Thesis-level catalog across chats — first-test DiscoverySpecs plus every idea we actually ran, scored strong / promising / weak / fail. Tuner SID/SMA remixes and BML grid cells are not listed (those stay on the research ledger).