📐gparse-cli

The groupant beside gparse/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

Components

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.

Parts

• gparse-cli-score
• gparse-cli-training
• gparse-cli-annotation

Model text

descriptant

Source-bound description

CLI dispatches train/eval/annotate/plan by arguments; plan parses positive budget before cfg and corpus. score counts only zipped sentence/token pairs, LAS additionally requires matching head, percentages divide max(n,1). train validates dev schedule, reads/selects corpora, optionally loads vocab/config/weights, allocates optimizer, writes model metadata, then runs shuffled batches with joint loss and warmup/decay. Full-dev strict improvements save weights; quick proxies cannot select checkpoint. annotate predicts, updates score comments, refills lexical fields and postverbal-not corrections then writes output.

Epoch 2026-10-31

Prospective model was recorded before edits. Added tests use bounded synthetic fixtures; production algorithms unchanged. Coverage is partial, not proof. No trained corpus, live teacher or throughput experiment was executed.

approximant

Forward approximant

CLI dispatches train/eval/annotate/plan by arguments; plan parses positive budget before cfg and corpus. score counts only zipped sentence/token pairs, LAS additionally requires matching head, percentages divide max(n,1). train validates dev schedule, reads/selects corpora, optionally loads vocab/config/weights, allocates optimizer, writes model metadata, then runs shuffled batches with joint loss and warmup/decay. Full-dev strict improvements save weights; quick proxies cannot select checkpoint. annotate predicts, updates score comments, refills lexical fields and postverbal-not corrections then writes output.

Domain: valid source-specific dimensions and resources stated in invariant; no automatic production substitution. Numerical observations are finite checks, not an end-to-end relative bound. Executable Rust surrogate: unsupported gap; existing exact tests are not surrogates. Future metric-specific caps may be 1/2/5/10%, never above 10%.

invariant

Invariant and cost

Training effects are nontransactional; partial metadata/checkpoint/log writes may survive errors. Score empty=all zero and truncation is intentional source behavior, not aligned-input verification. Default learning loss arc+rel+half(xpos+upos). No live training selected.

Cost

Corpus bytes and network inference/training dominate; plan O(S log S); scoring O(min paired tokens); optimizer state O(parameters), batch BT² and gradients. Epoch estimates can differ after sampling; no timing certificate.

Allocation, finite arithmetic, platform dimensions and external failures are not universally certified.

composition

Composition

CLI dispatches train/eval/annotate/plan by arguments; plan parses positive budget before cfg and corpus. score counts only zipped sentence/token pairs, LAS additionally requires matching head, percentages divide max(n,1). train validates dev schedule, reads/selects corpora, optionally loads vocab/config/weights, allocates optimizer, writes model metadata, then runs shuffled batches with joint loss and warmup/decay. Full-dev strict improvements save weights; quick proxies cannot select checkpoint. annotate predicts, updates score comments, refills lexical fields and postverbal-not corrections then writes output.

Scoped children

The byte-scoped children project distinct internal relations through the parent syntax file. The parent combines them in actual source order. Remaining declarations/tests outside child ranges belong to this parent. Their correctness does not establish external caller preconditions.

Interfaces and limits

Training effects are nontransactional; partial metadata/checkpoint/log writes may survive errors. Score empty=all zero and truncation is intentional source behavior, not aligned-input verification. Default learning loss arc+rel+half(xpos+upos). No live training selected.

Scores constrain counts over zipped pairs, not full dataset size; many unpaired suffixes and correct/incorrect arrangements give same percentages. Saved checkpoint is compatible with many earlier epochs below best mean and does not reveal all optimizer/RNG state.

models

Typed model catalog

behavior_complement

Zero or malformed plan budget excludes any cfg/corpus open. Incorrect head excludes LAS even if label matches. Proxy dev epoch cannot replace best checkpoint unless no dev files; equal full mean does not save. Annotation cannot preserve original raw formatting.

reverse_reconstruction

Scores constrain counts over zipped pairs, not full dataset size; many unpaired suffixes and correct/incorrect arrangements give same percentages. Saved checkpoint is compatible with many earlier epochs below best mean and does not reveal all optimizer/RNG state.

failure_precedence

Train dev_every/dev_quick validation precedes file reads. plan budget validation precedes cfg. annotate loads model before input/roles; output write follows all sentence edits.

checkpoint_selection

full_dev=(ep%dev_every==0 || ep==epochs); save iff (devs empty || full_dev) and mean>best. Empty dev set uses ep as mean.

schedule_state

lr_now=lr*min(step/warm,1)*max(1-step/total,0.02); total uses initial sampled batch count times epochs, not later exact counts.

metric_projection

UAS/LAS/XPOS/UPOS percentages over zipped pairs can hide length mismatches; blank sets divide by one.

tests

Checks and limits

wave30::score_zip_and_plan_failure_precedence; train/eval/annotate external effects remain unexecuted

Selected commands run twice debug and twice release; exact receipts and source-to-test mapping retained in wave30 gparse external run directory. Tests do not certify allocation failure, all NaN/overflow inputs, trained-model accuracy or corpus timing. Compiled mutants cover selected boundaries; per-source coverage is recorded in the external map, not presumed.

See source execution evidence for exact names and unexecuted paths.

evidence

Execution evidence — 2026-10-31

Source: gparse/src/main.rs

• wave30::score_zip_and_plan_failure_precedence: twice debug, twice release, all four exits 0.

Controls

1 source-local compiled assertion mutants; 1 observed control tests including dependency mutations. Mutant compile exit0, assertion exit101, byte-exact restoration and positive rerun exit0. Per-source controls are listed in external source-test-map.json, not implied for every source.

Limits

Only score and plan invalid/zero-budget helpers; live train/eval/annotate file effects unexecuted.

CUDA softmax comparisons use absolute tolerance2e-6. GELU grid maxima observed: values0.0004734993, derivatives0.00086838007, finite difference0.0000005534244635896357; respective strict caps5e-4/1e-3/1e-5. These are finite observations, not percentage profiles or throughput claims.

Receipts: external wave30 gparse run native-commands.json, results.json, controls.json and source-test-map.json. Sorted exact test names/outcomes match per profile; numeric diagnostics and timings need not be byte-identical. Strict scoped library Clippy reports ten pre-existing production lints; appended test blocks were formatted independently without production reformatting.

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