On Mars, the night lasts 14 hours. Solar panels produce nothing. Temperature drops to -80°C. The colony survives on what it stored during the day.

Every system has a night.

Cloud provider outage. Your API is down for 4 hours. If the answer is “error page,” you have zero thermal mass. If the answer is “cached data and queued writes,” you designed for the night.

Funding gap. Revenue dips. How long does the system survive on stored reserves?

Key person absence. The one engineer who understands the billing system is on vacation. Can the system survive two weeks?

The night is predictable. You know your cloud provider will go down. You know your funding will fluctuate. You know your team will have turnover. The night is coming. The only question is whether you stored enough energy.

Designing for the night: Measure your minimum viable power — what’s the bare minimum to stay alive. Size your reserves for the longest expected night. Build sleep modes — when reserves drop, voluntarily shed load. Make sunrise automatic — resume full operation without manual intervention.

The colony doesn’t panic at sunset. It was designed for this. Your system should work the same way. The night isn’t a surprise. It’s a design parameter.