Hi all, Quick update on the Axiom Health → Google trial. I want to get ahead of the questions Marcus will ask, so I'm getting this in front of you before our next sync. The thesis we're testing: Test Gemini for tier-2 customer support across our member portal. Goal: 35% deflection of tier-2 tickets in 60 days without compromising compliance review. What's clearly working so far: Member-eligibility questions: 'is this procedure covered?' — currently 18% of tier-2 volume, Prior-auth status lookups — 22% of tier-2 volume, currently bounced to a human in 90s, and Spanish-language support queries — 14% of member base, currently outsourced. Could be a deflection-on-deflection win. Together those represent the largest slice of the workload we've been trying to move. The risks we've surfaced — and are actively working: we can't put member-specific medical context into the model without a baa review, legal needs a logged audit trail for every model response that touches phi, and no standard pattern for hipaa-safe context injection in the gemini sdk. champion needs a docs page they can show legal. None of these are dealbreakers; each has an owner and a path forward. The decisions I need this week: Send champion the BAA-compliant reference architecture by Friday and Pull DevRel into a 30-min session with Marcus (CISO) to walk the security model. If we land this pilot, the adjacent opportunity is real: if tier-2 deflection works, champion is interested in gemini for clinical-trial recruitment screening — separate $1.2m opportunity. Worth flagging now so it doesn't surprise you in Q3 planning. Happy to walk any of this through live. The signals behind every line above are captured in the Trovii workspace if you want to drill in. — Park