Orchestrator Subsystem

The orchestrator subsystem (nav.nav_orchestrator) is the top-level driver that turns one observation into one NavResult. The driver class NavOrchestrator runs every registered NavModel’s feature extraction, applies the per-feature reliability gate, runs every feasible NavTechnique in two passes (prior-free, then prior-required), reconciles per-technique results into a final answer via the ensemble() combine, and emits a single NavResult carrying the headline offset plus the full diagnostic envelope.

The subsystem is more than a single class: it is a small package of cooperating dataclasses, helpers, and a pipeline. Each public component has its own page below.

Driver

• Orchestrator (NavOrchestrator)
  • Overview
  • Theory
    • Pipeline stages
    • Exception-sandbox discipline
    • Glob-pattern model and technique filters
    • Restrictions and assumptions
    • Sources of uncertainty
  • Configuration
  • Implementation
    • Call path
  • Examples

Per-image dataclasses

• Per-Image State (NavContext)
  • Overview
  • Theory
    • Restrictions and assumptions
    • Sources of uncertainty
  • Configuration
  • Implementation
  • Examples
• Final Output (NavResult)
  • Overview
  • Theory
    • Restrictions and assumptions
    • Sources of uncertainty
  • Configuration
  • Implementation
  • Examples
• Per-Feature Post-Mortem (NavFeatureSummary)
  • Overview
  • Theory
    • Restrictions and assumptions
    • Sources of uncertainty
  • Configuration
  • Implementation
  • Examples
• Reproducibility Envelope (Provenance)
  • Overview
  • Theory
    • Restrictions and assumptions
    • Sources of uncertainty
  • Configuration
  • Implementation
  • Examples

Per-image helpers

• Image Classifier (NavImageClassifier)
  • Overview
  • Theory
    • Restrictions and assumptions
    • Sources of uncertainty
  • Configuration
  • Implementation
  • Examples
• Per-Instrument Settings (InstrumentSettings)
  • Overview
  • Theory
    • Restrictions and assumptions
    • Sources of uncertainty
  • Configuration
  • Implementation
  • Examples
• Ensemble Combine (ensemble + EnsembleConfig)
  • Overview
  • Theory
    • Step 1 — drop spurious
    • Step 2 — drop at-edge
    • Step 3 — single-link Mahalanobis grouping
    • Step 4 — pick the highest summed-confidence group
    • Step 5 — precision-weighted merge
    • Step 6 — disagreement and conflict penalties
    • Step 7 — confidence-rank assignment
    • Restrictions and assumptions
    • Sources of uncertainty
  • Configuration
  • Implementation
  • Examples
• JSON Curation (build_metadata_dict)
  • Overview
  • Theory
    • Float rounding policy
    • Allow-list discipline
    • Restrictions and assumptions
    • Sources of uncertainty
  • Configuration
  • Implementation
  • Examples

The nav.nav_orchestrator.image_derivatives module also lives in this package; the orchestrator runs it once per image to populate the shared gradient and edge-distance-transform products on NavContext. Because the module exists to feed the distance-transform techniques, its dedicated page is filed under the techniques chapter at Image Derivatives (Shared Gradient and Edge DT).