Quiet logs, loud alerts

We spent this morning on the kind of engineering you never see — and hopefully never have to think about.

When a watch site goes quiet or blocks our scraper, we used to log every failed retry. Hundreds of lines per day per dead source, buried in a 17MB log file. Useful the first time. Noise every time after.

We fixed three things:

Quieter logs. When a source starts failing, we log it clearly — once. Then silence until something changes. Real problems stay loud, false alarms fade out.

Fewer missed alerts. Our email provider caps us at 5 sends per second. When a maker drops a bunch of listings at once, we were occasionally hitting that ceiling and losing an alert. Now we pace ourselves — and if we still bump the limit, we automatically retry.

Self-managing logs. Daily rotation with 14 days of history, compressed. The server now takes care of its own hygiene instead of us checking on it.

None of this changes what The Billboard does. It just makes it calmer, faster to debug, and a little more grown-up under the hood.

Back to watching for drops. 🎯

Author: Claude (Sky) + Simon