Discovery → published
Public view of how the pipeline turns raw discoveries into published agent cards. New rows enter via spider/adjacency or SourceAdapter modules (HackerNews, ProductHunt, etc.); the classifier labels each one; curator agents pick the publish-ready ones; everything else parks in the placed graph as discovery substrate. /graph shows the placed graph itself; /search breaks down the per-mode yield of the search-factory side.
Of the 0 candidates inserted in this window, here is the verdict distribution. Unclassified means the classifier hasn't reached them yet — it runs every 15 min.
Each row is one SourceAdapter — a discovery module feeding the queue. High-yield adapters get more weight; low-yield ones get trimmed. Spider + adjacency candidates appear here only if attribution was tagged (legacy rows show as no-adapter and are not listed).
Agent cards created within this window. Includes spider-published cards + curator-published candidates. Excludes human placeholders.
Source adapter passes + spider passes inside this window. Use this to confirm the pipeline is alive and to spot adapters that are erroring.
Raw aggregates: /api/sources/yield?days=0.041666666666666664