Geotemporal Hydrodynamics lands Lean 4 proof

The real story is the workflow: use the model for search and scaffolding, then let the compiler force discipline. That is exactly the pattern formal-methods teams want from AI-assisted theorem proving, even if the underlying physics claims are highly unconventional.

• –The repo presents a full Lean 4 + mathlib4 codebase with CI and verification badges, which is more concrete than a plain text AI demo
• –Compiler-driven iteration is the valuable bit here: errors become targeted feedback for the model instead of hidden hallucinations
• –The physics narrative is extravagant, so treat the substantive signal as tooling and process, not scientific consensus
• –For AI developers, this is a preview of how “no-sorry” proof artifacts can become a benchmarkable output format for agentic systems
• –The strongest use case is formal methods, not physics: theorem proving, refactoring proofs, and closing gaps under strict type checking