Concepts & architecture

Understand the theory powering calibrated explanations across binary and multiclass classification plus probabilistic and interval regression. Start with interpretation and alternatives, then follow the architecture threads that keep calibration guarantees intact.

For proofs, benchmarks, and citations that underpin these concepts, visit the Researcher hub hub before diving into individual guides.

Concept	Why it matters
Interpret explanations	Reuses notebook screenshots and walks through dual uncertainty plus the triangular plot.
Probabilistic & interval regression	Shows how calibrated probabilities and interval regression stay in lockstep across quickstarts and notebooks.
Alternatives & triangular plots	Explains how the triangular view pairs with rule tables for calibrated alternatives.
Architecture overview	Connects runtime components, caching, and plugin guardrails.
Error handling	Summarises runtime safeguards and expected exceptions.
Explanation structures	Documents the internal data structures used by CalibratedExplanation classes.
Guarded explanations	In-distribution filtering via conformal anomaly detection for trustworthy rules.
Calibrated interval semantics	Canonical semantics, assumptions, and non-guarantees for all three modes.
Terminology standardization	Summarises the shift to probabilistic regression terminology and compatibility guarantees.
Terminology: thresholded vs probabilistic regression	Clarifies language used across ADR-021 and user-facing regression docs.

• Probabilistic and interval regression
• Alternative explanations and ensured filters
• Architecture Overview
• Error handling and validation
• Explanation Structures
• Guarded explanations
• Calibrated interval semantics
• Terminology Standardization
• Terminology: thresholded vs probabilistic regression

Entry-point tier: Tier 3.