GasPal timing track record
Figures below measure the original timing-v1.0.0 RBOB/WTI 0.5% algorithm, which is now the private shadow comparator. The active Regional Heuristic Timing Beta is still accumulating prospective outcomes. No 85% claim has been established.
This page measures the original legacy v1.0.0 timing baseline: fill today or wait, NOT the active beta. Station savings are separate and come from comparing current nearby prices. Every day, GasPal records its timing prediction for Gainesville, Florida, and San Mateo, California. About a day later, we check what prices actually did.
GasPal works in other cities too. We picked these two cities for regular testing so we cannot choose only the places where our predictions happened to be right.
Counting wins, misses and small price moves…
What if GasPal gets a prediction wrong?
Wrong guesses will happen, and we show them. GasPal is an early timing model, not a guarantee. If gas prices move at least 2 cents per gallon in the opposite direction, we mark the prediction wrong. We keep the mistake in our public results. Smaller changes do not count as a meaningful win.
If you record a fill-up, we'll show the next-day price change when we have enough information. If prices fell when we said they would rise, we'll show that too. If we can't check, we'll clearly say the result isn't ready.
How we improve: we measure misses, collect optional receipt feedback, compare new candidate algorithms against the locked version, and release a clearly labeled upgrade when evidence supports it. The current algorithm does not automatically retrain after each visit.
How it is measured
Around lunchtime each day, we save the prediction and the typical price among nearby stations open to all drivers. We calculate this typical price as the middle price when available prices are sorted. Around one day later, we check that same kind of price again.
If GasPal says FUEL TODAY, it was the better price choice if prices rose afterward. If GasPal says WAIT, it was the better choice if prices fell. We count a clear win only if prices moved by at least 2 cents per gallon. We still record smaller changes and wrong calls.
Average result so far means the total price differences from checked predictions, divided by the number of predictions. Negative results count too. The result for a 14-gallon fill is that average multiplied by 14. It is a comparison, not guaranteed savings.
We cannot change a saved prediction after seeing what happened. That keeps the score honest.