Occam1976/commodities-dashboard

Commodity Pulse

A self-hosted dashboard for tracking physical commodity markets — energy, precious metals, PGMs, industrial metals, coal and uranium — in one place, with the context needed to read them rather than just the prices.

Live: https://occam1976-commodities-dashboard.hf.space

Why

Commodity data is scattered. Spot prices live on one site, futures curves on another, CFTC positioning in a weekly text file, LME warehouse stocks behind a daily PDF, and rhodium does not trade on an exchange at all. Pulling those together by hand, every day, to answer a simple question — is this market tight or loose right now? — is the kind of work that should be automated once.

This dashboard does that: a set of collectors write into a local SQLite database, and a Dash front end reads from it. No subscriptions, no vendor terminal.

What it shows

Overview

Every instrument in one sortable table: price, and the move over the last session, week, month and year to date. Two columns add the context that a raw price lacks — where the price sits inside its 52-week range (0% = the year's low, 100% = the high), and how far it has drifted from its 200-day moving average. Cards at the top call out the session's biggest mover in each direction and the best and worst performers of the year.

Below the table, cross-commodity ratios — gold/silver, platinum/palladium, platinum/gold, copper/gold, and the Brent–WTI spread — each with its historical range and a note on what a reading outside that range usually means.

Futures

The term structure for a given commodity, plotted across its real listed contracts, with the shape labelled automatically: contango (futures above spot — ample supply, storage being paid for) or backwardation (futures below spot — physical scarcity, a premium for immediate delivery). A table lists each expiry and its spread against the front contract, and a date picker replays how the curve looked on any earlier day the collector ran.

History

Candlestick charts for any instrument over the stored history.

COT

Weekly CFTC Commitments of Traders positioning — commercials against speculators — with each week's reading expressed as a percentile of the last three years, so an extreme is visible as an extreme.

News

A filtered feed of commodity headlines.

Coverage

Three of these have no tradable futures contract on the data source, and are marked in the interface as proxies rather than presented as the thing itself. Rhodium trades OTC only and is scraped from a daily dealer quote.

How it works

fetchers/          collectors, one per source, each independently runnable
  fetch_prices.py    spot, futures curves and FX          (yfinance)
  fetch_cot.py       CFTC positioning                     (cot_reports)
  fetch_eia.py       US energy inventories                (EIA API)
  fetch_lme.py       LME warehouse stocks                 (Nasdaq Data Link)
  fetch_rhodium.py   rhodium spot                         (scraped)
  fetch_news.py      headlines                            (RSS)

db/                SQLite schema and every read query in one place
dashboard/         Dash app; one module per tab under pages/
config.py          instruments, tickers, ratio thresholds

Collectors and interface are deliberately separate: each collector can be run, retried or scheduled on its own, and a source going down degrades one panel instead of the app. Scheduling is left to the host — launchd on macOS.

Stack: Python 3.11 · Dash 4 · Plotly · pandas · SQLite · Docker

Running it

git clone https://github.com/Link1976/commodities-dashboard.git
cd commodities-dashboard
pip install -r requirements.txt

python fetchers/fetch_prices.py --history 1095   # first run: backfill 3 years
python dashboard/app.py                      # http://localhost:7860

Or with Docker:

docker build -t commodity-pulse .
docker run -p 7860:7860 commodity-pulse

Configuration

Two collectors need a free API key, supplied through the environment:

Neither is required to start. If unset, those two collectors skip and log why; prices, futures, COT and news are unaffected. Never commit these values — set them in your shell, or as Space secrets when deploying.

Known limitations

• —The EIA and LME collectors are scheduled by the host, so they run locally but not inside the container.
• —Illiquid contracts can repeat a stale close for several sessions; the figures are shown as reported rather than smoothed.

Notes

Built for my own use. The interface is in Spanish; the code and this document are in English. Not investment advice — it reports public data and nothing more.