I've spent most of my time on used cars — 1.1 million listings, 73 regions, the usual churn of price drops and re-listings. So when property data started landing in the same dashboards I use for cars, I didn't expect it to hold my attention. It did, for reasons that had nothing to do with the plan.
Here's the log, roughly in the order I actually noticed things.
Day one: the number that doesn't move
First thing I check on any new feed is the total, because totals lie less than anything else in this business. Property sat at 601,702 listings. Fine. Then I split it by region and the number stopped looking like a global rollout and started looking like a single-country project with a rounding error attached: Japan alone accounted for 588,644 of those listings. That's 97.8% of everything in the feed.
For context, our car data spans 73 regions with no single one topping 6% of the total (France, our biggest, is 67,569 out of 1.1 million — about 6.1%). Property is running at sixteen times that concentration. I went back and forth on whether this was a bug in the region tagger before I remembered what actually drives it: Japan's own real estate portals list an enormous volume of akiya, the abandoned or near-abandoned rural homes the government has been trying to offload for years, often at prices that make a used Corolla look like a luxury purchase. The portals are thorough, machine-readable, and enormous. Everyone else's aren't, yet.
Day two: looking for the other 2.2%
I wanted to know what "everyone else" looked like, so I pulled the non-Japan regions out and lined them up.
Egypt edges out Germany, which surprised me — I'd have guessed a large, well-documented market like Germany's would dominate a random sample before Cairo did. Sri Lanka holds a steady four figures too. Then there's a cliff: the four Central American countries — Guatemala, Nicaragua, Panama, and El Salvador — sit at 80, 80, 80, and 78.
Day three: three identical numbers are never a coincidence
That cliff is what actually stopped me. Three countries landing on the exact same count isn't a market signal, it's a fingerprint. When Guatemala, Nicaragua, and Panama all show precisely 80 listings, that's not three economies converging on the same inventory size — that's one upstream source with a hard page cap of 80, feeding all three the same way. El Salvador's 78 is the tell that confirms it: close enough to be the same source, different enough to prove it isn't a copy-paste error in our own pipeline. Somewhere there's a regional real-estate aggregator that caps free result pages at roughly 80 listings per country, and until we go around it — direct-source instead of aggregator, or paginate past whatever wall it puts up — those four countries will read as identically tiny no matter what their real markets look like.
This is the unglamorous part of building any new dataset honestly: the first numbers you get aren't the market, they're the shape of your current sources. I'd rather find that out in week one than publish a "smallest real estate markets in Central America" post built on a scraper's page limit.
Day four: what actually surprised me
Not the Japan concentration — I'd half-expected that going in, given how much of our own property signal already came from Japanese sources on the car side of the business. What surprised me was how legible the data-quality problem was without any special tooling. You don't need a statistics background to spot three 80s in a row. You just need to look at the numbers in order, which is most of what this job actually is.
It also reframed something about our car data that I'd stopped noticing because I'm used to it: 73 regions with genuine spread, no region above 6.1%, 1,375 independent sources feeding it — that distribution isn't the natural state of a scraped dataset, it's eight years of deliberately going around exactly the kind of aggregator wall that just showed up on day three of the property feed. Every one of those 73 regions started out looking like Guatemala.
Where it sits now
Property isn't a product yet — it's a signal we're watching while the car side keeps running, one line in a longer roadmap toward wider real estate coverage. But a week of watching it fill in was a decent reminder of the order operations actually happen in: get the feed live, get suspicious of round numbers, find the wall, then go build a way past it. Japan will stop being 97.8% of this feed the same way France, not the UAE or Japan, ended up on top of our car counts — by adding sources until no single one can dominate the total by accident.
