Elamur Groupantsmova-whale-epoch
The groupant beside tools/whale/src/epoch.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.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.
| 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
The epoch driver (mova-whale epoch CONFIG): per document and epoch N, Mova writes a long cited text from the paper and its neighbourhood (compose::compose_epoch), registers it in the dossier, the Whale's own terminal writes from the given sources by path under a pinned prompt, its text is checked (citations only to given numbers, word bounds), registered (never public) and turned into claims; a per-document stop rule marks saturation. Source: tools/whale/src/epoch.rs.
Forward approximant
run(c, N, only, terminal): for each document of documents(c) (list | batches' assignments in order | batch::candidates), in order: skip if saturated before N; final states unchanged; sol_asking -> uncertain; else register original (epoch 0), read paper and neighbourhood texts, compose, cite, gate; fail -> mova_short|mova_gates; pass -> register mova -> (live) sol_inputs (original, mova, own earlier texts newest first, nearest key texts; hard caps) -> stage -> render -> status sol_asking -> ask -> error: sol_<kind>, stop the run | answer: check -> refused: sol_rejected | register sol, claims, (topic corpus), stop rule -> done|saturated.
Invariant
status.json is saved before every step; a question is asked at most once per (document, epoch) in a run root; a text below mova.words.min is never registered and nothing is asked for it; a consultant text with a citation outside the given numbers is never registered; generated texts are never public; one driver per run root (epoch.lock) and the terminal's holder.lock for a live run; documents are never ordered alphabetically.
Cost
Per document: Mova reads the paper and each neighbour text once (≈ 0.5–2 s per text; 7–27 s per paper with 12 neighbours, 2026-11-05) and composes in O(sentences × clauses) with the grammar gate per kept clause; one terminal question (minutes, the model's time); claims O(S × C × T).
Composition
Parent: whale-workers. Uses compose.rs (compose_epoch), reader.rs, gates.rs (Guard), knowledge.rs (optional knowledge), dossier.rs, claims.rs, terminal.rs (MovaTerminal, TerminalLease), batch.rs (candidates), topic.rs (sol-import/sol-claims child). Called from main.rs (epoch, epoch-status).
behavior_complement
Not done: retrying anything; re-asking after a limit, timeout, failure or an unrecorded answer (final for that document's epoch in its run root); building the world or the dossier's site-old/web texts; parallel documents or several terminals in one run; judging the consultant's text beyond citations, words and lexical support; saturation without a terminal step. Evidence status: source-derived model with finite checks, not a proof.
reverse_reconstruction
From the run root: per epoch and document the status (state, metrics, gates, refused reason, terminal error), the prompt and the given/left-out sources, the raw answer and the claims; from the dossier: each text with its inputs. The texts can be reproduced from the same inputs (Mova's text byte for byte; the consultant's not).
Epoch 2026-11-05 (created)
Measured on five batch papers: Mova 5,993–6,000 words on four, 25 on a paper without part headings; two byte-identical runs; fake consultant 169–185 claims per paper, 97–99 % cited.
Epoch 2026-11-05 (world tool)
Own material comes from the world's distance-0 entry (texts other than the original, of earlier epochs) or else from the dossier; consultant inputs are Inputs (a plain file or a text) and are staged into the terminal root whenever they are texts (always for the world store). Negative control: ignoring the distance-0 entry turns the_world_tool_gives_dossier_and_neighbourhood red.
tests.md
epoch::tests (fixture paper, fixture neighbourhood, fake terminal; 2026-11-05): unknown config fields refused at every level; the length gate keeps a short text out and asks nothing (control: a low minimum registers and asks); cited texts registered and a resume asks nothing; a citation to an unknown source refused; sol_asking becomes uncertain without a question; a limit stops the run and is final; epoch 2 reads and cites epoch 1 and saturates (control: a longer answer does not); two runs write the same text. Negative controls, red then restored: unknown citations accepted, re-asking after sol_asking, generated texts public, saturation off. measure_epoch (ignored) runs a real configuration with a reading fake terminal.
This page in Markdown → https://md.groupants.elamur.ai/free/en/mova/tools/whale/src/epoch.rs.groupant/