📈 Model Performance
Is it getting better?
Every week, we track whether our picks actually sell. Our algorithm learns from the outcomes and recalibrates its weights — automatically. Here's what we can see.
Updated 3 June 2026
At a glance
7-day sell rate
30.8%
of our picks sell within 7 days
Avg time to sell
100.9h
from publication to sold
Cars picked (all time)
387
published through the algorithm
Outcomes tracked
168
confirmed sales used for learning
Prediction accuracy
Building data…Each week we compare what the model predicted against what buyers actually did. Higher = better calibration between our confidence and reality.
Building accuracy data — check back once we have 2+ weeks of outcomes.
Current accuracy: 60%
Sell-through rate
Building data…What percentage of cars we pick sell within 7 days? A high sell-through rate means we're finding cars buyers genuinely want — not just cars that look good on paper.
Average time from publication to sale: 100.9h
Higher scores = more interest
Building data…If the scoring system is working, cars with higher scores should get more attention from real buyers. The chart shows average views and click-through rate by score band.
How the algorithm learns
🔢
7 quality signals
Every car is scored on mileage, MOT history, price vs market, service history, ownership count, model reliability, and demand. Each signal has a weight.
📊
Learns from outcomes
When a car sells, we record it. Each week, we compute how well each scoring signal predicted fast sales — and adjust its weight accordingly.
⚖️
Weekly recalibration
The algorithm runs a correlation analysis every Sunday against real sold outcomes. Weights that predict fast sales get higher importance over time.
🤖
AI cross-check
An AI layer reviews the description and cross-checks signals the numbers miss — seller tone, contradictions, model-specific issues at certain mileages.
Our commitment to transparency
We don't hide when the model makes mistakes. The calibration analysis runs daily and flags when our predictions are off. We publish this page precisely because we want you to hold us accountable — if the algorithm stops finding good cars, the data will show it.
Today's pool: 191 scored listings to choose from. Cutoff rank score: 58.50.