I don't have a good guide for you, but this is the lib you use to create custom React renderers:

https://www.npmjs.com/package/react-reconciler

ever heard of Wordle?

You mean that one game that actually managed to pull off the game-a-day scheme four years ago through clever design and simple gameplay? The game with the inherently player-hostile scheme that's still inspiring thousands of talentless hacks on the internet? The actual successful game that attracted millions of copycat losers? Nope, never heard of it.

Just wait until I tell you about your local newspaper’s Sunday crossword.

I built this game for fun. There are no ads, and I currently cover the LLM costs myself.

I’m not sure why you’re being so hostile toward the idea of a daily puzzle game. Wouldn’t it make more sense, ideologically, to direct your criticism at pay-to-win games designed to keep you playing for hours a day? This is a simple daily puzzle — a two-minute break that never asks for more of your time. How is that even remotely comparable?

You showed your game and asked for feedback. I gave you honest feedback: I won't play your "daily" game, because I think the "daily" aspect is an unnecessarily enforced, inherently player-hostile, mindlessly imitated anti-feature that's of no use to anyone.

Congratulations on making the game, I know it's hard to finish and release something. I wish you luck.

Thanks glad you like it!

But to answer your question - yes the string must contain only valid english letters and must be between 2-25 characters in length. If you can jailbreak the system with those constraints I salute you!

Others have tried and failed to jailbreak it. Give it a shot if you're keen and post results here.

It's on my list for sure. Thanks for the message!

Thank you! The solution is when user_word === ai_word, so there isn't technically an "answer." The only pre-determined word in the game is the first word the AI chooses, which is the same word for all players, each day.

This is called turn detection, and there are some great tools coming out to solve this recently. (One user mentioned Livekit’s turn detection model). I think in a years time we will see dramatic improvement.

If the turn detection model is small, could you run it on the edge and have like 10-50ms "shut the hell up" latency? That'd be nice.

God tier filtering. Do you mind sharing how you integrated AI into the filter system? Your "flagging peanut butter" example also makes me wonder if the LLM is tagging the product with a large number of attributes on each run so it's not prohibitively expensive.

Yea quite good! Here's mine:

Converge No. 11 (7/8)

peanut / *** shell / ascent jacket / climb parka / hike mountain / trail peak / path summit