schema_version: slac.sota_audit_protocol/v0.1
protocol_id: mid360-imu-lidar-extrinsic-vs-li-init-v2
scope:
  modalities:
  - imu
  - lidar
  quantities:
  - rotation
  - translation
  - time_offset
  category: targetless_imu_lidar
claim_text: On a hand-held MID360 recording of the RTK-SLAM dataset not used to develop
  the method (construction_seq2), Calibrex's full IMU-LiDAR extrinsic with the per-segment-gravity
  lever arm predicts held-out accelerometer-integrated motion at least as well as
  LI-Init's, and so does its lever arm alone.
minimum_dataset_families: 1
minimum_independent_rigs: 1
requirements:
- requirement_id: calibrex-estimate-construction_seq2
  phase: audit
  description: Calibrex returns a passing lever-arm estimate (construction_seq2)
  dataset_family: rtk_slam
  independent_rig: true
  rig_id: rtk-slam-mid360
  evidence_path: mid360_imu_lidar_v2_extrinsic_benchmark_construction_seq2.yaml
  evidence_sha256: 12084dff08070515ef350302d2b0f29c63d0aaec3d66294392ec910f6c1eca66
  method_id: calibrex_native
  statistic: failure_rate
  comparison: less_equal
  threshold: 0.0
  required: true
- requirement_id: calibrex-extrinsic-not-worse-than-li-init-construction_seq2
  phase: audit
  description: paired 95 % CI lower bound of the improvement over LI-Init >= 0
  dataset_family: rtk_slam
  independent_rig: false
  evidence_path: mid360_imu_lidar_v2_extrinsic_benchmark_construction_seq2.yaml
  evidence_sha256: 12084dff08070515ef350302d2b0f29c63d0aaec3d66294392ec910f6c1eca66
  method_id: calibrex_native
  reference_method_id: li_init
  metric: holdout_span_position_rms_m
  statistic: paired_improvement_ci95_low
  comparison: greater_equal
  threshold: 0.0
  required: true
- requirement_id: calibrex-lever-arm-not-worse-than-li-init-construction_seq2
  phase: audit
  description: paired 95 % CI lower bound of the improvement over LI-Init's lever
    arm (same rotation and clock offset) >= 0
  dataset_family: rtk_slam
  independent_rig: false
  evidence_path: mid360_imu_lidar_v2_lever_arm_benchmark_construction_seq2.yaml
  evidence_sha256: 4645d99fba8c599d5a01c9a559b8f89017d14f23f7157ca8e74f06d374c51db1
  method_id: calibrex_native
  reference_method_id: li_init_lever_arm
  metric: holdout_span_position_rms_m
  statistic: paired_improvement_ci95_low
  comparison: greater_equal
  threshold: 0.0
  required: true
- requirement_id: calibrex-absolute-fit-construction_seq2
  phase: audit
  description: mean held-out span position RMS <= 0.010 m
  dataset_family: rtk_slam
  independent_rig: false
  evidence_path: mid360_imu_lidar_v2_extrinsic_benchmark_construction_seq2.yaml
  evidence_sha256: 12084dff08070515ef350302d2b0f29c63d0aaec3d66294392ec910f6c1eca66
  method_id: calibrex_native
  metric: holdout_span_position_rms_m
  statistic: metric_mean
  comparison: less_equal
  threshold: 0.01
  required: true
provenance:
  generator: tools/build_mid360_imu_lidar_extrinsic_audit.py
  generator_version: 0.4.1
  command:
  - python
  - tools/build_mid360_imu_lidar_extrinsic_audit.py
  - docs/assets/mid360_imu_lidar_extrinsic_v2_scores
  - docs/assets
  - --round
  - v2
  source_sha256:
    preregistration: 9e4d79bff1df6caa7b74187d855e870a9204b30b835c63be978018b9908f8ac7
  created_at: '2026-09-30T08:47:04.976962+00:00'