pretickForecasting gas prices across dozens of markets.
A prediction-market companion app set out to give people betting on platforms such as Kalshi and Polymarket better information. Its focus was gas prices: historical settlement data, a live proxy, and a way to estimate how prices could move before a market settled.
An anonymous implementation study · September 2026
The data was available. Interpreting every market was harder.
The app brought together AAA price history and live GasBuddy observations. For the markets using AAA as their settlement source, GasBuddy provided a frequently updated proxy that users could follow during the day. Charts made the historical relationship and the latest movements visible.
But the app covered dozens of markets: national and state-level prices, with different settlement dates and time horizons. Understanding one chart was manageable. Building and maintaining a mental model for every market was much harder.
Each geography could have its own recurring behavior. A user might notice that a state's prices tended to move at a particular time or on a particular weekday, then need to work out whether today was following that pattern. For example, if Washington's historical prices rose most after 5 p.m. on Mondays, a flat Monday afternoon could mean something different there than in another state. That is an illustration of the kind of local pattern a user would need to investigate, rather than a measured finding reported here.
The missing piece was a reasonable path for the rest of the day.
Showing what had happened was useful. The app also wanted to give users an estimate of what might happen next: how could GasBuddy move over the remaining hours, given comparable historical days and the observations already available today?
Price histories can contain patterns across seasons, months, weekdays and intraday periods. The relevant history and timescale matter. For this task, the app focused on comparable daily sessions and today's price path, rather than asking users to mentally extend dozens of charts themselves.
Where Pretick entered the workflow
The app sent timestamped historical observations and today's observations to Pretick, along with the period it wanted to forecast. Historical days remained separate sessions. The API returned a series of predicted prices and timestamps that the app could place directly on its charts.
This gave each supported series a starting forecast without requiring a dedicated model for every state. Users could compare the projection with the actual observations, the historical glidepath and the prices implied by the market. The app could refresh that view as more data arrived.
The implementation uses a consensus forecast and also offers a separate Kronos opinion based on continuous price history. State-level forecasts use the same API and are cached for display, so opening a chart does not itself trigger another paid forecast.
The forecast also needed to keep up with the day.
A fixed historical projection could show a typical Monday, but it would stay unchanged as new prices came in. If today's path began to diverge from that pattern, users needed an updated estimate of the remaining hours.
A fresh request to Pretick could include the latest observed prices and produce a revised forecast from that point forward. The app could incorporate live changes without manually rebuilding a model for each market. This updates the forecast using new observations; it does not retrain the underlying AI model.
From a proxy forecast to a trading decision
Pretick forecasts the GasBuddy series. The app still handles the relationship between GasBuddy and AAA, the relevant geography, and each contract's settlement rules and timing. Those steps turn a projected proxy path into context for possible settlement prices.
For the user, the practical value is a forecast they can act on in their analysis immediately. If the projected path suggests a different outcome from the one reflected in market prices, that gives them a specific discrepancy to investigate as a potential trading edge. They can examine the underlying observations, compare model opinions and decide whether the difference supports a bet.
The aim was to make that first assessment easier across many markets, before a user had built a detailed model for each one. AI supplied a historical-data baseline; the app supplied the market context. The integration described here has not established a measured trading edge.
Read how timestamped observations and historical sessions are submitted to Pretick.
Source context: AAA fuel prices and GasBuddy gas prices. Platform names describe the intended audience; no affiliation or endorsement is implied.