Elamur Groupantscoref-it
The groupant beside coref/src/it.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 |
priority | priority | partial source derived model | priority |
candidate_rank | candidate_rank | partial source derived model | candidate_rank |
span_trim | span_trim | partial source derived model | span_trim |
effect_scope | effect_scope | partial source derived model | effect_scope |
Parts
Model text
descriptant
coref-it
Classify it/this/that as pleonastic, event or anaphoric and select a preceding clause for event reference.
State: Valid Sent forest, word index, sentence vector and PRAG_OFF environment snapshot.
Epoch 2026-10-31: prospective analysis preceded test additions. Production behavior is unchanged. A registry entry records partial source-bound modeling, not verification.
No exhaustive lexical/ablation coverage or arbitrary invalid public Sent tests; percentage surrogate unsupported.
approximant
Forward model
Other forms and missing parents return Anaphoric. For it, enabled weather, time/copula, extraposition and small-clause rules return Pleonastic first. it-event off returns Anaphoric next; event verb/copula/because and bare demonstrative rules follow. antecedent computes disjoint left predicates ranked (depth,-position), but returns root-left preceding-conj slice first, then ranked candidate, then previous sentence's first predicate root; trim edge punctuation and trailing cc.
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
Returned antecedent references existing sentence/predicate and inclusive bounds on a valid forest. Classification is categorical and environment-dependent. A previous root must be VERB or have cop child; no earlier-than-previous search occurs.
reading O(n) child scans plus lowercase bytes; antecedent O(n²) conservative for n candidate ancestor walks, O(n) trimming; O(1) auxiliary besides temporary strings. Public invalid indexes panic.
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 [3848,5703); Ordered classification predicates and ablation switches.
- antecedent: bytes [5703,7589); Preceding clause candidate ordering and punctuation bounds.
Parent relation: Other forms and missing parents return Anaphoric. For it, enabled weather, time/copula, extraposition and small-clause rules return Pleonastic first. it-event off returns Anaphoric next; event verb/copula/because and bare demonstrative rules follow. antecedent computes disjoint left predicates ranked (depth,-position), but returns root-left preceding-conj slice first, then ranked candidate, then previous sentence's first predicate root; trim edge punctuation and trailing cc.
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
With all switches unset, a subject it attached to rain is not Event/Anaphoric even if later event conditions might apply. Missing antecedent excludes inventing an event mention; a bare demonstrative adjacent to NOUN/PROPN/ADJ is excluded from that final event rule, not from earlier event predicates.
reverse_reconstruction
For reading r, preimage includes all valid sentence/index/environment triples satisfying the first enabled applicable branch. Event result alone cannot recover its predicate, antecedent or lexical form. For antecedent tuple a, all forests preserving winning candidate and exclusion/rank relations are compatible.
priority
Pleonastic precedes it-event switch; event branches precede bare demonstrative adjacency. PRAG_OFF uses exact comma tokens without trimming.
candidate_rank
root_left preempts left even though both are computed; left uses minimum depth then greatest position. Previous-sentence fallback uses first head=0 only.
span_trim
Trim trailing punctuation or cc while b>a; trim leading punctuation while a<b; a one-token punctuation span remains possible.
effect_scope
No filesystem writes; repeated environment reads mean a concurrently changed environment is outside deterministic snapshot assumptions. Closed lexical classes are heuristic, not semantic truth.
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:
- tests::it_readings_keep_it_from_nouns
- tests::wave27_event_antecedent_and_empty_evaluation
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 exhaustive lexical/ablation coverage or arbitrary invalid public Sent tests; percentage surrogate unsupported.
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::it_readings_keep_it_from_nouns. Compile exit0, assertion-run exit101, exact bytes restored, positive rerun exit0. Retained logs: control-it-compile.log, control-it-red.log, control-it-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.
This page in Markdown → https://md.groupants.elamur.ai/free/en/mova/coref/src/it.rs.groupant/