Elamur Groupantsmath-solver-orchestration
The groupant beside math/src/solve.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 | partial bounded 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 |
failure_precedence | failure_precedence | partial source derived model | failure_precedence |
representation_transport | representation_transport | partial source derived model | representation_transport |
resource_boundary | resource_boundary | partial source derived model | resource_boundary |
domain_boundary | domain_boundary | partial source derived model | domain_boundary |
Parts
Model text
descriptant
Source-bound description
Batch/workers clamp>=1, sort final records by id. A parses last answer/id. B retries up to max_retries+1, shares costs/tokens, correct=verified&&grade. Memos retain verified B regardless correctness. C trains10epochs, tries top3 same-method/template/count/different-id skeleton matches, gates then accepts else copies B fields with by_slm=false.
Epoch 2026-10-31: prospective model before bounded synthetic tests; production behavior unchanged. Teacher/Lean loops and append-failure path unexecuted; no accuracy claim.
approximant
Forward relation
Batch/workers clamp>=1, sort final records by id. A parses last answer/id. B retries up to max_retries+1, shares costs/tokens, correct=verified&&grade. Memos retain verified B regardless correctness. C trains10epochs, tries top3 same-method/template/count/different-id skeleton matches, gates then accepts else copies B fields with by_slm=false.
Domain: Synthetic parsers/batching/fallback only. Sort does not determine append order; duplicate ids collapse maps.
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
Unverified B excludes correct B. Empty memory excludes template success, but fallback may verified=true/by_slm=false. Append failures do not exclude returned records because ignored.
Ordering: Ask→parse→gate→retry. Memo config/verified before id lookup. C chooses unique numeric option via preliminary ungated-options run then reruns gate.
Cost: Teacher retries+plan costs; C trains then scores memory and<=3templates/problem. O(records+memory+plans); unchecked token/counter addition.
composition
Composition and epoch memory
curric→Expect→plan→Record→Memo→C. CLI must enforce held exclusion, run_c does not.
Teacher/Lean loops and append-failure path unexecuted; no accuracy claim.
Partial syntax and finite tests do not establish all caller assumptions.
models
behavior_complement
Unverified B excludes correct B. Empty memory excludes template success, but fallback may verified=true/by_slm=false. Append failures do not exclude returned records because ignored.
reverse_reconstruction
Summaries admit permutations/many assignments. Integer token sharing loses remainder. C fallback omits B trace/history, so cannot reconstruct B record. Memo requires matching verified B but not correct.
failure_precedence
Ask→parse→gate→retry. Memo config/verified before id lookup. C chooses unique numeric option via preliminary ungated-options run then reruns gate.
representation_transport
curric→Expect→plan→Record→Memo→C. CLI must enforce held exclusion, run_c does not.
resource_boundary
Teacher retries+plan costs; C trains then scores memory and<=3templates/problem. O(records+memory+plans); unchecked token/counter addition.
domain_boundary
Synthetic parsers/batching/fallback only. Sort does not determine append order; duplicate ids collapse maps.
tests
Evidence, 2026-10-31
Selected actual-source checks: cargo test -p math --lib solve::wave28. Exact commands, numeric statuses, normalized outcomes, compiled assertion controls and restoration receipts are in the external wave28 mathcontrol run.
Teacher/Lean loops and append-failure path unexecuted; no accuracy claim.
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/solve.rs.groupant/