Elamur Groupantsfish-executor-protocol
The groupant beside tools/fish/src/execute.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 |
| ast | partial | ../syntax.json |
| composition | partial source derived | composition |
| cost | partial source derived | cost |
| descriptant | partial source derived | descriptant |
| invariant | partial source derived | invariant |
| tests | finite fake cli controls | 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 | source derived finite checks | behavior_complement |
reverse_reconstruction | reverse_reconstruction | source derived finite checks | reverse_reconstruction |
Model text
model
Reply protocol and typed authority
forward
Domain: a bounded successful CLI stdout envelope. Parse type=result and is_error=false, then parse its result string as the strict AnnotationDraft. Reuse native source-bound validation; preserve each original producer comparison and draft-primary contextual findings. Observation: strict rejection or review-only receipt, never semantic approval.
behavior_complement
Missing fields, extra authority fields such as gold, stale task hashes, source-character mutations, incomplete target tables and noncanonical labels have no validated successor. This is a partial exclusion relation; semantically wrong but structurally valid annotations remain possible. Evidence: fake malformed-envelope, extra-field, target, source and label controls.
reverse_reconstruction
A validated receipt constrains task identity and admissible annotation tables. Multiple parse attachments, source-consistent token boundaries and hidden consultant computations remain compatible. Hashes constrain observations without recovering unique causes. Assumptions: trusted artifact files and collision-resistant binding.
invariant
Required schema/taskid/tasksha/ast_conllu/pragmatics_tsv/unresolved fields remain distinct. Native structure/source checks and semantic correctness remain different statuses. Original findings and reply-only context findings are retained; training_authorized is always false.
cost
At most 2MiB stdout is read into the result parser, followed by bounded native target/candidate checks. No model-quality, percentage-error or GPU gain is asserted.
composition
Envelope parsing feeds the exported annotation::validate_draft. The CLI transport and filesystem reservation are external to this fragment; a correct leaf parser alone does not prove the campaign.
descriptant
Unresolved findings are intended data rather than transport failure. Complete native tables remain mandatory; an incomplete answer is retained raw and rejected for review rather than silently completed.
tests
Finite fake-CLI controls are shared with execute.rs tests. Source-level prefix ordering is inspected; no exhaustive crash, provider or semantic proof.
This page in Markdown → https://md.groupants.elamur.ai/free/en/mova/tools/fish/src/execute.rs.groupant/protocol.groupant/