Elamur GroupantsโŒ•
โ˜ฐ

mova-whale-world-query

The groupant beside tools/whale-world/src/query.rs in the Mova repository: what the code computes, what it never does, how to go back from its outputs, its invariants, cost and checks.

Facts

Implementation material. A Mova code model describes this code at one source revision: partial models, finite checks and their stated scope. It is not a reviewed general exposition, and its labels are shown as written, never raised.

Components

componentlabelwhere
approximantpartialForward approximant
astpartialsyntax.json
compositionpartialComposition
costpartialCost
descriptantpresentDescriptant
invariantpartialInvariant
testsfinite evidencetests.md

Obernants

The typed relations around the forward model: the behaviour complement (what the code never does), the reverse reconstruction (every input compatible with an output) and specialised views.

idtypelabelwhere
behavior_complementbehavior_complementpartial source derived modelbehavior_complement
reverse_reconstructionreverse_reconstructionpartial source derived modelreverse_reconstruction

Model text

Descriptant

Queries for writers over world(e): entity (kinds, forms, mentions, docs, grouped expansions), explain (explain_score plus a text-kind bonus, two per document, duplicates dropped), related (cites, cited by, same author, similar, idf shared-entity share), neighbourhood (breadth-first nearest documents with texts; --content streams every sentence), sentences (round-robin diverse examples), context (entities with explanations from the neighbourhood, related documents, every statement of the neighbourhood, optional byte cut); paging by offset and k. Source: tools/whale-world/src/query.rs.

Forward approximant

entity: key -> bucket -> visible postings -> by_doc, grouped definitions. explain: entity postings U rarest-word postings -> visible -> sentence -> explain_score (+ case-folded retry) + kind bonus -> sort (score, doc, text, s) -> two per doc, dedup -> page. related: edges both ways + idf of shared doc entities -> score -> sort. nearest: BFS over related up to radius and max_docs. context: nearest -> doc entities -> explain_in(within neighbourhood, excluding doc) -> statements of texts_of each near doc -> cut an eighth of the larger list until it fits.

Invariant

Answers are deterministic (ties by doc, text, sentence); every citation names a document of docs.jsonl and a sentence of that text's shard; nothing in the store is dropped to answer.

Cost

entity 10 ms, explain 0.2-0.3 s, related 70 ms, sentences 18 ms, neighbourhood 73 ms, its content 0.2 s, context 0.9-1.1 s on the six-paper world (cold process, 2026-11-05); explain and context scan all candidate postings (no cap).

Composition

Parent: whale-world; reads world::World; uses topic_corpus::corpus::explain_score.

behavior_complement

Does not write the store, rank by truth, read texts of later epochs than asked, or include documents without visible texts in related; shared-entity strength counts general nouns too. Evidence status: source-derived model with finite checks, not a proof.

reverse_reconstruction

Each answer item carries {doc, text, s}, from which the sentence is recovered exactly from the world; scores do not recover their components beyond the listed edges.

tests.md

cargo test --manifest-path tools/whale-world/Cargo.toml --offline with WORLD_EN_MODEL=$MOVA_DATA/download/0/0.4/en.bin (9 unit tests: dossier append-only/select/changed files, licence negative controls, citation markers, importers incl. a topic corpus with a Sea Spider page and a consultant draft, entity extraction, LRU, two byte-identical world builds with cited queries, incremental epochs with old worlds intact and a changed file refused; 2026-11-05). Live 2026-11-05 on six papers: two independent imports and builds byte-identical.

These are finite fixtures and live samples; not a proof.

This page in Markdown โ†’ https://md.groupants.elamur.ai/free/en/mova/tools/whale-world/src/query.rs.groupant/