FrontierChem

Strategic thesis

Can unresolved chemistry be represented as executable search problems with unknown optima but objective partial scoring, such that agents make measurable cumulative progress—and computational improvements reliably escalate toward experimental discovery?

Honesty bounds

• Alpha demonstration toward computational MVP—not experimental discovery
• Synthetic fixtures; highest grade remains surrogate
• Local runner is not a security sandbox
• No active secret evaluation families or production leaderboard

Leverage / money path

• Own the open challenge protocol before weaker leaderboard norms freeze
• Hosted evaluator + attested receipts as future paid credibility layer
• Metered Elicit evidence (x402/BYOK) outside scored sandboxes
• Lab partnership + sponsored verticals after surrogate graduation

Reusable learnings

• Submission is a solver program, not a molecule essay
• Never inflate evidence grade; keep regimes incomparable when needed
• Paper is the map; private repo is the warehouse
• Public Pages hub recovers depth without forcing git auth on every scan