field,dtype,shape,unit,frame,version_added,description,part,source,stage
activity,str,[],-,-,1.2,"metadata: canonical snake_case activity (walk, run, stand, stand_one_leg, chair_rise_cycle, drop_jump, yoga_tree, step_down, stairs_down, sequence, ...); activity_sequence = parts joined by "">"" (e.g. walk>stand>stand_wide; the activity itself for a single activity); activity_modifier = a_pose / t_pose / idle / arms_raised / asymmetric / weak_legs from the N6 name. Han2023 ""stairs-down"" (one step down from a step, source_task step_down) is step_down; Tiziana2019 stairs-down stays stairs_down. The 2 Han ""walk"" trials of a step_down capture stay walk (not visually re-checked)",core,metadata,finalize
activity_class_v1,str,[],-,-,1.2,v1 activity class (renamed from activity_class; superseded by activity_name),core,build,finalize
activity_condition,str,[],-,-,1.1,v1.1 activity condition (N4 audit; kept for the QC static/pause rules),core,build,finalize
activity_confidence,str,[],-,-,1.1,confidence of the activity label (metadata_v12),core,build,finalize
activity_fix,str,[],-,-,1.2,metadata: v1.1 values and the fix id,core,metadata,finalize
activity_modifier,str,[],-,-,1.2,"metadata: canonical snake_case activity (walk, run, stand, stand_one_leg, chair_rise_cycle, drop_jump, yoga_tree, step_down, stairs_down, sequence, ...); activity_sequence = parts joined by "">"" (e.g. walk>stand>stand_wide; the activity itself for a single activity); activity_modifier = a_pose / t_pose / idle / arms_raised / asymmetric / weak_legs from the N6 name. Han2023 ""stairs-down"" (one step down from a step, source_task step_down) is step_down; Tiziana2019 stairs-down stays stairs_down. The 2 Han ""walk"" trials of a step_down capture stay walk (not visually re-checked)",core,metadata,finalize
activity_name,str,[],-,-,1.1,canonical activity name (metadata_v12),core,build,finalize
activity_proposed,str,[],-,-,1.1,v1.1 proposed activity (N4 audit),core,build,finalize
activity_recovered,str,[],-,-,1.1,"activity class (walking, running, standing, ...; metadata_v12)",core,build,finalize
activity_sequence,str,[],-,-,1.2,"metadata: canonical snake_case activity (walk, run, stand, stand_one_leg, chair_rise_cycle, drop_jump, yoga_tree, step_down, stairs_down, sequence, ...); activity_sequence = parts joined by "">"" (e.g. walk>stand>stand_wide; the activity itself for a single activity); activity_modifier = a_pose / t_pose / idle / arms_raised / asymmetric / weak_legs from the N6 name. Han2023 ""stairs-down"" (one step down from a step, source_task step_down) is step_down; Tiziana2019 stairs-down stays stairs_down. The 2 Han ""walk"" trials of a step_down capture stay walk (not visually re-checked)",core,metadata,finalize
addb_belt_speed_mps,<f8,[],m/s,-,1.2,median AddB toes velocity on frames with foot GRF >= 4 N/kg (0 below 0.15 m/s),core,qc,finalize
addb_heading_class,str,[],-,-,1.2,"'+x' | '-x' | '+z' | '-z' within 20 deg, else 'oblique'; q_gt pelvis tilt/list are anatomical only for '+x'",core,qc,finalize
addb_heading_deg,<f8,[],deg,-,1.2,"heading of the mean AddB forward vector Ry(q_gt[:,2]) x, atan2(z, x) in the AddB lab frame (0 = +x, 90 = +z)",core,qc,finalize
addb_joint_names,str,[20],-,-,1.2,AddB joint order of addb_joints_*,core,build,finalize
addb_joints_dyn,<f4,"[T,20,3]",m,skel_world,1.2,"AddB joint centres of the pass-2 model at the dyn fit target (q_gt, or the repaired target where dyn_target_source != 0; the fit target of poses_dyn); = addb_joints_kin where skel_fit_target_dyn == 'kin_retimed'",core,build,finalize
addb_joints_kin,<f4,"[T,20,3]",m,skel_world,1.2,AddB joint centres of the pass-0 model at q_kin (the fit target of poses),core,build,finalize
addb_limit_mask_dyn,|b1,"[T,11]",-,-,1.2,same for the dyn target: union of the pass-2 and pass-0 on-bound frames (pass 2 can leave the exact bound while inheriting the parked pose); = addb_limit_mask_kin on kin_retimed trials,core,build,finalize
addb_limit_mask_dyn_source,str,[],-,-,1.2,'pass2|pass0' (or 'kin_retimed'),core,build,finalize
addb_limit_mask_kin,|b1,"[T,11]",-,-,1.2,"per frame: the kin fit target (q_kin) DOF family sits on its AddB joint limit (parking), columns addb_limit_mask_names",core,build,finalize
addb_limit_mask_names,str,[11],-,-,1.2,"DOF families of addb_limit_mask_*: pro_sup_r/l, hip_r/l, knee_r/l, ankle_r/l, elbow_r/l, arm_l",core,build,finalize
addb_skel_joint_map,<i4,"[15,2]",-,-,1.2,"(AddB joint index, SKEL joint index) pairs used as direct fit targets",core,build,finalize
addb_skeleton_mass_kg,<f8,[],kg,-,1.2,mass of the b3d dynamics-pass skeleton as read (before any v1.2 rescale),core,qc,finalize
age,<f4,[],years,-,1.0,"age of the person, NaN = unknown (metadata_v12)",core,build,finalize
arm_event_ok_dyn,|b1,"[T,2]",-,-,1.2,"same for poses_dyn (the 2026-10-07 rows of re-fitted dyn trials are kept, conservative)",core,build,finalize
arm_event_ok_kin,|b1,"[T,2]",-,-,1.2,"columns (left, right): False on frames of an arm discontinuity of poses (2026-10-07 arm-jump diagnosis classes except A6, and the S20 re-scan of the final fits), +-0.1 s. AND into arm-DOF loss masks",core,build,finalize
b3d_path,str,[],-,-,1.0,AddBiomechanics b3d file (relative to the AddB dataset root),core,build,finalize
belt_mode,str,[],-,-,1.2,"metadata: constant / perturbed / mixed (Moore perturbation_onset or _offset) / start_or_stop (Moore) / none (off treadmill, or a standing segment on a stopped belt)",core,metadata,finalize
belt_perturbation,str,[],-,-,1.2,"metadata: treadmill: none / perturbed / unperturbed / perturbation_onset / perturbation_offset / start_or_stop (Moore2015); empty off-treadmill. Moore walking segments of the 31 captures with a confirmed phase fit are labelled from the fitted 480-s perturbation window (perturbed = inside, unperturbed = outside, onset/offset = the window starts/ends inside the segment); the 29 walking segments of subject6#T042 (fit not confirmed) keep the N6 label",core,metadata,finalize
belt_vx_mps,<f8,[],m/s,-,1.2,"that median stance-foot velocity (SKEL frame x, z) on treadmill trials, else 0",core,qc,finalize
belt_vz_mps,<f8,[],m/s,-,1.2,"that median stance-foot velocity (SKEL frame x, z) on treadmill trials, else 0",core,qc,finalize
betas,<f4,[10],-,-,1.0,"SKEL shape, one per person (v1.2 shape fit, identical in all trials of a person)",core,build,finalize
bmi,<f4,[],kg/m^2,-,1.0,mass_kg / height_m^2 (recomputed in v1.2),core,build,finalize
cadence_factor,<f8,[],-,-,1.2,metadata: vanderZee2022 Table 2 multipliers of f* and s* (after the D8 31/33 swap); NaN where self-selected or not vanderZee,core,metadata,finalize
cadence_spm,<f8,[],-,-,1.2,"metadata: steps per minute from sign flips of the fore-aft foot separation (AddB-model foot midpoints, hysteresis 0.10 m); NaN with < 3 flips (short trials)",core,metadata,finalize
cadence_spm_grf,<f8,[],steps/min,-,1.2,120 / stride_time_s_med (walking/running with usable force labels); NaN where qc_gait_event_inconsistent,core,finalize,finalize
caption_activity,str,[],-,-,1.2,"metadata: English verb phrase from activity, activity_modifier, activity_sequence, gait_style, support_side and path (e.g. ""stands on the left leg"", ""walks on the toes"", ""walks, then stands still"")",core,metadata,finalize
caption_condition_key,str,[],-,-,1.2,metadata: study-free key of the caption-visible condition: activity_sequence|activity_modifier|gait_style|support_side|path|setting|incline_label|belt_mode|speed_level,core,metadata,finalize
caption_gait_style,str,[],-,-,1.2,"metadata: caption phrases: ""on a treadmill"" / ""overground"" / ""in place"" / ""using a chair""; ""uphill at a 6% grade"" / ""downhill at a 4% grade"" (empty for level or unknown); Moore belt phrases per segment for confirmed captures and a capture-level phrase for subject6#T042; gait-style and support-side phrases",core,metadata,finalize
caption_height_m,<f8,[],-,-,1.2,"metadata: sex (never skel_gender; vanderZee2022 unknown, D15) and height_m",core,metadata,finalize
caption_incline,str,[],-,-,1.2,"metadata: caption phrases: ""on a treadmill"" / ""overground"" / ""in place"" / ""using a chair""; ""uphill at a 6% grade"" / ""downhill at a 4% grade"" (empty for level or unknown); Moore belt phrases per segment for confirmed captures and a capture-level phrase for subject6#T042; gait-style and support-side phrases",core,metadata,finalize
caption_mass_kg,<f8,[],kg,-,1.2,person mass for captions (D3prime; NaN = omitted),core,build,finalize
caption_ok,|b1,[],-,-,1.2,"metadata: label-level caption validity: caption_activity present, not eval_exclude_duplicate, and no KIN_COMPRESSED or ACTIVITY_MOTION_MISMATCH conflict code (SPEED_NOMINAL_DOUBTFUL does not block since the caption speed is measured). AND it with the QC usable mask at load time",core,metadata,finalize
caption_perturbation,str,[],-,-,1.2,"metadata: caption phrases: ""on a treadmill"" / ""overground"" / ""in place"" / ""using a chair""; ""uphill at a 6% grade"" / ""downhill at a 4% grade"" (empty for level or unknown); Moore belt phrases per segment for confirmed captures and a capture-level phrase for subject6#T042; gait-style and support-side phrases",core,metadata,finalize
caption_setting,str,[],-,-,1.2,"metadata: caption phrases: ""on a treadmill"" / ""overground"" / ""in place"" / ""using a chair""; ""uphill at a 6% grade"" / ""downhill at a 4% grade"" (empty for level or unknown); Moore belt phrases per segment for confirmed captures and a capture-level phrase for subject6#T042; gait-style and support-side phrases",core,metadata,finalize
caption_sex,str,[],-,-,1.2,"metadata: sex (never skel_gender; vanderZee2022 unknown, D15) and height_m",core,metadata,finalize
caption_speed_mps,<f8,[],-,-,1.2,"metadata: measured speed_mps rounded to 0.01 m/s and its speed_method (also for Carter walk_slow, D9); NaN / empty without a measured speed",core,metadata,finalize
caption_speed_source,str,[],-,-,1.2,"metadata: measured speed_mps rounded to 0.01 m/s and its speed_method (also for Carter walk_slow, D9); NaN / empty without a measured speed",core,metadata,finalize
caption_support_side,str,[],-,-,1.2,"metadata: caption phrases: ""on a treadmill"" / ""overground"" / ""in place"" / ""using a chair""; ""uphill at a 6% grade"" / ""downhill at a 4% grade"" (empty for level or unknown); Moore belt phrases per segment for confirmed captures and a capture-level phrase for subject6#T042; gait-style and support-side phrases",core,metadata,finalize
capture_id,str,[],-,-,1.2,"metadata: one AddB capture split into contiguous segments; prev/next = the contiguous neighbour (kinematics are NOT continuous at junctions, T3 verifier)",core,metadata,finalize
chair_contact,|b1,[T],-,-,1.2,"S25: subject seated on the instrumented chair: hysteresis of grf_contact_label (on > 0.05 BW_dyn, off < 0.03 BW_dyn, >= 0.05 s) on the vertical force of the contact_bodies 'pelvis' channel (Falisse2017); not ANDed with grf_frame_ok (the seated phase is mostly outside grf_valid_mask). All False without a chair channel. Not the manifest metadata column chair_contact (trial-level protocol flag)",core,qc,finalize
condition_key,str,[],-,-,1.2,"metadata: person-free protocol condition: study|activity|activity_sequence|activity_modifier|gait_style|support_side|path|setting|speed_nominal_kind|speed (speed_rel_factor for relative_to_norm, else speed_nominal_mps)|incline_grade_pct|belt_mode|cadence_factor|step_length_factor|step_width_target_m (""na"" = empty); vanderZee2022: 33 trial numbers -> 31 keys (trials 20-22 identical)",core,metadata,finalize
contact_bodies,str,[C],-,-,1.0,"contact order: 0 = calcn_r, 1 = calcn_l, 2 = pelvis (Falisse2017 chair)",core,build,finalize
contact_status,|i1,"[T,C]",code,-,1.2,"0 unloaded, 1 reliable CoP, 2 CoP filled, 3 untrusted CoP, 4 swing noise zeroed, 5 tail blanked, 6 airborne noise zeroed, 7 airborne but loaded, 8 crossover, 9 free moment zeroed",core,build,finalize
contact_torque,<f4,"[T,C,3]",N m,addb_lab,1.0,free moment per contact (v1.2 cleaned),core,build,finalize
contact_torque_aligned,<f4,"[T,C,3]",N m,skel_world,1.0,flip_xz(contact_torque),core,build,finalize
cop,<f4,"[T,C,3]",m,addb_lab,1.0,centre of pressure per contact (v1.2 cleaned); 0 where the force is 0,core,build,finalize
cop_align_gender,str,[],-,-,1.0,= skel_gender (v1.2),core,build,finalize
cop_align_gender_fallback,|b1,[],-,-,1.0,v1 field,core,build,finalize
cop_align_method,str,[],-,-,1.0,'flip_xz_v2',core,build,finalize
cop_align_offset,<f4,[3],m,skel_world,1.0,zero since v1.1,core,build,finalize
cop_aligned_foot,<f4,"[T,C,3]",m,skel_world,1.0,flip_xz(cop),core,build,finalize
cop_foot_method,str,[],-,-,1.0,cop_aligned_foot method,core,build,finalize
cop_foreaft_aligned,|b1,[],-,-,1.0,v1 field,core,build,finalize
cop_frame_ok,|b1,"[T,2]",-,-,1.2,"contact_status in {1,2,9} (trusted CoP) AND foot vertical GRF > 50 N AND GRF-OK(CoP) (grf_frame_ok without contact_status 3 frames; = the 1.2.0 definition) AND NOT qc_cop_fixed_point AND NOT qc_cop_plane_mismatch. contact_status absent (v1.1) -> the status test is skipped (assumed trusted)",core,qc,finalize
cop_frame_ok_frac,<f8,[],fraction,-,1.2,mean of cop_frame_ok,core,qc,finalize
cop_off_foot,|b1,"[T,2]",-,-,1.2,foot vertical GRF > 0.2 BW_dyn AND GRF-OK AND horizontal distance of the CoP to the AddB calcn->toes segment > 0.15 m,core,qc,finalize
direct_dof_mask,|b1,[37],-,-,1.0,DOFs that receive MC input noise,core,build,finalize
dof_names,str,[37],-,-,1.0,"AddB (OpenSim Rajagopal) DOF names, columns of q_* and tau_*",core,build,finalize
dof_names_opensim,str,[37],-,-,1.2,same list as dof_names (OpenSim coordinate names),core,build,finalize
dt,<f4,[],s,-,1.0,1 / fps,core,build,finalize
duration_s,<f4,[],s,-,1.0,"num_frames / fps (nominal rate, v1 convention); the time span of the release grid is (num_frames - 1) / fps_true_grid, about one frame shorter",core,build,finalize
duty_factor_l,<f8,[],-,-,1.2,"median stance time / stride time, left foot; NaN where qc_gait_event_inconsistent",core,finalize,finalize
duty_factor_r,<f8,[],-,-,1.2,"median stance time / stride time, right foot; NaN where qc_gait_event_inconsistent",core,finalize,finalize
dyn_arm_spliced,|b1,"[T,2]",-,-,1.2,"columns (left, right): dyn target arm DOFs come from the kin re-solved or IK arm (bits 2, 8, 32), so poses_dyn arms are not pass 2 there",core,build,finalize
dyn_root_jump_ok,|b1,[T],-,-,1.2,False on frames of class A6 (pass-2-only whole-body jump): drop the whole poses_dyn / trans_dyn frame,core,build,finalize
dyn_target_modified_frac,<f8,[],-,-,1.2,fraction of frames with any dyn_target_source bit,core,build,finalize
dyn_target_source,|u1,"[T,37]",bits,-,1.2,"per frame and AddB DOF: source of the dyn fit target when it is not q_gt. 1 kin q (GRF-invalid window / pass-2 root deviation), 2 left-arm splice from kin, 4 kin on pass-0 IK outlier, 8 left-arm position IK, 16 AddB acceleration-minimizing filter, 32 other difference from q_gt (R1 arm splice). 0 = target equals q_gt. All 0 on kin_retimed trials",core,build,finalize
eval_exclude_duplicate,|b1,[],-,-,1.2,metadata: exclude from evaluation and from sampling weights: qc_duplicate_drop (156) and the 8 Han2023 trimmed copies re-exported in a *_full trial (55-99 % of frames shared),core,metadata,finalize
eval_exclude_duplicate_reason,str,[],-,-,1.2,metadata: exclude from evaluation and from sampling weights: qc_duplicate_drop (156) and the 8 Han2023 trimmed copies re-exported in a *_full trial (55-99 % of frames shared),core,metadata,finalize
eval_grf_frame_ok,|b1,[T],-,-,1.2,"tau_frame_ok AND kin_frame_ok / grf_frame_ok AND kin_frame_ok (S19 F3: pelvis-flip frames excluded from evaluation, as in mc_frame_ok); combine with the trial flags eval_torque_trial_ok / eval_grf_trial_ok",core,qc,finalize
eval_grf_trial_ok,|b1,[],-,-,1.2,"force_label_usable AND NOT (qc_duplicate_drop OR qc_overlap_pair_drop OR static_calibration OR transition OR pause) AND NOT qc_grf_single_channel AND NOT qc_mass_dyn_vs_grf_flag (S19 F3 / U2) AND qc_grf_loaded_eval_frames > 0 (1.2.1, F1 R01b: at least one eval_grf_frame_ok frame with a foot vertical GRF > 20 N). Tiziana2019 overlap pairs are kept (they share no GRF-valid frame)",core,qc,finalize
eval_motion_trial_ok,|b1,[],-,-,1.2,force_label_usable AND NOT (qc_duplicate_drop OR qc_overlap_pair_drop OR static_calibration OR transition OR pause) AND NOT qc_overlap_motion_drop. qc_overlap_pair_drop = eval_exclude_duplicate AND NOT qc_duplicate_drop (the 8 Han2023 trimmed copies); qc_overlap_motion_drop = the Tiziana2019 overlapping-capture pair member with fewer grf_valid_mask frames (200 trials); both built by build_v12.py a,core,qc,finalize
eval_stratum,str,[],-,-,1.2,"'pathological' when health_status is not 'healthy' (D18: reported, never excluded), else 'healthy'",core,qc,finalize
eval_tau_frame_ok,|b1,[T],-,-,1.2,"tau_frame_ok AND kin_frame_ok / grf_frame_ok AND kin_frame_ok (S19 F3: pelvis-flip frames excluded from evaluation, as in mc_frame_ok); combine with the trial flags eval_torque_trial_ok / eval_grf_trial_ok",core,qc,finalize
eval_torque_trial_ok,|b1,[],-,-,1.2,"torque_label_usable (includes NOT qc_grf_single_channel) AND NOT (qc_duplicate_drop OR qc_overlap_pair_drop OR static_calibration OR transition OR pause) AND qc_root_residual_bw <= 0.1 BW (NaN treated as passing) AND NOT qc_mass_dyn_vs_grf_flag (S19 U2) AND qc_grf_loaded_eval_frames > 0 AND qc_tau_loaded_eval_frames > 0 (1.2.1, F1 R01b)",core,qc,finalize
extreme_tau_flag,|b1,[],-,-,1.0,v1.1 definition: tau_interior_absmax > extreme_tau_thresh; deprecated,core,build,finalize
extreme_tau_flag_gt,|b1,[],-,-,1.2,tau_interior_absmax_gt > 3000,core,build,finalize
extreme_tau_flag_mu,|b1,[],-,-,1.2,tau_interior_absmax_mu > 3000,core,build,finalize
extreme_tau_flag_sc,|b1,[],-,-,1.2,tau_interior_absmax_sc > 3000,core,build,finalize
extreme_tau_thresh,<f8,[],N m,-,1.0,3000,core,build,finalize
facing_hipline_deg,<f8,[],deg,-,1.2,"circular mean of the SKEL hip-line facing (forward = up x (femur_r - femur_l)), atan2(x, z) in the SKEL frame (0 = +z)",core,qc,finalize
fit_event_groups,str,[7],-,-,1.2,"pelvis (pose 0-2 + trans), leg_r (3-9), leg_l (10-16), spine (17-22), head (23-25), arm_r (26-35), arm_l (36-45)",core,build,finalize
fit_event_ok_dyn,|b1,"[T,7]",-,-,1.2,same for poses_dyn (events measured against its own repaired target),core,build,finalize
fit_event_ok_kin,|b1,"[T,7]",-,-,1.2,"per joint group (fit_event_groups): False on frames of a temporal event of poses found by the S20 re-scan of the final fits (fit-only spike, target-inherited spike, one-frame DOF step) or of a one-frame DOF step > pi (Euler branch flip), +-0.1 s; S25: head group also False around a one-frame head DOF step > 0.5 rad of the final poses after the head post-fit",core,build,finalize
fix_applied,str,[],-,-,1.1,"'|'-joined change codes: v1.1 codes, then v1.2 codes prefixed 'v12_' (CHANGELOG)",core,build,finalize
floor_height_m,<f8,[],m,skel_world,1.2,"median CoP height over trusted firm-stance frames (contact_status 1, vertical force > 0.5 BW_dyn, grf_valid_mask), NaN with fewer than 30 such frames. Derived from the force plates (CoP). On stairs / step activities it is the height of the loaded step surface, not of the floor. No floor normal is shipped (S19 U1)",core,build,finalize
floor_tilt_x_deg,<f8,[],deg,-,1.2,"least-squares plane y = a + b_x x + b_z z through firm-stance CoPs (SKEL frame, >= 30 frames); 1-D fit on the wider axis when the spread is < 0.05 m (x) / 0.02 m (z)",core,qc,finalize
floor_tilt_x_kin_deg,<f8,[],deg,-,1.2,"same least-squares floor-plane fit as floor_tilt_x_deg/floor_tilt_z_deg, but the vertical value is the SKEL kinematic sole height (Gs['sole']) instead of the CoP height, at the same (frame, foot) firm-stance samples",core,qc,finalize
floor_tilt_z_deg,<f8,[],deg,-,1.2,"least-squares plane y = a + b_x x + b_z z through firm-stance CoPs (SKEL frame, >= 30 frames); 1-D fit on the wider axis when the spread is < 0.05 m (x) / 0.02 m (z)",core,qc,finalize
floor_tilt_z_kin_deg,<f8,[],deg,-,1.2,"same least-squares floor-plane fit as floor_tilt_x_deg/floor_tilt_z_deg, but the vertical value is the SKEL kinematic sole height (Gs['sole']) instead of the CoP height, at the same (frame, foot) firm-stance samples",core,qc,finalize
foot_geometry_source,str,[],-,-,1.2,"'addb_fk_pass2' when the b3d was available, else 'none' (AddB-model fields NaN)",core,qc,finalize
footwear,str,[],-,-,1.2,"metadata: shod (Carter, Moore), barefoot (Tiziana), unknown elsewhere",core,metadata,finalize
force_invalid_from_frame,<i4,[],frame,-,1.2,"force invalid from this frame on (partial invalidation), -1 otherwise",core,build,finalize
force_label_usable,|b1,[],-,-,1.2,"not desync, where desync = the external input qc_skel_grf_desync_final when present (capture-propagated, qc_postprocess_v12.py; injected by the S18 build) else the own-trial qc_skel_grf_desync. Desync trials get grf_frame_ok, grf_frame_strict_ok, tau_frame_ok, mc_frame_ok, grf_contact_label and cop_frame_ok (and the eval_* frame aliases) forced to all-False",core,qc,finalize
force_label_usable_s8b,|b1,[],-,-,1.2,S8b verdict on the force labels (False: failed or stopped re-timing),core,build,finalize
forces_code_sha1,str,[],-,-,1.2,sha1 of the forces code,core,build,finalize
forces_fix_ids,str,[K],-,-,1.2,"forces workstream fix ids applied to this trial (F_contact_rule, ...)",core,build,finalize
forces_params_sha1,str,[],-,-,1.2,sha1 of this trial's forces decision parameters,core,build,finalize
forces_version,str,[],-,-,1.2,forces workstream version,core,build,finalize
fps,<f4,[],Hz,-,1.0,nominal release frame rate (use fps_true_grid for time stamps),core,build,finalize
fps_class,str,[],-,-,1.2,"'61hz' (fps <= 70), '100hz' (< 150), '201hz' (< 220), '250hz' (>= 220)",core,qc,finalize
fps_native_capture,<f4,[],Hz,-,1.0,native capture rate of the b3d trial,core,build,finalize
fps_true_grid,<f8,[],Hz,-,1.2,(T - 1) fps_native_capture / (N_native - 1): the true rate of the release grid; t_i = i / fps_true_grid. Equals mc_fps_noise,core,build,finalize
gait_style,str,[],-,-,1.2,metadata: normal / toe_walking / heel_walking / trunk_sway / medial_thrust / walking_poles / smooth / bouncy / crouch / turning,core,metadata,finalize
grf,<f4,"[T,C,3]",N,addb_lab,1.0,ground reaction force per contact (v1.2 cleaned),core,build,finalize
grf_aligned,<f4,"[T,C,3]",N,skel_world,1.0,flip_xz(grf),core,build,finalize
grf_complete_strides,<i8,[],strides,-,1.2,min over feet of (number of contact onsets - 1); onset = foot vertical GRF crosses 0.05 BW_dyn upward with both frames GRF-OK,core,qc,finalize
grf_contact_label,|b1,"[T,2]",-,-,1.2,"foot contact from GRF: hysteresis on foot vertical GRF (on > 0.05 BW_dyn, off < 0.03 BW_dyn, >= 0.05 s) AND GRF-OK; contact 0 = right, 1 = left",core,qc,finalize
grf_contact_ok,|b1,"[T,C]",-,-,1.2,"force label internally consistent on this frame and contact (forces workstream; not the same as ""force measured""). S25: False on every frame of a non-foot channel that carries zero force in the whole trial (unused channel, qc_extra_channel_zero). 1.2.1 (F1, R01a): also True on GRF-valid interior frames with contact_status 3 (loaded, CoP projected and not trusted): GRF validity no longer depends on CoP trust; CoP trust gates cop_frame_ok and tau_frame_ok only. In 1.2.0 these frames were False, so on fixed-point-CoP trials grf_frame_ok kept only unloaded frames",core,build,finalize
grf_contact_ok_source,str,[],-,-,1.2,"'forces_workstream' if grf_contact_ok was in the trial, else 'absent_all_true'",core,qc,finalize
grf_contact_state,|i1,"[T,2]",enum,-,1.2,"S19 F1: 1 = GRF-OK and in contact (grf_contact_label: hysteresis stance AND grf_frame_ok); 0 = GRF-OK (grf_frame_ok), not in contact; -1 = unknown: grf_valid_mask False, or GRF-valid but not grf_frame_ok (endpoint, tail, contact-inconsistent frame, every frame of a force-unusable trial). Before S19, GRF-valid frames outside grf_frame_ok were coded 0 even under load",core,qc,finalize
grf_foot_ok,|b1,"[T,2]",-,-,1.2.1,"1.2.1 (R05): per-foot GRF mask, columns (right, left): grf_frame_ok AND no foot flagged grf_force_in_air, grf_unmeasured_stance or cop_off_foot (frame level: likely plate misassignment, both feet dropped) AND foot k not grf_zero_force_grounded (per foot: a zero-force-grounded OTHER foot keeps the loaded foot). grf_foot_ok.all(1) == grf_frame_strict_ok",core,qc,finalize
grf_foot_ok_frac_l,<f8,[],fraction,-,1.2.1,"grf_foot_ok_frac_r = mean of grf_foot_ok[:, 0] (right foot); grf_foot_ok_frac_l = mean of grf_foot_ok[:, 1] (left foot)",core,qc,finalize
grf_foot_ok_frac_r,<f8,[],fraction,-,1.2.1,"grf_foot_ok_frac_r = mean of grf_foot_ok[:, 0] (right foot); grf_foot_ok_frac_l = mean of grf_foot_ok[:, 1] (left foot)",core,qc,finalize
grf_force_in_air,|b1,"[T,2]",-,-,1.2,"foot k in mid-swing (AddB: calcn > firm-stance height + 0.05 m AND toes > + 0.01 m AND calcn moving forward relative to the pelvis > 1.0 m/s along the gait direction; OR lowest foot point > firm height + 0.15 m) AND GRF-OK AND foot vertical GRF > 0.10 BW_dyn, runs >= 0.05 s. 1.2.1 (F1): GRF-OK here = grf_frame_ok (incl. contact_status 3 frames); the scalar qc_grf_*_frac keeps GRF-OK(CoP)",core,qc,finalize
grf_frame_ok,|b1,[T],-,-,1.2,"grf_valid_mask AND not frame 0 / T-1 AND not grf_tail_blank AND all(grf_contact_ok[t, used channels]) (grf_contact_ok from the forces output; absent = all True; S25: a non-foot channel with zero force in the whole trial, e.g. the Falisse2017 chair channel on walk / stairs, is unused and left out). 1.2.1 (F1, R01a): grf_contact_ok includes contact_status 3 frames (loaded, CoP projected / not truste",core,qc,finalize
grf_frame_ok_frac,<f8,[],fraction,-,1.2,mean of grf_frame_ok,core,qc,finalize
grf_frame_strict_ok,|b1,[T],-,-,1.2,"grf_frame_ok AND no foot flagged by grf_force_in_air, grf_unmeasured_stance or cop_off_foot",core,qc,finalize
grf_frame_strict_ok_frac,<f8,[],fraction,-,1.2,mean of grf_frame_strict_ok (NaN-free; equals grf_frame_ok_frac without b3d),core,qc,finalize
grf_missing_reason,|i1,[T],enum,-,1.2,AddB MissingGRFReason of the nearest native frame round(i (N-1)/(T-1)) (same map as grf_valid_mask); 0 = notMissingGRF; names in qc_v12.MISSING_GRF_REASONS,core,qc,finalize
grf_onset_censored,|b1,"[T,2]",-,-,1.2,"a foot's first hysteresis stance run is marked entirely True when it starts at frame <= 2 with foot vertical GRF at frame 1 > 0.10 BW_dyn (the capture very likely starts mid-contact, so frame 0 is not a real swing frame). Verified motivating case: Tiziana2019 single-plate-stance trials (579 stance runs, FU3-4); this S15a build found it also triggers, at lower rates, in several other (mostly contin",core,qc,finalize
grf_retimed_change_frame,<i4,[],frame,-,1.2,"first shifted frame of a change-point trial, -1 otherwise",core,build,finalize
grf_retimed_guard,|b1,[T],-,-,1.2,change-point guard band (force invalid),core,build,finalize
grf_retimed_ms,<f4,[T],ms,-,1.2,time shift applied to the force stream at each frame (0 = not shifted),core,build,finalize
grf_tail_blank,|b1,[T],-,-,1.2,frame blanked as a saturated tail artefact (contact_status 5),core,build,finalize
grf_tail_start,<i8,[],frame,-,1.2,first frame of grf_tail_blank (-1 = none),core,qc,finalize
grf_unmeasured_stance,|b1,"[T,2]",-,-,1.2,"AddB lowest foot point within 0.03 m of its firm-stance height AND belt-relative horizontal speed < 0.30 m/s AND GRF-OK AND foot vertical GRF < 0.05 BW_dyn, runs >= 0.10 s. 1.2.1 (F1): GRF-OK here = grf_frame_ok (incl. contact_status 3 frames); the scalar qc_grf_*_frac keeps GRF-OK(CoP)",core,qc,finalize
grf_valid_frac,<f8,[],fraction,-,1.2,"mean of grf_valid_mask (the AddB notMissingGRF flag), unchanged v1.1 definition",core,qc,finalize
grf_valid_mask,|b1,[T],-,-,1.0,AddB notMissingGRF on the release grid; on re-timed trials the shifted mask,core,build,finalize
grf_valid_mask_plate,|b1,[T],-,-,1.2,Carter2023 downhill: frames whose pass-2 force equals the plate force; all False elsewhere (see grf_valid_mask_plate_available),core,build,finalize
grf_valid_mask_plate_available,|b1,[],-,-,1.2,grf_valid_mask_plate was computed (Carter2023 downhill only),core,build,finalize
grf_valid_mask_plate_ok,|b1,[],-,-,1.2,grf_valid_mask_plate covers >= 90% of interior frames,core,build,finalize
grf_zero_force_grounded,|b1,"[T,C]",-,-,1.2,AddB foot on its ground level with zero force on a GRF-valid frame,core,build,finalize
has_uncertainty,|b1,[],-,-,1.0,"MC shipped as an uncertainty estimate (False: trial shorter than one 6 Hz period, sigma_tau == 0)",core,build,finalize
head_fit_json,str,[],-,-,1.2,"head post-fit record of poses (gauge, calibration trials, markers, at-limit fractions; '' when not head_postfit)",core,build,finalize
head_fit_json_dyn,str,[],-,-,1.2,head post-fit record of poses_dyn,core,build,finalize
head_postfit,|b1,[],-,-,1.2,head DOFs 23-25 and joint 13 come from the head post-fit to >= 2 head markers (both pose sets),core,build,finalize
heading_canonical_deg,<f8,[],deg,-,1.2,alias of heading_skel_deg: the ONE heading field a T2M user should read (SKEL frame). addb_heading_deg/addb_heading_class remain the separate AddB-lab-frame convention,core,qc,finalize
heading_canonical_frame,str,[],-,-,1.2,'skel_world',core,qc,finalize
heading_skel_deg,<f8,[],deg,-,1.2,"circular mean of SKEL poses[:,2]",core,qc,finalize
health_detail,str,[],-,-,1.2,"metadata: TKA right knee (JW, DM), TKA left knee (SC, PS; PS conflicts with the b3d tag Right), stroke with right-sided hemiparesis (Li2021)",core,metadata,finalize
health_status,str,[],-,-,1.1,"healthy / pathological group of the person (metadata_v12; sensitive, DETAILS.md ""Health status and sensitive information"")",core,build,finalize
height_m,<f4,[],m,-,1.0,recorded height,core,build,finalize
incline_grade_pct,<f8,[],-,-,1.2,"metadata: protocol grade: Carter flat and walking +1 %, uphill +6 %, downhill -4 %; overground walkways 0 %; NaN where not reported. incline_label (S16, NAMES_ACTIVITY-3): level / uphill / downhill / unknown (treadmill grade not reported), empty where not applicable (non-locomotion, stairs); Carter +1 % is level (was ""flat""), ""level_not_reported"" became unknown",core,metadata,finalize
incline_label,str,[],-,-,1.2,"metadata: protocol grade: Carter flat and walking +1 %, uphill +6 %, downhill -4 %; overground walkways 0 %; NaN where not reported. incline_label (S16, NAMES_ACTIVITY-3): level / uphill / downhill / unknown (treadmill grade not reported), empty where not applicable (non-locomotion, stairs); Carter +1 % is level (was ""flat""), ""level_not_reported"" became unknown",core,metadata,finalize
is_treadmill_kinematic,|b1,[],-,-,1.2,|median stance-foot velocity| > 0.15 m/s,core,qc,finalize
joint_names,str,[24],-,-,1.2,SKEL joint order of joints / joints_dyn (index 1 = femur_r),core,build,finalize
joints,<f4,"[T,24,3]",m,skel_world,1.2,SKEL joint centres of poses (order joint_names); joint 13 (head) from the head post-fit where head_postfit,core,build,finalize
joints_dyn,<f4,"[T,24,3]",m,skel_world,1.2,SKEL joint centres of poses_dyn,core,build,finalize
junction_dt_s,<f8,[],-,-,1.2,"metadata: time from the last frame of prev_uid to frame 0 of this trial = 1 native frame (1 / fps_native: 0.004 s at 250 Hz, 0.0083 s at 120 Hz, 0.01 s at 100 Hz); NaN without prev_uid. Chained windows still need the junction blended (poses/q jumps, TIME_ALIGN-7)",core,metadata,finalize
kin_frame_ok,|b1,[T],-,-,1.2,"not within +-0.1 s of a pelvis-flip frame of the label kinematics (q_gt; on trials with q_gt_usable False, i.e. re-timed trials, q_kin only) OR of q_kin (qc_q_gt_pelvis_tilt_gt60_frames / _q_kin_ above)",core,qc,finalize
kin_frame_ok_frac,<f8,[],fraction,-,1.2,mean of kin_frame_ok,core,qc,finalize
kin_hop_edge_frame_ok_r,|b1,[T],-,-,1.2,False on the hop-edge frames of qc_kin_hop_edge_branch_blend (right arm of poses),core,build,finalize
knee_clamp_frame_dyn,|b1,"[T,2]",-,-,1.2.1,"1.2.1 (R11, flag only): columns (right, left): the SKEL knee of poses / poses_dyn sits on its 0 bound (|knee_angle| < 1e-3 rad) while the AddB fit target knee (q_kin / q_gt walker_knee_*_0) is flexed >= 3 deg. forcebody_io.load_trial returns the one of the chosen pose set as knee_clamp_frame. Mask knee losses / knee captions there if needed; no refit",core,qc,finalize
knee_clamp_frame_kin,|b1,"[T,2]",-,-,1.2.1,"1.2.1 (R11, flag only): columns (right, left): the SKEL knee of poses / poses_dyn sits on its 0 bound (|knee_angle| < 1e-3 rad) while the AddB fit target knee (q_kin / q_gt walker_knee_*_0) is flexed >= 3 deg. forcebody_io.load_trial returns the one of the chosen pose set as knee_clamp_frame. Mask knee losses / knee captions there if needed; no refit",core,qc,finalize
label_conflict_codes,str,[],-,-,1.2,"metadata: pipe-separated: SPEED_NOMINAL_DOUBTFUL (660, Carter walk_slow), KIN_COMPRESSED (30, Carter P003 walking, F03), ACTIVITY_MOTION_MISMATCH (1, Li2021_S05_stand-apose_004) - kept as-is, not corrected here; F1 (1.2.1): SEGMENT_MOTION_MISMATCH (28 trials: Carter2023 static-trial stand segments with a 40-100 deg leg lift at a segment edge, Han2023 pause / sit / sub-1 s action segments without the labelled motion; T2M audit R17), flag only, labels and captions not changed",core,metadata,finalize
label_fix_code,str,[],-,-,1.2,"metadata: pipe-separated: PROTOCOL_SWAP_VZ31_33 (66, D8) and TIZIANA_TOE_HEEL (18, NAMES_ACTIVITY-1) - values actually corrected/relabelled in this build",core,metadata,finalize
marker_held,|b1,"[T,M]",-,-,1.2,marker bit-identical (all 3 coordinates) to the previous frame over a run longer than the nearest-frame resampling duplication ceil((T-1)/(N-1)); static lab markers are included,core,qc,finalize
marker_names,str,[M],-,-,1.0,marker names,core,build,finalize
marker_resample,str,[],-,-,1.2,'linear',core,build,finalize
markers,<f4,"[T,M,3]",m,addb_lab,1.0,"motion-capture markers, linear re-interpolation of the b3d native track (D14); NaN = missing",core,build,finalize
markers_aligned,<f4,"[T,M,3]",m,skel_world,1.1,flip_xz(markers),core,build,finalize
mass_dyn_scale,<f8,[],-,-,1.2,approved per-file rescale of the AddB link masses (1 unless rescaled),core,build,finalize
mass_dyn_used_kg,<f8,[],kg,-,1.2,mass_kg_dynamics used for every BW_dyn and per-kg value in this table,core,qc,finalize
mass_grf_method,str,[],-,-,1.2,estimator behind mass_kg_grf,core,build,finalize
mass_kg,<f4,[],kg,-,1.0,"recorded body mass (model / critic INPUT normalization, D21)",core,build,finalize
mass_kg_audited,<f8,[],kg,-,1.1,v1.1 audited mass (NaN where not audited),core,build,finalize
mass_kg_dynamics,<f8,[],kg,-,1.1,dynamics mass = mass_kg_dynamics_v11 x mass_dyn_scale (already applied). The ONLY mass for normalizing GRF / torque / sigma LABELS; never multiply by mass_dyn_scale again,core,build,finalize
mass_kg_dynamics_v11,<f8,[],kg,-,1.2,AddB pass-2 skeleton mass (v1.1 mass_kg_dynamics),core,build,finalize
mass_kg_grf,<f8,[],kg,-,1.2,per-trial GRF-implied mass estimate (NaN if no estimator applies),core,build,finalize
mass_source,str,[],-,-,1.2,source of caption_mass_kg / shape_mass_target_kg,core,build,finalize
mc_config_sha1,str,[],-,-,1.2,sha1 of the MC engine configuration,core,build,finalize
mc_deriv,str,[],-,-,1.2,noise differentiation scheme,core,build,finalize
mc_engine,str,[],-,-,1.0,MC inverse-dynamics engine,core,build,finalize
mc_forces_sha1,str,[],-,-,1.2,"sha1 of the forces file the MC read (= the shipped forces, checked)",core,build,finalize
mc_forces_source,str,[],-,-,1.2,forces version the MC read,core,build,finalize
mc_forces_version,str,[],-,-,1.2,forces version the MC read,core,build,finalize
mc_fps_noise,<f8,[],Hz,-,1.2,frame rate used for the noise band limit (= fps_true_grid),core,build,finalize
mc_frame_ok,|b1,[T],-,-,1.2,tau_frame_ok AND kin_frame_ok AND has_uncertainty (MC trials only). mu_tau and sigma_tau are finite on every such frame,core,qc,finalize
mc_grid,str,[],-,-,1.0,'release',core,build,finalize
mc_inputs_sha1,str,[],-,-,1.2,sha1 of the MC input arrays,core,build,finalize
mc_kinematics,str,[],-,-,1.2,"kinematics the MC ran on: 'q_gt', 'q_kin' or 'q_kin_pre_armfix' (verified by mc_inputs_sha1)",core,build,finalize
mc_mass_dyn_scale,<f8,[],-,-,1.2,= mass_dyn_scale,core,build,finalize
mc_missing_reason,str,[],-,-,1.2,'' or why has_uncertainty is False,core,build,finalize
mc_n_samples,<i4,[],-,-,1.2,MC samples,core,build,finalize
mc_noise_nbins,<i4,[],-,-,1.2,non-DC noise frequency bins below 6 Hz,core,build,finalize
mc_noise_rms_deg,<f4,[],deg,-,1.2,realized noise RMS,core,build,finalize
mc_nonfinite_frac,<f4,[],-,-,1.0,fraction of non-finite mu_tau entries,core,build,finalize
mc_raw_nonfinite_frac,<f4,[],-,-,1.2,fraction of non-finite raw MC samples,core,build,finalize
mc_seed_base,<i8,[],-,-,1.0,base RNG seed,core,build,finalize
mc_sigma_deg,<f4,[],deg,-,1.0,input angle-noise RMS,core,build,finalize
mc_source,str,[],-,-,1.0,MC input source,core,build,finalize
mc_version,str,[],-,-,1.2,'v1.2',core,build,finalize
mc_zero_noise_maxrel,<f8,[],-,-,1.2,zero-noise reproduction of tau_selfconsistent (max relative difference),core,build,finalize
mu_tau,<f4,"[T,37]",N m | N,addb_lab,1.0,MC mean torque,core,build,finalize
n_contacts,<i4,[],-,-,1.0,C,core,build,finalize
n_markers,<i4,[],-,-,1.0,M,core,build,finalize
n_segments_in_capture,<i8,[],-,-,1.2,"metadata: one AddB capture split into contiguous segments; prev/next = the contiguous neighbour (kinematics are NOT continuous at junctions, T3 verifier)",core,metadata,finalize
num_frames,<i4,[],frames,-,1.0,T,core,build,finalize
path,str,[],-,-,1.2,metadata: turn_right for Fregly2012 PS_rightturn4; empty otherwise (not stated in the sources),core,metadata,finalize
person_id,str,[],-,-,1.2,metadata: global person key <study>_<subject_id> (S<NN> alone repeats across studies),core,metadata,finalize
poses,<f4,"[T,46]",rad,skel_world,1.0,"SKEL pose, fitted to the AddB kinematics pass (pass 0 + 30 Hz = pass 1; arm re-solved where q_kin_source is addb_pass0_armfix_lp30); pose_source pass0. Head DOFs 23-25 from the head post-fit where head_postfit, else 0",core,build,finalize
poses_dyn,<f4,"[T,46]",rad,skel_world,1.2,"SKEL pose fitted to the AddB dynamics pass (pass 2), or to a repaired pass-2 target (skel_fit_target_dyn addb_pass2_armsplice / addb_pass2_repaired_v3a / addb_pass2_repaired_v4s; frames and DOFs in dyn_target_source); pose_source pass2. Equals poses where skel_fit_target_dyn == kin_retimed",core,build,finalize
poses_dyn_usable,|b1,[],-,-,1.2,"trial-level: poses_dyn may be used (False: force/mocap desync, D5/D22, S8b)",core,build,finalize
poses_dyn_usable_s8b,|b1,[],-,-,1.2,S8b verdict on poses_dyn,core,build,finalize
progression_heading_deg,<f8,[],deg,-,1.2,"heading of that median vector (SKEL frame, 0 = +z) when the speed > 0.3 m/s, else NaN",core,qc,finalize
progression_speed_mps,<f8,[],m/s,-,1.2,|median over kinematic-stance frames of (SKEL pelvis velocity - stance-foot centre velocity)|; kinematic stance = lower foot within 4 cm of the trial foot floor (2nd percentile),core,qc,finalize
protocol_condition,str,[],-,-,1.2,"metadata: source protocol condition: Carter file stem (flatrun_norm ...), Moore T-number, vanderZee trialN + Table 2 condition, Fregly trial stem, Hamner run200..run500",core,metadata,finalize
protocol_ref,str,[],-,-,1.2,"metadata: source-native protocol id, provenance only, never caption text: Carter file family (flatrun_norm), Hamner runNNN, Moore TNNN, vanderZee trialN, Fregly stem without the repetition number, otherwise the AddB trial stem",core,metadata,finalize
q_gt,<f4,"[T,37]",rad | m,addb_lab,1.0,AddB dynamics-pass (pass 2) joint angles; root translations in m. Force-aware (leakage policy); not usable where q_gt_usable is False,core,build,finalize
q_gt_usable,|b1,[],-,-,1.2,False on re-timed trials (pass 2 was fitted to the misaligned force),core,build,finalize
q_kin,<f4,"[T,37]",rad | m,addb_lab,1.2,"AddB kinematics pass (pass 0) filtered with the 30 Hz 2nd-order Butterworth filtfilt that AddB applies for pass 1, on the release grid. Not force-aware. In 875 trials the pass-0 arm is re-solved before the filter (q_kin_source). q_kin reproduces the stored AddB pass 1 in 8,218 trials (q_kin_pass1_match). q_kin_pass1_mismatch_reason gives the reason for the other 2,168 trials",core,build,finalize
q_kin_pass1_match,|b1,[],-,-,1.2,q_kin reproduces the AddB pass-1 positions (max abs < 1e-6 rad),core,build,finalize
q_kin_pass1_maxabs_diff_rad,<f4,[],rad,-,1.2,max |q_kin - AddB pass 1| on the native grid,core,build,finalize
q_kin_pass1_mismatch_reason,str,[],-,-,1.2,'match' or the documented reason (e.g. pass0_ik_outlier_repaired),core,build,finalize
q_kin_source,str,[],-,-,1.2,"'addb_pass0_lp30', or 'addb_pass0_armfix_lp30' (875 trials: arm IK re-solve of the parked / hopping pass-0 arm before the filter, skel_refit NOTES 7.4)",core,build,finalize
qc_addb_com_grf_lag_s,<f8,[],r / s,-,1.2,same with the AddB model COM (nimble getCOM of q_gt),core,qc,finalize
qc_addb_com_grf_time_offset,|b1,[],-,-,1.2,same test with the AddB COM (q_gt): an offset here means the force itself is misaligned with the dynamics pass,core,qc,finalize
qc_addb_com_grf_vert_r,<f8,[],r / s,-,1.2,same with the AddB model COM (nimble getCOM of q_gt),core,qc,finalize
qc_addb_yaw_rate_p99_dps,<f8,[],deg/s,-,1.2,"same for q_gt[:,2]",core,qc,finalize
qc_addb_yaw_rms_marker_deg,<f8,[],deg,-,1.2,"same for the AddB yaw heading atan2(-cos q2, sin q2)",core,qc,finalize
qc_angle_wrap_dofs,str,[],-,-,1.1,"'|'-joined q_gt rotational DOFs with a frame-to-frame jump > 1 rad, on the shipped q_gt (v1.1 definition, now recomputed; source wrap moved to qc_tau_gt_det_wrap_corrupt) [v1.2 QC recomputation; v1.1 value: provenance qc_angle_wrap_dofs_v11]",core,qc,finalize
qc_arm_event_classes_dyn,str,[],-,-,1.2,'|'-joined 2026-10-07 arm-event classes of poses_dyn,core,build,finalize
qc_arm_event_classes_kin,str,[],-,-,1.2,'|'-joined 2026-10-07 arm-event classes of poses ('' if none),core,build,finalize
qc_arm_unreliable,|b1,[],-,-,1.2,q_gt arm DOF (23-36) jump > 1 rad OR SKEL arm/scapula pose DOF (26-45) jump > 1 rad,core,qc,finalize
qc_belt_speed_change,|b1,[],-,-,1.2.1,"1.2.1 (R16): treadmill belt labelled constant (belt_mode) whose speed changes inside the trial in both pose sets (kinematic 2 s windows AND rolling 5-stance GRF windows below 85% or above 115% of the constant belt estimate, thresholds OURS of the T2M audit): vanderZee2022_S04_walk_040/044/056/080, deceleration at the trial end. Flag only",core,build,finalize
qc_belt_speed_change_suspected,|b1,[],-,-,1.2.1,"1.2.1 (R16): the same rule hit in one pose set only (8 trials: vanderZee2022 7, Carter2023 1); suspected, flag only",core,build,finalize
qc_calcn_above_floor_m,<f8,[],m,-,1.1,median SKEL calcn height minus CoP height on the same firm frames (v1.1 definition) [v1.2 QC recomputation; v1.1 value: provenance qc_calcn_above_floor_m_v11],core,qc,finalize
qc_chair_channel_active,|b1,[],-,-,1.2,S25: any chair_contact frame (the chair channel carries the seated body weight somewhere in the trial),core,qc,finalize
qc_contact_torque_max_firm_nm,<f8,[],N m,-,1.2,max |contact_torque| (vector norm) over firm frames of both feet,core,qc,finalize
qc_contact_torque_med_firm_nm,<f8,[],N m,-,1.2,median |contact_torque| over firm frames,core,qc,finalize
qc_cop_fixed_point,|b1,[],-,-,1.1,CoP is a near-fixed lab point (raw plate CoP unusable); loaded frames status 3,core,build,finalize
qc_cop_foot_dist_m,<f8,[],m,-,1.1,"median horizontal distance of flip_xz(cop) to the SKEL calcn->toes segment on firm frames (mass_kg, grf_valid_mask, <= 200 evenly spaced frames; v1.1 definition) [v1.2 QC recomputation; v1.1 value: provenance qc_cop_foot_dist_m_v11]",core,qc,finalize
qc_cop_foot_shift_baseline_ms,<f8,[],ms,-,1.2,"QC post-process (qc_postprocess_v12.py), per-subject CoP-shift baseline",core,qc_post,finalize
qc_cop_foot_shift_dev_ms,<f8,[],ms,-,1.2,"QC post-process (qc_postprocess_v12.py), per-subject CoP-shift baseline",core,qc_post,finalize
qc_cop_foot_shift_dist_m,<f8,[],m,-,1.2,the pooled median CoP-segment distance at that best shift (compare with qc_cop_foot_dist_m at lag 0),core,qc,finalize
qc_cop_foot_shift_ms,<f8,[],ms,-,1.2,"integer frame shift s in +-0.4 s of the SKEL calcn->toes segment (kin fit) against flip_xz(cop) that minimises the pooled median horizontal CoP-segment distance, on the qc_cop_foot_dist_m firm frames; positive = kinematics later than the force. Meaningful on treadmill trials only (FU3-3: overground the curve is flat over a stance). Diagnostic, no flag",core,qc,finalize
qc_cop_heel_offset10_m,<f8,[],m,-,1.2,"median, over loaded stance runs (foot vertical GRF > 0.2 BW_dyn, >= 3 frames) of both feet, of the CoP position at 10% of the run along the SKEL calcn->toes axis, measured from the calcn joint",core,qc,finalize
qc_cop_measured_frac,<f8,[],fraction,-,1.2,"same denominator, share with contact_status == 1 (measured, reliable)",core,qc,finalize
qc_cop_off_foot_frac,<f8,[],fraction,-,1.2,frames with cop_off_foot on any foot / frames with a foot > 0.2 BW_dyn and GRF-OK,core,qc,finalize
qc_cop_plane_mismatch,|b1,[],-,-,1.2,"need_mesh AND >= 10 firm frames AND |qc_skel_sole_minus_cop_m| > 0.10 m (Tiziana2019 stairs: CoP stays on the plate plane, ~0.15-0.20 m below the loaded foot on the instrumented step)",core,qc,finalize
qc_cop_status_source,str,[],-,-,1.2,"'forces_workstream' when contact_status was shipped, else 'absent'",core,qc,finalize
qc_cop_travel_m,<f8,[],m,-,1.1,"median horizontal CoP travel from the first to the last frame of firm-stance runs >= 5 frames (firm with mass_kg, grf_valid_mask; v1.1 definition, now with the corrected mass) [v1.2 QC recomputation; v1.1 value: provenance qc_cop_travel_m_v11]",core,qc,finalize
qc_cop_untrusted_flag,|b1,[],-,-,1.2,qc_cop_untrusted_frac > 0.30,core,qc,finalize
qc_cop_untrusted_frac,<f8,[],fraction,-,1.2,"share of loaded frames (grf_valid_mask AND foot vertical GRF > 20 N) whose contact_status is 3 (projected) or 7 (air-loaded, untrusted); NaN when contact_status is absent",core,qc,finalize
qc_duplicate_drop,|b1,[],-,-,1.1,byte-identical GRF duplicate of another trial (drop for evaluation),core,build,finalize
qc_duplicate_of,str,[],-,-,1.1,trial this one duplicates,core,build,finalize
qc_dyn_arm_spliced,|b1,[],-,-,1.2,dyn_arm_spliced has a True frame,core,build,finalize
qc_dynamics_mass_ratio,<f8,[],-,-,1.1,mass_kg_dynamics / mass_kg,core,build,finalize
qc_extra_channel_zero,|b1,[],-,-,1.2,"S25: a non-foot contact channel exists (Falisse2017 chair, contact_bodies 'pelvis') but carries exactly zero force in the whole trial (walk / stairs): its grf_contact_ok column is False and grf_frame_ok ignores it",core,build,finalize
qc_floor_slope_facing_deg,<f8,[],deg,-,1.2,slope of that plane along the facing direction (facing_hipline),core,qc,finalize
qc_floor_slope_facing_kin_deg,<f8,[],deg,-,1.2,floor_tilt_*_kin_deg slope along the facing direction (facing_hipline_deg),core,qc,finalize
qc_gait_event_inconsistent,|b1,[],-,-,1.2,"|cadence_spm_grf / cadence_spm - 1| > 0.10 (OURS, S19 F1) where both exist; the GRF gait scalars are then NaN and the trial has no rows in gait_events.csv",core,finalize,finalize
qc_grf_contact_bad_frames,<i8,[],frames,-,1.2,GRF-valid non-endpoint frames removed by grf_contact_ok,core,qc,finalize
qc_grf_endpoint_valid_v11,<i8,[],frames,-,1.2,number of the two endpoint frames that grf_valid_mask marks valid (their GRF is 0),core,qc,finalize
qc_grf_foot_error,|b1,[],-,-,1.1,v1.1 QC flag (kept unchanged): GRF assigned to the wrong foot,core,build,finalize
qc_grf_force_in_air_frac,<f8,[],fraction,-,1.2,frames with grf_force_in_air on any foot / GRF-OK frames,core,qc,finalize
qc_grf_frame_tilt_deg,<f8,[],deg,-,1.2,raw downhill-candidate GRF angle (diagnostic); NaN when not a candidate,core,build,finalize
qc_grf_frame_tilted,|b1,[],-,-,1.2,"qc_grf_frame_tilt_deg is present AND the forces tilt_deg (rotation actually applied) == 0. A rotated file's own tilt_deg stores the pre-rotation angle it corrected (not 0), so this correctly excludes the rotated files",core,qc,finalize
qc_grf_invalid_force_frac,<f8,[],fraction,-,1.2,fraction of GRF-invalid frames whose summed |GRF| > 1 N (force present but masked),core,qc,finalize
qc_grf_kin_n_events,<i8,[],events,-,1.2,number of toe-off plus onset events used by the two fields above,core,qc,finalize
qc_grf_kin_onset_lag_ms,<f8,[],ms,-,1.2,same at GRF onsets (rise above 0.10 BW_dyn) against the marker speed falling through 0.2 m/s; onsets inside a grf_onset_censored run are skipped (NaN when none remain),core,qc,finalize
qc_grf_kin_toeoff_lag_ms,<f8,[],ms,-,1.2,"median over GRF toe-off events (foot vertical GRF falls below 0.10 BW_dyn) of (nearest frame where min(heel, forefoot) belt-relative marker speed rises through 0.2 m/s) - (GRF event frame), searched within +-0.3 s; positive = kinematics later. Markers: FOOT_MK table (v_per_trial.py:60-71), name heuristic as fallback; 15 Hz zero-phase marker low-pass",core,qc,finalize
qc_grf_load_frac_l,<f8,[],fraction,-,1.2,same for the left foot,core,qc,finalize
qc_grf_load_frac_r,<f8,[],fraction,-,1.2,fraction of interior GRF-OK frames with right-foot vertical GRF > 0.1 BW_dyn,core,qc,finalize
qc_grf_loaded_eval_frames,<i8,[],frames,-,1.2.1,1.2.1 (R01b): eval_grf_frame_ok frames with a foot vertical GRF > 20 N; eval_grf_trial_ok and eval_torque_trial_ok require > 0,core,qc,finalize
qc_grf_loaded_foot_lower_frac,<f8,[],fraction,-,1.2,"GRF single-support frames (one foot firm, other < 0.05 BW_dyn): fraction where the loaded foot lowest point is below the other foot",core,qc,finalize
qc_grf_loaded_foot_moving_frac,<f8,[],fraction,-,1.2,firm frames where the loaded AddB foot lowest point moves > 0.5 m/s horizontally relative to the belt,core,qc,finalize
qc_grf_mask_matches_b3d,|b1,[],-,-,1.2,grf_valid_mask equals (grf_missing_reason == notMissingGRF) on every frame,core,qc,finalize
qc_grf_motion_r,<f8,[],r,-,1.1,deprecated v1.1 QC (see qc_skel_com_grf_vert_r),core,build,finalize
qc_grf_n_strides,<i8,[],strides,-,1.2,"strides (onset to onset of the hysteresis contact, 0.4-2.5 s, fully GRF-OK) used for qc_mass_grf_implied_kg",core,qc,finalize
qc_grf_newton_seg_lag_ms,<f8,[],ms,-,1.2,"median over GRF-valid segments (>= 0.25 s) of the lag maximising r(8 Hz total vertical GRF, 8 Hz vertical acceleration of the pelvis-marker centroid), keeping segments with r > 0.7; positive = kinematics later. Kinematic signal: PELVIS_MK markers (v_lag.py:24-30) when >= 2 are present and >= 90% finite, else SKEL trans y (kin fit); never q_gt or trans_dyn (both follow the force)",core,qc,finalize
qc_grf_newton_seg_n,<i8,[],segments,-,1.2,number of segments behind qc_grf_newton_seg_lag_ms,core,qc,finalize
qc_grf_newton_seg_src,str,[],-,-,1.2,'pelvis_markers' or 'skel_trans' (fallback),core,qc,finalize
qc_grf_onset_censored_frac,<f8,[],fraction,-,1.2,mean of grf_onset_censored.any(1),core,qc,finalize
qc_grf_plate_offset_bw_c0,<f8,[],BW,-,1.2,plate re-zero offset removed from contact 0; NaN when not re-zeroed,core,build,finalize
qc_grf_plate_offset_bw_c1,<f8,[],BW,-,1.2,plate re-zero offset removed from contact 1; NaN when not re-zeroed,core,build,finalize
qc_grf_plate_offset_l_bw,<f8,[],BW_dyn,-,1.2,median foot vertical GRF on GRF-OK mid-swing frames of that foot (>= 10 frames),core,qc,finalize
qc_grf_plate_offset_r_bw,<f8,[],BW_dyn,-,1.2,median foot vertical GRF on GRF-OK mid-swing frames of that foot (>= 10 frames),core,qc,finalize
qc_grf_single_channel,|b1,[],-,-,1.2,both feet loaded on one force channel (forces workstream list or rule),core,build,finalize
qc_grf_single_channel_rule,|b1,[],-,-,1.2,the load-fraction part of qc_grf_single_channel alone,core,qc,finalize
qc_grf_single_channel_source,str,[],-,-,1.2,"'forces_workstream' when the forces output shipped its own qc_grf_single_channel (precedence over the QC rule), else 'qc_rule'",core,qc,finalize
qc_grf_swing_wrench_p95_nmkg,<f8,[],N m/kg,-,1.2,95th percentile over mid-swing GRF-OK frames of |(cop - calcn) x grf + contact_torque| / mass_kg_dynamics (spurious wrench on the swing foot),core,qc,finalize
qc_grf_tail_blank_frames,<i8,[],frames,-,1.2,number of grf_tail_blank frames,core,qc,finalize
qc_grf_unmeasured_stance_frac,<f8,[],fraction,-,1.2,frames with grf_unmeasured_stance on any foot / GRF-OK frames,core,qc,finalize
qc_grf_valid_lowload_frac,<f8,[],fraction,-,1.2,walking only: fraction of GRF-OK frames with total vertical GRF < 0.5 BW_dyn (force missing),core,qc,finalize
qc_grf_vertical_p01_bw,<f8,[],BW(mass_kg),-,1.1,"1st percentile of per-contact vertical GRF / (mass_kg g) on grf_valid_mask frames (v1.1 definition, build_v11.py:337-340); superseded by qc_grf_plate_offset_* [v1.2 QC recomputation; v1.1 value: provenance qc_grf_vertical_p01_bw_v11]",core,qc,finalize
qc_grf_wrong_foot_frac,<f8,[],fraction,-,1.2,firm frames (both feet pooled) whose CoP is closer to the other AddB foot segment by more than 0.05 m,core,qc,finalize
qc_grf_zero_force_grounded_source,str,[],-,-,1.2,"'forces_workstream' when grf_zero_force_grounded was shipped, else 'absent_all_false'",core,qc,finalize
qc_grf_zero_valid,|b1,[],-,-,1.2,no GRF-valid frame (grf_valid_mask all False),core,qc,finalize
qc_h3d_contact_agreement,<f8,[],fraction,-,1.2,"agreement of HumanML3D foot contacts (squared 20 fps displacement of talus or toes < 0.002 m^2, belt removed) with GRF contact (foot vertical GRF > 0.05 BW_dyn) on GRF-OK 20 fps pairs, both feet",core,qc,finalize
qc_h3d_heading_err0_deg,<f8,[],deg,-,1.2,the same error at frame 0,core,qc,finalize
qc_h3d_heading_err_frac_gt20,<f8,[],fraction,-,1.2,fraction of frames whose HumanML3D facing (up x ((femur_r - femur_l) + (humerus_r - humerus_l))) is > 20 deg from progression_heading,core,qc,finalize
qc_head_at_limit_frac_dyn,<f8,[],-,-,1.2,same for poses_dyn,core,build,finalize
qc_head_at_limit_frac_kin,<f8,[],-,-,1.2,fraction of frames with a head DOF of poses at a bsm.osim neck limit (NaN when not head_postfit); > 0.05 = head orientation unreliable on those frames (full audit D2),core,build,finalize
qc_head_untracked,|b1,[],-,-,1.2,"NOT head_postfit: < 2 observed head markers, head DOFs 23-25 stay 0 (rigid with the thorax at SKEL neutral; neck line 3-10 deg forward of the markers in head-marker studies). Mask DOFs 23-25 and joint 13 in losses",core,build,finalize
qc_heading_err_marker_deg,<f8,[],deg,-,1.2,median |wrap(SKEL hip-line facing - first-R/first-L ASIS heading)|,core,qc,finalize
qc_heading_err_progression_deg,<f8,[],deg,-,1.2,|wrap(facing_hipline - progression_heading)|,core,qc,finalize
qc_heading_rms3_dyn,|b1,[],-,-,1.2,same for poses_dyn,core,build,finalize
qc_heading_rms3_kin,|b1,[],-,-,1.2,"pelvis heading r dropped > 0.1 vs the first v1.2 fit AND detrended heading RMS > 3 deg (OURS, NOTES 7.16)",core,build,finalize
qc_kin_armfix_edge,|b1,[],-,-,1.2,kin target has an A4 armfix window-edge artifact (20 trials; frames in arm_event_ok_kin),core,build,finalize
qc_kin_hop_edge_branch_blend,|b1,[],-,-,1.2,"kin target hop window edge joins two Euler branches of the right arm (3 trials, skel_refit NOTES 7.15)",core,build,finalize
qc_knee_clamp_frac_dyn,<f8,[],fraction,-,1.2.1,max over sides of the mean of knee_clamp_frame_kin / knee_clamp_frame_dyn,core,qc,finalize
qc_knee_clamp_frac_kin,<f8,[],fraction,-,1.2.1,max over sides of the mean of knee_clamp_frame_kin / knee_clamp_frame_dyn,core,qc,finalize
qc_knee_locked,|b1,[],-,-,1.1,SKEL knee frozen (std < 1e-9 rad) on either side (redefined from v1.1 identically-zero) [v1.2 QC recomputation; v1.1 value: provenance qc_knee_locked_v11],core,qc,finalize
qc_marker_held_frac,<f8,[],fraction,-,1.2,marker_held entries / finite marker entries,core,qc,finalize
qc_marker_held_markers,<i8,[],markers,-,1.2,markers with any held run,core,qc,finalize
qc_mass_dyn_consistent,|b1,[],-,-,1.2,|mass_kg_dynamics_v11 * mass_dyn_scale - mass_kg_dynamics| < 1e-3 kg when both inputs exist (True otherwise),core,qc,finalize
qc_mass_dyn_vs_grf,<f8,[],ratio,-,1.2,qc_mass_grf_implied_kg / mass_kg_dynamics,core,qc,finalize
qc_mass_dyn_vs_grf_flag,|b1,[],-,-,1.2,|qc_mass_dyn_vs_grf - 1| > 0.05,core,qc,finalize
qc_mass_grf_implied_kg,<f8,[],kg,-,1.2,GRF-implied mass: median over >= 3 strides of mean total vertical GRF / (g + delta qd_ty / stride time); else impulse balance over GRF-OK runs >= 1 s (>= 2 s in total); else NaN,core,qc,finalize
qc_mass_grf_method,str,[],-,-,1.2,'strides' | 'impulse' | 'none',core,qc,finalize
qc_mass_grf_ratio,<f8,[],ratio,-,1.1,"mean total vertical GRF / g over grf_valid_mask frames / mass_kg (v1.1 definition, > 50 valid frames); superseded by qc_mass_grf_implied_kg [v1.2 QC recomputation; v1.1 value: provenance qc_mass_grf_ratio_v11]",core,qc,finalize
qc_mass_kg_grf_vs_implied,<f8,[],ratio,-,1.2,forces-output mass_kg_grf / qc_mass_grf_implied_kg (cross-check of the two implementations),core,qc,finalize
qc_mass_outlier_flag,|b1,[],-,-,1.2,trial's GRF-implied mass deviates from its file's ratio (forces stage 2),core,build,finalize
qc_mass_shape_vs_dyn,|b1,[],-,-,1.2,|mass_kg_dynamics / qc_skel_mass_from_volume_kg - 1| > 0.05 (need_mesh only; False when the SHAPY mass estimate is unavailable),core,qc,finalize
qc_mass_source_gap,<f8,[],-,-,1.2,"metadata: per-person mass table of S7 joined on person_id; caption_mass_kg = shape_mass_target_kg = recorded mass except Moore2015 (GRF calibration pose) and Falisse2017_S01 (caption NaN, D4)",core,metadata,finalize
qc_mc_lag_frames,<f8,[],frames,-,1.2,"k in -2..2 maximising the median over direct DOFs of corr(mu_tau(t), tau_sc(t+k)) on interior GRF-OK frames (MC trials); 0 expected after the v1.2 MC recompute",core,qc,finalize
qc_mc_lag_gain,<f8,[],r,-,1.2,correlation at the best k minus at k = 0,core,qc,finalize
qc_mc_mu_sc_corr,<f8,[],r,-,1.2,that median correlation at k = 0,core,qc,finalize
qc_mc_mu_sc_rms_nmkg,<f8,[],N m/kg,-,1.2,RMS of (mu_tau - tau_sc) over joint DOFs 6-36 on mc_frame_ok frames / mass_kg_dynamics (0 up to MC noise after the recompute),core,qc,finalize
qc_mc_nonfinite,<i8,[],entries,-,1.2,non-finite entries of mu_tau plus sigma_tau,core,qc,finalize
qc_mc_sigma_frame01_ratio,<f8,[],ratio,-,1.2,median direct-DOF sigma at frames 0-1 / median on interior GRF-OK frames (about 0.2-0.3 with the v1.1 start-of-trial defect),core,qc,finalize
qc_mc_sigma_negative,<i8,[],entries,-,1.2,number of sigma_tau entries < 0,core,qc,finalize
qc_mc_sigma_zero_frac,<f8,[],fraction,-,1.2,"fraction of frames failing the ""sigma_tau > 0 on every direct DOF"" test (MC trials only, i.e. mu_tau/sigma_tau shipped)",core,qc,finalize
qc_newton_n,<i8,[],frames,-,1.2,"frames used by the Newton checks: +-0.1 s neighbourhood GRF-OK, not within 0.1 s of the ends",core,qc,finalize
qc_overlap_motion_drop,|b1,[],-,-,1.2,"Tiziana2019 overlapping-capture pairs (qc_overlap_of): the member with fewer grf_valid_mask frames (tie: higher segment_index, then later trial_uid); excluded from eval_motion_trial_ok only (the pairs share no GRF-valid frame, so force evaluation keeps both)",core,build,finalize
qc_overlap_of,str,[],-,-,1.2,"metadata: partner trial_uid:shared seconds for release trials sharing native marker frames (Tiziana2019 400, Han2023 19)",core,metadata,finalize
qc_overlap_pair_drop,|b1,[],-,-,1.2,eval_exclude_duplicate AND NOT qc_duplicate_drop: the Han2023 trimmed copies of a *_full re-export (8 trials); excluded from every eval_* trial flag,core,build,finalize
qc_pass0_newton_r,<f8,[],r,-,1.2,"qc_addb_com_grf_vert_r computed with q_kin (pass 0 + 30 Hz filter) instead of q_gt (pass 2), on the same pass-2-scaled AddB skeleton; NaN when q_kin is absent (v1.1) or shape-mismatched",core,qc,finalize
qc_pass2_equals_pass1,|b1,[],-,-,1.2,max |q_gt - q_kin| < 1e-6 rad (both in the same 37-DOF space); False when q_kin is absent (v1.1),core,qc,finalize
qc_pass2_marker_rms_rise_mm,<f8,[],mm,-,1.2,"S19 F5: AddB marker RMS rise from the kinematics pass to the dynamics pass: median over the native b3d frames of SubjectOnDisk.getTrialMarkerRMSs(trial, dyn_pass) minus the same for pass 0 (dyn_pass = 2, else the last pass of the trial), in mm; NaN without a b3d",core,qc,finalize
qc_pass2_pass0_jc_tv_p95_mm,<f8,[],mm,-,1.2,"95th percentile of the time-varying part (per-joint temporal mean removed) of |addb_joints_dyn - addb_joints_kin| (D33, v1.2-only, SKEL frame); NaN when absent",core,qc,finalize
qc_pose_wrap_n_dyn,<i4,[],frames,-,1.2,same for poses_dyn,core,build,finalize
qc_pose_wrap_n_kin,<i4,[],frames,-,1.2,"one-frame SKEL DOF steps > pi in poses (Euler branch flips; masked in fit_event_ok_kin, not unwrapped)",core,build,finalize
qc_q_gt_abs_gt_2pi_dofs,str,[],-,-,1.2,those DOFs,core,qc,finalize
qc_q_gt_abs_gt_2pi_frames,<i8,[],entries,-,1.2,"number of (frame, rotational DOF) entries with |q_gt| > 2 pi",core,qc,finalize
qc_q_gt_pelvis_euler_gt90_frames,<i8,[],frames,-,1.2,frames with |q_gt pelvis_tilt| or |pelvis_list| > 90 deg,core,qc,finalize
qc_q_gt_pelvis_tilt_gt60_frames,<i8,[],frames,-,1.2,"frames whose pelvis y-axis is more than 60 deg from world +y, i.e. cos(q_gt[:,0]) cos(q_gt[:,1]) < cos 60 deg (AddB EulerFreeJoint intrinsic ZXY; FU2-3 rule, 23 v1.1 walking/running trials). S15b fix: the S15a version tested |tilt| > 60 OR |list| > 60 separately (22 trials)",core,qc,finalize
qc_q_kin_abs_gt_2pi_dofs,str,[],-,-,1.2.1,those DOFs,core,qc,finalize
qc_q_kin_abs_gt_2pi_frames,<i8,[],entries,-,1.2.1,"1.2.1 (R12): number of (frame, rotational DOF) entries with |q_kin| > 2 pi (same test as qc_q_gt_abs_gt_2pi_frames, on the fit target of poses); -1 when q_kin is absent",core,qc,finalize
qc_q_kin_pelvis_tilt_gt60_frames,<i8,[],frames,-,1.2,"same on q_kin (D20, pass 0 + 30 Hz filter); -1 when q_kin is absent or shape-mismatched (v1.1)",core,qc,finalize
qc_q_kin_wrap,|b1,[],-,-,1.2.1,"1.2.1 (R12): qc_q_kin_abs_gt_2pi_frames > 0 (68 trials, all on acromial (shoulder) DOFs). The build sets arm_event_ok_kin False on the frames of a wrapped arm DOF (+-0.1 s), and arm_event_ok_dyn too on kin_retimed trials. Flag and mask only, q_kin is not unwrapped",core,qc,finalize
qc_qdd_corr_2diffq,<f8,[],r,-,1.2,"mean over DOFs of corr(qdd_gt, np.gradient twice of q_gt), all frames (exact v1.1 qdd_fix_corr_2diffq formula, fix_qdd_batch.py), recomputed on the shipped arrays. S19 F4: own name so the v1.1 fix record qdd_fix_corr_2diffq is never overwritten",core,qc,finalize
qc_qgt_pelvis_grf_vert_r,<f8,[],r,-,1.2,same r with the q_gt pelvis ty acceleration,core,qc,finalize
qc_root_residual_bw,<f8,[],BW_dyn,-,1.2,"median |tau_selfconsistent[3:6]| / BW_dyn on loaded frames (GRF-OK, interior, total vertical GRF > 0.05 BW_dyn)",core,qc,finalize
qc_root_residual_flight_bw,<f8,[],BW_dyn,-,1.2,"median on flight frames (GRF-OK, interior, |total GRF| < 20 N; >= 5 frames)",core,qc,finalize
qc_root_residual_hicks_ok,|b1,[],-,-,1.2,qc_root_residual_hicks_peak_ratio <= 0.05 (and >= 5 loaded frames),core,qc,finalize
qc_root_residual_hicks_peak_ratio,<f8,[],ratio,-,1.2,max residual force / max |total GRF| on loaded frames,core,qc,finalize
qc_root_residual_moment_flight_nmkg,<f8,[],N m/kg,-,1.2,same on flight frames,core,qc,finalize
qc_root_residual_moment_nmkg,<f8,[],N m/kg,-,1.2,"median |M| / mass_kg_dynamics on loaded frames, M = J_w^-T tau_sc[0:3] = world residual moment about the pelvis origin, J_w = [e_z, Rz(q0) e_x, Rz(q0) Rx(q1) e_y] (checked equal to nimble getWorldJacobian to 1e-16)",core,qc,finalize
qc_root_residual_p95_bw,<f8,[],BW_dyn,-,1.2,95th percentile of the same,core,qc,finalize
qc_shape_bone_err_gt20mm,|b1,[],-,-,1.2,qc_shape_bone_err_mm > 20 mm,core,qc,finalize
qc_shape_bone_err_mm,<f8,[],mm,-,1.2,"mean absolute error, over thigh_r/shank_r/thigh_l/shank_l (the 4 DIRECT-only bones), of median-over-frames SKEL bone length minus the AddB pass-2 bone length",core,qc,finalize
qc_skel_acc_glitch_20fps,|b1,[],-,-,1.2,qc_skel_acc_max_20fps > 200 m/s^2,core,qc,finalize
qc_skel_acc_max_20fps,<f8,[],m/s^2,-,1.2,"max over joints and native frames t of |x(t + 0.05 s) - 2 x(t) + x(t - 0.05 s)| / 0.05^2 (SKEL joints linearly interpolated): the 20 fps acceleration maximised over all sampling phases, so a one-frame spike cannot fall between samples",core,qc,finalize
qc_skel_ankle_at_limit_frac,<f8,[],fraction,-,1.2,max over sides of the fraction of frames with the SKEL ankle within 1e-3 rad of the fitter lower bound (-0.785),core,qc,finalize
qc_skel_ankle_reaches_limit,|b1,[],-,-,1.2,qc_skel_ankle_at_limit_frac > 0,core,qc,finalize
qc_skel_body_below_floor_3cm_frac,<f8,[],fraction,-,1.2,fraction of the ~50 Hz subgrid frames whose lowest skin vertex is > 3 cm below the median firm-stance CoP height,core,qc,finalize
qc_skel_com_grf_lag_r,<f8,[],s / r,-,1.2,"lag within +-1 s maximising r(GRF(t), a(t + lag)) (positive = motion lags force); lags keeping < 80% of the frames skipped; among lags within 0.01 of the best r the smallest |lag| (periodic gait); and the r there",core,qc,finalize
qc_skel_com_grf_lag_s,<f8,[],s / r,-,1.2,"lag within +-1 s maximising r(GRF(t), a(t + lag)) (positive = motion lags force); lags keeping < 80% of the frames skipped; among lags within 0.01 of the best r the smallest |lag| (periodic gait); and the r there",core,qc,finalize
qc_skel_com_grf_lag_s_dyn,<f8,[],mixed,-,1.2,"D0 poses_dyn duplicate of the kin-target fields of the same name (without _dyn): computed on poses_dyn/trans_dyn instead of poses/trans. qc_skel_grf_desync is explicitly NOT duplicated here (must stay on `poses` only, S15b). All NaN/False when poses_dyn/trans_dyn are absent (v1.1)",core,qc,finalize
qc_skel_com_grf_time_offset,|b1,[],-,-,1.2,SKEL motion time-shifted against the GRF: |best lag| >= 0.1 s AND best r - lag-0 r >= 0.1 AND best r >= 0.9 AND >= 1 s of Newton frames,core,qc,finalize
qc_skel_com_grf_vert_r,<f8,[],r,-,1.2,"Pearson r of 6 Hz total vertical GRF and m_dyn (a_y + g), a_y = second difference of the 6 Hz SKEL COM (skin-mesh volume centroid, uniform density: exact on a ~50 Hz subgrid, spline-corrected rigid-skinning estimate between)",core,qc,finalize
qc_skel_com_grf_vert_r_dyn,<f8,[],mixed,-,1.2,"D0 poses_dyn duplicate of the kin-target fields of the same name (without _dyn): computed on poses_dyn/trans_dyn instead of poses/trans. qc_skel_grf_desync is explicitly NOT duplicated here (must stay on `poses` only, S15b). All NaN/False when poses_dyn/trans_dyn are absent (v1.1)",core,qc,finalize
qc_skel_com_grf_vert_rms_bw,<f8,[],BW_dyn,-,1.2,RMS of the vertical Newton residual,core,qc,finalize
qc_skel_com_grf_vert_rms_bw_dyn,<f8,[],mixed,-,1.2,"D0 poses_dyn duplicate of the kin-target fields of the same name (without _dyn): computed on poses_dyn/trans_dyn instead of poses/trans. qc_skel_grf_desync is explicitly NOT duplicated here (must stay on `poses` only, S15b). All NaN/False when poses_dyn/trans_dyn are absent (v1.1)",core,qc,finalize
qc_skel_d2_max_m,<f8,[],m,-,1.2,max over joints and frames of |second difference| of SKEL joints at the native rate (= acceleration / fps^2; a one-frame spike of height s gives 2 s),core,qc,finalize
qc_skel_elbow_at_upper_frac,<f8,[],fraction,-,1.2,max over sides of the fraction of frames at the elbow upper fitter bound (2.356 rad),core,qc,finalize
qc_skel_elbow_zero,|b1,[],-,-,1.2,SKEL elbow identically 0 on either side,core,qc,finalize
qc_skel_fit_err_mm,<f8,[],mm,-,1.1,mean joint error of poses vs its fit target (fit2_w_mm),core,build,finalize
qc_skel_fit_err_mm_dyn,<f8,[],mm,-,1.2,mean joint error of poses_dyn vs its fit target (fit2_w_mm),core,build,finalize
qc_skel_fit_err_rise_mm,<f8,[],mm,-,1.2,"qc_skel_fit_err_mm_dyn - qc_skel_fit_err_mm: SKEL joint-centre fit error (fit2_w_mm, each fit against its own AddB joint-centre target) of the dyn fit minus the kin fit; 0 on kin_retimed trials; NaN unless both are present. Named qc_pass2_marker_rms_rise_mm before S19 (it is not a marker RMS)",core,qc,finalize
qc_skel_foot_collapse,|b1,[],-,-,1.1,v1.1 metadata flag,core,build,finalize
qc_skel_frozen_dofs,str,[],-,-,1.2,'|'-joined names of skel_dof_frozen DOFs,core,qc,finalize
qc_skel_grf_desync,|b1,[],-,-,1.2,"own-trial force/mocap desync gap rule on `poses`/`trans` (kin fit, D0; never poses_dyn): activity_recovered in {walking, running} AND qc_skel_com_grf_vert_r < 0.8 AND (qc_qgt_pelvis_grf_vert_r - qc_skel_com_grf_vert_r) >= 0.1. Cross-trial forward propagation along capture_id is done by qc_postprocess_v12.py (qc_skel_grf_desync_final)",core,qc,finalize
qc_skel_grf_desync_final,|b1,[],-,-,1.2,force/mocap desync used by the QC masks: NOT force_label_usable_s8b,core,build,finalize
qc_skel_grf_desync_propagated,|b1,[],-,-,1.2,QC post-process desync_final on the v1.2 poses (diagnostic),core,finalize,finalize
qc_skel_grf_newton_mismatch,|b1,[],-,-,1.2,"v1.2 gap rule on poses (propagated along the capture when the post-process ran) OR the S8b soft flag: SKEL root path not consistent with the GRF, no time-lag evidence; force labels are kept",core,finalize,finalize
qc_skel_grf_newton_mismatch_s8b,|b1,[],-,-,1.2,"gap-rule trial without time-lag evidence (soft flag, S8b)",core,build,finalize
qc_skel_head_gt30_frac,<f8,[],fraction,-,1.2,fraction of frames with any head DOF beyond 30 deg,core,qc,finalize
qc_skel_head_wrist_zero,|b1,[],-,-,1.2,SKEL head (3) and wrist (4) DOFs identically 0,core,qc,finalize
qc_skel_height_err_m,<f8,[],m,-,1.1,"SKEL T-pose height (|y(v411) - y(v3466)| of the shaped template) minus height_m (v1.1 definition, per trial) [v1.2 QC recomputation; v1.1 value: provenance qc_skel_height_err_m_v11]",core,qc,finalize
qc_skel_knee_at_zero_frac,<f8,[],fraction,-,1.2,max over sides of the fraction of frames with |SKEL knee| < 1e-3 rad,core,qc,finalize
qc_skel_lr_asym_ankle_deg,<f8,[],deg,-,1.2,"(SKEL ROM_R - ROM_L) - (q_gt ROM_R - ROM_L), ROM = p95 - p5",core,qc,finalize
qc_skel_lr_asym_hipadd_deg,<f8,[],deg,-,1.2,"(SKEL ROM_R - ROM_L) - (q_gt ROM_R - ROM_L), ROM = p95 - p5",core,qc,finalize
qc_skel_lr_asym_hipflex_deg,<f8,[],deg,-,1.2,"(SKEL ROM_R - ROM_L) - (q_gt ROM_R - ROM_L), ROM = p95 - p5",core,qc,finalize
qc_skel_lr_asym_knee_deg,<f8,[],deg,-,1.2,"(SKEL ROM_R - ROM_L) - (q_gt ROM_R - ROM_L), ROM = p95 - p5",core,qc,finalize
qc_skel_mass_from_volume_kg,<f8,[],kg,-,1.2,qc_skel_volume_rest_m3 x 985 kg/m^3,core,qc,finalize
qc_skel_mass_volume_ratio,<f8,[],ratio,-,1.2,qc_skel_mass_from_volume_kg / mass_kg,core,qc,finalize
qc_skel_pelvis_tilt_at_bound_frac,<f8,[],fraction,-,1.2,fraction of frames with SKEL pelvis_tilt <= fitter lower bound (-0.35) + 1e-4 rad,core,qc,finalize
qc_skel_pelvis_yaw_dev_frac_gt15,<f8,[],fraction,-,1.2,fraction of frames with |dev| > 15 deg,core,qc,finalize
qc_skel_pelvis_yaw_dev_max_deg,<f8,[],deg,-,1.2,"max |dev|, dev = circular difference (SKEL poses[:,2] - q_gt[:,2]) minus its circular mean",core,qc,finalize
qc_skel_pelvis_yaw_frozen,|b1,[],-,-,1.2,pelvis_rotation frozen,core,qc,finalize
qc_skel_pose_jump_dofs,str,[],-,-,1.2,SKEL pose DOFs (except pelvis_rotation) with a frame-to-frame jump > 1 rad,core,qc,finalize
qc_skel_pose_jump_max_rad,<f8,[],rad,-,1.2,largest such jump,core,qc,finalize
qc_skel_scapula_pinned,|b1,[],-,-,1.2,"any of the 6 scapula DOFs (abduction/elevation/upward_rot, both sides) sits within 1e-3 rad of a real (finite) v1.2 fit bound on > 50% of frames",core,qc,finalize
qc_skel_scapula_pinned_dyn,|b1,[],mixed,-,1.2,"D0 poses_dyn duplicate of the kin-target fields of the same name (without _dyn): computed on poses_dyn/trans_dyn instead of poses/trans. qc_skel_grf_desync is explicitly NOT duplicated here (must stay on `poses` only, S15b). All NaN/False when poses_dyn/trans_dyn are absent (v1.1)",core,qc,finalize
qc_skel_scapula_pinned_frac,<f8,[],fraction,-,1.2,the largest per-DOF at-bound fraction among the 6 scapula DOFs,core,qc,finalize
qc_skel_scapula_pinned_frac_dyn,<f8,[],mixed,-,1.2,"D0 poses_dyn duplicate of the kin-target fields of the same name (without _dyn): computed on poses_dyn/trans_dyn instead of poses/trans. qc_skel_grf_desync is explicitly NOT duplicated here (must stay on `poses` only, S15b). All NaN/False when poses_dyn/trans_dyn are absent (v1.1)",core,qc,finalize
qc_skel_scapula_uprot_at_bound_frac,<f8,[],fraction,-,1.2,max over sides of the fraction of frames with scapula_upward_rot within 1e-3 rad of the fitter bound -0.5236,core,qc,finalize
qc_skel_scapula_uprot_const_at_bound,|b1,[],-,-,1.2,scapula_upward_rot frozen at that bound on either side,core,qc,finalize
qc_skel_sole_minus_cop_m,<f8,[],m,-,1.2,"median over firm frames of (lowest skin vertex of that foot - CoP height); foot vertices = dominant skin weight talus/toes, exact LBS with pose blend shapes",core,qc,finalize
qc_skel_stance_skate_frac,<f8,[],fraction,-,1.2,"per stance run (foot vertical GRF >= 4 N/kg, GRF-OK, >= 5 frames), belt = median toes velocity; fraction of frames whose slower of calcn/toes belt-relative horizontal speed > 0.5 m/s",core,qc,finalize
qc_skel_stance_skate_frac_dyn,<f8,[],mixed,-,1.2,"D0 poses_dyn duplicate of the kin-target fields of the same name (without _dyn): computed on poses_dyn/trans_dyn instead of poses/trans. qc_skel_grf_desync is explicitly NOT duplicated here (must stay on `poses` only, S15b). All NaN/False when poses_dyn/trans_dyn are absent (v1.1)",core,qc,finalize
qc_skel_stepwidth_ratio,<f8,[],ratio,-,1.2,"walking/running: SKEL step width (calcn lateral offsets from the pelvis joint along the hip axis, firm frames) / heel-marker step width",core,qc,finalize
qc_skel_volume_posed_med_m3,<f8,[],m^3,-,1.2,median posed skin-mesh volume over the ~50 Hz subgrid,core,qc,finalize
qc_skel_volume_rest_m3,<f8,[],m^3,-,1.2,volume of the closed SKEL skin mesh with the trial betas in the rest pose,core,qc,finalize
qc_skel_vs_qgt_err_mm,<f8,[],mm,-,1.1,"mean 3D distance over frames and the 15 DIRECT joints between SKEL FK and flip_xz(AddB joint centres of q_gt), no alignment [v1.2 QC recomputation; v1.1 value: provenance qc_skel_vs_qgt_err_mm_v11]",core,qc,finalize
qc_skel_vs_qgt_err_mm_dyn,<f8,[],mixed,-,1.2,"D0 poses_dyn duplicate of the kin-target fields of the same name (without _dyn): computed on poses_dyn/trans_dyn instead of poses/trans. qc_skel_grf_desync is explicitly NOT duplicated here (must stay on `poses` only, S15b). All NaN/False when poses_dyn/trans_dyn are absent (v1.1)",core,qc,finalize
qc_skel_vs_qgt_err_rr_mm,<f8,[],mm,-,1.2,same after subtracting the pelvis joint per frame,core,qc,finalize
qc_skel_yaw_clamped_frac,<f8,[],fraction,-,1.2,fraction of frames with SKEL pelvis_rotation within 1e-6 rad of its trial min or max (1.0 if frozen),core,qc,finalize
qc_skel_yaw_offset_marker_deg,<f8,[],deg,-,1.2,"circular mean of (unwrapped SKEL poses[:,2] - unwrapped ASIS heading); ASIS heading from mean right minus mean left ASIS markers (markers_aligned), >= 100 frames",core,qc,finalize
qc_skel_yaw_offset_marker_p95_deg,<f8,[],deg,-,1.2,"95th percentile over frames of |wrap(poses[:,2] - ASIS heading)|",core,qc,finalize
qc_skel_yaw_r_marker,<f8,[],r,-,1.2,Pearson r of mean-removed unwrapped SKEL yaw and ASIS heading,core,qc,finalize
qc_skel_yaw_rate_p99_dps,<f8,[],deg/s,-,1.2,"99th percentile of |d poses[:,2]| * fps",core,qc,finalize
qc_skel_yaw_rms_marker_deg,<f8,[],deg,-,1.2,RMS of their difference (means removed),core,qc,finalize
qc_tau_cop_unreliable,|b1,[],-,-,1.1,tau_selfconsistent and MC used that fixed-point CoP (= qc_cop_fixed_point),core,build,finalize
qc_tau_gt_ankle_stance_mean_nmkg,<f8,[],N m/kg,-,1.2,"mean of the joint torque over frames with that foot vertical GRF > 0.1 BW_dyn, GRF-OK, |tau_gt|,|tau_sc| < 1e4 N m (both feet pooled), / mass_kg_dynamics; hip = hip flexion DOF",core,qc,finalize
qc_tau_gt_det_cop_unreliable,|b1,[],-,-,1.1,tau_gt_det used an unreliable CoP (v1.1),core,build,finalize
qc_tau_gt_det_wrap_corrupt,|b1,[],-,-,1.2,"tau_gt_det corrupted by the source pelvis-yaw wrap: 'q_gt' listed in the v1.1 fix_applied (P4 rebuilt the yaw; 1,045 Carter downhill trials)",core,qc,finalize
qc_tau_gt_det_wrap_evidence,|b1,[],-,-,1.2,tau_gt_det has a joint frame > 1e4 N m (data evidence; not specific to the wrap),core,qc,finalize
qc_tau_gt_extreme_frames,<i8,[],frames,-,1.2,"frames with max over joint DOFs 6-36 of |tau| > 3000 N m (all frames); gt = tau_gt_det, sc = tau_selfconsistent, mu = mu_tau",core,qc,finalize
qc_tau_gt_extreme_frames_grfok,<i8,[],frames,-,1.2,same on interior GRF-OK frames,core,qc,finalize
qc_tau_gt_extreme_frames_interior,<i8,[],frames,-,1.2,same on interior frames,core,qc,finalize
qc_tau_gt_hip_stance_mean_nmkg,<f8,[],N m/kg,-,1.2,"mean of the joint torque over frames with that foot vertical GRF > 0.1 BW_dyn, GRF-OK, |tau_gt|,|tau_sc| < 1e4 N m (both feet pooled), / mass_kg_dynamics; hip = hip flexion DOF",core,qc,finalize
qc_tau_gt_huge_frames,<i8,[],frames,-,1.2,frames with max joint |tau| > 1e4 N m,core,qc,finalize
qc_tau_gt_knee_stance_mean_nmkg,<f8,[],N m/kg,-,1.2,"mean of the joint torque over frames with that foot vertical GRF > 0.1 BW_dyn, GRF-OK, |tau_gt|,|tau_sc| < 1e4 N m (both feet pooled), / mass_kg_dynamics; hip = hip flexion DOF",core,qc,finalize
qc_tau_gt_sc_stance_sign_mismatch,|b1,[],-,-,1.2,any of ankle / knee / hip: sign of the gt stance mean differs from the sc stance mean,core,qc,finalize
qc_tau_loaded_eval_frames,<i8,[],frames,-,1.2.1,1.2.1 (R01b): eval_tau_frame_ok frames with a foot vertical GRF > 20 N; eval_torque_trial_ok requires > 0,core,qc,finalize
qc_tau_mu_extreme_frames,<i8,[],frames,-,1.2,"frames with max over joint DOFs 6-36 of |tau| > 3000 N m (all frames); gt = tau_gt_det, sc = tau_selfconsistent, mu = mu_tau",core,qc,finalize
qc_tau_mu_extreme_frames_grfok,<i8,[],frames,-,1.2,same on interior GRF-OK frames,core,qc,finalize
qc_tau_mu_extreme_frames_interior,<i8,[],frames,-,1.2,same on interior frames,core,qc,finalize
qc_tau_mu_huge_frames,<i8,[],frames,-,1.2,frames with max joint |tau| > 1e4 N m,core,qc,finalize
qc_tau_sc_ankle_stance_mean_nmkg,<f8,[],N m/kg,-,1.2,"mean of the joint torque over frames with that foot vertical GRF > 0.1 BW_dyn, GRF-OK, |tau_gt|,|tau_sc| < 1e4 N m (both feet pooled), / mass_kg_dynamics; hip = hip flexion DOF",core,qc,finalize
qc_tau_sc_extreme_frames,<i8,[],frames,-,1.2,"frames with max over joint DOFs 6-36 of |tau| > 3000 N m (all frames); gt = tau_gt_det, sc = tau_selfconsistent, mu = mu_tau",core,qc,finalize
qc_tau_sc_extreme_frames_grfok,<i8,[],frames,-,1.2,same on interior GRF-OK frames,core,qc,finalize
qc_tau_sc_extreme_frames_interior,<i8,[],frames,-,1.2,same on interior frames,core,qc,finalize
qc_tau_sc_hip_stance_mean_nmkg,<f8,[],N m/kg,-,1.2,"mean of the joint torque over frames with that foot vertical GRF > 0.1 BW_dyn, GRF-OK, |tau_gt|,|tau_sc| < 1e4 N m (both feet pooled), / mass_kg_dynamics; hip = hip flexion DOF",core,qc,finalize
qc_tau_sc_huge_frames,<i8,[],frames,-,1.2,frames with max joint |tau| > 1e4 N m,core,qc,finalize
qc_tau_sc_knee_stance_mean_nmkg,<f8,[],N m/kg,-,1.2,"mean of the joint torque over frames with that foot vertical GRF > 0.1 BW_dyn, GRF-OK, |tau_gt|,|tau_sc| < 1e4 N m (both feet pooled), / mass_kg_dynamics; hip = hip flexion DOF",core,qc,finalize
qc_tau_sc_q_kin_pre_armfix,|b1,[],-,-,1.2,"guard flag: tau_selfconsistent, mu_tau and sigma_tau of a re-timed trial computed on q_kin before the arm IK re-solve (mc_kinematics = q_kin_pre_armfix). False for every trial: the 3 affected vanderZee2022 trials were recomputed on the final q_kin in S20",core,build,finalize
qc_tau_wrap_corrupt,|b1,[],-,-,1.1,deprecated v1.1 QC,core,build,finalize
qc_trans_frame_alt_m,<f8,[],m,-,1.2,same without flip_xz (large on travelling trials; small on treadmill trials where the root stays near the origin),core,qc,finalize
qc_trans_frame_check_m,<f8,[],m,-,1.2,median |SKEL pelvis joint - flip_xz(q_gt pelvis translation)| (1-2 cm when trans is in the SKEL frame),core,qc,finalize
qc_trans_frame_ok,|b1,[],-,-,1.2,qc_trans_frame_check_m <= 0.25 m,core,qc,finalize
qc_trans_jump_m,<f8,[],m,-,1.1,largest frame-to-frame |delta trans| (v1.1 definition) [v1.2 QC recomputation; v1.1 value: provenance qc_trans_jump_m_v11],core,qc,finalize
qc_trans_jump_m_dyn,<f8,[],mixed,-,1.2,"D0 poses_dyn duplicate of the kin-target fields of the same name (without _dyn): computed on poses_dyn/trans_dyn instead of poses/trans. qc_skel_grf_desync is explicitly NOT duplicated here (must stay on `poses` only, S15b). All NaN/False when poses_dyn/trans_dyn are absent (v1.1)",core,qc,finalize
qc_upper6_rise10_dyn,|b1,[],-,-,1.2,same for poses_dyn,core,build,finalize
qc_upper6_rise10_kin,|b1,[],-,-,1.2,"own-target upper-6 joint error of poses rose by > 10 mm vs the first v1.2 fit (OURS, NOTES 7.16); QC flag, not an exclusion",core,build,finalize
qc_upper_body_unobserved,|b1,[],-,-,1.2,"marker_names is non-empty and none of them matches a torso/shoulder/elbow/wrist token (exact-token match after stripping a leading L/R, or a SHOULDER/ELBOW/WRIST/CLAV/ACRO substring); verified True for vanderZee2022_S09 (133 trials, lacks LAC/RAC/LEP/REP/LWR/RWR present in other vanderZee subjects) and False for vanderZee2022_S01",core,qc,finalize
qd_gt,<f4,"[T,37]",rad/s | m/s,addb_lab,1.0,pass-2 velocities,core,build,finalize
qd_kin,<f4,"[T,37]",rad/s | m/s,addb_lab,1.2,velocities of q_kin (AddB pass-1 convention),core,build,finalize
qdd_fix_corr_2diffq,<f8,[],r,-,1.0,"corr(qdd_gt, second difference of q_gt), v1.1 health check",core,build,finalize
qdd_fix_method,str,[],-,-,1.0,v1.1 qdd_gt correction method,core,build,finalize
qdd_fix_native_Tn,<i4,[],frames,-,1.0,native b3d trial length N,core,build,finalize
qdd_gt,<f4,"[T,37]",rad/s^2 | m/s^2,addb_lab,1.0,pass-2 accelerations (v1.1 corrected),core,build,finalize
qdd_kin,<f4,"[T,37]",rad/s^2 | m/s^2,addb_lab,1.2,accelerations of q_kin (AddB pass-1 convention),core,build,finalize
qdd_spike_frame,|b1,[T],-,-,1.2,~tau_plausible_frame dilated +-2 frames,core,qc,finalize
regen_note,str,[],-,-,1.0,v1 regeneration note,core,build,finalize
release_version,str,[],-,-,1.1,'v1.2': build version of the trial file format; unchanged in 1.3.0 (the dataset version is the manifest column dataset_version),core,build,finalize
retime_source,str,[],-,-,1.2,"re-timing provenance, 'none' when not re-timed",core,build,finalize
retime_version,str,[],-,-,1.2,"re-timing version, '' when not re-timed",core,build,finalize
s8b_decision,str,[],-,-,1.2,"S8b outcome: 'not_candidate', 'no_evidence', 'retime_constant', 'retime_changepoint', 'partial_invalidate', 'stop'",core,build,finalize
segment_index,<i8,[],-,-,1.2,"metadata: one AddB capture split into contiguous segments; prev/next = the contiguous neighbour (kinematics are NOT continuous at junctions, T3 verifier)",core,metadata,finalize
segment_role,str,[],-,-,1.2,metadata: main / transition / pause_within_task / static_calibration,core,metadata,finalize
setting,str,[],-,-,1.2,metadata: treadmill / overground / in-place / chair,core,metadata,finalize
setting_motion,str,[],-,-,1.2,"metadata: motion-only setting: treadmill (belt >= 0.15 m/s and net pelvis travel < 0.3 m/s), overground (net >= 0.3 m/s), stationary",core,metadata,finalize
sex,str,[],-,-,1.0,recorded sex of the person: male / female / unknown (metadata_v12),core,build,finalize
sex_resolved,str,[],-,-,1.1,"resolved sex; equal to sex since 1.3.0 (1.2.x kept the v1.1 value, male, for the 225 trials of the 11 female Tiziana2019 persons; v1.1 value in the manifest column sex_resolved_v11)",core,build,finalize
sex_source,str,[],-,-,1.2,metadata: male / female / unknown; source priority paper > b3d; Tiziana2019 11 persons (225 trials) female per Lencioni Demo_Anthro; vanderZee2022 unknown (not reported),core,metadata,finalize
shape_mass_target_kg,<f8,[],kg,-,1.2,person mass target of the shape fit,core,build,finalize
sigma_tau,<f4,"[T,37]",N m | N,addb_lab,1.0,"MC torque std: kinematic sensitivity on q_gt (pass 2; q_kin on re-timed trials), not the uncertainty of poses; normalize by mass_kg_dynamics only",core,build,finalize
skel_dof_at_lower_frac,<f4,[46],fraction,-,1.2,fraction of frames with the SKEL DOF within 1e-3 rad of the fitter lower bound (qc_fit_bounds.json = COMPREHENSIVE_POSE_LIMITS of the refit code),core,qc,finalize
skel_dof_at_upper_frac,<f4,[46],fraction,-,1.2,same at the upper fitter bound,core,qc,finalize
skel_dof_frozen,|b1,[46],-,-,1.2,SKEL pose DOF with std over the trial < 1e-9 rad,core,qc,finalize
skel_dof_valid,|b1,[46],-,-,1.2,"True if the DOF is free in the v1.2 fit (lower bound < upper bound). False = pinned to 0 during the main fit (head DOFs 23-25, wrist DOFs 34, 35, 44, 45). The head DOFs are then set by the head post-fit where head_postfit is True (1,948 trials). The wrist DOFs stay 0",core,build,finalize
skel_fit_code_version,str,[],-,-,1.2,"fit code version of poses (skel_refit FIT_CODE_VERSION; v1.2-fixround-2026-10-03c or -04d), from full12_kin/final_provenance.csv",core,build,finalize
skel_fit_code_version_dyn,str,[],-,-,1.2,"fit code version of the dyn fit (-03c pure pass 2, -07e arm splice, -08i repaired v3a, -09l repaired v4s), from full12_dyn/final_provenance.csv",core,build,finalize
skel_fit_metrics_json,str,[],-,-,1.2,"fit metrics of poses (JSON: fit2_w_mm, pelvis orientation errors, ...)",core,build,finalize
skel_fit_metrics_json_dyn,str,[],-,-,1.2,fit metrics of poses_dyn (JSON),core,build,finalize
skel_fit_source_run,str,[],-,-,1.2,"skel_refit run folder that produced poses (full12_kin, full12_kin_fix47, full12_kin_fix48, full12_kin_fixr3)",core,build,finalize
skel_fit_source_run_dyn,str,[],-,-,1.2,"skel_refit run folder of the dyn fit (full12_dyn, full12_dyn_e, full12_dyn_f, full12_dyn_g)",core,build,finalize
skel_fit_target,str,[],-,-,1.2,'kin_pass0_lp30',core,build,finalize
skel_fit_target_dyn,str,[],-,-,1.2,"'addb_pass2' (pure pass 2), 'addb_pass2_armsplice' (R1: kin re-solved arm spliced into pass 2), 'addb_pass2_repaired_v3a' (kin root/body in GRF-invalid and pass-2 root-deviation windows, kin values on pass-0 IK outliers, left-arm splice), 'addb_pass2_repaired_v4s' (v3a + left-arm position IK + AddB acceleration-minimizing filter on all 37 DOFs), or 'kin_retimed' (re-timed trials: poses_dyn = poses)",core,build,finalize
skel_fit_version,str,[],-,-,1.2,SKEL refit version string of poses,core,build,finalize
skel_fit_version_dyn,str,[],-,-,1.2,SKEL refit version string of poses_dyn,core,build,finalize
skel_gender,str,[],-,-,1.1,SKEL model gender used by both fits,core,build,finalize
skel_gender_source,str,[],-,-,1.2,'sex_label' (skel_gender = recorded sex) or 'height_rule' (sex unknown),core,build,finalize
skel_pose_names,str,[46],-,-,1.2,SKEL pose DOF names (columns of poses),core,build,finalize
source_task,str,[],-,-,1.2,metadata: task of the original capture (e.g. taichi for a standing segment of a taichi capture),core,metadata,finalize
speed_level,str,[],-,-,1.2,"metadata: slow / normal / fast where a source names the level (Carter running slow/norm/fast, Carter walk slow/fast; source protocol) or the Tiziana k-means level (natural -> normal; source derived_OURS); empty otherwise (Moore and vanderZee give speeds, not level names). speed_level_source used to hold the k-means description, now a token",core,metadata,finalize
speed_level_derived,str,[],-,-,1.2,metadata: Tiziana2019 level walking only: slow / natural / fast from a within-subject 1-D k-means of log speed (19 of 21 subjects with >= 6 trials; centres >= 10 % apart); toe/heel relabels excluded (D10),core,metadata,finalize
speed_level_source,str,[],-,-,1.2,"metadata: slow / normal / fast where a source names the level (Carter running slow/norm/fast, Carter walk slow/fast; source protocol) or the Tiziana k-means level (natural -> normal; source derived_OURS); empty otherwise (Moore and vanderZee give speeds, not level names). speed_level_source used to hold the k-means description, now a token",core,metadata,finalize
speed_method,str,[],-,-,1.2,"metadata: measured horizontal speed: treadmill belt speed from the AddB-model stance foot (running: GRF firm-stance frames; walking: kinematic single support; other estimator as fallback); overground: median 1 s moving-average pelvis speed. FK of q_gt on the b3d DYNAMICS-pass skeleton, not SKEL",core,metadata,finalize
speed_mps,<f8,[],-,-,1.2,"metadata: measured horizontal speed: treadmill belt speed from the AddB-model stance foot (running: GRF firm-stance frames; walking: kinematic single support; other estimator as fallback); overground: median 1 s moving-average pelvis speed. FK of q_gt on the b3d DYNAMICS-pass skeleton, not SKEL",core,metadata,finalize
speed_nominal_kind,str,[],-,-,1.2,"metadata: absolute_protocol (Hamner 2-5 m/s, Carter fixed 12 km/h and walk 3/5 km/h, Moore 0.8/1.2/1.6, vanderZee Table 2) / relative_to_norm (Carter running slow/norm/fast) / self_selected / self_selected_levels (Tiziana natural-faster-slower series) / none",core,metadata,finalize
speed_nominal_mps,<f8,[],-,-,1.2,"metadata: protocol speed where the source defines it; label slow/norm/fast/fixed (Carter), fixed_belt (Hamner, Moore, vanderZee), self_selected, self_selected_varied (Tiziana)",core,metadata,finalize
speed_rel_factor,<f8,[],-,-,1.2,"metadata: 0.9 / 1.0 / 1.1 of the participant norm speed, Carter running only",core,metadata,finalize
step_length_factor,<f8,[],-,-,1.2,metadata: vanderZee2022 Table 2 multipliers of f* and s* (after the D8 31/33 swap); NaN where self-selected or not vanderZee,core,metadata,finalize
step_width_m,<f8,[],-,-,1.2,metadata: median mediolateral distance between the foot midpoints of successive contralateral stances (AddB model),core,metadata,finalize
stride_time_s_med,<f8,[],s,-,1.2,median heel-strike to heel-strike time of the same foot from grf_contact_state 0 -> 1 transitions (strides 0.4-2.5 s with no unknown (-1) frame inside); NaN where qc_gait_event_inconsistent,core,finalize,finalize
study,str,[],-,-,1.0,"source study id (Hamner2013 since 1.3.0; the AddBiomechanics source folder keeps the spelling Hammer2013, see b3d_path)",core,build,finalize
study_citation,str,[],-,-,1.1,citation key of the source study,core,build,finalize
subject_base,str,[],-,-,1.2,metadata: subject label of the source study (e.g. P002 for person Carter2023_S02); person_id is the release person key,core,metadata,finalize
subject_id,str,[],-,-,1.1,S<NN> within the study,core,build,finalize
subject_id_raw,str,[],-,-,1.0,AddB subject folder,core,build,finalize
subject_pure,str,[],-,-,1.0,AddB subject folder (v1 field),core,build,finalize
support_side,str,[],-,-,1.2,"metadata: one-leg / asymmetric tasks only (stand-oneleg, yoga-tree, yoga-treearms, ballet-legraise, hop, stand-kneecurl, dropjump-asym, squat-asym; 155 trials): right if the median share of vertical GRF on calcn_r over GRF-valid frames with total > 100 N exceeds 0.8, left if below 0.2, else both; unknown without such frames; empty for every other trial",core,metadata,finalize
support_side_grf_share,<f8,[],-,-,1.2,"metadata: one-leg / asymmetric tasks only (stand-oneleg, yoga-tree, yoga-treearms, ballet-legraise, hop, stand-kneecurl, dropjump-asym, squat-asym; 155 trials): right if the median share of vertical GRF on calcn_r over GRF-valid frames with total > 100 N exceeds 0.8, left if below 0.2, else both; unknown without such frames; empty for every other trial",core,metadata,finalize
t2m_motion_eval_ok,|b1,[],-,-,1.2,aliases of eval_motion_trial_ok / eval_torque_trial_ok,core,qc,finalize
t2m_torque_eval_ok,|b1,[],-,-,1.2,aliases of eval_motion_trial_ok / eval_torque_trial_ok,core,qc,finalize
tau_frame_ok,|b1,[T],-,-,1.2,GRF-OK(CoP) (grf_frame_ok without frames where a used contact has contact_status 3: tau depends on the CoP) AND frame in 5..T-6 AND max over joint DOFs 6-36 of |tau_selfconsistent| <= 3000 N m. The recommended torque loss mask. For evaluation use eval_tau_frame_ok (= tau_frame_ok AND kin_frame_ok),core,qc,finalize
tau_frame_ok_frac,<f8,[],fraction,-,1.2,mean of tau_frame_ok,core,qc,finalize
tau_gt_det,<f4,"[T,37]",N m | N,addb_lab,1.0,AddB native dynamics-pass torque (not reproducible from the shipped fields),core,build,finalize
tau_gt_det_usable,|b1,[],-,-,1.2,False on re-timed trials,core,build,finalize
tau_interior_absmax,<f8,[],N m (mixed),-,1.0,"v1.1 definition: max over 37 DOFs and interior frames (5-frame trim) of max(|tau_gt_det|, |tau_selfconsistent|); deprecated, use tau_interior_absmax_sc",core,build,finalize
tau_interior_absmax_gt,<f8,[],N m,-,1.2,max |tau_gt_det| over joint DOFs 6-36 and interior frames,core,build,finalize
tau_interior_absmax_mu,<f8,[],N m,-,1.2,max |mu_tau| over joint DOFs 6-36 and interior frames,core,build,finalize
tau_interior_absmax_sc,<f8,[],N m,-,1.2,max |tau_selfconsistent| over joint DOFs 6-36 and interior frames,core,build,finalize
tau_plausible_frame,|b1,[T],-,-,1.2,max over lumbar+arm DOFs (index 20-36) of |tau_selfconsistent| / mass_kg_dynamics <= 10 N m/kg,core,qc,finalize
tau_sc_corr_taugt,<f8,[],r,-,1.0,"mean over DOFs of r(tau_selfconsistent, tau_gt_det) (QC definition at finalize: GRF-OK(CoP) frames = grf_frame_ok without contact_status 3 frames)",core,build,finalize
tau_sc_corr_taugt_v11def,<f8,[],r,-,1.2,tau_sc_corr_taugt on grf_valid_mask (v1.1 definition); validation only,core,qc,finalize
tau_sc_kinematics,str,[],-,-,1.2,"kinematics under tau_selfconsistent: 'q_gt', or q_kin on re-timed trials",core,build,finalize
tau_sc_method,str,[],-,-,1.0,tau_selfconsistent method,core,build,finalize
tau_selfconsistent,<f4,"[T,37]",N m | N,addb_lab,1.0,"release ID engine on the shipped kinematics (tau_sc_kinematics) + cleaned forces, link masses scaled by mass_dyn_scale",core,build,finalize
tilt_deg,<f8,[],deg,-,1.2,"rotation applied to the GRF frame (Carter2023 inclined files), 0 otherwise",core,build,finalize
torque_label_usable,|b1,[],-,-,1.2,"force_label_usable AND NOT qc_tau_wrap_corrupt AND NOT qc_tau_cop_unreliable AND NOT qc_grf_foot_error (all three v1.1 fields, read as optional inputs) AND NOT qc_grf_single_channel (S19 F3: one plate carries both feet, so the per-foot GRF and the foot assignment of tau_selfconsistent are wrong)",core,qc,finalize
trans,<f4,"[T,3]",m,skel_world,1.0,"SKEL root translation of poses (pelvis joint = trans + rest offset, DETAILS.md ""Conventions"")",core,build,finalize
trans_dyn,<f4,"[T,3]",m,skel_world,1.2,SKEL root translation of poses_dyn,core,build,finalize
trans_frame,str,[],-,-,1.2,'skel_world': trans / trans_dyn are the SKEL root translation in the SKEL world frame,core,build,finalize
trial_idx,<i4,[],-,-,1.0,trial index inside the AddB b3d file,core,build,finalize
trial_name,str,[],-,-,1.1,"AddB trial name (may contain capture dates or initials, DETAILS.md ""Health status and sensitive information"")",core,build,finalize
trial_uid,str,[],-,-,1.1,"trial key (v1.1 name; in 1.3.0 the 40 Hammer2013_* keys became Hamner2013_*, see trial_uid_v12 and uid_map_v12_to_v13.csv)",core,build,finalize
trial_uid_v1,str,[],-,-,1.1,v1 key <AddB subject>__trial<NNNN>,core,build,finalize
trunk_event_ok_dyn,|b1,[T],-,-,1.2,same for poses_dyn,core,build,finalize
trunk_event_ok_kin,|b1,[T],-,-,1.2,False on frames of class A3 (AddB poor IK trunk shake) of poses; lumbar / thorax DOFs 17-22,core,build,finalize
activity_condition_v11,str,[],-,-,1.2,"v1.1 mixed condition string, provenance only (replaced by setting, gait_style, path, ...)",manifest,table,finalize
activity_name_v12,str,[],-,-,1.2,activity_name with the Tiziana2019 toe / heel walking relabel applied (18 trials),manifest,table,finalize
activity_proposed_n4_v11,str,[],-,-,1.2,"v1.1 activity_proposed (8-class N4 vocabulary), provenance only",manifest,table,finalize
age_source,str,[],-,-,1.2,"source of age (b3d subject header, the same person's other file, paper, or unknown)",manifest,table,finalize
belt_perturbation_n6,str,[],-,-,1.2,Moore2015 belt token of the N6 activity audit (perturbed / unperturbed / start_or_stop); belt_perturbation starts from it and takes the fitted perturbation-window label where the phase fit is confirmed,manifest,table,finalize
belt_perturbation_source,str,[],-,-,1.2,source of belt_perturbation,manifest,table,finalize
cadence_nominal_spm,<f8,[],steps/min,-,1.2,protocol target cadence (vanderZee2022 conditions); NaN elsewhere,manifest,table,finalize
chair_contact,|b1,[],-,-,1.2,trial-level protocol metadata: a chair is used in the trial. Not the frame-level npz field chair_contact [T] (seated mask of the Falisse2017 chair channel),manifest,table,finalize
footwear_source,str,[],-,-,1.2,"source of footwear (paper methods text, or not stated)",manifest,table,finalize
gait_style_source,str,[],-,-,1.2,"source of gait_style: activity_name_v12 lookup, or assumed (5 Fregly2012 trials)",manifest,table,finalize
health_detail_source,str,[],-,-,1.2,source of health_detail (b3d subject tag or paper),manifest,table,finalize
incline_source,str,[],-,-,1.2,source of incline_grade_pct / incline_label,manifest,table,finalize
mass_kg_dynamics_med,<f8,[],kg,-,1.2,median mass_kg_dynamics over the person's trials; per-person mass table (build_mass_per_person_v12.py),manifest,table,finalize
mass_kg_gaitdynamics,<f8,[],kg,-,1.2,"GaitDynamics WEIGHT_KG_OVERWRITE value (Moore2015, Uhlrich2023 only)",manifest,table,finalize
mass_kg_grf_gait_v12,<f8,[],kg,-,1.2,median mass_kg_grf over the person's walking / running trials; per-person mass table (build_mass_per_person_v12.py),manifest,table,finalize
mass_kg_grf_stand,<f8,[],kg,-,1.2,median GRF-implied mass over the person's standing trials (outliers vs the gait median dropped); per-person mass table (build_mass_per_person_v12.py); uses the GRF,manifest,table,finalize
mass_kg_published,<f8,[],kg,-,1.2,person mass reported by the source paper or dataset where available,manifest,table,finalize
mass_kg_recorded,<f8,[],kg,-,1.2,person mass recorded in the AddB subject data (per-person mass table (build_mass_per_person_v12.py)),manifest,table,finalize
mass_published_type,str,[],-,-,1.2,"kind of mass_kg_published (e.g. paper_table1_same_as_recorded, paper_table1_measured_at_calib_pose, not_available)",manifest,table,finalize
n_prov_keys,<i8,[],-,-,1.2,number of fields in provenance_v11/<npz_path> (build bookkeeping),manifest,table,finalize
next_uid,str,[],-,-,1.2,trial_uid of the next segment of the same AddB capture (segment_index + 1); empty if none,manifest,table,finalize
npz_path,str,[],-,-,1.2,trial file path relative to the release root (trials/<study>/S<NN>/<trial_uid>.npz),manifest,table,finalize
npz_path_v11,str,[],-,-,1.2,trial file path in v1.1 (equal to npz_path),manifest,table,finalize
prev_uid,str,[],-,-,1.2,"trial_uid of the previous segment of the same AddB capture (capture_id, segment_index - 1); empty if none",manifest,table,finalize
qc_gap_rule_own_v12,|b1,[],-,-,1.2,own-trial v1.2 gap rule of this trial before the capture propagation (finalize bookkeeping; see qc_skel_grf_desync),manifest,table,finalize
setting_source,str,[],-,-,1.2,"source of setting: the N6 activity-audit condition token (motion checked, see setting_motion)",manifest,table,finalize
sex_raw_v11,str,[],-,-,1.2,"v1.1 sex string as shipped (mixed codes: male, female, m, f, unknown)",manifest,table,finalize
sex_resolved_v11,str,[],-,-,1.2,"v1.1 resolved sex (male / female / unknown), provenance of sex",manifest,table,finalize
speed_1s_max_mps,<f8,[],m/s,-,1.2,"same, maximum over 1 s windows",manifest,table,finalize
speed_1s_min_mps,<f8,[],m/s,-,1.2,locomotion only: minimum over 1 s windows of the measured speed (treadmill: AddB forward-kinematics belt speed; overground: pelvis speed); NaN otherwise,manifest,table,finalize
speed_belt_grf_mps,<f8,[],m/s,-,1.2,treadmill belt speed from AddB forward kinematics of the foot on GRF firm-stance frames (uses the GRF; leaky for force tasks),manifest,table,finalize
speed_belt_kin_mps,<f8,[],m/s,-,1.2,treadmill belt speed from AddB forward kinematics of the single-support foot,manifest,table,finalize
speed_nominal_note,str,[],-,-,1.2,"note on speed_nominal_mps (e.g. doubtful Carter2023 walk_slow file-naming value, perturbed Moore2015 belts)",manifest,table,finalize
speed_nominal_source,str,[],-,-,1.2,source of speed_nominal_mps and speed_protocol_label,manifest,table,finalize
speed_pelvis_1s_mps,<f8,[],m/s,-,1.2,median over 1 s windows of the AddB pelvis horizontal speed (overground speed_mps),manifest,table,finalize
speed_protocol_label,str,[],-,-,1.2,protocol speed label of the source (slow / norm / fast / fixed / fixed_belt / self_selected / self_selected_varied); empty where none,manifest,table,finalize
speed_ratio_to_nominal,<f8,[],-,-,1.2,speed_mps / speed_nominal_mps,manifest,table,finalize
split_addb,str,[],-,-,1.2,"force-benchmark split (train / test), person level. Redrawn on 2026-10-09 by the stratified person-level search (DETAILS.md ""Splits""); no longer the split of the AddBiomechanics source data (the metadata column dictionary text ""inherited AddBiomechanics split"" is outdated)",manifest,table,finalize
split_addb_val,str,[],-,-,1.2,"force-benchmark validation split: val persons drawn from the split_addb train persons, train = the rest of split_addb train, test = split_addb test",manifest,table,finalize
split_t2m,str,[],-,-,1.2,"text-to-motion split (train / val / test / patho), person level (stratified, 2026-10-09)",manifest,table,finalize
split_t2m_group,str,[],-,-,1.2,healthy / pathological_TKA / pathological_stroke,manifest,table,finalize
step_length_nominal_m,<f8,[],m,-,1.2,protocol target step length (vanderZee2022 conditions); NaN elsewhere,manifest,table,finalize
step_width_target_m,<f8,[],m,-,1.2,protocol target step width (vanderZee2022 conditions); NaN elsewhere,manifest,table,finalize
trial_uid_m11,str,[],-,-,1.2,"v1.1 manifest trial_uid used as the build join key (equal to trial_uid except the 40 Hamner2013 trials, renamed from Hammer2013 in 1.3.0)",manifest,table,finalize
trial_uid_v11,str,[],-,-,1.2,"v1.1 trial name, metadata join key (equal to trial_uid except the 40 Hamner2013 trials, renamed from Hammer2013 in 1.3.0)",manifest,table,finalize
trial_uid_v12_proposed,str,[],-,-,1.2,"renumbered name that a v1.2-only naming rule would give; NOT applied (it would reuse a v1.1 name for a different trial); differs from trial_uid in 2 trials (equal to trial_uid except the 40 Hamner2013 trials, renamed from Hammer2013 in 1.3.0)",manifest,table,finalize
uid_act_token_stale,|b1,[],-,-,1.2,the activity token inside trial_uid no longer matches activity_name_v12 (20 trials; the name is kept for key compatibility),manifest,table,finalize
usable_provisional,|b1,[],-,-,1.2,"provisional trial usability of the metadata workstream (audit flags, duplicates, no GRF) used by the split scripts; superseded by force_label_usable and the eval_* trial flags",manifest,table,finalize
beta_0,<f8,[],-,-,1.2,SKEL shape coefficient 0 of the person (= betas[0] in every trial of the person),persons,table,finalize
beta_1,<f8,[],-,-,1.2,SKEL shape coefficient 1 of the person (= betas[1] in every trial of the person),persons,table,finalize
beta_2,<f8,[],-,-,1.2,SKEL shape coefficient 2 of the person (= betas[2] in every trial of the person),persons,table,finalize
beta_3,<f8,[],-,-,1.2,SKEL shape coefficient 3 of the person (= betas[3] in every trial of the person),persons,table,finalize
beta_4,<f8,[],-,-,1.2,SKEL shape coefficient 4 of the person (= betas[4] in every trial of the person),persons,table,finalize
beta_5,<f8,[],-,-,1.2,SKEL shape coefficient 5 of the person (= betas[5] in every trial of the person),persons,table,finalize
beta_6,<f8,[],-,-,1.2,SKEL shape coefficient 6 of the person (= betas[6] in every trial of the person),persons,table,finalize
beta_7,<f8,[],-,-,1.2,SKEL shape coefficient 7 of the person (= betas[7] in every trial of the person),persons,table,finalize
beta_8,<f8,[],-,-,1.2,SKEL shape coefficient 8 of the person (= betas[8] in every trial of the person),persons,table,finalize
beta_9,<f8,[],-,-,1.2,SKEL shape coefficient 9 of the person (= betas[9] in every trial of the person),persons,table,finalize
mass_kg_recorded,<f8,[],kg,-,1.2,person mass recorded in the AddB subject data (per-person mass table (build_mass_per_person_v12.py)),persons,table,finalize
shape_fit_bone_err_mean_mm,<f8,[],mm,-,1.2,mean absolute bone-length error of the shape fit against the pass-0 AddB bones (bone_abs_err_mean_mm),persons,table,finalize
shape_fit_height_err_cm,<f8,[],cm,-,1.2,height error of the shape fit (shape_mass/shapes_v12_final_persons.csv height_err_cm),persons,table,finalize
shape_fit_loss,<f8,[],-,-,1.2,final objective of the shape fit (loss_final),persons,table,finalize
shape_fit_mass_err_rel,<f8,[],-,-,1.2,relative mass error of the shape fit (mass_err_rel),persons,table,finalize
shape_fit_multistart_flag,|b1,[],-,-,1.2,shape-fit restarts disagree in their final loss (loss_spread_flag),persons,table,finalize
shapy_chest_girth_m,<f8,[],m,-,1.2,chest girth of the shaped mesh at vertex 6489,persons,table,finalize
shapy_height_m,<f8,[],m,-,1.2,height of the shaped SKEL mesh: |y(HeadTop vertex 411) - y(HeelLeft vertex 3466)|,persons,table,finalize
shapy_hip_girth_m,<f8,[],m,-,1.2,hip girth of the shaped mesh at vertex 1210,persons,table,finalize
shapy_mass_kg,<f8,[],kg,-,1.2,shapy_volume_m3 x 985 kg/m^3,persons,table,finalize
shapy_volume_m3,<f8,[],m^3,-,1.2,volume of the shaped SKEL mesh,persons,table,finalize
shapy_waist_girth_m,<f8,[],m,-,1.2,waist girth of the shaped mesh at vertex 3501,persons,table,finalize
split_addb,str,[],-,-,1.2,"force-benchmark split of the person (stratified, 2026-10-09; DETAILS.md ""Splits"")",persons,table,finalize
split_addb_val,str,[],-,-,1.2,force-benchmark validation split of the person (train / val / test),persons,table,finalize
split_t2m,str,[],-,-,1.2,text-to-motion split of the person (train / val / test / patho),persons,table,finalize
split_t2m_group,str,[],-,-,1.2,healthy / pathological_TKA / pathological_stroke,persons,table,finalize
activity_confidence_v11,str,[],-,-,1.1,v1.1 value of activity_confidence,provenance,v1.1,finalize
activity_name_v11,str,[],-,-,1.1,v1.1 value of activity_name,provenance,v1.1,finalize
activity_recovered_v11,str,[],-,-,1.1,v1.1 value of activity_recovered,provenance,v1.1,finalize
age_v11,<f4,[],-,-,1.0,v1.1 age (provenance),provenance,v1.1,finalize
betas_v1,<f4,[10],-,-,1.1,v1.1 betas_v1 (provenance),provenance,v1.1,finalize
betas_v11,<f4,[10],-,-,1.0,v1.1 value of betas,provenance,v1.1,finalize
bmi_v11,<f4,[],-,-,1.0,v1.1 bmi (provenance),provenance,v1.1,finalize
contact_torque_aligned_v11,<f4,"[T,C,3]",N m,skel_world,1.0,v1.1 value of contact_torque_aligned,provenance,v1.1,finalize
contact_torque_v11,<f4,"[T,C,3]",N m,addb_lab,1.0,v1.1 value of contact_torque,provenance,v1.1,finalize
cop_align_gender_v1,str,[],-,-,1.1,v1.1 cop_align_gender_v1 (provenance),provenance,v1.1,finalize
cop_align_gender_v11,str,[],-,-,1.0,v1.1 value of cop_align_gender,provenance,v1.1,finalize
cop_aligned,<f4,"[T,C,3]",m,skel_world,1.0,v1.2 duplicate of cop_aligned_foot kept for v1.1 loaders,provenance,build,finalize
cop_aligned_foot_v11,<f4,"[T,C,3]",m,skel_world,1.0,v1.1 value of cop_aligned_foot,provenance,v1.1,finalize
cop_aligned_v11,<f4,"[T,C,3]",-,-,1.0,v1.1 cop_aligned (provenance),provenance,v1.1,finalize
cop_v11,<f4,"[T,C,3]",m,addb_lab,1.0,v1.1 value of cop,provenance,v1.1,finalize
direct_dof_mask_v11,|b1,[37],-,-,1.0,v1.1 value of direct_dof_mask,provenance,v1.1,finalize
extreme_tau_flag_v11,|b1,[],-,-,1.0,v1.1 value of extreme_tau_flag,provenance,v1.1,finalize
fix_applied_v11,str,[],-,-,1.1,v1.1 value of fix_applied,provenance,v1.1,finalize
grf_aligned_v11,<f4,"[T,C,3]",N,skel_world,1.0,v1.1 value of grf_aligned,provenance,v1.1,finalize
grf_v11,<f4,"[T,C,3]",N,addb_lab,1.0,v1.1 value of grf,provenance,v1.1,finalize
grf_valid_mask_v1,|b1,[T],-,-,1.1,v1.1 grf_valid_mask_v1 (provenance),provenance,v1.1,finalize
grf_valid_mask_v11,|b1,[T],-,-,1.0,v1.1 value of grf_valid_mask,provenance,v1.1,finalize
has_uncertainty_v11,|b1,[],-,-,1.0,v1.1 value of has_uncertainty,provenance,v1.1,finalize
health_status_v11,str,[],-,-,1.1,v1.1 value of health_status,provenance,v1.1,finalize
marker_names_v11,str,[K],-,-,1.0,v1.1 marker_names (provenance),provenance,v1.1,finalize
markers_aligned_v11,<f4,"[T,K,3]",-,skel_world,1.1,v1.1 markers_aligned (provenance),provenance,v1.1,finalize
markers_v11,<f4,"[T,K,3]",-,addb_lab,1.0,v1.1 markers (provenance),provenance,v1.1,finalize
mc_engine_v11,str,[],-,-,1.0,v1.1 mc_engine (provenance),provenance,v1.1,finalize
mc_grid_v11,str,[],-,-,1.0,v1.1 mc_grid (provenance),provenance,v1.1,finalize
mc_nonfinite_frac_v11,<f4,[],-,-,1.0,v1.1 mc_nonfinite_frac (provenance),provenance,v1.1,finalize
mc_seed_base_v11,<i8,[],-,-,1.0,v1.1 mc_seed_base (provenance),provenance,v1.1,finalize
mc_sigma_deg_v11,<f4,[],-,-,1.0,v1.1 mc_sigma_deg (provenance),provenance,v1.1,finalize
mc_source_v11,str,[],-,-,1.0,v1.1 mc_source (provenance),provenance,v1.1,finalize
mu_tau_v11,<f4,"[T,37]",N m | N,addb_lab,1.0,v1.1 value of mu_tau,provenance,v1.1,finalize
n_markers_v1,<i4,[],-,-,1.1,v1.1 n_markers_v1 (provenance),provenance,v1.1,finalize
n_markers_v11,<i4,[],-,-,1.0,v1.1 n_markers (provenance),provenance,v1.1,finalize
percentile_values,<i4,[7],%,-,1.0,"percent levels of percentiles: [1, 5, 25, 50, 75, 95, 99]",provenance,build,finalize
percentiles,<f4,"[7,T,37]",N m | N,addb_lab,1.0,"v1.2 MC torque percentiles (moved from core, S19 U3)",provenance,build,finalize
percentiles_v11,<f4,"[7,T,37]",-,-,1.0,v1.1 percentiles (provenance),provenance,v1.1,finalize
poses_v11,<f4,"[T,46]",rad,skel_world,1.0,v1.1 value of poses,provenance,v1.1,finalize
qc_angle_wrap_dofs_v11,str,[],-,-,1.1,v1.1 value of qc_angle_wrap_dofs,provenance,v1.1,finalize
qc_calcn_above_floor_m_v11,<f8,[],m,-,1.1,v1.1 value of qc_calcn_above_floor_m,provenance,v1.1,finalize
qc_cop_fixed_point_v11,|b1,[],-,-,1.1,v1.1 value of qc_cop_fixed_point,provenance,v1.1,finalize
qc_cop_foot_dist_m_v11,<f8,[],m,-,1.1,v1.1 value of qc_cop_foot_dist_m,provenance,v1.1,finalize
qc_cop_travel_m_v11,<f8,[],m,-,1.1,v1.1 value of qc_cop_travel_m,provenance,v1.1,finalize
qc_dynamics_mass_ratio_v11,<f8,[],-,-,1.1,v1.1 value of qc_dynamics_mass_ratio,provenance,v1.1,finalize
qc_grf_vertical_p01_bw_v11,<f8,[],BW,-,1.1,v1.1 value of qc_grf_vertical_p01_bw,provenance,v1.1,finalize
qc_knee_locked_v11,|b1,[],-,-,1.1,v1.1 value of qc_knee_locked,provenance,v1.1,finalize
qc_mass_grf_ratio_v11,<f8,[],-,-,1.1,v1.1 value of qc_mass_grf_ratio,provenance,v1.1,finalize
qc_skel_dj_gap_mm,<f8,[],-,-,1.1,v1.1 qc_skel_dj_gap_mm (provenance),provenance,v1.1,finalize
qc_skel_fit_err_mm_v11,<f8,[],mm,-,1.1,v1.1 value of qc_skel_fit_err_mm,provenance,v1.1,finalize
qc_skel_height_err_m_v11,<f8,[],m,-,1.1,v1.1 value of qc_skel_height_err_m,provenance,v1.1,finalize
qc_skel_vs_qgt_err_mm_v11,<f8,[],mm,-,1.1,v1.1 value of qc_skel_vs_qgt_err_mm,provenance,v1.1,finalize
qc_tau_cop_unreliable_v11,|b1,[],-,-,1.1,v1.1 value of qc_tau_cop_unreliable,provenance,v1.1,finalize
qc_trans_jump_m_v11,<f8,[],m,-,1.1,v1.1 value of qc_trans_jump_m,provenance,v1.1,finalize
qdd_gt_original_corrupt,<f4,"[T,37]",-,-,1.0,v1.1 qdd_gt_original_corrupt (provenance),provenance,v1.1,finalize
release_version_v11,str,[],-,-,1.1,v1.1 value of release_version,provenance,v1.1,finalize
sex_v11,str,[],-,-,1.0,v1.1 value of sex,provenance,v1.1,finalize
sigma_tau_v11,<f4,"[T,37]",N m | N,addb_lab,1.0,v1.1 value of sigma_tau,provenance,v1.1,finalize
skel_gender_v11,str,[],-,-,1.1,v1.1 value of skel_gender,provenance,v1.1,finalize
tau_interior_absmax_v11,<f8,[],N m (mixed),-,1.0,v1.1 value of tau_interior_absmax,provenance,v1.1,finalize
tau_sc_corr_taugt_v11,<f8,[],-,-,1.0,v1.1 tau_sc_corr_taugt (provenance),provenance,v1.1,finalize
tau_sc_method_v11,str,[],-,-,1.0,v1.1 value of tau_sc_method,provenance,v1.1,finalize
tau_selfconsistent_v11,<f4,"[T,37]",N m | N,addb_lab,1.0,v1.1 value of tau_selfconsistent,provenance,v1.1,finalize
trans_v11,<f4,"[T,3]",m,skel_world,1.0,v1.1 value of trans,provenance,v1.1,finalize
trans_v1_addb_frame,<f4,"[T,3]",-,-,1.1,v1.1 trans_v1_addb_frame (provenance),provenance,v1.1,finalize
trial_uid_v12,str,[],-,-,1.3,trial_uid of release 1.2.x (differs from trial_uid only for the 40 Hamner2013 trials); also in name_map,manifest,table,finalize
dataset_version,str,[],-,-,1.3,ForceBody dataset version of this table ('1.3.0'); the trial-file field release_version stays 'v1.2',manifest,table,finalize