Search the curated corpus with lexical retrieval first.

Title + abstract lexical searchDeterministic ordering

Search v1 is intentionally narrow: lexical retrieval over titles and abstracts, plus practical filters for narrowing the current corpus. Semantic assist can come later, but it is not part of Search v1's scope.

Search v1 scope: dedicated lexical search, practical filters, and clean handoff into dossier and ranking views. When ranking family filtering is active, the API resolves and returns one explicit run context.

Query surface

Search controls

What Search v1 supports

Lexical retrieval over title + abstract.
Filters for year, scope, venue/source, topic label, and ranking family filter.
Stable ordering: lexical rank, then year, citations, and work id.
Run metadata appears only when the search depended on ranking state.

Topic labels are imported metadata and can be noisy; use them as coarse navigation hints, not authoritative classifications.

Workflow map

How search moves through Research Radar

Find the right candidate set

Start with lexical retrieval and narrow with year, venue, topic, and scope filters until the candidate set is focused enough to inspect.

Inspect the dossier

Open the paper dossier to review abstract, metadata, ranking presence, and adjacent papers without losing the corpus framing.

Move into signals

Use Recommended and Evaluation to understand why a result matters now and how it sits inside the exact resolved ranking run when family filtering is active.

Lexical results

Run a query to search the corpus

Start with a concrete phrase from the literature, then narrow by topic or venue if the first pass is too broad. Search v1 only claims lexical retrieval today, so the surfaced rows should be interpretable from their text.

Retrieval is lexical over each paper's stored title and abstract (word/phrase match, deterministic ordering). It is not semantic vector search, hybrid retrieval, or LLM-ranked results.