Struct PlannerOptions 

pub struct PlannerOptions {
    pub min_followup_probability: f32,
    pub bytes_per_token: u32,
    pub latency_penalty_ms: Option<u32>,
    pub dollar_penalty: Option<f32>,
}

Hard knobs the solver does not (yet) read from AdaptiveConfig. Keeping them as a separate struct so the API stays additive.

Fields

min_followup_probability: f32

Minimum FollowUpLink.probability for the link to be admitted. Default 0.5 — corpus mining shows ~half of edges are usable signals, the other half are noise.

bytes_per_token: u32

Bytes-per-token assumption for the budget gate. We use 4 to match the existing pipeline heuristic; once Paper 2’s accurate tokenizer ships in this branch we switch to that.

latency_penalty_ms: Option<u32>

Latency penalty knee — calls with cost_model.latency_ms_p50 ≥ this get their value-score halved. Picked at 5000 ms because any prefetch slower than that is unlikely to land before the LLM organically asks for it. None = latency-unaware (default).

dollar_penalty: Option<f32>

Dollar penalty knee — calls with cost_model.dollars ≥ this get their value-score halved. Picked at $0.10 / call. None = dollar-unaware (default).

Implementations

impl PlannerOptions

pub fn cost_aware() -> Self

Enable both latency and dollar awareness with sensible defaults (5 s knee, $0.10 knee). Useful in deployments where prefetch fan-out hits a paid API rate-limit or a slow upstream.

Trait Implementations

impl Clone for PlannerOptions

fn clone(&self) -> PlannerOptions

Returns a duplicate of the value.

fn clone_from(&mut self, source: &Self)

Performs copy-assignment from source.

impl Debug for PlannerOptions

fn fmt(&self, f: &mut Formatter<'_>) -> Result

Formats the value using the given formatter.

impl Default for PlannerOptions

fn default() -> Self

Returns the “default value” for a type.

impl Copy for PlannerOptions