Using Ai To Find The Pekin Marigold Festival Medallion

I built a tool to find the Medallion. See here for details of the hunt:

https://pekinchamber.com/marigold-medallion-hunt/

So, for my prediction for the 2026 Pekin Marigold Medallion Hunt (made after Clue #5, before Clue #6).

I'm posting the SHA-256 hash of a text file containing my predicted location plus a random nonce.

After the medallion is found I'll post the file; anyone can run sha256sum on it and check it matches.

SHA-256: 374d6cbc5227a6225365dbee07b4f38e77ff2a3069d5166ed6938a68214b1a52

It was found where I predicted (and looked)

The medallion was found on Labor Day at the Dirksen Park RC Field area. I had predicted Dirksen park as early as Wednesday. But didn't think about posting proof until after Clue #5 (Friday I think).

I actually went to the RC field on Sunday before Labor Day, and we looked for about 15 minutes, but my kids were getting tired and disengaged, haha.

See the proof below.

All this was commited to GitHub so we have third party timestamps, and I also posted on Facebook (now deleted though).

Analysis

Overall, the hunt is extremely tractable if you treat it as a constraint problem instead of a riddle. Pekin, while not a small town, isn't large by any means. And the number of locations in town than it can be hidden is fairly small.

I collected data on all previous hunts that I could find and put it in DB. Every clue, every year, and where the medallion actually turned up. Once it's all sitting in one place a few things jump out.

The board is tiny. It has to be public property, it has to be somewhere a few hundred people can trample through without anybody getting sued, and you have to be able to walk to it. That's a couple dozen parks and locations, and the same handful of them keep coming up year after year.

The clues also lean on the same tricks over and over. Synonyms you're supposed to add together - H2O plus "an advantage" gets you EDGEWATER. Homophone puns, like deer and dear. Numbers that are really something else: "four" and "total nine" gives you 459 (not even accurate probably 🤣), which is Dirksen Park's acreage, and is also I-L-Y on a phone keypad. Local history name drops, which in Pekin means Everett Dirksen sooner or later. Once you have that trick list written down, a clue stops being a riddle and becomes a filter.

Filtering is the part the AI is good at, and it's the part we are bad at. A human reads Clue #4, gets a hunch, and then spends the rest of the week bending every new clue to fit the hunch. The model has no ego about it. Give it the clue text, the candidate list, and the history, and it'll score every location against every clue and hand back the one that fits the best. Dirksen Park was a contendor at Clue #4 and nothing after that moved it much except for a slight detour to the Dirksen statue, potentially.

Where it quits helping is the last hundred feet. It gets you a location (at least this year) and I was standing in the right corner of it on Sunday and walked off because everybody was hot and bored. The thing was still sitting there on Monday, haha.

I'm not sure there's a fix for this, either. Make the clues undecodable and the hunt turns into pure luck, which is worse. It's just that the search space is small enough for a excel file to hold all the data. At one point I was commenting that the prize should be bigger, but what if there was actually no prize at all? The prize is the hunt and journey and the friends we make along the way (lol).