Vibe-code your first hockey data setup

You are my senior data engineer and hockey analyst. Help me build my first local hockey
analytics setup from a file I already have. Work step by step and keep everything runnable
on my laptop — no cloud accounts, as few files as possible.

My data: I have a file called [FILENAME.csv] that I exported from [SOURCE]. Before writing
any code, ask me to paste the header row (column names) and 2–3 example rows so you know the
real schema. Do not assume columns.

Then do the following, explaining each step in one or two sentences:

1. Architecture: propose the simplest possible setup — a single DuckDB database file plus one
   Python script — and explain the bronze → silver → gold idea in one sentence.
2. Load (bronze): write build.py that installs what it needs (duckdb, pandas), loads
   [FILENAME.csv] into a "bronze" table, and prints the row count and column types.
3. Clean (silver): create a "silver" view that fixes data types, standardizes player/team
   names, and drops obviously broken rows.
4. Metrics (gold): create "gold" views with beginner-friendly hockey metrics from whatever
   columns I actually have — e.g. per-player games, goals, assists, points, shots, and
   points-per-60 if ice time exists. If I have shot coordinates, add a simple distance-based
   shot-quality estimate, name it "xg_estimate", and clearly label it a rough proxy, not a
   real model.
5. Ask questions: give me 5 example SQL queries against the gold views (top scorers, best
   shot rates, etc.).
6. Dashboard: write a small Streamlit app (app.py) with a team/player filter and 2–3 charts
   (top scorers bar chart, shots-vs-points scatter) reading from the DuckDB file, plus the
   exact commands to run it.

Rules: every script must run top-to-bottom without manual edits once I give you my columns;
add comments only where they explain a non-obvious choice; and tell me honestly whenever a
metric is an approximation. Start by asking me for my column names and a couple of sample rows.