Early binding means applying rules and constraints as early as possible—before or during retrieval. In stores, it filters out the wrong items (e.g., out-of-stock, forbidden, wrong region) before ranking, so results stay clean and fast.
Early binding commits certain decisions before ranking—often at index time or pre-filter retrieval. Typical examples are ACL/security trimming, region/assortment gating, hard filters (in-stock/price bounds), or precomputed features (synonyms expansions, normalized attributes).
Early binding applies non-negotiable constraints up front, protecting privacy, speed, and quality. Keep it for hard rules; leave preferences and context to later stages.
Early vs late binding?
Early = commit constraints before ranking/retrieval; late = decide during/after ranking (e.g., boosts, personalization).
Does early binding hurt recall?
Only if misconfigured. Limit it to hard, objective rules.
What about synonyms?
Use late binding for intent-dependent expansions; early only for safe, unambiguous ones.