Elamur Groupantsmova-nautilus-replay
The groupant beside tools/nautilus/src/replay.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 source derived | Forward approximant |
| ast | partial | syntax.json |
| composition | partial source derived | Composition |
| cost | partial source derived | Cost |
| descriptant | present | Descriptant |
| invariant | partial source derived | 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
model
mova-nautilus-replay
Descriptant
Replay: measure the loop on copies with earlier real answers. earlier(roots) reads tasks/.md and answers/.txt, the dictionary name of each question (both templates), answers by (name, word); run copies each configured layer's asked rows to OUT/layers, runs work::Loop with a Replayer terminal (draft: new: <old sense>; layer: the asked fields from the old columns, pick new when a sense field is written) and one store OUT/senses.jsonl, and reports questions, bytes, words per question, native picks (and their agreement with the earlier teacher's text), junk, thin, numbered/same/new/restates. Source: tools/nautilus/src/replay.rs.
Forward approximant
roots -> earlier questions and answers -> per layer: subset copy -> Loop steps until idle (min_gap 0, operator word 'replay', no ledger) -> proposals -> counts -> JSON report.
Invariant
Reads the terminal roots and the configured files only; writes only under OUT; the same inputs give the same report and store bytes (checked twice on 2026-11-04).
Cost
One loop run per layer over the copied rows; O(rows x examples) planning.
Composition
Parent: nautilus. Uses animal::Config, work::Loop, teacher::Config, improve::teacher_fields, senses.
behavior_complement
Never sends anything to a teacher or terminal; never writes the live folders or the layer files; the agreement figure is a lexical proxy, not gold. Evidence status: source-derived model with finite checks, not a proof.
reverse_reconstruction
The report recovers per-layer counts; OUT/state holds every replayed journal and proposal. The earlier answers are not reconstructed from it.
Epoch 2026-11-04
First model, written with the source; prospective: on new runs the native share stays near the replay's 10% and restatements near 10% of written senses.
tests
Checks
Command: cargo test --release --manifest-path tools/nautilus/Cargo.toml --offline --locked with MOVA_OCTOPUS_TEST_MODEL=$MOVA_DATA/download/0/0.4/en.bin.
2026-11-04: dictionary_names_of_both_templates; two replay runs on a snapshot of six instances' 129 questions gave byte-identical reports and stores. Finite fixtures; no teacher or model was called; not a proof.
This page in Markdown → https://md.groupants.elamur.ai/free/en/mova/tools/nautilus/src/replay.rs.groupant/