foot on walking and running trials with usable force labels (343,553 rows, 8,940 trials), frame on the release grid.",
+      "contentUrl": "https://huggingface.co/datasets/kwonjoon/ForceBody/resolve/main/gait_events.csv",
+      "encodingFormat": "text/csv",
+      "sha256": "b881938d4de958cdd0ca1f5dc9b93024abe60e3977b24a04123d71f166f37843"
+    },
+    {
+      "@type": "cr:FileObject",
+      "@id": "name_map.csv",
+      "name": "name_map.csv",
+      "description": "trial_uid to AddBiomechanics source file and trial (10,386 rows). The split column is the AddBiomechanics source split, not split_addb.",
+      "contentUrl": "https://huggingface.co/datasets/kwonjoon/ForceBody/resolve/main/name_map.csv",
+      "encodingFormat": "text/csv",
+      "sha256": "fb36b21172281cb04a4b2da9435c370b7e258a29c4ded6a9a2898a380683516a"
+    },
+    {
+      "@type": "cr:FileObject",
+      "@id": "name_map.json",
+      "name": "name_map.json",
+      "description": "Same content as name_map.csv as a JSON list.",
+      "contentUrl": "https://huggingface.co/datasets/kwonjoon/ForceBody/resolve/main/name_map.json",
+      "encodingFormat": "application/json",
+      "sha256": "6c941dcab0863a075bbc54d3cb4c3724ed1e020bc28e44dbb7068902896f176a"
+    },
+    {
+      "@type": "cr:FileObject",
+      "@id": "uid_map_v12_to_v13.csv",
+      "name": "uid_map_v12_to_v13.csv",
+      "description": "The 40 trial keys and paths renamed in 1.3.0 (Hammer2013 to Hamner2013).",
+      "contentUrl": "https://huggingface.co/datasets/kwonjoon/ForceBody/resolve/main/uid_map_v12_to_v13.csv",
+      "encodingFormat": "text/csv",
+      "sha256": "b8426a679c0413c6489f3b31cc315a64f7fd3604c86f4029d4a9c1a930359c02"
+    },
+    {
+      "@type": "cr:FileObject",
+      "@id": "marker_vertex_map.json",
+      "name": "marker_vertex_map.json",
+      "description": "Motion-capture marker name to SKEL skin vertex index used by the fit.",
+      "contentUrl": "https://huggingface.co/datasets/kwonjoon/ForceBody/resolve/main/marker_vertex_map.json",
+      "encodingFormat": "application/json",
+      "sha256": "7eff6ed14682f68c613b1c29b4f1ccd82d552080e277cbc5fef784eadaeb9b65"
+    },
+    {
+      "@type": "cr:FileObject",
+      "@id": "models/models_index.csv",
+      "name": "models/models_index.csv",
+      "description": "Index of the 756 OpenSim model files (source b3d file, processing pass and path).",
+      "contentUrl": "https://huggingface.co/datasets/kwonjoon/ForceBody/resolve/main/models/models_index.csv",
+      "encodingFormat": "text/csv",
+      "sha256": "69fdc999d627fb19d23c69ea9d0674f01d6942ec862a489f224f0a6282ba31ca"
+    },
+    {
+      "@type": "cr:FileObject",
+      "@id": "models/body_segments.csv",
+      "name": "models/body_segments.csv",
+      "description": "Segment masses and local centre-of-mass positions of every body of the 756 OpenSim models.",
+      "contentUrl": "https://huggingface.co/datasets/kwonjoon/ForceBody/resolve/main/models/body_segments.csv",
+      "encodingFormat": "text/csv",
+      "sha256": "1c6b55efd914f4b14061153ee63e2a6edbd4b0a5d0a7847c98b750b065317d30"
+    },
     {
       "@type": "cr:FileObject",
       "@id": "README.md",
@@ -226,16 +289,16 @@
       "description": "Dataset card.",
       "contentUrl": "https://huggingface.co/datasets/kwonjoon/ForceBody/resolve/main/README.md",
       "encodingFormat": "text/markdown",
-      "sha256": "3b408cbe34c1a2175930c9c4d8838c693d1a4e7de2bd3f2f86c4bd122d52be8a"
+      "sha256": "ed649ef1f9e3559d9f8b58146ed6bc9c4ac2767d9f92a6fed3a2e0174e572e07"
     },
     {
       "@type": "cr:FileObject",
       "@id": "DETAILS.md",
       "name": "DETAILS.md",
-      "description": "Full documentation: masks, conventions, known issues.",
+      "description": "Full documentation of masks, conventions and known issues.",
       "contentUrl": "https://huggingface.co/datasets/kwonjoon/ForceBody/resolve/main/DETAILS.md",
       "encodingFormat": "text/markdown",
-      "sha256": "d9038e42c5d2a3b0cd5b1e7724b054e815dc5310e565115f94700adc04be1b93"
+      "sha256": "e075125ee16eb51ab25a6d5391fa5de3b3ba28b51cf90a3d545b22d406471134"
     },
     {
       "@type": "cr:FileObject",
@@ -244,16 +307,16 @@
       "description": "Release history.",
       "contentUrl": "https://huggingface.co/datasets/kwonjoon/ForceBody/resolve/main/CHANGELOG.md",
       "encodingFormat": "text/markdown",
-      "sha256": "06b558385413550b600736dcdbd8672319172353d38569ee96e599c9102168a9"
+      "sha256": "8bf63ae1e17d57dd5c5d38e5e5f93d05e85b62c8415591a2097ddf0965bd99a6"
     },
     {
       "@type": "cr:FileObject",
       "@id": "MD5SUMS.txt",
       "name": "MD5SUMS.txt",
-      "description": "MD5 of every file in the release.",
+      "description": "MD5 of every data file and of README.md, DETAILS.md and CHANGELOG.md.",
       "contentUrl": "https://huggingface.co/datasets/kwonjoon/ForceBody/resolve/main/MD5SUMS.txt",
       "encodingFormat": "text/plain",
-      "sha256": "3b6e7529d7d95532446ca28b32b07de8d1280cc2a3d60f8f842718bd47919cd2"
+      "sha256": "0737a29f0226122572e8a2850b6948b6df52e87710c24fef1545b7b687379562"
     },
     {
       "@type": "cr:FileSet",
@@ -262,7 +325,7 @@
       "containedIn": {
         "@id": "huggingface-repo"
       },
-      "description": "One NumPy .npz per trial at trials/<study>/S<NN>/<trial_uid>.npz (10,386 files). Fields are listed in SCHEMA_v1.3.csv.",
+      "description": "One NumPy .npz per trial at trials/<study>/S<NN>/<trial_uid>.npz (10,386 files, 547 core fields each). Fields are listed in SCHEMA_v1.3.csv.",
       "encodingFormat": "application/x-npz",
       "includes": "trials/*/*/*.npz"
     },
@@ -273,7 +336,7 @@
       "containedIn": {
         "@id": "huggingface-repo"
       },
-      "description": "AddBiomechanics OpenSim models per subject, used for the torques.",
+      "description": "AddBiomechanics scaled OpenSim models, one per source b3d file and processing pass (passes 0 and 2), 756 files for 378 b3d files. The pass-2 model is the skeleton of the torque labels. models/models_index.csv maps each file to its b3d file and pass.",
       "encodingFormat": "application/gzip",
       "includes": "models/*/*.osim.gz"
     }
@@ -286,7 +349,7 @@
       "key": {
         "@id": "trials/trial_uid"
       },
-      "description": "One record per trial, from manifest.csv. Arrays (poses, betas, trans, grf_aligned, tau_selfconsistent, sigma_tau, frame masks) are in the trial npz at npz_path.",
+      "description": "One record per trial, from manifest.csv. Arrays (poses, poses_dyn, betas, trans, grf_aligned, tau_selfconsistent, sigma_tau, frame masks) are in the trial npz at npz_path.",
       "field": [
         {
           "@type": "cr:Field",
@@ -352,7 +415,7 @@
           "@type": "cr:Field",
           "@id": "trials/split_addb_val",
           "name": "split_addb_val",
-          "description": "Validation fold inside train.",
+          "description": "Benchmark split with the validation subjects (train, val, test). val holds 12 subjects of the split_addb train set.",
           "dataType": "sc:Text",
           "source": {
             "fileObject": {
@@ -487,7 +550,7 @@
           "@type": "cr:Field",
           "@id": "trials/health_status",
           "name": "health_status",
-          "description": "Health status of the subject.",
+          "description": "Health status of the subject (healthy, TKA for total knee arthroplasty, or stroke). Health information.",
           "dataType": "sc:Text",
           "source": {
             "fileObject": {
@@ -498,6 +561,21 @@
             }
           }
         },
+        {
+          "@type": "cr:Field",
+          "@id": "trials/health_detail",
+          "name": "health_detail",
+          "description": "Health detail of the subject (healthy, TKA left knee, TKA right knee, or stroke, right-sided hemiparesis). Health information.",
+          "dataType": "sc:Text",
+          "source": {
+            "fileObject": {
+              "@id": "manifest.csv"
+            },
+            "extract": {
+              "column": "health_detail"
+            }
+          }
+        },
         {
           "@type": "cr:Field",
           "@id": "trials/num_frames",
@@ -517,7 +595,7 @@
           "@type": "cr:Field",
           "@id": "trials/fps",
           "name": "fps",
-          "description": "Frame rate (Hz).",
+          "description": "Nominal release frame rate (Hz). Use fps_true_grid for time stamps.",
           "dataType": "sc:Float",
           "source": {
             "fileObject": {
@@ -528,6 +606,36 @@
             }
           }
         },
+        {
+          "@type": "cr:Field",
+          "@id": "trials/fps_native_capture",
+          "name": "fps_native_capture",
+          "description": "Capture rate of the source trial (Hz). It differs from fps in the 8,848 resampled trials.",
+          "dataType": "sc:Float",
+          "source": {
+            "fileObject": {
+              "@id": "manifest.csv"
+            },
+            "extract": {
+              "column": "fps_native_capture"
+            }
+          }
+        },
+        {
+          "@type": "cr:Field",
+          "@id": "trials/fps_true_grid",
+          "name": "fps_true_grid",
+          "description": "True rate of the release grid (Hz). Frame i is at time i / fps_true_grid.",
+          "dataType": "sc:Float",
+          "source": {
+            "fileObject": {
+              "@id": "manifest.csv"
+            },
+            "extract": {
+              "column": "fps_true_grid"
+            }
+          }
+        },
         {
           "@type": "cr:Field",
           "@id": "trials/duration_s",
@@ -547,7 +655,7 @@
           "@type": "cr:Field",
           "@id": "trials/has_uncertainty",
           "name": "has_uncertainty",
-          "description": "1 if sigma_tau is available.",
+          "description": "True if Monte Carlo sigma_tau was computed. False for 4 trials shorter than one 6 Hz period, where sigma_tau is 0.",
           "dataType": "sc:Boolean",
           "source": {
             "fileObject": {
@@ -607,7 +715,7 @@
           "@type": "cr:Field",
           "@id": "trials/eval_stratum",
           "name": "eval_stratum",
-          "description": "Evaluation stratum of the benchmark.",
+          "description": "Evaluation stratum of the benchmark (healthy, or pathological for the TKA and stroke subjects).",
           "dataType": "sc:Text",
           "source": {
             "fileObject": {