P3.0: all four verticals as declarative kits

[ocularminds]

Roadmap item

P3.0 — Phase 3 opens the way the roadmap always insisted: verticals are data, never forks. All four ship at once; the pilot choice becomes an activation + hardware decision.

What a kit is

• Metric catalog: snake_case types with units and typical ranges (drives simulators and dashboards)
• Labeled rule pack: operations language, not comparisons — 'Bin needs collection', 'Frost risk', 'ISO 10816 zone D — investigate', 'Pump possibly dry-running'
• Field notes: the roadmap's watch-outs, including the water-actuation safety boundary

Mechanics

• GET /kits, GET /kits/:key; POST /admin/sites/:id/apply-kit instantiates the pack per site, idempotent on exact (metric, op, threshold) duplicates — a coarser dedupe was caught by the test suite shadowing a kit rule behind an unrelated threshold
• Rules gain human labels (migration 011), surfaced in alerts
• A templated Grafana dashboard (metric/site variables) serves any vertical — no per-vertical dashboard forks either
• Simulator CLI: npm run simulate -- --kit waste --site 1 --admin $TOKEN --breach enrolls a virtual sensor and streams signed random-walk readings — pilots rehearse before hardware exists; three concurrent runs = M3's '3 nodes' simulated

Verification

• 59/59 tests: kit/metric referential consistency, sim-walk clamping, apply-kit idempotency incl. the threshold-dedupe fix, and a waste-kit rule firing end to end with its label on the alert
• Live 3-node simulator demo was prepared but the local Docker daemon went away mid-run; it's a one-command rerun once the stack is up (feature itself is fully pipeline-verified in tests)

🤖 Generated with Claude Code