Elamur Groupantsfish-swarm-cli
The groupant beside tools/fish-swarm/src/main.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 |
| cost | symbolic source derived | cost |
| descriptant | present | domain |
| invariant | partial source derived | invariant |
| tests | finite checks | checks |
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 with finite checks | behavior-complement |
reverse-reconstruction | reverse_reconstruction | source derived with finite checks | reverse-reconstruction |
heartbeat-wait | temporal_boundary | source derived with finite checks | heartbeat-wait |
Model text
domain
Exactly two CLI arguments: run|status and canonical config path. The library owns queue and process semantics. No language/model provider choice is hardcoded. Trusted cooperative storage and a conforming executor are assumed.
forward
Parse command and immutable configuration. Status reads the current snapshot and probes the cooperative lease; it does not dispatch. Run acquires Supervisor, repeatedly executes at most one step, exits on stopped/deadline_reached, fails on uncertain/failed, and otherwise sleeps min(config.poll_seconds,5). Queue exhaustion therefore waits without replay until a boundary. The forward vidtvorant is the command/state-to-loop-control mapping; language output is outside its observations.
invariant
Status never becomes run. Failed or uncertain step cannot advance to another dispatch. Terminal campaign states exit; completed_review_only is merely a per-job state. Actual wait is at most5seconds subject to OS scheduling. Library errors propagate to stderr and process exit1 rather than implicit retry.
cost
For k steps, cost is sum of library step costs plus k bounded sleeps; process startup and idle polling are not a measured throughput advantage.
behavior-complement
Type behavior_complement. The command relation excludes unknown verbs, wrong arity, status-triggered execution and continuation after uncertainty/failure. Counterexample to treating every successful step as loop termination: completed_review_only and queue_exhausted continue polling. Evidence: source branch inspection plus the library's fake controls; CLI branch behavior is not separately integration-tested.
reverse-reconstruction
Type reverse_reconstruction. Exit1 has compatible causes including config parse, lease failure, uncertain history and child failure; stderr refines but does not uniquely reconstruct system history. Exit0 has status-query or STOP/deadline preimages. No unique model computation or semantic success follows.
heartbeat-wait
Type temporal_boundary. For configured p in1..60, nominal idle wait w=min(p,5). Thus w<=5, while observed heartbeat spacing can exceed w because scheduling and storage latency are external. Counterexample to a hard5second guarantee: blocking filesystem operations. Source-derived arithmetic, no hard real-time proof.
checks
Native cargo test compiles this CLI; thirteen library controls test the delegated scheduling transitions. No real model or deployment is exercised.
epoch-memory
A per-job completed state must never be admitted by a priority observer as campaign-terminal quiescence. Configurable provider/deadline values remain outside CLI identity.
This page in Markdown → https://md.groupants.elamur.ai/free/en/mova/tools/fish-swarm/src/main.rs.groupant/