The App That Grades Its Own Forecast

A Forecast No One Checks

Most people open a weather app with a simple goal: they need to decide what to wear, whether to mow the lawn this afternoon, or whether to move a weekend get-together indoors. A cloud icon, a temperature, and a probability of precipitation appear on the screen. The day plays out, the weather does what it does, and the next morning a fresh forecast appears on the screen. No one ever goes back to look at what the app claimed yesterday.

Ask any mainstream weather service how often it actually gets it right. In most cases, you will not find an answer. The forecast is generated and presented, and once the day has passed it quietly slips into obscurity.

Yet weather forecast accuracy is the one thing that should matter most. It marks the line between a practical planning tool and a mere generator of random guesses. If you do not know how reliable the data in front of you is, you cannot plan around it with any confidence. Weather InTouch takes the opposite approach: instead of sweeping yesterday’s mistakes under the rug, it puts evaluation front and center.

What Gets Compared

Meaningfully measuring forecast reliability takes more than a vague feeling that yesterday was fairly nice. The process requires strict discipline and systematic data logging.

When a location is served, the application records what it expected for the following day; where no record of its own exists, it falls back to the archive of earlier model runs. Once that day concludes and the data provider’s analysis of what actually happened arrives, the two datasets are placed side by side. The building block of the evaluation is the next-day forecast. In other words, the evaluation answers a simple question: what did we say yesterday about today, and what actually happened?

A single measurement means very little, however. In meteorology, something unusual can happen — a weather front might stall unexpectedly against mountains, causing the forecast for that day to fail. Conversely, a stable summer high-pressure system might linger for a week, when nearly any prediction would hit the mark. That is why overall forecast accuracy is calculated using rolling windows, specifically over the past seven and thirty days.

The seven-day view shows how the system performs under current weather patterns. The thirty-day window provides a long enough horizon to prevent a single unusually good or bad day from skewing the overall picture. Furthermore, the results are not tracked for a single city in isolation; they are aggregated across all monitored locations. The aggregate therefore says whether the system is reliable across the board, not just at one favorite spot.

Rain: Hitting the Mark or Crying Wolf

Temperature forecasts tend to be fairly accurate, and an error of two degrees rarely ruins anyone’s day. Precipitation is the real stumbling block. When an app claims ninety percent forecast accuracy, that figure can be deeply misleading. If it only rained three times in a given month and the system simply predicted dry conditions every single day, it would achieve an outstanding score on paper — without ever predicting a single drop of rain.

That is why precipitation requires a strict distinction between two completely different types of errors:

The first type is a missed forecast: rain actually occurred, but the forecast failed to catch it. The user left their umbrella at home, went for a bike ride without a rain jacket, or left the skylights open and got soaked. This metric tracks the proportion of actual rainy days that the system successfully flagged in advance.

The second type of error is a false alarm. The app warned of heavy rain, people canceled outdoor gatherings or postponed pouring a concrete foundation, but the sun shone all day.

For users, these two scenarios present fundamentally different problems. A single accuracy percentage cannot distinguish between them, often disguising false alarms as harmless caution. When you separate the two metrics, the truth becomes clear: does the service cry wolf unnecessarily, or does it overlook incoming rain?

Timing Is Half the Answer

Knowing that it will rain on Tuesday is only half the information you need. In everyday life, knowing when it will start is just as critical. If rain is expected in the afternoon, you can still mow the lawn in the morning. But if precipitation arrives at nine in the morning, those plans fall apart.

The system therefore does not treat rain as a simple yes-or-no for the whole day; it also evaluates timing accuracy. Specifically, it calculates by how many minutes the actual onset of precipitation differed from the predicted time.

The direction of this discrepancy matters significantly. There is a huge difference between rain arriving two hours late and two hours early. When a shower arrives later, it usually just means you rushed home unnecessarily. When it arrives earlier, it catches you unprepared halfway through your trip. Tracking the direction of this timing error tells users how to read the estimated times.

Closely tied to this is what we call hourly overlap. Consider a day when the app forecasted rain in the morning, between 8:00 and 11:00 AM. In reality, the morning was sunny, and a heavy downpour did not arrive until 5:00 PM. From a full-day perspective, a rainy day was correctly predicted. In reality, however, the system was not right for a single hour. That is why we evaluate how far the predicted rainy hours overlap the hours when rain actually fell.

Specific Days, Not Just Averages

Statistics and long-term averages provide a helpful overview, but they can feel abstract to everyday users. You only truly begin to trust a monthly accuracy score when you can connect it to your own experience from last week.

Alongside aggregate metrics, the app displays a detailed breakdown of specific recent days. The view is straightforward: one column shows what the app predicted, and the next shows what actually occurred at that location.

You can easily pull up last Tuesday, for example, and see that the forecast called for rain at 2:00 PM and a temperature of 19 degrees. In reality, the rain started at 3:00 PM and the thermometer read 20 degrees. Seeing specific logs like this gives the statistics a human face. You immediately recall your own experience and can judge for yourself whether that discrepancy made any real difference in daily life.

This commitment to transparent data fits the broader design of the platform. If you want to see what Weather InTouch can do overall, you will find that this focus on practical usability and clarity runs through every feature, from alerts to automations.

Why Make It Public

Most meteorological projects guard their internal error statistics closely. Publishing data on where a forecast missed takes courage and confidence in the quality of the product. Why choose complete openness?

The primary reason is building trust. No meteorological model in the world predicts the weather with one hundred percent certainty. The atmosphere is a dynamic, volatile system where a tiny shift can develop into an unexpected storm. Claiming that our data is infallible would be dishonest. It is far better to be upfront: here are the numbers for the last seven and thirty days, and here are our limits.

When users can see how the forecast performed at the locations the app tracks, and what types of errors to expect, they can make far better decisions. Instead of blind trust or total skepticism, they get a realistic picture. They can trust that when the system signals a high risk of rain, it is not a random guess, but a number whose hit rate the app measures and publishes. For a closer look at how this concept was developed, check out the project case study.

Public evaluation also serves as relentless feedback for ongoing development. Once performance is visible to every visitor, there is no room for compromises or for ignoring systematic errors. Every improvement and every dip is immediately reflected in the numbers.

If you are curious about how the forecast performs today and how accurately yesterday’s predictions matched reality at the tracked locations, you can explore the current evaluation directly on the Weather InTouch website.