Elamur Groupants⌕
☰

prag-schema

The groupant beside prag/src/schema.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

componentlabelwhere
approximantpartial source modelforward
astpartialsyntax.json
compositionpartial source modelcomposition
costpartial source modelcost
descriptantpartial source modelepoch-memory
invariantpartial source modelinvariant
testspartial source modelchecks

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.

idtypelabelwhere
prag-schema-behavior-complementbehavior_complementpartial source modelbehavior-complement
prag-schema-reverse-reconstructionreverse_reconstructionpartial source modelreverse-reconstruction
prag-schema-couplingconstraint_relationpartial source modelcoupling

Parts

Model text

model

File composition

forward

Tagged operations Closed(E,s), Get(f,L), Set(f,L,c), Parse(columns), Tsv(L), Json(L). Closed enums have cardinalities 8,16,3,3; six learned fields have cardinalities 8,16,2,2,3,3. Parsing seven supplied columns trims each, validates in form/act/indirect/hedge/polarity/voice/means order, and returns the first error or labels. TSV emits six names/bits and verbatim means. Default is decl/assert/false/false/none/none/dash.

behavior-complement

U consists of tagged operation inputs paired with outputs of the matching operation, including errors or panic observations. An inner complement excludes successful parses with wrong count, unknown categorical strings, non-bits, empty trimmed means or more than 30 whitespace words; excludes setter outputs changing an unselected coordinate or means; excludes a successful invalid-class setter. These concrete exclusions are source-derived, not established for all Rust executions by the finite tests.

reverse-reconstruction

Parse preimages retain all whitespace-padded columns producing the same trimmed labels. For a field-vector observation, means is unconstrained and cannot be reconstructed. TSV has a unique label preimage on canonical valid row labels, but parse(Tsv(L)) loses edge whitespace outside that domain and can fail on embedded TAB. JSON and TSV are distinct encodings.

invariant

Shared field order and class number couple parser, setter, serializer and model consumers. Canonical means domain D requires nonempty trimmed text, <=30 words, and no TAB/CR/LF. Parse(split_tab(Tsv(L)))=L on D. It is not a law for every publicly constructible Labels.

coupling

Shared field order and class number couple parser, setter, serializer and model consumers. Canonical means domain D requires nonempty trimmed text, <=30 words, and no TAB/CR/LF. Parse(split_tab(Tsv(L)))=L on D. It is not a law for every publicly constructible Labels.

composition

The closed table, coordinate and codec fragments share exact enums and field order. Classifier indices require the corresponding classes list; schema is not directly used by infer.rs. Agree uses string tables and does not validate this schema. Native tests exercise schema operations; caller inspection is not an executed cross-file join.

cost

O(B) parser/serializer time and output storage in input/output bytes B; fixed finite enum tables. Field operations O(1); classes allocates O(k), k<=16. No performance measurement or gain claim.

checks

Six wave13_models tests plus one prior schema test run twice per debug/release profile; evidence.md at the file root binds receipts. Literal independent tables cover 4608 categorical rows, 34 valid coordinate updates, 12 invalid boundaries, 127 first-error masks, five word counts and Unicode whitespace fibers. JSON coverage executes Form variants and default Labels; arbitrary serde error cases are untested.

epoch-memory

2026-10-30: prospective model recorded before edits. Old Boolean class 2 silently overwrote true with false; invalid_class_panics_before_mutation failed. The narrow guard preserves unit return and valid indices, consistently panicking before mutation. The prospective note estimated 36 valid classes; actual table sum is 34. The word-limit > to >= production mutant compiles and is rejected by rejection_order_and_word_boundary, then exact source is restored. No escaping/API redesign was introduced: noncanonical means remains an explicit transport limit. Syntax generation repeats exactly; macro bodies remain opaque, although methods are exercised natively. No live parser, teacher, corpus, network, GPU, training, filesystem-durability or full-system claim.

evidence

Executed evidence, 2026-10-30

Seven schema tests (six new) pass twice in debug and twice in release with identical sorted names/results. Old Boolean-index witness fails on Indirect 2. Word-limit production mutant compiles and fails the named boundary assertion; exact source restoration passes. Source-bound partial syntax generated twice identically. Receipt set octopus-wave13-schema-2026-10-30. SHA-256:

Structural checks use an external full-base registry with a validation-only wrapper over the unchanged base tree and this leaf. Both reports pass and are identical; ancestor integration remains the root agent's task. An earlier disconnected scaffold and a leaf-only registry were rejected; neither was a source/model failure or a passing check.

This page in Markdown → https://md.groupants.elamur.ai/free/en/mova/prag/src/schema.rs.groupant/