Elamur Groupantstopic-corpus
The groupant beside models/topic-corpus.groupant 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
| component | label | where |
|---|---|---|
| approximant | partial source derived | Forward |
| ast | not materialized | |
| composition | partial source derived | Composition |
| cost | partial source derived | Cost |
| descriptant | partial source derived | Descriptant |
| invariant | partial source derived | Invariant |
| tests | finite evidence | tests |
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.
| id | type | label | where |
|---|---|---|---|
behavior_complement | behavior_complement | partial source derived finite checks | behavior_complement |
reverse_reconstruction | reverse_reconstruction | partial source derived finite checks | reverse_reconstruction |
Model text
model
topic-corpus
Descriptant
Topic graphs and closed topic corpora for the paper-describing Whales (our Owls): a paper's neighbourhood from the library (stage 1), download requests for the most-cited missing open copies through the library's own driver (2), a lazy cross corpus annotated by Mova with a bounded parse cache and a shared hard-cases list (3), the Whale seed hook, pages through Sea Spider kept internal (4), cited descriptions (5), consultant drafts turned into claims followed to originals (6), and the paper's small world. Crate tools/topic-corpus (mova-topic-corpus), a standalone workspace depending on mova-whale (reader, knowledge store, terminal adapter), mova-seaspider (store reader), mova-octopus (relay), world and en.
Forward
library -> graph (neighbourhood) -> missing (ranked, routed, request dir; nothing started) | corpus (register cross corpus, parse hot, parse others on request) -> world (paper's small world) / seed (Whale knowledge, under learner.lock) / web (Sea Spider queue and import) / sol (terminal draft -> claims -> download and page requests feeding missing and web) -> cite (draft.json + knowledge provenance + corpus -> cited description).
Invariant
Every corpus document is internal; only open-licensed library texts are public-admissible; Sea Spider pages and their trees, consultant drafts and our notes never go public (operator rule 2026-11-04, enforced in licence, corpus::assert_public, public_export and cite --public). No stage starts a download stream, a terminal session or a fetch by itself.
Cost
Library load about 10 s (186k export rows); one OpenAlex snapshot scan per request (10-20 s); hot build about 16 s for 12 documents (2,508 sentences); further parses lazy and cached under a byte cap.
Composition
Leaves are file groupants under tools/topic-corpus/src/*.rs.groupant. Correct leaves do not prove the wiring: the interaction checks are the fixture tests that run graph -> missing -> corpus -> seed/web/cite/sol/world on one fixture library, and the live sample of 2026-11-04 on bryant-2017. The Whale hook writes the Whale's knowledge store through mova_whale::knowledge::Store (no Whale code change); the parent is crab-orchestration, beside whale-workers and seaspider.
behavior_complement
Domain: all configured runs. The crate does not: download (the library driver does), fetch pages (Sea Spider does), run a consultant without an explicit command and an untaken terminal lease, write a Whale's store while it learns, publish anything, admit a web page, draft or note to a public output, or judge claims beyond word overlap. Counterexample: a web record forged as public-admissible is refused by assert_public.
reverse_reconstruction
From the corpus registry every original is recoverable (path, Sea Spider URL and store, zstd original) with its hash; from a request directory every candidate with its route; from claims every citation's resolution. Parses are recoverable only while cached. Evidence status: source-derived with finite checks.
tests
Checks
cargo test --manifest-path tools/topic-corpus/Cargo.toml --offline with WORLD_EN_MODEL=$MOVA_DATA/download/0/0.4/en.bin (18 unit tests; fixture library, fake papers binary, fake terminal, local Sea Spider store; two identical runs on 2026-11-04). Without the model the model-dependent tests report a skip.
Negative controls 2026-11-04 (each red, then restored): public admission without the origin condition; assert_public without the Sea Spider/draft/note refusals; seeding without the learner lock; shadow-library filter off; public citing that ignores refusals; cache eviction ignoring the hot pin.
Live 2026-11-04 on bryant-2017-automatic-annotation-evaluation-error-types: graph twice byte-identical (10 s); missing --dry twice identical (201 candidates, 23 queued); build (80 registered, 12 hot, 2,508 sentences, 309 hard cases); world twice identical; cite of owl-1 round 28 (247 sentences); sol claims from a stand-in draft (6 claims).
This page in Markdown → https://md.groupants.elamur.ai/free/en/mova/models/topic-corpus.groupant/