Struct TextEngine 

pub struct TextEngine { /* private fields */ }

Loaded text-projection models (SPO tagger + optional English SPLADE + optional gazetteer), reusable across files.

Implementations

impl TextEngine

pub fn load( ml_bundle: &Path, splade_dir: Option<&Path>, ) -> Result<TextEngine, Box<dyn Error + Send + Sync>>

pub fn enable_growth(&mut self, path: impl Into<PathBuf>)

Enable the growing gazetteer: merge any prior overlay at path, then learn newly-discovered multi-word entities during ingest and persist them back with Self::save_overlay.

pub fn save_overlay(&self) -> usize

Persist the learned overlay (no-op unless growth was enabled). Returns the learned-entry count.

pub fn enable_tuned( &mut self, dir: &Path, base_dir: &Path, tokenizer: &Path, max_len: usize, ) -> Result<(), Box<dyn Error + Send + Sync>>

Attach a tuned tagger (a save() directory from step 2) so ingest emits typed facets + polarity. base_dir is the encoder’s HF snapshot (config.json) and tokenizer its tokenizer.json.

pub fn enable_relations( &mut self, dir: &Path, spec: &VocabularySpace, ) -> Result<(), Box<dyn Error + Send + Sync>>

Attach Head C (step 5) to the tuned tagger so ingest emits bound relation tokens.

pub fn is_tuned(&self) -> bool

True when a tuned tagger is driving the entity/epistemic tiers.

pub fn project_situation(&mut self, sentence: &str) -> Situation

Project one sentence into a full Situation — tokens, numeric fields, and infon polarity. This is the ingest entry point that preserves everything the tuned tagger knows; project_sentence remains for callers that only need tokens.

pub fn project_sentence( &mut self, sentence: &str, ) -> (Vec<String>, Vec<(String, f64)>)

Project one sentence → (sorted de-duplicated tokens, numeric quantity fields). QTY spans that canonicalise (via units) become (field, si_value) pairs for (num …) range queries.

pub fn tokens_for(&mut self, sentence: &str) -> Vec<String>

Project one sentence to a sorted, de-duplicated token set (tokens only).

pub fn project_text(&mut self, text: &str, sink: &mut dyn FnMut(Situation))

Drive a sink with one Situation per sentence of text.