Changelog

Every material change to the data, bench, or site — dated, newest first.

DATE	TYPE	WHAT CHANGED
2026-07-12	[MEMBER]	Member system live: accounts (email + password, double opt-in), verified-member company up/down voting with a VOTES ranking, comments with upvotes, and a research pipeline where super admins canonize third-party submissions into the Beef Score (turning it green). Self-serve account deletion and admin user management with full contribution preview; deleting a user removes every contribution and reverts any score it changed.
2026-07-12	[SITE]	Facet rail (filter by grade, raised category, antibiotic/hormone status, organic, welfare cert, price band — filter state lives in the URL and is shareable), side-by-side compare view (up to 4 companies, differences-only toggle), and this changelog page shipped.
2026-07-12	[PRICE]	Cost baseline set: lowest verified benchmark ribeye $/lb among listed companies is flagged [BASELINE] on the board. Recalculates whenever a new verified low lands.
2026-07-12	[BENCH]	Bench 1.0 published: taste 30% / healthy 25% / cost 20% / happy 15% / texture 10%. All scores render yellow — agent-researched, awaiting user + third-party verification.
2026-07-12	[QA]	18 corrections applied before publication: 7 verification-tier downgrades on claims that would have over-rendered a verified mark, aging values normalized across 6 companies (flash-freezing is not aging), 4 prices flagged as third-party-sourced pending first-party confirmation.
2026-07-12	[RESEARCH]	Round-1 agent research landed: 270 sourced claims, 25 benchmark ribeye prices ($13.98–$89.00/lb observed), 135 graded values across 27 companies. Every claim carries its source URLs, evidence class, and verification tier. Rastelli's observed storewide sold-out — listed, on inventory watch.
2026-07-12	[LAUNCH]	beefshopper.com live: 27 home-delivery beef companies listed, served from Cloudflare's edge. Costco excluded by policy (baseline concept deleted — owner ruling AP6).
2026-07-11	[DESIGN]	v1 design adopted: parametric-catalog layout, evidence-class model (self-claimed / user-reported / third-party), yellow→green score confidence, owner-canon research pipeline. Full spec and adversarial gap report in the repo.