Struct BlockMaxScoreExecutor 

pub struct BlockMaxScoreExecutor<S: ScoringIterator> { /* private fields */ }

Unified Block-Max MaxScore executor for top-k retrieval

Combines three optimizations from the literature into one executor:

1. MaxScore partitioning (Turtle & Flood 1995): terms split into essential (must check) and non-essential (only scored if candidate is promising)
2. Block-max pruning (Ding & Suel 2011): skip blocks where per-block upper bounds can’t beat the current threshold
3. Conjunction optimization (Lucene/Grand 2023): progressively intersect essential terms as threshold rises, skipping docs that lack enough terms

Works with any type implementing ScoringIterator (text or sparse). Replaces separate WAND and MaxScore executors with better performance across all query lengths.

Implementations

impl<S: ScoringIterator> BlockMaxScoreExecutor<S>

pub fn new(scorers: Vec<S>, k: usize) -> Self

Create a new executor with exact search (heap_factor = 1.0)

pub fn with_heap_factor(scorers: Vec<S>, k: usize, heap_factor: f32) -> Self

Create a new executor with approximate search

heap_factor controls the trade-off between speed and recall:

• 1.0 = exact search
• 0.8 = ~20% faster, minor recall loss
• 0.5 = much faster, noticeable recall loss

pub fn execute(self) -> Vec<ScoredDoc>

Execute Block-Max MaxScore and return top-k results

Algorithm:

1. Partition terms into essential/non-essential based on max_score
2. Find min_doc across essential scorers
3. Conjunction check: skip if not enough essential terms present
4. Block-max check: skip if block upper bounds can’t beat threshold
5. Score essential scorers, check if non-essential scoring is needed
6. Score non-essential scorers, group by ordinal, insert results