Elamur Groupantsprag-main-sentences
The groupant beside prag/src/main.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 | finite observations | 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 |
|---|---|---|---|
prag-main-sentences-behavior-complement | behavior_complement | partial source model | behavior-complement |
prag-main-sentences-reverse-reconstruction | reverse_reconstruction | partial source model | reverse-reconstruction |
Model text
forward
Split after maximal runs of ASCII .!?; trim and discard empty pieces; append subsequent pieces without alphabetic Unicode characters to the preceding piece separated by one space.
Domain: finite supplied valid token/tree/label data or isolated saved run files as scoped above. Observation: returned values, named errors, serialized TSV/JSON and public path composition. Operator: the scoped production functions at the manifest source hash. Approximation semantics: partial source-derived behavioral relation, exact on specified fixtures; no numerical approximation cap.
invariant
Punctuation runs remain attached; nonletter first piece remains a first output. This is heuristic segmentation, not a general sentence tokenizer.
cost
O(text bytes) time and storage.
behavior-complement
Splitting punctuation runs or dropping a trailing emoticon is excluded; decimal/abbreviation semantic handling is not claimed. Universe: scoped inputs paired with candidate observations. This inner exclusion model concerns code behavior, not just test failure; outside the stated assumptions remains unknown.
reverse-reconstruction
Whitespace trimming and inserted joining spaces lose input spacing; all strings yielding the same merged pieces are compatible antecedents. Target relation is {(y,x): F(x)=y} within the domain; no implemented complete preimage solver or injectivity claim.
composition
Preserve shared schema class, source voice, document identity/order and transport descriptor across producer/consumer boundaries. File headers and filtering wrap delegated library computation. Test fragments use public APIs; source module exports alone establish availability only.
checks
Five binary and two library wave17_models tests exercise literal options/numbers, Unicode segmentation, supplied CoNLL-U role positions, restricted/full world equality, saved-call metadata, schema/serialization/gate and public role/world wiring. Debug and release are each repeated twice; named outcomes compared. Main duplicate-key first-wins mutation and library transport substitution are compiled rejection controls. See file-root receipts.md.
epoch-memory
2026-10-30: prospective inspection identified false Claude provenance in gate_report. The old binary fails the Codex transport assertion; repair centralizes descriptor without renaming opus APIs. Compatibility accepts interleaved options, duplicate overwrite and --prefixed values; this behavior is intentionally preserved. No optimization or task-score gain claimed.
limitations
Supplied fixtures are finite; no proof of linguistic correctness, complete exported API coverage or live dispatch. Environment-derived model/effort and role-alone selection are not varied. Syntax projection omits macro bodies and semantic resolution; groupant source hash binds exact bytes.
This page in Markdown → https://md.groupants.elamur.ai/free/en/mova/prag/src/main.rs.groupant/sentences.groupant/