Struct ResearchStep 

pub struct ResearchStep { /* private fields */ }

An agent that proposes experiments, runs them, and decides when to stop.

Implementations

impl ResearchStep

pub fn config_hash(&self) -> CacheKey

Content-addressable hash of this step’s configuration.

Every field that is not #[soma(skip_hash)], so adding one cannot silently leave it out of the journal key.

impl ResearchStep

pub fn new( model: impl Into<String>, objective: impl Into<String>, pipeline: Graph, ) -> Self

A researcher that pursues objective by experimenting on pipeline, asking model what to try next. Starts with an empty seed and 20 iterations; adjust with with_history and with_max_iterations.

pub fn with_history(self, records: Vec<ExperimentRecord>) -> Self

Seed the loop with what is already known. An agent that starts from an empty pool repeats work someone already did.

pub fn with_max_iterations(self, n: usize) -> Self

Cap the loop at n experiments. The cap is the budget, not the goal — the agent may Conclude well before reaching it.

pub fn completed(&self, ctx: &StepCtx<'_>) -> Vec<ExperimentRecord>

Every experiment behind this run, seed first, newest last.

Rebuilt from the turns rather than accumulated in a field. A step is a description of behaviour, not a place to keep state: derive the history and a replay reconstructs exactly the history the original run had, because it replays exactly the same results.

Trait Implementations

impl Serialize for ResearchStep

fn serialize<__S>(&self, __serializer: __S) -> Result<__S::Ok, __S::Error>

where __S: Serializer,

Serialize this value into the given Serde serializer.

impl Step for ResearchStep

fn config_hash(&self) -> CacheKey

Derived: the pipeline under investigation is part of the key too, which the hand-written version left out — two research loops over different graphs shared a journal.

fn meta(&self) -> StepMeta

This step’s [StepMeta]: turn budget, journaling, schemas, placement.

fn poll(&self, ctx: &StepCtx<'_>) -> Result<Transition>

Advance one turn.