Flight Deck — Ducklington Pond

[Example — this was iterative, not one-shot]

Build me a MotherDuck Dive: a playful, duck-themed demo of something real my
team runs at Hamilton Beach Brands — a single-pane "platform cockpit" (we call
it Mission Control) that an analytics engineer opens every morning. Keep the
tone fun, but make the data-engineering details realistic and accurate.

PERSONA: "Ducklington Pond," a maker of whimsical duck appliances (Down-Dryer
3000, Quack-Pot Slow Cooker, Vermi-Blend Worm Processor, Egg-cellent Incubator
Pro, Pond-to-Table Purifier…). 100% synthetic, safe to share publicly.

STEP 1 — Seed a MotherDuck database `ducky_brews` with time-indexed synthetic
tables. The hero interaction is a TIME DIAL that scrubs 3 business days, business
hours only (7am–4pm) — anchor it to today/last-business-day ±1 so it always
reads current. ~30 hourly steps (idx 0..29). Tables: a clock; a YTD daily
revenue series (always shown in full); as-of KPI series (gross margin, units,
forecast accuracy) where margin goes MISSING during an incident; the 10 duck
products; a full dbt test registry (mostly passing); Dagster jobs; comms
(AI-summarized Outlook+Slack threads — data requests, project wins, and incident
breadcrumbs); a technical lineage (source→flight→model→metric); and an "arch"
table describing the real plumbing. Each incident SIGNAL has an onset AND a
resolve hour so issues appear AND clear as you scrub.

Bake in TWO incidents you discover by exploring (don't scream them): (1) a stale
supplier feed — our Flight still calls a deprecated v2 API so a model goes 31h
stale while the job still reports SUCCESS; (2) OMS v3.2 ships unit_price/cogs as
dollars not cents, so margin inflates 100× and a dbt quality gate BLOCKS the
publish → Published "Gross margin" quietly goes to "—". Each should light up a
different mix of dbt + Health + comms.

STEP 2 — Build the Dive as a dark "cockpit" (slate #0f172a, Inter-ish, gradient
green→blue→gold wordmark "Flight Deck", teal accent #2dd4bf). Six tabs:
- Published: clean exec KPIs + full-YTD revenue line + product table; the only
  tell is data going "missing" when a gate blocks.
- dbtCentral: full test registry (X passing / Y warning / Z blocked), failures
  expand to show impacted published metric.
- Health: Dagster jobs + freshness (a job can be SUCCESS-but-stale).
- Issues: one card per incident joining the dbt test + Health alert + offending
  Flight + downstream metric + comms breadcrumb + a numbered action plan, all
  gated by the dial (ACTIVE → CLEARED).
- comms: the AI thread-summaries, newest first, tagged request/project/incident.
- ReadMe: the "get real" tab — "playful duck demo, but we actually do this at
  HBB" — explaining how MotherDuck Flights + MCP servers + Claude Code converge
  into one pane (Jira/ADO/Dagster/dbt via MCP, sales-wire via Flight, etc.).

REQUIREMENTS: use useSQLQuery for every visual; wrap all numerics in N(Number());
format dates in SQL with strftime; fully-qualified double-quoted table names;
plain tables only (no ducklake calls in the Dive runtime); and declare the
database via an exported REQUIRED_DATABASES share so public viewers can load it.
Make it feel like a quiet, comfy product, not a busy BI dashboard.