Post-processing Module

The spine.post module refines the objects and predictions coming out of model execution and construction. It collects configurable processors for physics cleanup, detector matching, calibration-aware corrections, and truth-aware bookkeeping.

Post-processing tools for ML-based neutrino reconstruction.

This module transforms raw ML model outputs into high-level physics quantities, serving as the bridge between deep learning predictions and physics analysis.

Core post-processing management:

PostManager coordinates configurable reconstruction cleanup pipelines.

Major processing categories:

Particle reconstruction, including direction, energy, PID, and vertex refinement.
Detector matching for optical flashes, CRT hits, and trigger information.
Event-level classification, containment studies, and quality metrics.

Output products:

Reconstructed particles and interactions with refined kinematics.
Truth matching associations and quality flags.
Detector-matching metadata and calibration-aware quantities.

Example configuration

post:
  chain:
    - name: ParticleBuilder
      track_min_length: 3.0
    - name: EnergyReconstructor
      method: csda
    - name: FlashMatcher
      time_window: 10.0

This module is essential for converting raw ML outputs into analysis-ready physics objects with proper uncertainties and quality assessments.

Module Index

This package sits after construction in the standard pipeline and before analysis or writing, making it the main place for converting raw reconstruction output into analysis-ready quantities.

`base`	Contains base class of all post-processors.
`manager`	Manages the operation of post-processors.
`factories`	Construct a post-processor module class from its name.
`template`	Post-processor module template.
`crt`	CRT module.
`optical`	Optical module.
`reco`	Reconstruction post-processor modules.
`trigger`	Trigger information processing module.
`truth`	Reconstruction quality evaluation module.