Elamur Groupantsmova-hard-cases-eel
The groupant beside tools/hard-cases/src/eel.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 | Forward approximant |
| ast | partial | syntax.json |
| composition | partial | Composition |
| cost | partial | Cost |
| descriptant | present | Descriptant |
| invariant | partial | Invariant |
| tests | finite 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 |
Model text
Descriptant
mova_eel::TrainingData over a track's harvested hard-case data and one round of the public Eel engine (trainer::run once) with its gates unchanged; the internal track must not install into the deployed model directory. Source: tools/hard-cases/src/eel.rs.
Forward approximant
ready() = (sha256 over the six learn files' hashes, missing ids) when train and dev are nonempty and witness.json binds the current train file; status -> generation; export -> immutable copy + generation re-check; round(track) = refuse internal==deployed model_dir and missing hard: prefix, else trainer::run(eel[track], once) and read its receipt.
Invariant
An export's generation equals the copied files' generation; not ready -> no engine call; the engine's agreement, held-out, dev, EWT and install gates apply unchanged.
Cost
File hashes O(bytes); the round is the engine's CUDA cost.
Composition
Implements the public Eel TrainingData contract; the engine admits hard: ids through fresh_train_prefixes. Parent hard-cases.
behavior_complement
Does not train by itself, run in a loop, install into the deployed directory for the internal track, or count unwitnessed records as training data. Evidence status: source-derived model with finite checks, not a proof.
reverse_reconstruction
The engine's generation directory keeps the export, receipt and candidate; from a receipt the exported files are recoverable by their hashes.
Tests
eel::tests (2026-11-04): readiness red controls (nothing, no dev, no witnesses, witnesses of another train file), status/export generation equality, a changed train file not ready. Finite fixtures and live samples; not a proof.
This page in Markdown → https://md.groupants.elamur.ai/free/en/mova/tools/hard-cases/src/eel.rs.groupant/