Module quant_graph_grounding 

QuantGraph mode — selective grounding / repair against a NQuin fact graph.

Aggressive quantization can drop nuance; this module is the neuro-symbolic counterpart: after the LLM proposes text, high-stakes fact patterns are checked against an in-process fact graph (NQuin triples + human repair strings).

Gated by crate::inference_modes::quant_graph_grounding_enabled. Does not run in Portable or CudaTc modes.

Graph model

Each fact is a parity-valid NQuin:

• subject = place / entity (q_hash)
• predicate = q42:capitalOf (or other relation)
• object = answer entity hash
• context = q42:grounding-fact plus cold-path strings for prompt needles, answer tokens, and repair text.

Expand later: load from SPARQL / Wellfair graph / CBOR-LD package.

Structs

GroundingFact

Cold-path fact: NQuin + strings for matching and repair.

GroundingResult

Result of a grounding pass.

Constants

CTX_GROUNDING

Graph context for all grounding facts.

P_CAPITAL_OF

Predicate IRI for capital facts (hashed into quins).

Functions

export_fact_quins

Export all fact quins (for tests / graph dump / later SPARQL seed).

fact_count

Number of facts currently registered.

ground_generation

Unconditional grounding pass (tests / CLI).

load_facts_from_tsv

Load facts from a TSV file path. Returns count added (merge by reason).

lookup_capital_object

Lookup object hash for a subject entity if a capital fact exists (graph API).

maybe_ground_generation

Apply quant-graph grounding when the mode is active; otherwise identity.

parse_facts_tsv

Parse TSV lines: reason \t needles; \t answer_ok; \t repair \t place_iri \t city_iri

register_capital_fact

Convenience: capital-of fact from IRIs + match strings.

register_fact

Register (or replace by reason id) a grounding fact. Cold path.

reset_fact_store_to_defaults

Reset store to seed defaults (tests) — re-reads bundled TSV when present.

seed_facts_from_bundled

Re-seed from bundled TSV (or QUALIA_GROUNDING_FACTS), replacing current store.