Prod deploy catch-up — migrations → promote develop→production → verify

Single ordered pass to ship everything merged to develop this session. Migrations FIRST, then promote code, then verify. (Per .claude/rules/branches.md.)

A. Migrations → prod first (else new code 500s)

1. SSH tunnel to prod DB (separate terminal): ssh -N -L 5433:<prod-supabase-db-ip>:5432 vps_prod &
2. git pull origin develop
3. export PROD_DB_URL='postgresql://postgres:<PASS>@127.0.0.1:5433/postgres?sslmode=disable'
4. ./scripts/push-migrations.sh — idempotent; applies what's pending. Expected new: 20260525113000_options_ranking_drop_default.sql (the priority migration 20260524120000 was already applied earlier).

B. Promote code (auto-deploys on vps_prod)

1. git checkout production && git merge develop -m "Merge develop into production: preference_score MCP + P4/P6 agents, ranking fix, extension search + sort-by-property, extension build guard"
2. git push origin production → push auto-deploys.

C. Verify prod

1. curl -sS -o /dev/null -w "%{http_code}\n" https://pickism.com → 200
2. Prod MCP get_option(01118eef) → now returns preference_score (the reaction) alongside ranking/property_rankings.
3. Log in as admin, confirm /app/admin/subjects and /app/admin/harness render.

Adjacent (not prod-blocking — when you have a moment)

• Extension: create apps/extension/.env.production (vars in task create-appsextensionenvproduction…), then ping Claude → build:prod + zip pickism-ext.zip.
• P5 calibration: Claude can run the live calibrator box-local (scrape + Haiku) → real calibration_2026Q2.json, then merge PR #142.
• Ground-truth: rate the 17 SKE subjects (gates the harness) — task yan-mustdo.

You supply the two placeholders (<prod-supabase-db-ip>, prod DB <PASS>); everything else is literal.