Elamur Groupantsmova-whale-claims
The groupant beside tools/whale/src/claims.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
| component | label | where |
|---|---|---|
| approximant | partial | Forward approximant |
| ast | partial | syntax.json |
| composition | partial | Composition |
| cost | partial | Cost |
| descriptant | present | Descriptant |
| invariant | partial | Invariant |
| tests | finite evidence | tests.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.
| id | type | label | where |
|---|---|---|---|
behavior_complement | behavior_complement | partial source derived model | behavior_complement |
reverse_reconstruction | reverse_reconstruction | partial source derived model | reverse_reconstruction |
Model text
Descriptant
Numbered citations in a consultant's text: split (body sentences with their citation numbers, and the numbered reference list; shared with mova-topic-corpus sol.rs), check (every citation and reference entry must name a given source, and some sentence must cite), claims (each sentence followed to the sentences of the sources it cites) and claim_key (a claim's identity across epochs). Source: tools/whale/src/claims.rs.
Forward approximant
split(text) -> ([(sentence without brackets, sorted citation numbers)], {n -> reference line}); a heading «References/Bibliography/Sources/Works cited» opens the list, the next heading closes it; ranges «1-3» expand when shorter than 20. check(text, given) = Err(Unknown(sorted numbers not given)) | Err(Uncited) | Ok(split). claims(prefix, body, given, texts, min_support): per sentence the best content-word Jaccard over the cited sources' sentences; status uncited (no number) | sourced (best >= min_support) | weak.
Invariant
check never accepts a number outside the given set, in the body or the reference list; claims are in body order with ids prefix-c0001…; support names a cited source only; deterministic (BTreeMap/BTreeSet, ties keep the first sentence).
Cost
O(text) to split; claims O(S × C × T) for S sentences, C citations each, T sentences per cited source (word sets built once per source).
Composition
Parent: whale-workers. Used by epoch.rs (the consultant's text of an epoch) and, through split, by tools/topic-corpus sol.rs (draft claims → library sources).
behavior_complement
Not done: entailment or factual checking (support is lexical overlap), resolving reference lines to library works (topic-corpus does that), reading citation styles other than bracketed numbers ([Smith 2020] or footnotes are ignored, so such a text counts as uncited). Evidence status: source-derived model with finite checks, not a proof.
reverse_reconstruction
From a claim: its sentence text, cited ids and the supporting sentence with its overlap; from check's refusal: the offending numbers. The original text is not recoverable from claims (bracket contents and line breaks are dropped).
Epoch 2026-11-05 (created)
split moved here from topic-corpus sol.rs unchanged (its tests there still pass); measured on fake answers of 4,500 words: 169–185 claims per paper.
tests.md
claims::tests (2026-11-05): citations to unknown sources refused in the body and in the reference list, no citation refused; claims sourced / weak / uncited by support. Negative control: the unknown-number check off turns claims::tests and epoch::tests::a_citation_to_an_unknown_source_is_refused red. topic-corpus sol::tests::entries_and_split covers split.
This page in Markdown → https://md.groupants.elamur.ai/free/en/mova/tools/whale/src/claims.rs.groupant/