Dev log

Why a 2027 "used car" sometimes shows up in a Cyprus search

One Tuesday morning, AllCars started showing 2027 model-year cars. Plus a 2028. Plus one with 575 million kilometres on the clock. Here's why, and the gate I added so it doesn't happen again.

If you're shopping for a used car in Cyprus, the last thing you want is a search result that's just wrong. Future-dated model years. Mileages that suggest a car has driven to the moon and back. Descriptions that contain nothing but a town name. Garbage at the top of your filter is a fast way to lose trust in a tool.

For a small window, that's exactly what a few of the results looked like. Somewhere upstream, a parser had decided that the trailing four digits of a numeric blob were the year. Sometimes that blob was a price. Sometimes it was a phone number. The number always passed because, hey, four digits.

The fix wasn't in the parser

I could've gone hunting for the parser bug. I didn't. Upstream sources change shape all the time, and any parser will eventually drift. Instead I added a validation gate at the door: before any listing reaches the index, it has to make physical sense.

Year more than one ahead of "now"? Reject. Year before 1900? Reject. Mileage in seven figures? Reject. Description that's just a town name with no numbers in it? Reject. None of these checks know why a value is wrong — they just know what's physically impossible, and they refuse to let those records through.

What you see — and don't see

You don't see year-2028 listings in your filters. You don't see "deals" on cars with 800-million-kilometre odometers. You don't see "Limassol" sitting in a description column. The validation gate eats them at ingest.

The bonus: the rest of the pipeline gets simpler. Stages downstream don't need defensive logic for the obviously broken cases, because the obvious cases never make it that far. AllCars stays clean for you because it stays strict at the door.