Over the past two months we shipped a series of accuracy upgrades to the Taziky evaluation engine — the system that estimates what a car is worth from auction results and market data. Here's a plain-language tour of what changed.
Trim matching got a lot more precise
A "Civic Touring" and a "Civic LX" are thousands of dollars apart, but auction data doesn't always label trims cleanly. The engine now recognizes many more trim variants and matches your car against comparable sales of the same trim far more often, instead of falling back to a broad model-wide bucket. Fewer broad buckets means estimates built from cars that actually resemble yours.
Model mix-ups fixed
Manufacturer databases and market listings sometimes disagree on what a model is even called — a "Cruze Limited" is a Cruze, but to a computer those look like two different cars. We swept through these mismatches and taught the engine the aliases, so cars that used to get a "not enough data" answer or a skewed estimate now match their true comparables. The number of affected lookups dropped by roughly 90%.
Every defect is priced individually
This is the biggest change. When a car has auction damage codes — a branded title, major damage, missing keys, a prior taxi life — many pricing tools apply one blanket "damaged car" discount. We don't anymore.
Each specific defect now carries its own price impact, measured from how cars with that exact defect actually sold. A missing key is a small fixed cost. A branded title is a deep percentage cut. Flood risk, structural damage, a voided factory warranty — each one is priced on its own, and they combine the way real buyers stack them. Two cars with the same model and mileage but different defect lists will get meaningfully different estimates, as they should.
A simpler, honest value range
We replaced our old statistical value band with a clean ±10% indicative range around the estimate. Why? Testing showed the old wide bands looked precise but weren't more useful. The new range is deliberately labeled indicative: it tells you the neighborhood the price lives in, not a statistical guarantee. For final bidding decisions, the estimate itself plus the car's specific condition report is what to lean on.
Confidence you can actually read
Every estimate now ships with a 0–100 confidence rank alongside the LOW / MEDIUM / HIGH label. It reflects how many true comparables we found, how similar they were, and whether independent pricing methods agreed with each other. A 85-confidence estimate on a common Camry means something different from a 40 on a rare trim — now you can see that at a glance.
What's next
We're continuing to work on estimate accuracy for cars with sparse comparable sales — the hardest and most valuable cases. If an estimate ever looks off, use the feedback button on the result page; those reports go straight into our review queue.