Rules¶

Rules are deterministic check functions grouped by validation area. Each rule returns Issue objects with severity, check name, message, optional file, optional row, and metadata.

Files¶

Rule	Severity	Purpose
`check_required_files`	error	Require `metadata.json` and `calibration.json`
`check_empty_files`	error	Report empty CSV files
`check_missing_sensor_files`	error	Ensure declared sensors have matching `sensors/<name>.csv` files
`check_broken_paths`	error	Ensure CSV `path` references exist
`check_duplicate_filenames`	warning	Report duplicate filenames in the dataset tree

Metadata¶

Rule	Severity	Purpose
`check_metadata_schema`	error	Validate `metadata.json` shape
`check_declared_sensors_exist`	error	Ensure declared sensors are loaded
`check_duration_matches_timestamps`	warning	Compare metadata duration to observed sensor timestamp span
`check_dataset_version_present`	warning	Require a non-empty version string

Timestamps¶

Rule	Severity	Purpose
`check_monotonic_timestamps`	error	Require monotonically increasing timestamps
`check_duplicate_timestamps`	warning	Report duplicate timestamps
`check_large_timestamp_gaps`	warning	Report gaps above `timestamp_gap_threshold_sec`

When timestamp checks and sync checks run together, DatasetLint deduplicates overlapping burst-gap findings.

Sensors¶

Rule	Severity	Purpose
`check_sensor_columns`	error	Require sensor-specific columns
`check_sensor_dimensions`	error	Validate camera width and height
`check_sensor_frequency`	warning	Compare observed rate to `expected_sensor_rates`
`check_missing_frames`	warning	Detect likely missing frames from timestamp gaps

Column expectations:

camera: timestamp, path, width, height
IMU: timestamp, ax, ay, az, gx, gy, gz
GPS: timestamp, lat, lon, alt
other sensors: timestamp

Camera CSVs may use filename as a compatibility alias when path is absent.

Synchronization¶

Rule	Severity	Purpose
`check_sensor_time_overlap`	error or warning	Require overlapping sensor timestamp ranges and warn on low overlap ratio
`check_timestamp_offset`	warning	Report stream start offsets above `max_pairwise_sync_gap_sec`
`check_pairwise_sync_gap`	warning	Report pairwise median sync gaps above threshold
`check_missing_frame_bursts`	warning	Report burst-sized timestamp gaps
`check_frequency_stability`	warning	Report frame interval jitter above `frequency_jitter_ratio`

Calibration¶

Rule	Severity	Purpose
`check_calibration_exists`	error	Require calibration entries for declared sensors
`check_intrinsics_shape`	error	Require camera intrinsics to be 3x3
`check_intrinsics_values`	error	Require finite positive focal lengths
`check_extrinsics_shape`	error	Require translation[3] and rotation_quat[4]
`check_quaternion_norm`	error	Require normalized quaternions

Labels¶

Rule	Severity	Purpose
`check_label_columns`	error	Require detection label columns
`check_label_confidence_range`	error	Require confidence in `[0, 1]`
`check_label_geometry`	error	Require positive width and height
`check_label_timestamps_match_sensor_range`	error	Require labels inside observed sensor time range
`check_track_id_consistency`	warning	Report track class switches above threshold
`check_label_bbox_jumps`	warning	Report large bounding box center jumps
`check_label_missing_timestamps`	warning	Report likely missing labels inside a track
`check_duplicate_track_id_timestamp`	error	Report duplicate `track_id` at the same timestamp
`check_short_tracks`	warning	Report tracks shorter than `label_min_track_length`
`check_label_size_changes`	warning	Report abrupt bounding box size changes

Detection labels require:

timestamp, track_id, class, x, y, width, height, confidence

Trajectories¶

Rule	Severity	Purpose
`check_trajectory_columns`	error	Require trajectory columns
`check_trajectory_finite_values`	error	Require finite numeric values
`check_unrealistic_speed`	warning	Report speeds above `max_speed_mps`
`check_unrealistic_acceleration`	warning	Report acceleration above `max_accel_mps2`
`check_yaw_range`	warning	Require yaw in `[-pi, pi]`
`check_stationary_dataset`	warning	Report nearly stationary trajectories

Trajectory CSVs require:

timestamp, x, y, yaw, vx, vy

Configuration¶

Thresholds, per-rule enable/disable, and per-rule severity overrides are documented in Configuration. issue_regression_severity is separate and applies to diff regression classification.