Search: how matching works

Matching and ranking answer different questions

Open Search and enter the language behind your market question. CredScape first uses indexed full-text matching to decide which active offerings belong in the matched set. It then applies semantic ranking and a cross-encoder rerank to order that set by meaning, even when the strongest offering does not repeat your exact wording.

This separation keeps the count stable. Ranking can change the order, but it cannot add or remove an offering from the matched set. If semantic ranking is temporarily unavailable, every matching offering remains included and counted. The page states that it has fallen back to keyword relevance until ranking recovers.

Read counts and tiers together

The count is exact for the query and active filters. Search can return both course and program kinds. Because a program bundles courses that may also appear separately, the header reports the two tiers side by side instead of adding them into a third figure.

Relevance is available when a query or skill drives the set. You can choose another sort where the interface offers it. Exports use a stable lexical order so the same dataset can be reproduced rather than inheriting a ranking that may move between requests.

Shape the matched set with filters

Every active filter narrows membership before the count is shown. Multiple values inside one facet act as alternatives, while selections across facets must all be satisfied. Exclusions remove the named value unless an explicit inclusion creates a direct contradiction, in which case the inclusion wins.

Search spans all markets, provider types, and catalogues by default. A market selected elsewhere does not constrain it. Apply the catalogue, institution location, provider, or provider-type controls here when those boundaries are part of your question.