Elamur Groupantsmova-elephant-suspect
The groupant beside tools/elephant/src/suspect.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 |
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.
| relation | 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
What counts cannot decide: suspicious edges (compound shadows and class outliers) and approach nodes put as bounded questions, strict parsing of the answers, and review-only proposals. Source: tools/elephant/src/suspect.rs.
Forward approximant
edges: shadows (a NOUN whose amod modifier is a VERB v with an obj or nsubj edge v -> noun, both counts >= 8) ranked first by the smaller count, then class outliers (a dependent sharing no top-8 isa class with the others of a coherent list, PMI >= median, count >= 8) by PMI x log2 count; known triples skipped. edge_task / node_task: bounded text. parse_edges / parse_nodes: fixed columns, known items and kinds, a compound phrase ending with one of the edge's words. keep: proposals.jsonl with provenance, wrong.proposed.tsv in the lists' columns, once per item.
Invariant
Deterministic order; every proposal is review_only and carries the question and answer hashes; nothing is written to the hand lists or the graph.
Cost
Shadows: O(sum of noun amod lists x two edge lookups); outliers: O(groups x 16 x 8); 7 977 suspects on the 50 000-article view in 0.36 s, identical twice (2026-11-06).
Composition
Parent: elephant. Called by work and main (suspects).
Behavior complement
Asks nothing itself; admits nothing. Evidence status: source-derived model with finite checks, not a proof.
Reverse reconstruction
proposals.jsonl recovers which answer gave which proposal (hashes, profile, time).
Tests
Checks
suspect::tests: strict parsing, kept once, the lists' own format; work test: the planted eat -> disorder edge found; measured recall: 10 of 23 hand-listed compound edges present in the 50 000-article graph among the suspects (class outliers alone: 1). These are finite fixtures and measurements; not a proof.
History
2026-11-06 · document
Epoch 2026-11-06
First model, written with the source (the Elephant made incremental and an animal on Mova).
Provenance
This page in Markdown → https://md.groupants.elamur.ai/free/en/mova/tools/elephant/src/suspect.rs.groupant/