Elamur GroupantsโŒ•
โ˜ฐ

prag-frames-md-hebrew

The groupant beside prag/src/frames_md.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
astpartial../syntax.json
compositionpartial source modelcomposition
costpartial source modelcost
descriptantpartial source modelepoch-memory
invariantpartial source modelinvariant
testsbounded native and inspected../model.md#evidence

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-frames-md-hebrew-behavior-complementbehavior_complementpartial source modelbehavior-complement
prag-frames-md-hebrew-reverse-reconstructionreverse_reconstructionpartial source modelreverse-reconstruction
prag-frames-md-hebrew-mid-collisionidentity_projectionpartial source modelmid-collision
prag-frames-md-hebrew-field-transportcolumn_projectionpartial source modelfield-transport

Model text

Subject: prag/src/frames_md.rs SHA-256 023ca98fa461e9d1545915abd8071ecb542cc61999bbf5a054b712e5edf9b5e3. All entries share this exact revision. Status: partial source-derived relations, finite native evidence for pure seams, source inspection for effects. Domain: finite supplied inputs under the stated resource/arithmetic premises. Exclusions are inner source-derived families, not a complete behavioral complement. Reverse relations retain compatible antecedents but no complete inverse solver is implemented.

forward

hwords skips comments/fences, short rows, range/decimal IDs and unparsable integer IDs. Keep columns 0,1,2,6,5,9; malformed head becomes zero. kv uses first exact key=. bare deletes U+0591..U+05AF. MId map strips leading nondigits, last duplicate wins. Semicolon Frame parts inherit current role; unresolved IDs skip. Negation joins integer head to verb.id for four exact lemmas.

behavior-complement

Universe: supplied text/words and proposed projections. Exclude accepted range/decimal IDs, preserving deleted cantillation, resolving an absent MId, or a row with empty Frame. First-key/last-MId collision policies exclude their opposites.

reverse-reconstruction

hwords loses ignored columns and rejected lines; invalid and explicit zero heads coincide. bare loses accent positions. Rendering loses MId prefixes and shadowed duplicates. Full original verse reconstruction is ambiguous.

invariant

Accepted row order and field indices persist despite extra columns. Current role persists after an unresolved ID. Unique MIds and valid dependency trees are not enforced.

cost

O(B+W log W+VW+P log W+output). Multiple roles may render one word repeatedly.

mid-collision

For normalized key 2, a later record shadows earlier old for all participant references. References themselves are not prefix-normalized. Exact-row native fixture checks later new and inherited A0.

field-transport

Observe six HWord fields; extra/ignored columns remain arbitrary in reverse. Invalid head maps to zero whereas invalid id drops the row. Synthetic CoNLL-U checks feats/misc transport without parser execution.

composition

Integer dependency-head keys drive negation while normalized MId strings drive roles. hwords and hebrew_table are separate public APIs joined by inspected site callers.

epoch-memory

Wave14 first modeling epoch, 2026-10-30: preserves rendering and tuning selection semantics. Two public pure tuning seams are now actual production-path calls. Six rendering tests and four tuning tests record field loss, tolerance/duplicate behavior, strict selection thresholds and domain limits. No bug repair or speed claim. Existing semantics retained include newline passthrough, first key/last MId collision, zip truncation and unchecked score arithmetic. Evidence is finite; no parser, teacher, filesystem corpus, journal-backed tune, optimizer training or GPU execution.

limitations

Partial syntax binds source bytes; macro bodies remain opaque and semantic AST is absent. Native/structural checks are not universal proofs or full-system coverage. No approximation error budget or automatic substitution is claimed. File evidence distinguishes executed pure paths and inspected effects.

This page in Markdown โ†’ https://md.groupants.elamur.ai/free/en/mova/prag/src/frames_md.rs.groupant/hebrew.groupant/