This was generated by AI during triage.
Split from #170 (originally split from #158, which fixed the tag-vocabulary dump but explicitly deferred this).
Problem: discover.Options.Limit defaults to 200 candidates per planned page (enchiridion-go/internal/discover/discover.go), applied per-page in Discover(). A multi-chunk ingest plan multiplies candidate volume linearly per chunk, with no filtering or summarization available today. This burns tokens in the calling agent's context, the same class of problem #158 fixed for the vocabulary dump — but candidates carry classification hints (duplicate/refines/related/distinct, summary, tags, volatility, supersession) an agent actually needs, so this isn't a simple "add a filter flag" fix like the vocabulary one was.
Needs its own design pass — e.g. whether to cap total candidates across the whole plan rather than per-page, drop low-value fields from HintDistinct candidates, or offer a summarized/paginated output shape. Not yet triaged into a concrete fix.
Split from #170 (originally split from #158, which fixed the tag-vocabulary dump but explicitly deferred this).
Problem:
discover.Options.Limitdefaults to 200 candidates per planned page (enchiridion-go/internal/discover/discover.go), applied per-page inDiscover(). A multi-chunk ingest plan multiplies candidate volume linearly per chunk, with no filtering or summarization available today. This burns tokens in the calling agent's context, the same class of problem #158 fixed for the vocabulary dump — but candidates carry classification hints (duplicate/refines/related/distinct, summary, tags, volatility, supersession) an agent actually needs, so this isn't a simple "add a filter flag" fix like the vocabulary one was.Needs its own design pass — e.g. whether to cap total candidates across the whole plan rather than per-page, drop low-value fields from
HintDistinctcandidates, or offer a summarized/paginated output shape. Not yet triaged into a concrete fix.