Module abductive 

Abductive inference — Peirce’s “inference to the best explanation”.

Given observed effects, find the hypotheses that would account for them. This library (split per CLAUDE.md §10) covers the full abductive cycle:

• chain explanation (abductive_explanation) — walk explanatory edges back to a root;
• minimal explanation (minimal_explanation) — the parsimonious set of roots covering a set of observations (Peirce’s economy of hypotheses);
• counter-abduction (counter_abduction) — aggressively prune refuted hypotheses;
• probabilistic abduction (probabilistic) — Bayesian scoring / ranking of hypotheses;
• ATMS (atms) — assumption-based truth maintenance: minimal environments + nogoods.

Zero-heap throughout (bounded chains, caller-supplied out buffers, bitset environments).

Re-exports

pub use atms::env_subset;

pub use atms::holds_in;

pub use atms::is_nogood;

pub use atms::label_add;

pub use atms::label_holds;

pub use atms::Environment;

pub use probabilistic::bayesian_posteriors;

pub use probabilistic::best_hypothesis;

pub use probabilistic::Hypothesis;

Modules

atms

Assumption-based Truth Maintenance System (de Kleer’s ATMS).

probabilistic

Probabilistic abduction — Bayesian scoring and ranking of competing hypotheses.

Constants

MAX_ABDUCTION_DEPTH

Max backward-chaining depth for abductive explanation (bounded, zero-heap).

Functions

abductive_explanation

Abductive inference — walk BACKWARD along explanatory edges (hypothesis →explains→ effect, predicate == explains) from an observed effect to the root hypothesis that accounts for it. Returns that root, or None if the observation has no explanation in the rule set. Zero-heap.

counter_abduction

Counter-abduction: from candidates, drop every hypothesis that has been refuted (ruled out by an observation, or contradicted by an established fact), writing the survivors into out. Returns the surviving count — aggressive pruning of contradictory hypotheses. Zero-heap.

minimal_explanation

Minimal explanation generation (Peirce’s parsimony): the DISTINCT root hypotheses that together explain every observation in observations, written into out. A single hypothesis accounting for several observations appears once (the smallest covering set under the chain model). Observations with no explanation are skipped. Returns the count. Zero-heap.