The weighting system is the core of Lumenare Search's relevance ranking. It combines indexed field weights, multi-pass matching, and optional boost layers to decide which eligible result appears first.
Current release: 2.2.01
How Weights Work
During indexing, Lumenare stores per-field occurrence counts for each term. At search time, those counts are multiplied by their configured weights and by the match-pass multiplier.
| Field | Default | Notes |
|---|---|---|
| Title | 10 | Highest default signal for descriptive titles |
| Excerpt | 5 | Medium signal for summaries |
| Content | 1 | Baseline body-content signal |
| Category | 5 | Configured through taxonomy weights |
| Tag | 3 | Configured through taxonomy weights |
| Author | 2 | Author display-name matches |
| Custom fields | 3 | Only configured meta keys are indexed |
| Attachment content | 1 | Only used when document indexing is enabled |
Setting a field weight to 0 excludes that field from indexing and scoring. Changes between positive weights apply at search time without reindexing. Changing a field from 0 to a positive weight requires reindexing because its occurrences may not have been stored.
These weights control the keyword score. They do not create embedding vectors or change the embedding model. Keyword indexing, passage preparation, and embedding analysis are separate steps.
Match Passes
Lumenare searches the inverted index in progressively broader passes:
| Pass | Match Type | Default Multiplier |
|---|---|---|
| Exact term | Term equals the indexed term | 10x |
| Prefix | Indexed term starts with the query term | 5x |
| Suffix | Reversed-term lookup for suffix matches | 2x |
| Fuzzy | Levenshtein typo tolerance, when enabled | 2x |
Earlier passes win. If a post matches in the exact pass, later prefix/suffix/fuzzy passes do not replace that score for the same post.
When match mode is All, exact/prefix/suffix passes require every retained query term to match before a post joins the candidate set. Configured stop words are removed first. Supplementary synonym and fuzzy passes can add candidates, so All is not a strict Boolean guarantee across the complete pipeline.
Boost Pipeline
After the base score is computed, optional ranking layers can adjust the score:
- Synonyms add scaled exact-match scores for configured synonym groups.
- Trending boosts queries that have become popular recently.
- Phrase boosting applies a separate 2× full-query title-adjacency boost or the configured multiplier (default 1.5) for a phrase in both the query and title. It checks the top 100 candidates when at least two keywords remain.
- Freshness gives newer or recently updated content a configurable lift.
- Audio attachment boost gives posts with ListenUp audio metadata a small lift by default when ListenUp is installed.
- Click boost uses query-specific CTR from impressions/clicks, falling back to raw clicks when needed.
- Post-type weights can emphasize or soften whole content types.
- Query promotions boost already-eligible results for exact queries from Search Intelligence actions.
The minimum relevance threshold is applied after these boosts.
Optional meaning reranking then compares eligible keyword candidates with compatible passage vectors. In 2.2.00 it considers at most 200 posts and 800 passages, with a default 0.6 lexical / 0.4 semantic blend. Explicit live meaning search is limited to 20 posts. This step reorders existing candidates; it cannot retrieve a post outside the keyword candidate set or rescue a query with no eligible hits.
Exact title matches and decisive keyword leads retain ranking protections. When only some candidates have compatible vectors, uncovered candidates keep their original positions; missing or incompatible vectors leave keyword ranking available. A high meaning score does not prove an answer: cited excerpts and AI answers still need supporting source passages.
Choose search behavior independently for each assigned Search Profile. Full results, live search, blocks, shortcodes, and Ask Site can use different profiles. An answer card uses its results surface's profile, rather than switching to the Ask Site profile.
Example
A post titled "WordPress Security Best Practices" with category "Security" and tag "WordPress" will score strongly for wordpress security because both terms appear in the title, one appears in the tag, and one appears in the category.
With default field weights, the title matches dominate the base score. If the full phrase appears in the title, the phrase boost can lift it further. If the post was recently updated, the freshness boost can also help it win close ties.
Sorting
Relevance is the default order. The results filters also let visitors choose:
- Newest first (
date_desc) - Oldest first (
date_asc)
When relevance remains selected, date can still act as a secondary tie-breaker through the configured secondary sort.
Explicit date sorting uses the requested date order instead of meaning reranking and does not request a query embedding. Facets and query suggestions also remain separate from meaning reranking.
Next Steps
Ready to optimize your weights? Continue to the Tuning Guide →