Struct PipelineConfig 

pub struct PipelineConfig {

    pub domain: String,
    pub size: usize,
    pub seed: u64,
    pub difficulty: String,
    pub output_format: String,
    pub run_id: String,
    pub entity_plans: Vec<EntityPlan>,
    pub hard_neg_types: Vec<HnTypeConfig>,
    pub hard_neg_ratio: f64,
    pub noise_passes: usize,
    pub graph_enabled: bool,
    pub graph_format: String,
    pub skip_ground_truth: bool,
}

Fields

domain: String

size: usize

seed: u64

difficulty: String

output_format: String

run_id: String

entity_plans: Vec<EntityPlan>

hard_neg_types: Vec<HnTypeConfig>

hard_neg_ratio: f64

noise_passes: usize

Number of independent noise passes applied to each duplicate copy — see the doc comment on schema::DifficultySettings::passes for why this (not the noise_types category count) is the actual difficulty lever between tiers.

graph_enabled: bool

graph_format: String

skip_ground_truth: bool

--skip-ground-truth: skip ground-truth classification entirely and don’t write the _ground_truth file at all (see gt::GtAccumulator). The GT file is about a third of the dataset’s size, so skipping it saves that much write I/O plus the per-row classification — for stress-test workloads that only need the dataset itself. Defaults to false (absent from the JSON build_pipeline_config constructs) so every existing caller keeps producing ground truth unchanged. Incompatible with --graph, which needs the post-GT cluster map to emit duplicate- cluster edges — validated in main.rs, not here.

Trait Implementations

impl Debug for PipelineConfig

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

Formats the value using the given formatter.

impl<'de> Deserialize<'de> for PipelineConfig

fn deserialize<__D>(__deserializer: __D) -> Result<Self, __D::Error>

where __D: Deserializer<'de>,

Deserialize this value from the given Serde deserializer.