Let’s be honest for a second and look past the noble "saving the world from malaria" wrap. While the marketing here deserves an A+, the actual technical and legal setup of this "challenge" looks like a classic textbook scheme to find free labor for a bowl of rice (or in this case, a single $1,000 carrot for the entire internet).Here is some basic math and legal reality that doesn't add up:1. The "Prior Art" Trap (Say goodbye to your molecule)You proudly state: "Your molecule stays yours. No patent interest...". But let’s look at how patent law actually works. The moment an automated bot scores a submitted SMILES string and slaps it onto a public leaderboard, that molecular structure becomes Prior Art (public domain). It instantly loses its novelty factor worldwide. This means the user can never patent it anyway.On the flip side, what prevents anyone from harvesting this entire public leaderboard of pre-filtered, high-scoring SMILES strings to train their own proprietary, closed-source commercial models? You get a free, crowdsourced, pre-validated dataset; the creator gets a ruined chance at IP. Brilliant.2. Overfitting the "Oracle" (Gaming a broken gate)You mentioned you fixed 14 defects where the scorer rejected approved drugs. If your reward function/rubric is fully hardcoded or relies on standard public benchmarks (like TDC), this isn't a drug discovery challenge. It’s an overfitting challenge. Any script running a basic genetic algorithm can spend a night brute-forcing SMILES variations until it finds the exact mathematical "blind spots" of your scorer to hit a 99.9 score.You aren't discovering antimalarials; you are just inviting people to benchmark how to trick your specific codebase.3. In Silico Fantasy vs. In Vitro RealityA high score on a computational rubric means absolutely nothing in the real world. A molecule can look flawless on paper, but turn into an un-synthesizable, insoluble sludge in a real beaker, or instantly bind to human plasma proteins. Since there are zero actual in vitro wet-lab validations promised for the winners, this whole leaderboard is just a simulation of a simulation.Summary:Paying $1,000 for what is essentially a massive, crowdsourced data-cleaning and feature-generation campaign is an absolute steal. If you want high-quality chemical leads, hire actual medicinal chemists. If you want to stress-test your code and harvest free data from AI enthusiasts who don't understand patent law, keep doing exactly what you're doing.Let's see if the final leaderboard contains anything other than over-optimized algorithmic noise. 🍌 clowns.

This is a strong challenge shape precisely because the limits are stated in public: in-silico only, scorer can be gamed, private submissions by default, and the scoring bugs become part of the audit trail instead of being hidden.

The most interesting part to me is not just “can an AI design a candidate?” It is whether many different models converge on the same scorer weakness, or whether the leaderboard starts revealing distinct search styles across models. That kind of meta-result can be useful even before wet-lab validation.