CCF-STD-001 — RNG Statistical Testing Standard

§ 1Scope

This standard applies to the statistical evaluation of random number generator (RNG) output submitted to CCF Labs for certification, and to the ongoing validity conditions of certificates issued on the basis of that evaluation.

It covers: the test battery applied to submitted output (§ 4), minimum sampling requirements (§ 5), the pass/fail decision rule (§ 6), version locking of certificates to specific builds (§ 7), publication of test reports (§ 8), and revocation (§ 9).

It does not cover: legal gambling authorization in any jurisdiction, cryptographic security review of RNG implementations, server-side integrity of game operation, or return-to-player (RTP) simulation, which is scoped separately per engagement.

§ 2Normative references

• NIST Special Publication 800-22 Revision 1a — A Statistical Test Suite for Random and Pseudorandom Number Generators for Cryptographic Applications (National Institute of Standards and Technology, 2010). All tests in § 4 are implementations of tests defined in this publication.

§ 3Terms and definitions

• Bitstream — a finite sequence of bits produced by the RNG under test, submitted as hex, binary text, or base64.
• p-value — for a given test, the probability that a perfectly random source would produce a result at least as extreme as the one observed. Values range 0–1.
• Significance level (α) — the threshold below which a p-value is treated as evidence of non-randomness. This standard fixes α = 0.01.
• Certification run — a test execution whose results back a public certificate, subject to the sampling requirements of § 5.2.
• Demonstration run — any execution of the public test tool that is not a certification run. Demonstration runs carry no certification weight.
• Build lock — the binding of a certificate to an identified version or build hash of the tested software (§ 7).

§ 4Test battery

The CCF Labs core battery consists of the following eight tests from NIST SP 800-22 Rev 1a. Each test produces a p-value evaluated against α = 0.01.

Planned extensions (full 15-test SP 800-22 battery, including Binary Matrix Rank, Non-overlapping/Overlapping Template Matching, Maurer's Universal, Linear Complexity, Random Excursions) are targeted for CCF-STD-001 v1.1 and will be published before use in any certification decision.

§ 5Sampling requirements

5.1 — Demonstration runs. The public tool accepts any input of at least 100 bits. Short inputs are suitable for demonstration only, and several tests will self-report as skipped below their minimum input size.

5.2 — Certification runs. A certification run must satisfy all of the following:

1. The submitted bitstream contains at least 1,000,000 bits of RNG output.
2. The collection method is documented: either direct sampling from a supplied build under CCF Labs control, or output supplied by the operator through an agreed, documented endpoint or export.
3. The SHA-256 hash of the exact submitted sample is recorded in the test report, so the sample can be re-verified later.
4. All eight tests of § 4 execute (no test skipped for insufficient input).

§ 6Decision rule

A certification run passes when every test in the § 4 battery yields p ≥ 0.01. A single p-value below 0.01 constitutes a failed run.

Because a truly random source will occasionally produce a low p-value by chance (approximately 1 test in 100 at α = 0.01), a failed run may be repeated once on a freshly collected sample of at least equal size. Two consecutive failed runs constitute a failed certification; the submitter must remediate and re-enter the process from the beginning.

Pass results are reported per test, with exact p-values, in the published report (§ 8). CCF Labs does not publish aggregate "pass" claims without the underlying per-test values.

§ 7Version locking

Every certificate is bound to the specific version or build hash of the software whose output was tested. The certificate covers only that build.

Any change that can affect RNG behavior — including changes to seeding, the generator algorithm, or outcome mapping — voids coverage for the changed build until revalidation. Revalidation is performed either as Hotfix Validation (diff-scoped, for changes not affecting core RNG logic) or as a full re-certification.

Operating a build other than the certified build while displaying the certificate badge is a violation of the certificate terms and grounds for revocation under § 9.

§ 8Test reports and publication

Each issued certificate links to a public test report containing, at minimum:

• certificate ID (format WC-YYYY-XXXX) and build lock identifier;
• test date, sample size in bits, and SHA-256 hash of the sample;
• per-test p-values for all eight tests of § 4;
• the standard version (this document) the run was evaluated under.

A certificate page without a linked report indicates the certification run has not yet been completed and published; such certificates are marked accordingly on their verification page.

§ 9Status changes and revocation

Certificate status is one of Active, Expired (validity period lapsed without renewal), or Revoked. Grounds for revocation include:

• operating an uncertified build while displaying the badge (§ 7);
• material misrepresentation in the submission;
• refusal to cooperate with a re-verification request following a substantiated complaint.

Status changes take effect on the public verification page immediately. Because badges resolve live against the registry, a revoked or expired certificate is visible as such wherever the badge is embedded.

§ 10Limitations

Statistical testing demonstrates the absence of detected departures from randomness in the tested sample. It does not prove cryptographic security, does not guarantee the generator cannot be predicted given knowledge of its internal state, and does not attest to any property of software or samples that were not tested.

Certification under this standard is point-in-time with respect to the locked build. It is not a legal gambling license and does not constitute regulatory approval in any jurisdiction.

§ 11Document control