Struct ScoreCollector 

pub struct ScoreCollector {
    pub k: usize,
    /* private fields */
}

Efficient top-k collector using min-heap (internal, scoring-layer)

Maintains the k highest-scoring documents using a min-heap where the lowest score is at the top for O(1) threshold lookup and O(log k) eviction. No deduplication — caller must ensure each doc_id is inserted only once.

This is intentionally separate from TopKCollector in collector.rs: ScoreCollector is used inside MaxScoreExecutor where only (doc_id, score, ordinal) tuples exist — no Scorer trait, no position tracking, and the threshold must be inlined for tight block-max loops. TopKCollector wraps a Scorer and drives the full DocSet/Scorer protocol, collecting positions on demand.

Fields

k: usize

Implementations

impl ScoreCollector

pub fn new(k: usize) -> Self

Create a new collector for top-k results

pub fn threshold(&self) -> f32

Current score threshold (minimum score to enter top-k)

pub fn insert(&mut self, doc_id: DocId, score: f32) -> bool

Insert a document score. Returns true if inserted in top-k. Caller must ensure each doc_id is inserted only once.

pub fn insert_with_ordinal( &mut self, doc_id: DocId, score: f32, ordinal: u16, ) -> bool

Insert a document score with ordinal. Returns true if inserted in top-k. Caller must ensure each doc_id is inserted only once.

pub fn would_enter(&self, score: f32) -> bool

Check if a score could potentially enter top-k

pub fn would_enter_candidate( &self, doc_id: DocId, score: f32, ordinal: u16, ) -> bool

Check whether this fully identified candidate ranks ahead of the current worst retained entry, including deterministic tie breaks.

pub fn len(&self) -> usize

Get the conceptual heap length, including virtual threshold sentinels.

pub fn real_len(&self) -> usize

Number of real results retained, excluding threshold sentinels.

pub fn is_empty(&self) -> bool

Check if collector is empty

pub fn seed_threshold(&mut self, initial_threshold: f32)

Seed the threshold from a cross-segment shared value.

Logically fills unused slots and replaces retained entries below the new floor with virtual dummy entries. This can be called repeatedly while another segment raises the shared threshold; equal-scoring real candidates win the deterministic doc-id tie break over sentinels.

pub fn into_sorted_results(self) -> Vec<(DocId, f32, u16)>

Convert to sorted top-k results (descending by score). Filters out sentinel entries (doc_id == u32::MAX) from threshold seeding.