Projects · 03

Orbital Watch

Satellite passes over a ground station next to live space weather, on a static site that reports how fresh its own data is.

Open the live site Read the code

Right now

Checking…

Read from /api/status after the page loads. How it works

My part
My own project, built with Claude Code. Its design decisions, and the failures that forced them, are written up in the repository's DECISIONS.md.
Result
Each of the seven upstream sources is marked fresh, stale or failed on the page, with the reason. A build with every source unreachable still finishes and says so, and CI checks that on every push.
Stack
  • Python build
  • Cloudflare Pages
  • A Cloudflare Worker on a schedule
  • GitHub Actions
  • pytest
The Orbital Watch dashboard. A header marked FRESH with the build time, an operational assessment reading geomagnetic conditions nominal, five space-weather tiles and a chart of recent planetary Kp values.
orbital.rodrigopimenta.com on 18 November 2026. Taken with headless Chrome at 1440 by 900, at twice the pixel density.

The idea

A monitoring system must report its own degradation as honestly as it reports its data. Every number on the page carries its source and age. Every upstream source is classified fresh, stale or failed against published thresholds, and no fallback is ever allowed to claim freshness it does not have.

The orbital mechanics are not the interesting part. What happens when things break is.

How it works

A Python build fetches element sets and space weather, computes satellite positions and passes, judges the health of every source, and writes JSON files into the build output. The page loads those files and renders. The browser never contacts an upstream API.

That buys three things. An upstream outage cannot take the site down: the build falls back to a cache or a committed snapshot, every panel still renders, and the health panel says the data is stale and why. Hosting is static files on a CDN. And “last updated” is measured, not asserted, because every fetch is recorded with its outcome and time.

Every network call goes through one function. It has a hard timeout, catches failure, records the outcome, and never raises, so no caller can abort the build by forgetting a try.

Why it is on this site twice

The status panel on the home page follows the same rule. Its service reads this project’s own health file, and when a check fails the panel says “could not check” and gives the reason. It does not guess.

Architecture · scroll sideways for all of it →

How Orbital Watch gets its data A scheduled refresher on GitHub Actions is the only thing that calls Celestrak; it commits a snapshot to the repository. A Cloudflare Worker on a schedule triggers a Cloudflare Pages build every two hours and checks that the previous build produced fresh data. The build reads the snapshot, fetches NOAA space weather, computes passes, judges the health of each source and writes JSON. The result is static files. The browser reads those files and never calls an upstream API. Celestrak element sets snapshot refresher GitHub Actions, commits it NOAA SWPC space weather Cloudflare Worker every 2 h · dead-man check build Cloudflare Pages fetch · compute passes judge each source write JSON static files HTML, data/health.json and four more JSON files browser never calls upstream
A scheduled build fetches, computes and judges its own sources, then writes static files. The browser never calls an upstream API, so an upstream outage cannot take the page down. Drawn from README, "Architecture" in the repository.

Where these claims come from