Elamur Groupantscoref-corefud
The groupant beside coref/src/corefud.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 |
| composition | partial | composition |
| descriptant | present | descriptant |
| invariant | partial | invariant |
| 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 |
event_stack | event_stack | partial source derived model | event_stack |
coordinate_transport | coordinate_transport | partial source derived model | coordinate_transport |
serialization_order | serialization_order | partial source derived model | serialization_order |
effect_boundary | effect_boundary | partial source derived model | effect_boundary |
Parts
Model text
descriptant
coref-corefud
CorefUD bracket events, sentence-bounded mention stacks, normalized MISC output and tree-derived span heads.
State: Document sentences with optional Entity strings; read stack (eid, output index); writer mutates raw UdDoc rows/comments.
Epoch 2026-10-31: prospective analysis preceded test additions. Production behavior is unchanged. A registry entry records partial source-bound modeling, not verification.
No arbitrary malformed public rows/spans, allocation failure or full grammar proof. Writer zip truncation and unrefreshed derived views are modeled limitations.
approximant
Forward model
Scan values left to right. '(' emits Open(eid, first-hyphen remainder), ')' immediately emits Close for singletons; other text through ')' emits Close. Open appends (sent,start,MAX); Close removes the rightmost equal eid and sets inclusive end, rejecting no-open before cross-sentence. Final nonempty stack errors. Writer opens descending end then eid, closes descending start then eid; head field=head-start+1. span_head chooses external-parent candidates, minimum depth, nonpunctuation, lowest index; no candidates returns start.
Domain: acyclic, in-range document views from checked parsing unless explicitly discussing permissive adapters; stable environment during each call; finite addressable allocations. Public vectors can be malformed and panic.
No runtime substitute or numerical error budget is established.
invariant
Invariant and cost
For successful read, all emitted ends are assigned in the opening sentence. Writer valid-domain spans have start<=head<=end<sentence length and file_idx maps to the same raw words. strip removes only Entity/Bridge/SplitAnte fields across all rows. Output ordering does not establish mention disjointness.
B annotation bytes, M mentions, N words: reading O(B+M²) worst case for rightmost-search/remove; writing filters all mentions per sentence and per word, conservative O(B+NM log M), O(B+M) temporary/output storage; span_head O(span width).
Integer counters/lengths use Rust native arithmetic with no separate overflow certificate. Allocation exhaustion, panics and I/O failures are not experimentally characterized.
composition
Source-fragment composition
Fragments project exact byte ranges of this source; shared types/imports outside ranges remain prerequisites. Models compose through actual caller/callee state, not an error-budget sum.
- reading: bytes [1097,4115); Event scanner and rightmost matching stack produce opening-order mentions; syntax errors and stack errors return no vector.
- writing: bytes [4115,7492); Output brackets and selected raw MISC fields are normalized; strip acts on all raw rows, write only zipped document words.
- head: bytes [7492,8251); Tree candidate set -> minimum depth -> punctuation exclusion -> first index, with start fallback.
Parent relation: Scan values left to right. '(' emits Open(eid, first-hyphen remainder), ')' immediately emits Close for singletons; other text through ')' emits Close. Open appends (sent,start,MAX); Close removes the rightmost equal eid and sets inclusive end, rejecting no-open before cross-sentence. Final nonempty stack errors. Writer opens descending end then eid, closes descending start then eid; head field=head-start+1. span_head chooses external-parent candidates, minimum depth, nonpunctuation, lowest index; no candidates returns start.
Integration: checked document tree -> mention spans/features -> ordered resolver partition -> eval visibility -> CorefUD rows or existing score module. Source-specific exceptions and projection losses remain in models.md.
models
Typed models
behavior_complement
For well-formed Document, successful read excludes an unmatched close, final open or completed cross-sentence mention. write excludes old Entity/Bridge/SplitAnte on processed words; it does not exclude stale annotations in documents omitted by zip truncation. strip excludes these fields globally but preserves other raw fields.
reverse_reconstruction
R(y)={documents whose event fold produces ordered spans/eids/etype y}. Additional non-Entity MISC and discarded fields after etype are free. After rebuilding Document Entity views from written raw rows, valid sentence-bounded spans with stack-compatible same-eid nesting and writer-generated ids recover span/eid pairs (with the serialized e prefix), not original annotations or note/head text. Merely in-range spans are insufficient: crossing same-eid spans [0,2] and [1,3] read back as [0,3] and [1,2]. Existing Document Entity views remain stale after write. Head choice has multiple compatible trees.
event_stack
Close(e) binds max index in open stack with eid=e, so nested reused ids close inner first; unknown closes fail before final-open checking.
coordinate_transport
Span sent/start/end are zero-based inclusive; serialized relative head is one-based. No validation prevents underflow for head<start or out-of-range row indexing.
serialization_order
Open key=(-end,eid), close key=(-start,eid); singleton closes inline. Remaining MISC sorts by lowercase field key; equal-key stable order retained.
effect_boundary
write updates sequentially using docs.zip(per_doc), assumes valid raw indexes and does not refresh token views. strip is infallible only for structurally valid rows; neither provides rollback.
No executable percentage surrogate: categorical state/effect semantics currently lack a useful cheaper observation metric. Supported future caps are 1%, 2%, 5%, 10%; no profile is claimed here.
tests
Evidence — 2026-10-31
Safe synthetic command: env -u COREF_X -u COREF_MD -u PRAG_OFF cargo test -p coref tests:: -- --skip score::. Two debug and two release executions are retained externally with numeric exit status and sorted named outcomes. Selected checks:
- corefud::tests::parse_gum_values
- tests::wave27_entity_boundaries_and_head_ties
- tests::write_read_round_trip
- tests::resolver_ignores_gold_entities
See source-test-map.json and controls.json in the wave27 coref run receipts. The library declaration file has compilation evidence only. No corpus, teacher, GPU or speed measurement.
Partial scope: No arbitrary malformed public rows/spans, allocation failure or full grammar proof. Writer zip truncation and unrefreshed derived views are modeled limitations.
Measured receipts
Final source state: 13 named tests (12 library, 1 binary) passed twice debug and twice release, 52 total executed test instances; all four command exits0 and sorted named outcomes identical. Five added tests are four library boundary tests plus one main load test. The pre-existing score tests and groupant_transfers were filtered out.
Compiled actual-source control: tests::wave27_entity_boundaries_and_head_ties. Compile exit0, assertion-run exit101, exact bytes restored, positive rerun exit0. Retained logs: control-corefud-compile.log, control-corefud-red.log, control-corefud-restored.log.
Nine source AST projections were generated twice identically. Shared ancestor declarations are deferred to root integration; local owned leaves have no structural diagnostics. No timings or general accuracy claim.
Strict scoped Clippy: exit101 on four pre-existing production lints (corefud unnecessary_sort_by, doc if_same_then_else and clone_on_copy, mention collapsible_if). Those source files are byte-identical to the base. No unrelated lint refactor was made.
Review witness, 2026-10-31: an external Rust binary linked the actual coref/doc public APIs against bounded four-word synthetic input. It confirmed crossing same-eid spans [0,2],[1,3] reconstruct as [0,3],[1,2], existing Document Entity caches remain stale after raw write, and retree forces output word DEPS to _ while preserving the original input. Compile and execution exited0. This single debug witness is additional to the paired suite, not a new paired/release test. Source and numeric receipts: annotation wave27 external run, coref-review-witness.rs and coref-review-witness*.exit/log.
This page in Markdown → https://md.groupants.elamur.ai/free/en/mova/coref/src/corefud.rs.groupant/