What a public information platform has to survive

In July 2026 a community platform we built for Guam went from almost no traffic to thousands of residents in about forty-eight hours, because a typhoon was coming. It kept growing after the storm passed. The numbers were satisfying. The lessons were more useful.

It has to work on the worst connection someone has

The week a storm approaches is the week bandwidth degrades, power becomes uncertain, and everyone is on a phone. That single fact should determine the entire architecture. Pages have to be small and fast on a bad mobile connection. Anything requiring a download or an install is a barrier at the worst possible moment. The platform has to open in the browser the resident already has, on the phone already in their hand, with whatever signal is left.

This is not a place for heavy frameworks or clever loading. It is a place for pages that render immediately and degrade gracefully. The most sophisticated engineering decision here is restraint.

The missing layer was the residents

Official sources are authoritative and necessarily slower. A utility confirms an outage after it has been verified. Meanwhile the person on the street already knows their village is dark. The gap between those two moments is where people are most anxious and least served.

Letting residents file structured, village-tagged reports closed that gap. Not free-text comments, which become unusable within an hour, but structured reports with a category and a location so they can be counted and mapped. During the storm the overwhelming majority documented power outages, concentrated in a handful of villages, which drew a live outage map faster than any single official channel published one. The map was made by the island about the island.

Residents did not only consume information. They produced the most current picture available.

Honesty is a safety feature

The moment you accept community reports you inherit a serious obligation. Every resident-submitted item has to be labeled as exactly that, with the actual authorities named plainly, so nobody confuses a neighbor's observation with a utility confirmation or a weather service warning. Moderation has to happen before something misleading spreads, not after.

This is where we use AI as a working component rather than a headline: sorting and flagging incoming reports so a human reviews the right ones first. The model triages. It does not decide what the public sees.

What we would tell anyone building something similar

• Build for the phone and the bad connection first. Everything else is secondary.
• Structure what you collect. Categories and locations turn reports into a map; free text turns them into noise.
• Label the source of every claim, and never let your platform imply official status it does not have.
• Moderate before publication, and give a person the final say.
• Launch before you need it. The platform that goes live during the emergency is already late.

That last one deserves emphasis. The platform sat quiet for days after launch. That quiet period is not failure, it is readiness, and it is the only way the thing is in place when conditions change. Guam gets a typhoon season every year. The infrastructure for it should be standing before the forecast does.

The platform is independent and free, presented by a Guam newsroom, and it is not an official government service. During an emergency, official guidance from the National Weather Service, the Guam Power Authority, the Guam Waterworks Authority, and Homeland Security is authoritative. For life-threatening emergencies, call 911.