I designed Forecast from one premise: a market is an inference machine, and capital is its compute. The objective is information discovery per unit of capital-time: how much information a dollar can carry, and for how long. This one objective contains the others. The venue that maximizes it gives traders the best execution, gives each dollar the most leverage and reuse, and pays the pool behind them the most yield for the least risk. They rise together. That is the flywheel. Perps make traders spend that capital on defense before any liquidation fires. Traders idle margin, cut size, shorten horizons, and drop views that may be right at the horizon but can’t survive the path. When liquidation fires, it destroys information twice. It deletes a funded view before its horizon. Then it turns the view into a forced order. The order says nothing about btc. An account ran out of collateral, so the venue closed it. Every open position is then marked against a price that order moved. In a cascade the loop closes: forced orders move the price, the price forces more orders, and for those minutes the market’s estimate of bitcoin is an estimate of its own margin engine. Forecast removes the obligation. A position is a bounded claim, bought outright. Maximum loss and the price of capital-time are fixed at creation. The claim can trade near zero and recover. It cannot create debt, so the venue never stops, seizes, or socializes anything. Options also bound loss. But an option has a writer, and a writer can be margined, squeezed, and blown out. Forecast has no writer. The far side of every claim is one pooled reserve that stays funded in every reachable state. A trader who pays for Friday keeps Friday. Whoever discovers the most information per unit of capital-time takes the informed flow, then the depth, then the price itself. I’m building Forecast to be that market.

We designed the Forecast kernel to be both immutable and formally provable end to end. The proof stack now contains 383 machine-checked Lean theorems. The gate rejects placeholders and any proof that relies on more than two standard axioms. A second layer verifies the exact bytecode that Base will execute in KEVM so the proof does not stop at source code and trust the compiler. Every run must also catch a forbidden storage write that we deliberately inject. Miss it, and the run is invalid. This guards against formal verification’s most dangerous failure mode: a valid proof of the wrong claim. Medusa completed 5m calls against nine safety properties this morning. Echidna completed its 5m by noon. Neither found a violation. Both are coverage gated: if they miss a path we require, the run fails. Then we attacked the test suite itself. We generated 1,813 mutant contracts, each containing one deliberate defect, and ran the full suite against every one. The suite killed 1,707. We reviewed the survivors by hand and recorded why each survived. The campaign is now running again on the machine pictured, 540 mutants in. Everything closes into one verification ledger tied to one exact commit. That is the commit we launch.

This week’s 🧭 spotlight is @forecastfdn by @joeyroth Forecast adds leverage to prediction markets, giving traders a new way to express conviction on outcomes It’s built to make prediction markets more capital-efficient and more dynamic for active traders ♾️