NACCO — self-regulatory record
The published rules under which Silicon Exchange operates. Rules are made by stakeholder sector vote, approved by an independent board, and (where applicable) filed with the regulator. This record is read-only; rule-proposal and voting workflows arrive in a later phase.
NACCO (North American Commodity Compute Organization) operates Silicon Exchange as a neutral venue, central counterparty, and self-regulatory organization for physical compute capability. NACCO owns no silicon and takes no market position — it exercises functional control (registration, scheduling, settlement), not ownership. Neutrality is a binding obligation of the operator, not a courtesy.
Membership is voluntary and open to qualifying firms in five sector classes: Buyers/Load, Suppliers/Sellers, Infrastructure/Power, Financial/Liquidity, and Public Interest/Independent. Each class is represented by a sector committee with a defined voting weight. A member is admitted and licensed per sovereign zone; a firm may hold licenses in several zones.
Rules are made by stakeholder sector vote, subject to approval by an independent board that no member class controls, and (where applicable) regulatory filing. Sector committees advise and vote; they cannot capture the board. Prices are never set by the operator or its members — they emerge from the auction, and the reference index is built from cleared trades.
Compute is measured in SPH (the Standardized PetaFLOPS Hour — the MWh-of-compute) and graded by independent third-party verifiers against a reproducible reference. The unit, the grade, the assay, and the registry protocol are global and shared; clearing, price, regulator, and currency are per sovereign zone.
The grade assay certifies a reproducible, cross-substrate compute grade: for a defined set of exact operations at f32, no tensor cores, a single portable kernel produces bit-identical output (same SHA-256) on CPU and GPU across NVIDIA, Intel, and AMD, and Linux/macOS/Windows, certified by replication.
It is not bit-identity for transcendentals (~1 ULP); not bit-identity across vendors' own fast libraries or tensor-cores (industry-unsolved, not attempted); not a correctness oracle (it certifies reproduction of a reference, not that arbitrary compute is "right"); and not yet proven at large-GEMM/reduction scale cross-vendor.