Elamur Groupantsmova-whale-topic
The groupant beside tools/whale/src/topic.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 topic corpus in the learning loop: runs the frozen mova-topic-corpus binary (which depends on this crate, so it is a child process, never linked) to build an assigned paper's closed corpus and world, seed term facts into the Whale's knowledge before each round's questions, and write each round's cited description with a bibliography. Off unless configured. Source: tools/whale/src/topic.rs.
Forward approximant
run(cmd, args): spawn EXECUTABLE cmd CONFIG args, drain stdout/stderr on threads, kill after timeout_seconds; {ok, ms, result JSON | exit + stderr tail | timeout}. build(paper) = build, then world when build succeeded. seed(paper, whale, state_root). cite(paper, round_dir, state_root, whale, out [--public]).
Invariant
A child's failure or timeout is returned as a value and never stops learning; nothing is retried; the learner releases learner.lock only for the seed and stops (lock_lost) when it cannot take it back; paths are absolute.
Cost
One child per call; build and world per assigned paper (minutes for a large neighbourhood); seed and cite per round (seconds each, loading the library and corpus).
Composition
Parent: whale-workers. Used by batch.rs (plan: build on assignment, topic_corpus events) and learn.rs (seed at round start, cite after every round's files). The topic-corpus mechanism itself is tools/topic-corpus.
behavior_complement
Not done: downloads, web pages, Sol drafts (manual steps of the topic-corpus README), checking the child's output beyond JSON parsing, caching between rounds. cite resolves pod items (pod:…) as the paper itself (it knows terminal and topic-corpus provenance only). Evidence status: source-derived model with finite checks, not a proof.
reverse_reconstruction
The journal holds each child's command, status, timing and result; the corpus and cited.md are on disk.
Epoch 2026-11-04 (created)
Tested with a fake binary (a shell script whose seed refuses a held learner.lock); no live corpus run from the loop yet.
tests.md
topic::tests::children_report_success_failure_and_timeouts; batch::tests::topic_corpus_is_built_seeded_and_cited (2026-11-04): build and world on assignment, seed with the lock released (the fake refuses a held lock: red if the learner keeps it), cited.md per round.
This page in Markdown → https://md.groupants.elamur.ai/free/en/mova/tools/whale/src/topic.rs.groupant/