When I posted about Dispatch Zero a month ago, I mentioned that the app was built from a town of about four hundred people, and that small-town coverage was the part of the system that worried me most. The Wikipedia tier kept things from breaking entirely (every small town has at least something on Wikipedia, usually a historic building), but in my own town, the app would find the Opera House on a first dispatch and then have almost nothing else to send. After the ninety-day re-entry filter kicked in, you’d get the Opera House again, and again.

That was the gap. This post is about how I closed it a bit.

The Federal Dataset That Used to Have Everything

The United States Geological Survey maintains a dataset called the Geographic Names Information System, or GNIS. It’s public domain, exhaustive, and goes back decades. For most of its history it catalogued not just natural features (lakes, summits, streams) but also cultural ones — churches, cemeteries, post offices, schools, dams, bridges, towers, mines, parks, trails, waterfalls. Tens of thousands of named places per state, with coordinates. Exactly what I needed.

Then I went to download it, and discovered that in 2021, USGS overhauled GNIS and removed all of the cultural feature classes from the active dataset. Modern GNIS is purely topographic — useful if you want to know where every named ridge and bay in Washington is, useless if you want to know where the small-town churches are.

A Detour Through the Internet Archive

The Wayback Machine had a snapshot of the pre-overhaul Washington file from March 2020. The data is about six years old now, which sounds bad but mostly isn’t — buildings don’t move, churches and cemeteries close very slowly, and the structure of a small town in 2020 is essentially the structure of a small town in 2026.

I wrote an importer that reads the legacy GNIS format, filters by feature class and a name-safety pass (the existing “school / academy / children” exclusion still applies, so we don’t dispatch people to schools), and upserts into a new fourth tier of the discovery system: a local PostGIS query that runs after the existing OpenStreetMap and Wikipedia tiers have all come up empty.

Three New Categories

Cemeteries naturally fit the existing historic bucket. Churches got their own church category back when this work started. The new feature types needed somewhere to live, and I tried to be thoughtful about it rather than just dumping everything into one bin:

• park — parks, named waterfalls, trailheads. The outdoor-scenic bucket. The handler’s briefing voice can lean into that.
• infrastructure — dams, bridges, fire lookout towers. Engineering landmarks. A dispatch to a small-town truss bridge reads very differently from a dispatch to a cemetery.
• civic — post offices. In a town of four hundred, the post office often is the most photographable building. It deserves its own framing.

I left out schools (safety), hospitals (operational), generic “Building” entries (too noisy, no way to distinguish a historic landmark from someone’s house), and named natural features like summits and ridges (the labelled point is often a long hike from any reasonable photo subject).

By the Numbers

Running the importer against the full state file produced 6,710 new dispatch targets across Washington:

• 2,693 churches
• 1,642 parks, waterfalls, and trailheads
• 1,265 dams, bridges, and towers
• 571 cemeteries
• 539 post offices

Concretely, what this changes: a town of about seventeen hundred that previously had zero in-range dispatch targets from local data now has eight churches and a cemetery within walking distance, plus the surrounding county’s worth of dams, parks, and bridges within a short drive. My own town, population four hundred, finally has at least one more in-range target: the post office, ninety-seven meters from the center of town. It’s not Manhattan-level density, but it is double what we had before.

Phantom Places

Six-year-old data has some lying in it. Buildings get demolished. Coordinates get entered wrong. GNIS occasionally has features that were never quite there to begin with, a name on a topographic map that nobody can locate on the ground today. The first dispatch the new system produced for my area was to “Armstrong Lake Dam,” which does not actually appear to exist, at least not where it says it is. There’s no lake, no water, and definitely no dam.

Rather than try to clean the dataset preemptively, I built the fix into the post-mission survey. When you give a location a thumbs-down, you now get two structured options for why:

• “I couldn’t find it.” soft signal. Could be GPS, could be bad signage, could be a coordinate error, could just be a miss in the dark.
• “It’s not really there.” hard signal. The place is gone, demolished, fenced off, or never existed.

Two “not really there” reports from independent users automatically flag the place, which removes it from any further dispatches and queues it for review. One report is not enough, GPS has bad days and so do people. The existing rule (three of the last five ratings being thumbs-down for any reason) also still applies, so consistently disappointing places get flagged on the general-negativity signal even without a specific reason.

Nothing gets automatically deleted. The auto-action stops at “flag for review.” Final retirement is a manual decision after looking at the photos and the reasons. That guards against the failure mode where a single coordinated user could nuke real places.

What’s Next

Other states are the obvious next step, the same importer works against any state’s GNIS file, and I’ll roll out Oregon and Idaho once I’ve spent more time with the Washington data. Beyond that, the GNIS file is one source; there are others (HIFLD’s places-of-worship layer for currently-active churches, Overture Maps for general POI coverage) that I want to evaluate next.

The backend for the new survey reason codes is live; the frontend chip picker is the next thing I’ll build. Until that ships you can still give a thumbs-down through the existing survey, it just won’t carry the fast-flag signal yet.

Longer-term, I want a way for users to add their own community places, a “pin this and I’ll add it” path for the things that no public dataset will ever know about. The local diner, the painted shed at the edge of town, the bridge nobody officially named. That’s where rural coverage actually finishes.

Try It

https://dispatchzero.ataary.com

Still very much in development. Sign up, request a dispatch, and if it sends you to a dam that isn’t there, the system now knows what to do with that.