Elamur Groupants⌕
☰

mova-whale-run

The groupant beside tools/whale/src/whale.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
approximantpartialForward approximant
astpartialsyntax.json
compositionpartialComposition
costpartialCost
descriptantpresentDescriptant
invariantpartialInvariant
testsfinite evidencetests.md

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

Model text

Descriptant

One Whale run: pinned task queue, at most one teacher call per task, contract gates and immutable versioned bundles. Source: tools/whale/src/whale.rs. Public, domain-agnostic: concrete prompts, schemas, gate parameters and deployments come from pinned configuration.

Forward approximant

prepare: load Spec and teacher identity, validate items (kind, format, binding rule, metadata), read pinned export rows, compute admission, extract context, render prompt, write per-task prompt/context/metadata and a manifest with disjoint worker indices (index mod workers). run_with: honor STOP/PAUSE and storage floor, take the worker lease, pick the first own task without receipt, refuse unresolved reservations, withhold local-only items when the kind says so, verify source/prompt/context hashes, create the reservation, call the teacher once, store the reply, then admit (identity check, verbatim guard, projection, validate, publish) into a completed or quarantined receipt. publish: originality on the raw reply, status rewrite, artifacts, staged directory renamed into store/<prefix>/<kind>/<sha(id)>/<version>. project_quarantined_lengths: re-gate retained length quarantines without a teacher.

Invariant

Per task: at most one reservation and one teacher call; outcomes in {completed, quarantined, uncertain, local_only_no_remote_teacher}; uncertain or unreceipted reservations stop the worker; a published version directory is never overwritten; identity changes require a new run root; original receipts/replies are never modified by projection.

Cost

Preparation O(sum source bytes + prompts); per task one teacher call plus O(reply) validation and O(corpus) originality; status O(tasks).

Composition

Uses spec, source, contract, originality and util; launched by pool (workers) and swarm survey lanes (one-item runs).

behavior_complement

Excluded: calls without teacher/spec configuration, altered template/spec, changed sources, duplicate leases, unknown receipt outcomes, replies from another model/provider, any contract/originality failure (quarantine, no reprompt). Counterexample: a failed transport records uncertain and a second run makes no call. Evidence status: source-derived model with finite checks, not a proof.

reverse_reconstruction

A bundle manifest pins task, reply, spec and artifact hashes, so its reply and prompt are recoverable from the retained job directory; a receipt alone admits many replies; a quarantine reason partitions the failing gate but not the reply content. No unique inverse is claimed.

tests.md

16 tests with a fake teacher: missing configuration, altered template, word profile measures, disjoint immutable outputs, terminal quarantine, transport uncertainty, local-only withholding, pinned export, EPUB adapter, pause/stop, projection receipts and isolation, binding scope, queue bounds/unknown receipts, originality tail, lease and source changes.

Command: cargo test --manifest-path tools/whale/Cargo.toml --offline --locked (74 tests, fake teachers/executables; two identical runs on 2026-11-03). These are finite fixtures; no live run, model call or deployment was exercised.

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