Project restrictions

Project restrictions

These rules apply to all contributors and automated agents.

Non-negotiable

1. Deterministic — no randomness; same input ⇒ same output, always.
2. No stochastic Miller–Rabin — no random bases, no “probably prime” engines as the core.
3. No prime libraries as the implementation (e.g. primesieve, sympy.isprime as the engine).
4. Allowed: NumPy / Numba for speeding up our trial division and helpers.
5. Correctness model
   • $n < 2^{64}$: exact 30030-wheel trial division up to $\lfloor\sqrt{n}\rfloor$
   • larger $n$: small-factor trial, then AKS (may be slow for huge primes)

Why not “just use MR”?

Fixed witness Miller–Rabin is deterministic only on proven finite ranges (e.g. 64-bit with a known base set). That does not give a uniform finite-base proof for every natural number. This project optimizes under the stricter goal: deterministic for all $n$ in theory, with a fast engineered path for 64-bit inputs.

Agent files in the main repo

• .github/copilot-instructions.md
• .github/AGENT_BRIEFING.md
• CONTRIBUTING.md