How Lumenfold cut rollback time from 40 minutes to 4 seconds
Lumenfold deploys 60 times a day to 4 million readers. Their rollback used to be a 40-minute redeploy. Now it is a toggle, and their on-call rota is boring again.

Lumenfold publishes long-form journalism to four million readers a month, from a monorepo that deploys about sixty times a day. Two years ago, undoing a bad change meant reverting the commit and waiting for a full pipeline: 40 minutes on a good day.
The incident that started it
A new paywall check shipped on a Friday evening and locked out every subscriber in Canada. The fix was one line. Getting it to production took 52 minutes, most of it waiting on tests for code nobody had touched.
We did not have a code problem. We had a time-to-undo problem, and you cannot test your way out of that. — Dana Okafor, VP Engineering
What changed
Every user-facing change now ships behind a Driftwire flag, starting at 1% of readers in one region. The paywall team owns a kill switch per payment provider, and the platform team owns one for the whole subscription service.
- Rollback time: from 40 minutes to 4 seconds
- Deploys per day: from 22 to 60
- Incidents that reached more than 5% of readers: from 9 a quarter to 1
What they would tell another team
Start with the kill switches, not the experiments. The first time someone turns off a broken integration from their phone in four seconds, the rest of the organisation asks for flags on its own.
Want numbers like these? Book a demo and bring your worst rollback story.

Written by
Mara Sato
Developer Advocate
Mara writes the guides and talks to the teams who use Driftwire every day, then writes those conversations down as case studies.
More posts by Mara Sato