WxDuel Game News

Posted on January 26, 2026 by testuser

This past weekend's storm brought some unexpected (actually expected!) challenges, including a power outage that impacted time to work on this page. Work is continuing, and several updates are now live.

Technical Improvements

I have been focused on several "under the hood" improvements to enhance site performance and accuracy. This update includes fixes to our statistics calculations and refinements to the methods used for retrieving model data.

New Feature: User Hits Ranking

I am excited to introduce a new results section. This page ranks users based on their "hits"—specifically, how many times they successfully nail the actual maximum or minimum temperature forecast. Head over to the rankings to see who is currently leading the pack!

WxDuel on YouTube

In addition to site updates, I am beginning to create content for the WxDuel YouTube channel. I plan to use this space for game development updates, live streams, and periodic weather updates. While everything is still very much a work in progress, well to be honest, it is an experiment in learing.

^ Back to Top

Posted on January 6, 2026 by testuser

Happy New Year! Rolling into 2026, I've been working under the hood of WxDuel. The biggest addition is a brand new way to measure accuracy: Hits Leaderboards.

"Hits" vs. Errors

While our standard MAE (Mean Absolute Error) scores tell us how close a forecast was on average, they don't capture the thrill of nailing a forecast exactly. The new Combined Hits (Max + Min) leaderboard tracks how many times a user predicts the exact observed temperature (0° error) across all 7 cities in a game. Check out the Standings page to see who is racking up the most perfect "7/7" games!

Engine Tuning & New Models

I worked on several updates to the scoring engine to squash bugs and ensure verification data is correct. You might have noticed new automated players joining the fray recently. These new additions will likely round out the automated players we will use.

Meet "testuser"

You may have spotted a player named testuser climbing the ranks. This isn't your standard bot! It's an experimental hybrid account simulating a human forecaster. It generates forecasts using a weighted randomization of model averages, but with small manual "tweaks" applied to mimic human intuition.

There is a lot more to come in 2026. Stay tuned and happy forecasting!

^ Back to Top

Posted on December 30, 2025 by testuser

As noted in last night's update, Atlanta's temperature took a plunge, hitting 32°F just before midnight! This sharp drop showed that there was more work to be done in ensuring the automated data processing for the NWS forecast was correct.

I have now implemented corrections to how the NWS forecast is retrieved to ensure it handles non-diurnal temperature trends more accurately in the future.

Finally, I'm excited to announce the arrival of two powerful new AI-driven players:

• AIGFS (00Z): NOAA's new deterministic AI model.
• AIGEFS (00Z): The ensemble version, providing a mean forecast from the AI-driven ensemble system.

Both are now live on the leaderboards and available for comparison in the detailed stats. Happy forecasting!

^ Back to Top

Posted on December 29, 2025 by testuser

Atlanta is experiencing a textbook non-diurnal temperature trend today. After starting with mild temperatures this morning, a strong cold front is driving readings down throughout the day—plummeting from the 50s down into the 20s by tonight. This creates a challenging forecast scenario where the daily high occurs in the morning rather than the afternoon.

Just in time for this transition, Iincorporated our recent non-diurnal testing protocols into the core engines for the NBM, NWS, and NBP players. These updates significantly improve how our automated players handle "wrong-way" temperature curves and rapid airmass changes. With the automated players better tuned for complex transitional weather, the baseline for victory has been raised.

The uncertainty in handling this airmass creates a 15-degree spread in the low temperature forecast! While some models like the CMCE and GEFS are hugging the climatological normal near 37°F, others like NB1 are aggressively undercutting the guidance, predicting a plunge to 23°F. Part of the reason for the high ensemble values is that I am only able to pull three hourly temperature data from the grib files. Thus, we cannot capture the entire temperature swing.

I’ll be glued to the observations this evening watching the temperatures drop, waiting to see just how low they go before the official climate day ends. It feels just like watching a ball game go down to the wire in the final seconds—which is exactly what will make playing WxDuel so exciting!

^ Back to Top

Posted on December 25, 2025 by testuser

Merry Christmas, WxDuelers!

Snuck in a nice update to WxDuel today. My goal with this project has always been to answer a simple question: "Who really knows what the weather will do?" To make that comparison more interesting, I've significantly expanded the field of automated "Benchmark Players" that satisfy the competitive field.

Introducing Three New Challengers

I'm excited to add three new automated forecasting bots to the mix. These models cover different resolutions and strategies, giving us a much wider range of guidance on the dashboard:

• NAM (North American Mesoscale): A high-resolution (12km) regional model. I've set this bot to run on the 12Z cycle to capture fresh morning guidance. It's often great at resolving local terrain effects that the coarser Global models might miss.
• HRRR (High-Resolution Rapid Refresh): The gold standard for short-term convection. Running at a 3km resolution on the 12Z cycle, this model adds detailed insights into precipitation and temperature trends for the immediate future.
• GEFS (Global Ensemble Forecast System): This was the trickiest one to build. Rather than using a single "deterministic" run, I programmed the GEFS bot to calculate the Ensemble Mean from the full dataset (~31 members). This "consensus" approach smooths out outliers and typically provides a more stable forecast than any single run. I am not quite sure how this approach will work, it will take a little monitoring to see how it turns out.

Site Improvements

To support these new players, I've pushed several updates to the interface (Version 2025.12.25 Alpha):

• The Past Weather charts now feature trend lines for all six active bots (NWS, NBM, GFS, NAM, HRRR, GEFS), so you can visualize model agreement (or disagreement!) at a glance.
• The Daily Forecast Entry table is expanded. You can now see side-by-side guidance from all models for Max Temp, Min Temp, and Precipitation before locking in your numbers.
• I also squashed a few layout bugs to ensure the data columns align perfectly.

I look forward to seeing how these new ensemble-based and high-resolution strategies stack up against the NWS—and against you—on the leaderboard!

I think the next task will be to take a look at the full NBM probaility suite and get some of that date included into mix.

^ Back to Top

Posted on December 21, 2025 by testuser

Happy Winter Solstice! Today is the shortest day of the year, which feels appropriate because I’ve spent most of it staring at code while the sun was barely up. I spent this Sunday doing a bit of "housekeeping" on the site, focusing on fixing bugs and making the data tools more useful. What I worked on today: New View Modes: I’ve created an Analysis Lab so you can actually toggle how you see your history. I am not sure if I will continue to call it the Analysis Lab, we'll see. You can now switch between seeing your daily averages (to see your overall trend) or a granular view of every individual city forecast. You reach the Analysis Lab from the Rankings page and clicking on the individual user. Raw Data Access: I added a "View Raw Data" option to the analysis lab. If you want to audit your own performance or see exactly why your error spiked on a certain day, you can now see the full table of your forecasts side-by-side with what actually happened. I spent a good chunk of time fixing bugs in the scoring engine. As we hit the "bottom" of the year's daylight today, the data on the site is finally starting to look a lot clearer. I’m still tweaking some of the newer features and the layout, but the core engine is running much smoother now. Starting tomorrow, the days start getting longer again. Hopefully, that means more daylight to catch those tricky winter temperature swings! It also means going back to a the real job which means a slow down on introducing new things here. For now I plan to let the bots forecast and verify data to ensure the scripts that generate statistics and rankings are working correctly. The data certainly is off for the bots, with some errors and bad entries noted. The will shake out as the week progresses.

- kmc

^ Back to Top

Posted on December 20, 2025 by testuser

Hi everyone, and welcome to WxDuel.

I built this site as a fun way to track forecast accuracy without all the noise. It’s pretty simple: predict the high temperature and precipitation, and see how close you get to the actual observations.

What's running now?

Right now, the site is in Alpha Mode. I have the automated system running daily games for 7 cities, mostly just to make sure the scoring math works correctly.

You can see the NWS and NBM (National Blend of Models) bots already posting their forecasts on the Results Page. It's interesting to see how they compare day-to-day. Note that some of the results will be inaccurate at times as I tweak formulas and how results are calculating. In this Alpha Mode expect data to drop off and reset at times. Bear with me as I get things spun up to be ready to accept real players.

Coming Soon

I'm working on adding a "Climatology" bot next, just to show what the average weather for that day usually looks like. Once I iron out a few more bugs, I'll open up registration so actual people can play along.

Thanks for stopping by!

- kmc

^ Back to Top