Why now?

Compiling logic directly to polynomial semantics removes the arithmetization layer that dominates proving cost in every other proving system. That was the reason to build a new language for this purpose, and it is also why the result cannot be lifted into an existing stack: a zkVM cannot delete the machine it emulates and remain one.

The consequence is measured and published. On the same machine, the same fib(10,000) claim, and the same fast-doubling algorithm on every system, the zkFOL prover completes in 11 milliseconds against 3.7 to 49 seconds across RISC Zero 3.0.5 and SP1 6.3.1's proof modes, and verifies in under 4 milliseconds. On a structural workload, sudoku validity, the gap holds and the statement proved is stronger: zkFOL shows a valid grid exists while disclosing none of it, where the comparator programs publish the grid they check. The full tables, the code every system ran, and every caveat are on the performance page, reproducible from the public benchmark record.

The axis that matters throughout this documentation is the prover envelope: proving time, memory, and the hardware proving runs on. The prover fits on ordinary hardware, so ordinary participation needs neither specialist machines nor a paid market in proofs.

The same machinery carries privacy natively. Every transaction already carries a proof, so hiding the witness while proving the rule was followed is what the system does by default, and private state with public verifiability is available without a separate privacy layer.