roundtrip-hebrew-row-codec-prediction
The groupant beside tools/roundtrip/src/blocks/site_heb.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 | approximant |
| ast | partial | ../syntax.json |
| cost | symbolic source derived | invariant-and-cost |
| descriptant | present | descriptant |
| invariant | partial | invariant-and-cost |
| tests | executed bounded checks | 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.
| id | type | label | where |
|---|---|---|---|
behavior_complement | behavior_complement | partial source derived model | behavior_complement |
reverse_reconstruction | reverse_reconstruction | partial source derived model | reverse_reconstruction |
token_number | token_number | partial source derived model | token_number |
range_emission | range_emission | partial source derived model | range_emission |
given_priority | given_priority | partial source derived model | given_priority |
head_clamp | head_clamp | partial source derived model | head_clamp |
Model text
descriptant
Description and epoch memory
Token cuts select analyses by precedence, create ranges for multiword analysis, number nonpunct tokens, emit fixed ten columns and derived metadata. External head selects contextual offset first and clamps.
Epoch 2026-10-31
Prospective model retained externally before edits. Existing production bytes preserved; only bounded inline tests appended to six owned files. Models distinguish byte reconstruction with residual from semantic understanding and persistent learning. Finite tests and compiled mutants support selected boundaries only. Source inspection identified noninjective projections and explicit admissibility restrictions; no bug repair or performance claim.
Combiner input is this source snapshot, syntax projection, models, costs and retained receipts. Proposed next epoch: target the stated unexecuted branches with bounded witnesses before expanding any claim. AST remains partial, macros opaque; no name resolution or proof.
approximant
Forward relation
Token cuts select analyses by precedence, create ranges for multiword analysis, number nonpunct tokens, emit fixed ten columns and derived metadata. External head selects contextual offset first and clamps.
Domain: Ten fields per row; unique canonical FEATS/MISC keys for pair identity, '_' empty convention, delimiter-free nonempty analyses for serialization. Finite counts/products/indices, signed offsets avoiding overflow; supplied lexicon only, no global load.
No executable percentage surrogate is supplied: these categorical codec/effect observations have no justified lossy metric. Future caps are 1/2/5/10% only after defining metric, zero behavior and differential reference.
catalog
Source-bound typed models
Each section is a partial source-derived relation on the bound snapshot and declared domain; finite checks are falsification evidence, not a universal proof. Observation horizon is one call unless history is explicitly named.
behavior_complement
Punctuation carries preceding token number at least1 and emits no MId. Offset zero always yields root; no global resource loading in predict_with.
Domain and held state: Ten fields per row; unique canonical FEATS/MISC keys for pair identity, '_' empty convention, delimiter-free nonempty analyses for serialization. Finite counts/products/indices, signed offsets avoiding overflow; supplied lexicon only, no global load.
Status: partial source-derived model. Oracle: the named scoped tests in tests.md; unexercised branches remain inspection hypotheses.
reverse_reconstruction
Rows do not identify whether analysis came from given/exact/bare/heuristic; metadata retains parsed numeric verse fields but loses original numeric spelling and extra verse suffix.
Domain and held state: Ten fields per row; unique canonical FEATS/MISC keys for pair identity, '_' empty convention, delimiter-free nonempty analyses for serialization. Finite counts/products/indices, signed offsets avoiding overflow; supplied lexicon only, no global load.
Status: partial source-derived model. Oracle: the named scoped tests in tests.md; unexercised branches remain inspection hypotheses.
token_number
A token all-PUNCT does not increment n; num=max(n,1).
Domain and held state: Ten fields per row; unique canonical FEATS/MISC keys for pair identity, '_' empty convention, delimiter-free nonempty analyses for serialization. Finite counts/products/indices, signed offsets avoiding overflow; supplied lexicon only, no global load.
Status: partial source-derived model. Oracle: the named scoped tests in tests.md; unexercised branches remain inspection hypotheses.
range_emission
Only >1 analysis words generate range row at first word id.
Domain and held state: Ten fields per row; unique canonical FEATS/MISC keys for pair identity, '_' empty convention, delimiter-free nonempty analyses for serialization. Finite counts/products/indices, signed offsets avoiding overflow; supplied lexicon only, no global load.
Status: partial source-derived model. Oracle: the named scoped tests in tests.md; unexercised branches remain inspection hypotheses.
given_priority
given[ti] replaces exact and heuristic analyses without lexical validation.
Domain and held state: Ten fields per row; unique canonical FEATS/MISC keys for pair identity, '_' empty convention, delimiter-free nonempty analyses for serialization. Finite counts/products/indices, signed offsets avoiding overflow; supplied lexicon only, no global load.
Status: partial source-derived model. Oracle: the named scoped tests in tests.md; unexercised branches remain inspection hypotheses.
head_clamp
External offset sum clamps to0..nw; inner references are caller-validity responsibility.
Domain and held state: Ten fields per row; unique canonical FEATS/MISC keys for pair identity, '_' empty convention, delimiter-free nonempty analyses for serialization. Finite counts/products/indices, signed offsets avoiding overflow; supplied lexicon only, no global load.
Status: partial source-derived model. Oracle: the named scoped tests in tests.md; unexercised branches remain inspection hypotheses.
tests
Actual checks
blocks::site_heb::wave32::patch_fields_and_canonical_pair_fibersblocks::site_heb::tests::affixes_cut_an_unknown_tokenblocks::site_heb::tests::a_learned_tree_comes_back_with_its_patches
The listed parent/source tests exercise only their named mechanisms; other model statements are source inspection. Sixteen selected actual roundtrip library tests passed twice debug and twice release (64 exact-name executions), with identical sorted test outcomes. Eight compiled production-behavior mutants, one per source, failed assertions; exact originals were restored and the selected check passed. No assertion-only mutants or surviving compiled mutants. An initial mutation selector was ambiguous and stopped before writing that source; corrected selector retained separately.
Raw logs, numeric statuses, original/restored hashes, prospective.md, native-commands.json and syntax pairs: $MOVA_DATA/runs/octopus-wave32-rt_blocks-2026-10-31. No live model/corpus/site/learning workflow or full pipeline was executed.
invariant-and-cost
Invariants and cost
Punctuation carries preceding token number at least1 and emits no MId. Offset zero always yields root; no global resource loading in predict_with.
Rows do not identify whether analysis came from given/exact/bare/heuristic; metadata retains parsed numeric verse fields but loses original numeric spelling and extra verse suffix.
Cost
O(W²+T*W+input bytes+lookup work), output proportional to rows and fields; bounded compose recursion depth2.
Integer additions/products and allocation sizes must fit native types. Floating costs require finite values; no throughput or all-input guarantee follows from synthetic tests.
This page in Markdown → https://md.groupants.elamur.ai/free/en/mova/tools/roundtrip/src/blocks/site_heb.rs.groupant/prediction.groupant/