Elamur Groupants⌕
☰

math-plan-gate

The groupant beside math/src/plan.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

componentlabelwhere
approximantpartialapproximant
astpartialsyntax.json
compositionpartialcomposition
descriptantpresentdescriptant
invariantpartialinvariant
testspartial bounded evidencetests

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.

idtypelabelwhere
behavior_complementbehavior_complementpartial source derived modelbehavior_complement
reverse_reconstructionreverse_reconstructionpartial source derived modelreverse_reconstruction
failure_precedencefailure_precedencepartial source derived modelfailure_precedence
representation_transportrepresentation_transportpartial source derived modelrepresentation_transport
resource_boundaryresource_boundarypartial source derived modelresource_boundary
domain_boundarydomain_boundarypartial source derived modeldomain_boundary

Parts

Model text

descriptant

Source-bound description

Nonunderscore ids bind values. Execution error records and breaks; false checks accumulate but later steps continue; answer only if no failures. Gate can leave Some answer with failures. verified iff failures empty and answer present. Linear solve before polynomial; dependent inclusive enumeration; bounded claim falsification. Panic becomes empty error trace.

Epoch 2026-10-31: prospective model before bounded synthetic tests; production behavior unchanged. Finite checks not proof; no meaningful percentage surrogate for categorical gate.

approximant

Forward relation

Nonunderscore ids bind values. Execution error records and breaks; false checks accumulate but later steps continue; answer only if no failures. Gate can leave Some answer with failures. verified iff failures empty and answer present. Linear solve before polynomial; dependent inclusive enumeration; bounded claim falsification. Panic becomes empty error trace.

Domain: Bounded synthetic plans. Empty count/sum/list identities, min/max/first errors; empty claim range list evaluates one case. run catches rational panics, not all resource failures.

Executable Rust approximant unsupported: categorical/effect observations have no useful percentage metric yet. Future caps1/2/5/10% only; no automatic substitution.

invariant

Invariant and cost

Execution/check failures exclude verified. Some answer does not exclude gate failure. Only integer type explicitly enforced. Nonnumeric option may pass numerical check when no matching numeric option.

Ordering: Execute→checks→bind; exec error breaks, check failure suppresses answer. Linear domain unknown defaults true, nonlinear unknown domain errors during root filtering.

Cost: Step+trace bytes plus delegated arithmetic; enumerate product of ranges, max visited steps, list100000, timeout poll every4096 visited cases. claim instead caps condition-passing cases and has no leaf-level timeout poll; evaluator checks remain delegated. Independent checks add work; no hard global wall-clock/resource isolation.

composition

Composition and epoch memory

JSON→calc AST→Env→Trace text→grade/memory. Verification is internal consistency, not NL correctness.

Finite checks not proof; no meaningful percentage surrogate for categorical gate.

Partial syntax and finite tests do not establish all caller assumptions.

models

behavior_complement

Execution/check failures exclude verified. Some answer does not exclude gate failure. Only integer type explicitly enforced. Nonnumeric option may pass numerical check when no matching numeric option.

reverse_reconstruction

Trace constrains compatible plans/envs but loses metadata and computation paths. Verified answer admits wrong problem models that pass internal checks. vars_of retains last duplicate id index.

failure_precedence

Execute→checks→bind; exec error breaks, check failure suppresses answer. Linear domain unknown defaults true, nonlinear unknown domain errors during root filtering.

representation_transport

JSON→calc AST→Env→Trace text→grade/memory. Verification is internal consistency, not NL correctness.

resource_boundary

Step+trace bytes plus delegated arithmetic; enumerate product of ranges, max visited steps, list100000, timeout poll every4096 visited cases. claim instead caps condition-passing cases and has no leaf-level timeout poll; evaluator checks remain delegated. Independent checks add work; no hard global wall-clock/resource isolation.

domain_boundary

Bounded synthetic plans. Empty count/sum/list identities, min/max/first errors; empty claim range list evaluates one case. run catches rational panics, not all resource failures.

tests

Evidence, 2026-10-31

Selected actual-source checks: cargo test -p math --lib plan::tests. Exact commands, numeric statuses, normalized outcomes, compiled assertion controls and restoration receipts are in the external wave28 mathcontrol run.

Finite checks not proof; no meaningful percentage surrogate for categorical gate.

No live teacher/API, trained model, corpus/held-out, GPU, DB or expensive benchmark run. Parent syntax generated twice identically. Scope remains partial.

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