pub fn apply_floor(hits: &mut Vec<Hit>)Expand description
Adaptive confidence floor — a presentation filter, applied at the surface (MCP/CLI), NOT in the retrieval/eval path. The engine always returns the full ranked list so the eval can measure rank-2+ recall honestly; the surface uses this to drop trailing noise before showing results.
When a confident hit exists (Exact/Strong/Ambiguous), the trailing Fallback-band hits are
noise, not answers — drop them (the “1 strong + 7 fallback” cleanup). When nothing is confident
(a genuinely weak query, or garbage), the ranked list is kept untouched so a weak-but-real answer
still surfaces and garbage still reads as rejected. The TOP hit’s band is never changed.