Elamur Groupantsworld-mwe-prediction
The groupant beside world/src/mwe.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 source model | forward |
| ast | partial | ../syntax.json |
| composition | partial source model | composition |
| cost | partial source model | cost |
| descriptant | partial source model | epoch-memory |
| invariant | partial source model | invariant |
| tests | partial source model | checks |
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 |
|---|---|---|---|
world-mwe-prediction-behavior-complement | behavior_complement | partial source model | behavior-complement |
world-mwe-prediction-reverse-reconstruction | reverse_reconstruction | partial source model | reverse-reconstruction |
world-mwe-prediction-knob-threshold | constraint_relation | partial source model | knob-threshold |
world-mwe-prediction-suppression-order | priority_relation | partial source model | suppression-order |
Model text
Configuration gates and fixed-lexicon span/pair assembly.
forward
Contiguous flag activates global form/lemma matching. Accept spans of >=2 with allowed entry kind, a VERB and content count>=min(configured minimum,span length). Arc flag scans retained order: ignore zero, out-of-range and self heads; require obj/compound:prt/iobj plus verbal parent and enabled global idiom category. Sort each two-position pair and suppress it if already contained in an output span.
invariant
Under aligned arrays and valid global spans, every output has at least two distinct indices and a VERB. Arc output pairs are sorted; global spans retain global ordering. The code assumes aligned public arrays, and cannot certify arbitrary manually malformed Sent values.
cost
Beyond opaque global matching, O(MN+NP*K) conservative scanning bound for M candidate spans and P outputs of size K, plus ordered string lookup costs. Output size and allocation are unbounded by knobs alone.
composition
The dense head field produced by read_cupt is consumed as h-1. Global filtering occurs before local gates, so rejecting a chosen longest global expression does not recover a suppressed shorter one. Tests cover disabled configuration and invalid-head rejection without activating global dictionaries.
epoch-memory
2026-10-30 first lexical epoch: old native witnesses returned raw heads [0,10,99], accepted noun-only overlap and duplicate token IDs, and panicked on an out-of-range head. Narrow fixes translate unambiguous IDs, omit invalid/missing arcs, and enforce the documented distinct/shared-verb condition. Prediction-order greedy matching remains intentionally unchanged; historical tuning scores can differ. No corpus retuning or measured speed claim.
checks
Native filter mwe::wave10_models; see ../evidence.md for fragments or evidence.md at the root. Finite supplied fixtures and independent oracles; no live parser, corpus, teacher or global dictionary activation. Partial syntax has opaque macro bodies and no semantic AST. No executable percentage surrogate or speed claim.
behavior-complement
Universe: aligned Sent, fixed lexical result and cfg, outputs. Inner absent transitions include arc emitted for zero/out-of-range/self head, nonverbal parent, unsupported relation or disabled category. Unknown lexical entries and contextual idiom senses remain outside the model.
Subject: current source hash and this node's declared scope. Relation describes one terminating call over the finite inputs stated above, with fixed dependencies and successful allocation. Status: partial source-derived model; named native fixtures support only their finite domain. Unobserved effects remain unknown.
reverse-reconstruction
For a predicted pair retain any valid verbal-head/object-or-particle attachment with an enabled matching idiom category, and any accepted contiguous length-two match yielding that pair. The vector alone does not identify dictionary source or whether an arc was suppressed by a larger span.
Subject: current source hash and this node's declared scope. Relation describes one terminating call over the finite inputs stated above, with fixed dependencies and successful allocation. Status: partial source-derived model; named native fixtures support only their finite domain. Unobserved effects remain unknown.
knob-threshold
Boolean controls compare >0.5; absent, zero, negative and NaN controls are false. min_content is numeric and capped at candidate span length, so a configured minimum above the length does not forbid that span if every token is content. Knobs() defaults differ from a missing-key cfg.
Subject: current source hash and this node's declared scope. Relation describes one terminating call over the finite inputs stated above, with fixed dependencies and successful allocation. Status: partial source-derived model; named native fixtures support only their finite domain. Unobserved effects remain unknown.
suppression-order
Arc pair q is appended iff no previous output o contains every q member. Contiguous results are inserted first; later arcs may be blocked by them. This prevents some duplicate pairs but is not a general deduplicated set or maximal matching objective.
Subject: current source hash and this node's declared scope. Relation describes one terminating call over the finite inputs stated above, with fixed dependencies and successful allocation. Status: partial source-derived model; named native fixtures support only their finite domain. Unobserved effects remain unknown.
This page in Markdown → https://md.groupants.elamur.ai/free/en/mova/world/src/mwe.rs.groupant/prediction.groupant/