VOC & research bank

Source-gap rule: any persona carrying >10% of spend with <2% of bank rows fires an ingest alert, not a creative conclusion. At seed, P04 grandparent-gift (~25% of real spend) had 3 verbatims and P03 christian-mom (behind the real 240d Faith Kit runner) had 1 — both fired. Post-fix shares: P04 7.9% · P03 6.4% of bank representative. The bank must out-sample where the money already is, or the V component scores blind.

Frequency in the bank ≠ purchase-decision weight. The seed lane is general parenting talk; the objection board below re-ranks within purchase-context threads only (the real ICP source-of-truth ordering), and the live review/ticket pipes skew purchase-context — which is why lost-pieces and price are the fastest risers despite low seed counts. Both rankings feed V: frequency as the recency-weighted count (e^−days/90), gold flags as logged manual multipliers. Pattern-v1 keyword families are versioned with the thresholds — counts re-measure on every run, never hand-edited.

27 sampled rows of 2,644 — the live table paginates at 50/page and every chip is a registry-FK filter, so any slice (persona × objection × source × sentiment) is one query at any volume. Date column: "h 05-29" = harvest ingest date (post dates not captured by SERP harvest; recency weight uses post date when a pipe provides one, ingest date otherwise). Weight = e^−days/90; the whole seed lane carries 0.73 at the Jun 26 scoring date, live-pipe rows spread 0.73–1.00. Live-pipe rows are excluded from this static preview — every quote shown is from the verified 05-29 harvest; representative rows carry counts, never invented verbatims. Pin sends a verbatim into the Brief Factory as a named evidence slot (the brief skeleton requires 5 VOC verbatims per concept) and writes the used-in-brief back-reference here — circulation is queryable in both directions, and an unpinned gold flag older than 30 days surfaces in the Wednesday curation pass.