With civic data, being right is the whole job

It also has a nearer, plainer job, because the present is where most people are stuck. Who's your House representative? What does a county commission actually do? When's your next election, and who's on the ballot? The information exists — all of it is public, sitting in government records — but it's scattered across thousands of agency sites in a hundred incompatible formats, which for a normal person is the same as not existing. Public and unfindable is a peculiarly modern way for information to be missing.

So CivicsMatter is two things pointed the same direction: a legibility layer over government as it is right now — representatives, governors, counties, districts, states — and the first bricks of that much larger historical record. I built it because I genuinely love the subject, and because an informed citizenry is a kind of infrastructure nobody quite owns building. The word I keep coming back to is legibility, not access: what's missing isn't the records but the layer that makes a thousand mismatched sources trustworthy at a glance.

But civic information turned out to be different from the other things I map, and the difference is the whole story. It's what the industry calls YMYL — “your money or your life” — the category that Google, and plain common sense, hold to a higher standard. Being wrong about a barbecue joint's hours is an annoyance. Being wrong about who represents someone, or what their rights are, or when they can vote, is a different kind of wrong. And I learned that the hard way, by being wrong.

At one point, thousands of my county pages confidently displayed a sentence that read: “The county seat is .” Just a period, where a fact was supposed to be. The data to fill it didn't exist in my pipeline for most counties, and the template printed the hole anyway, in a clean, authoritative-looking sentence that said nothing at all. On most sites that's a cosmetic bug. On a civic site it's a small betrayal: you came to learn something true about the place you live, and I handed you a blank dressed up as knowledge. The failure wasn't the missing datum — data is always missing somewhere. The failure was that the template treated absence as a value, publishing the empty string as though it were an answer.

Around the same time, a Google core update swept through and demoted a huge swath of those thin templated pages — the corpus went from real traffic to almost none. And I want to be honest about how I feel about that, because it's the center of this whole essay: it was right to. Not just algorithmically right — right, full stop. I had built pages that were thin because the facts underneath them were thin, and a thin civic page is worse than no page at all. The algorithm and the moral stakes were telling me the same thing, at the same time, and both were correct.

So I did something that felt awful and was obviously the right call: I took thousands of my own pages down. Deliberately delisted the thin ones rather than keep them up padding a number. I didn't hard-delete them — I noindexed them and pulled them from the sitemap, so the work can come back the day the facts behind it are real. But the arithmetic is simple — fewer pages that are true beat a big corpus that's mostly hollow — and a platform that wants the authority to tell you about your own government has to earn it. You do not earn it with volume.

The county-seat blank rewired how I publish anything programmatic. The old default was optimistic — assemble the page, and a missing field would mostly be fine. The new default is suspicious: a page earns publication only when every load-bearing field it needs is present and sourced; a required fact that's missing withholds the page rather than rendering it with a hole. This is the same quality gate I now run in front of every programmatic build I own, and the habit underneath it is fail-loud, never fail-quiet. I would rather a citizen find nothing than a hollow sentence, which is a lie the reader can't see.

The part nobody tells you about civic data is that acquiring it is the easy part — the records are public. The hard part is reconciliation: the same office named three ways across three counties, boundaries redrawn so one file contradicts another, a name that changed at some point in 250 years, an official who appears twice under two spellings. Raw civic records are a commodity anyone can download. The moat is the cleaning — the unglamorous discipline of making the download trustworthy enough to publish. When the sources conflict and I can't yet tell which is right, the honest move is to withhold, not to average two guesses into a false certainty.

That reframed the entire project for me. For most of what I build, the guiding question is “is this useful, is this interesting.” For CivicsMatter, accuracy isn't one feature among several — it's the entire permission to exist. If I can't be confident a fact is right, it doesn't go up. Everything I'm doing on it now points there: better sources, more automation around verification, a pipeline that would genuinely rather show you nothing than show you a blank pretending to be a fact.

I'll be straight about what CivicsMatter is, because the honest version is more interesting than the flattering one. It's a programmatic platform — built with the same engine as my other sites, generating pages at scale from structured data. A skeptic could fairly ask whether a programmatic build has any business informing people about their democracy. That's the tension the whole project lives inside. My answer is that the method is neutral; what matters is what you hold above it. A platform that prunes itself for truth is a different thing from one that farms pages for traffic. The engine is identical. The discipline is not.

I didn't build this as a business, at least not first. There's a version where it earns money eventually — but only after it's consistently, confidently, accurately comprehensive. Monetizing an unreliable civic platform gets the order exactly backwards, and on this subject, more than most, the order is the ethics.

So the work now is depth and reach at once — the national spine first: presidential, Senate, and House elections and the people who won and lost them, then outward toward governors, cabinets, officeholders, and eventually every local seat I can source. And there's a convergence I can already see coming. So much of the early republic was decided in war — the Revolution, 1812, the Civil War — so the civic history I'm building here and the military history I've mapped in The War Atlas are, in the end, the same story told from two sides. Both live or die on the same discipline: refusing to print a precise-looking number I can't stand behind. At some point those two projects meet.

I still watch the Magic Wall for fun; that hasn't changed. What changed is that I finally understand what I'm building. Not a big website about government — a promise. That when you come to it to learn something true about the people who govern you, it's true. Everything else is downstream of that. The county-seat sentence isn't a blank anymore. That's a small thing. It's also the entire thing.