Introduced new truth labeling scheme with additional classes to facilitate fused ATL data products and improve machine learning model signal context; Converted bathypathfinder to C++, added logic to broaded bathymetry classifications in dense regions of point cloud; Converted C-SHELPh algorithm to C++; Converted MedianFilter algorithm to C++; Added new water surface classification algorithm based on dynamic median absolute deviation threshold; Added dynamic along-track partitioning logic to ensure strict compute resource limits for more robust and reliable processing in sliderule; Added new empirical constituent model (Denclue) and new trained constituent models (Hilbert Curve, SCBMLD, and RadSlice); Upgraded ensemble training framework to use trained constituent model multi-class probabilities in the feature set; Re-designed the final classification logic; Re-designed the blunder detection post processing logic to filter false positives using a connected-components algorithm; Overhauled the training pipeline to ensure out-of-fold predictions are used to train the ensemble; Overhauled the validation pipeline to ensure nested out-of-fold predictions are used to validate all constituent models and the ensemble; Added new light attenuation (Kd) estimation algorithm and new wind speed / surface roughness estimation algorithm; Removed external data dependencies from pipeline; Upgraded the cBlue derived uncertainty look up tables that incorporate an improved Monte-Carlo position uncertainty simulation Architecture; Removed the following /gtx/ parameters: invalid_kd, invalid_wind_speed, low_confidence_flag, sensor_depth_exceeded; Added the following /gtx/ parameters: segment_id, kd, surface_roughness; ITRF2014 was replaced with ITRF2020.