Struct SeedDiscovery 

pub struct SeedDiscovery { /* private fields */ }

A seed discovery run — rapid iteration to find optimal parameters

Implementations

impl SeedDiscovery

pub fn new(role: &str) -> Self

pub fn run_iteration( &mut self, params: TileParams, trajectory: &[(f64, f64)], ) -> IterationScore

Run a single seed iteration with given parameters on a trajectory.

The trajectory is a series of (x, y) sensor readings. The evaluation scores how well the agent handles the trajectory.

pub fn run_sweep(&mut self, trajectory: &[(f64, f64)], n_variations: usize)

Run a sweep of parameter variations (the seed experimentation).

Generates parameter variations around the current best and evaluates them.

pub fn crystallize(&self) -> DiscoveryTile

Crystallize the discovered pattern into a tile.

This is the key operation: extract the inner logic from the iteration history and compress it into a self-contained tile that can condition future models.

pub fn refine(&mut self, trajectory: &[(f64, f64)], n_variations: usize)

Refine: run another sweep centered on the current best. Each refinement tightens the search around the optimum.