Words first.
No hidden meaning layer.
Seek separates text into letter-and-number words, lowercases them, and removes combining accent marks. Query words are deduplicated. Punctuation separates words rather than acting as an advanced query operator.
Exact mode compares normalized whole words. Prefix mode also accepts words that begin with a query alternative, so rank can match ranking. It does not correct misspellings, perform stemming, or infer a related concept.
All-word mode requires every query group to match somewhere in title, summary, or body. Any-word mode requires at least one. An empty query browses the faceted record set.
Keep the filter logic visible.
Selected record types combine with OR. Selecting Guide and Reference permits either. With no type checked, all three types are eligible. Topic is one exact selection, and it combines with the type group and query using AND.
The counts beside type choices are calculated from the current query and topic before type selection. They show how many matching records each type could contribute, not the size of the whole unfiltered library.
Make vocabulary a choice.
A synonym group such as search, find, lookup expands any member into those alternatives. Each query group still counts once per field, even if several alternatives appear there.
Use one group per line, two to six single words, and at most eight groups. Repeated words across groups are rejected to avoid ambiguous overlapping rules. Multiword phrases are not supported. Clearing the textarea disables synonyms.
These groups are configured associations. They are not generated by an AI model and do not guarantee that the words are interchangeable in every context.
See every point
in the result.
For each query group, title matches receive the title weight, summary matches receive the summary weight, and body matches receive one point. Repeated occurrences in one field do not multiply the score. A configured preferred topic adds two points to eligible records.
Sort by total score descending. When recent-date tie-breaking is enabled, equal scores use the fixed sample updated date descending. Stable record ID breaks remaining ties. The score is not a probability, quality rating, or measure of semantic understanding.
A query group matching title and body with title weight 5 scores 6 points before any topic preference. Changing title weight to 2 makes the same evidence worth 3 points.
Keep the path predictable.
Press / outside an input to focus the page’s search field. The shortcut does not interrupt typing in form controls. Press Escape while the main query has focus to clear just the query, retaining facets.
Use Tab to reach native checkboxes, selects, and result links. Enter follows the focused result link. Seek does not disguise a list of links as an inaccessible custom command palette.
Save locally.
Export deliberately.
Save settings validates every weight and synonym group before writing to this browser. If storage is unavailable, the previously applied settings remain active and the form explains the failure. Invalid saved data falls back to defaults.
Configuration export validates the current form and downloads JSON. Dataset export contains only the original fictional records. There is no external index creation, provider request, API key, telemetry, query logging, or AI call.
The dataset is deliberately small and the engine is a transparent demonstration. No production performance or scale claim is made.