Elamur Groupantsmova-hard-case-list
The groupant beside tools/hard-case-list/src/lib.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 | Forward approximant |
| ast | partial | syntax.json |
| composition | partial | Composition |
| cost | partial | Cost |
| descriptant | present | Descriptant |
| invariant | partial | Invariant |
| tests | finite evidence | 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 |
Model text
Descriptant
The one shared hard-cases list (mova-hard-case-v2): append-only found/learned/open events per parser (en, gparse) keyed by the first 16 hex of SHA-256 of the trimmed text; reads v1 lines; record, sync_source, recheck, state. Source: tools/hard-case-list/src/lib.rs.
Forward approximant
state(lines) = fold over events: found sets the case's record and opens the event's parser view; learned/open set that view. record(found, parser) appends found for keys not open for the parser (or open without the licence the new report brings). sync_source(source, current) appends found for new keys and learned for that source's open keys missing from current. recheck(check) appends learned for judged-well open cases and open for learned cases judged bad (all=true); unjudged cases (None) unchanged.
Invariant
Append-only under an exclusive flock with fsync; one key per trimmed text; a parser view changes only by an event naming that parser; other sources' cases are never closed by a source sync; nothing is training-authorized.
Cost
O(L) per operation for L list lines (the whole file is read); appends are one write.
Composition
Used by tools/topic-corpus (hardcases.rs), peta/tools/wikicorpus (check.rs) and tools/hard-cases (wave, judge, main). Parent: hard-cases.
behavior_complement
Does not delete or compact lines, does not decide training admission, does not judge a sentence itself (callers pass the judge), does not authenticate writers. Counterexample guarded by tests: a topic-corpus sync never learns a Wikipedia case. Evidence status: source-derived model with finite checks, not a proof.
reverse_reconstruction
From the list, each case's latest found record (text, source, document pointer, licence, judges) and per-parser history are recoverable; the failed parse is recoverable only when draft_conllu was recorded (Wikipedia). Many event orders map to one state: no unique inverse.
Tests
hard-case-list unit tests (2026-11-04, 3): v1 lines read as v2 with per-parser views; record dedupes, licence completion, source-scoped sync and reopening; recheck learns, reopens regressions (all=true), honours the source filter and leaves unjudged cases unchanged (red controls). Finite fixtures and live samples; not a proof.
This page in Markdown → https://md.groupants.elamur.ai/free/en/mova/tools/hard-case-list/src/lib.rs.groupant/