court-measure
The groupant beside yom-hadin/src/measure.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 | descriptant |
| cost | partial | model.md |
| descriptant | present | descriptant |
| invariant | partial | Invariant |
| tests | finite native 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 |
denominator_projection | denominator_projection | partial source derived model | denominator_projection |
shift_control | shift_control | partial source derived model | shift_control |
recall_existential | recall_existential | partial source derived model | recall_existential |
gold_row_loss | gold_row_loss | partial source derived model | gold_row_loss |
Parts
Model text
descriptant
Own-rule selection and synthetic gold precision/recall accounting.
Epoch 2026-10-31
Prospective review preceded test edits. Runtime logic unchanged; three sources gain tests. Synthetic witnesses do not validate legal conclusions.
Composition: sample selects matching owner jurisdiction and sentence membership. judge_rule matches force and any act/condition verb, authority ASCII-insensitive. precision maps sentence through supplied order+shift, omits unavailable gold, excludes '?' authority denominator; recall counts each expected entry once if any sample matches.
Retained limits: partial syntax, opaque macros; no live corpus, teacher, parser model, production filesystem or frozen gold. Complete semantics and executable approximants remain open.
approximant
Forward model
Own-rule selection and synthetic gold precision/recall accounting.
sample selects matching owner jurisdiction and sentence membership. judge_rule matches force and any act/condition verb, authority ASCII-insensitive. precision maps sentence through supplied order+shift, omits unavailable gold, excludes '?' authority denominator; recall counts each expected entry once if any sample matches.
Not an executable substitute. Percentage approximation is unsupported without a useful metric; future profiles limited to 1/2/5/10% require actual-reference comparison.
models
Typed models: Own-rule selection and synthetic gold precision/recall accounting.
Subject: yom-hadin/src/measure.rs, SHA-256 c08d407318b3b1f8c70263d6ce08fd1325f4bbe5bb08c8bab82530a5c57511c7. Domain: bounded finite synthetic rules/cases; valid indices and successful allocation unless a section narrows this. Each relation observes implementation, not legal authority. Status: partial source-derived model, finite evidence mapped in tests.md.
behavior_complement
Excluded outputs include judging a sample absent from order, counting inherited rules as own when r.jur!=j.id, and unknown '?' authority increasing authority denominator. Metric correctness does not establish real legal precision.
Operator: stated relation on this source projection. Assumptions: named domain and unchanged dependencies. Evidence: tests.md maps finite observations; other branches remain source review only.
reverse_reconstruction
Formatted precision recovers compatible integer counts up to rounding and supplied sample size, not unique rules/gold. Recall hits ignore duplicate extracted matches; many corpora share the same formatted percentage.
Operator: stated relation on this source projection. Assumptions: named domain and unchanged dependencies. Evidence: tests.md maps finite observations; other branches remain source review only.
denominator_projection
judged counts only available ordered gold; fa uses judged, authority uses known gold among judged, both treats unknown authority as unconstrained. No single shared denominator is valid.
Operator: stated relation on this source projection. Assumptions: named domain and unchanged dependencies. Evidence: tests.md maps finite observations; other branches remain source review only.
shift_control
For matching order position k, compare gold[order[(k+shift)%len]]; absence of position returns None before modulo, so empty order is safe. Duplicate order names use first occurrence.
Operator: stated relation on this source projection. Assumptions: named domain and unchanged dependencies. Evidence: tests.md maps finite observations; other branches remain source review only.
recall_existential
For each expected(force,verbs), hit iff some same-sentence extracted rule has force and intersects verbs across acts+conditions. One extracted rule can satisfy multiple expected rows; no bijective matching.
Operator: stated relation on this source projection. Assumptions: named domain and unchanged dependencies. Evidence: tests.md maps finite observations; other branches remain source review only.
gold_row_loss
read_gold overwrites duplicate sentence keys, ignores notes, missing second column means no rules, malformed force/verb errors. gold_order retains repeated sentence order independently; these representations can disagree.
Operator: stated relation on this source projection. Assumptions: named domain and unchanged dependencies. Evidence: tests.md maps finite observations; other branches remain source review only.
tests
Checks and evidence
External run label: octopus-wave31-court-2026-10-31; exact commands in native-commands.json, per-source observations in source-test-map.json, numeric exits and named outcomes in raw logs.
19 exact selected native tests twice debug and twice release execute actual linked code on bounded hand-built cases. CLI covers argument helpers only. Full extraction/model initialization, actual jurisdiction loading, protocol disk errors, resource exhaustion and malformed public lattices are unexecuted. Module declarations have compile evidence only.
Three compiled source-behavior mutants: count weight, precedence tie, modal tie; each requires compile success, named assertion failure, exact byte/hash restoration and positive rerun. Legacy fixture interventions count separately. Final observations: external evidence-summary.md.
This node observes: Own-rule selection and synthetic gold precision/recall accounting.. Unmapped models remain source-derived/prospective, not verified by the whole test count.
The exact per-source names and limits are retained in source-evidence.md.
invariant-and-cost
Invariant and cost
Invariant
Metrics are descriptive formatted counts of supplied synthetic data; sample borrows original rules. Empty denominators format0%, and unknown authority can count jointly correct with force+act.
Cost
sample O(R log G); precision O(SO + SVGverbs), recall O(GentriesS*V); rows/strings allocated. shift addition must fit usize. No frozen assessment run.
Symbolic bounds, not timings. Resource exhaustion is not tested.
source-evidence
Source observation map
Actual in-memory sample/precision/recall/judge_rule; no frozen gold or read_gold filesystem.
- wave31_extraction_adjacency_disputes_and_metric_denominators
This page in Markdown โ https://md.groupants.elamur.ai/free/en/mova/yom-hadin/src/measure.rs.groupant/