Skip to content

Find and rank

One need, several candidates, and an honest place for each. This page is about how candidates are judged and compared; Find a part is the flow that brings them in.

Built so far

One candidate at a time. A listing brought back for a search request is read into fields (from its text by Muse Spark, from a screenshot by the extractor, or typed in as your word), dated, and checked against the need by the profile's rules: pass, fail or unknown per rule, with the candidate's provenance and, for a measured value, its uncertainty. The ladder gives one verdict, and the search page names the state it puts the candidate in: a documented candidate, needs evidence, conflicting evidence, rejected, or withdrawn. Every unknown on the candidate's side is a question to ask the seller.

Several side by side, by hand. The candidates table on a search lists every listing brought back with its state, its asking price, the day it was seen and where it came from, so the documented ones stand out and the rejected ones say why. Open any of them for its checks. A candidate seen more than CANDIDATE_DAYS ago is marked too old to trust; Recheck runs it against the need as it stands now.

Listings before they are candidates. What you paste back from a Marketplace search, or read from a screenshot, is ranked first without a model: fits when the search plan's pattern matches, no exclusion word appears and the price is within the top price; maybe when the pattern does not match or the price is unclear; does not fit on an exclusion word or a price over the limit. Fits come first, then maybe, then the rest, cheaper first within each, with the matched words highlighted and the reason written out. Check fit turns one into a candidate. See Bring back listings.

What a fit score is not. There is none. A candidate is not ranked by a number a model produced; it is placed by which rules passed, which failed and which are still unknown, and the asking price is shown beside that, not folded into it.

Not yet

A comparison view across the candidates of one need, with the fit, the price and the open questions of each in one table, is under Next steps.