BENCHMARKS

Measure first.
Rank second.

Chromatic Engines treats benchmark claims as reproducible evidence bundles rather than isolated timing screenshots.

METHODOLOGY

A benchmark needs its conditions.

Performance numbers are meaningful only when the comparison is reproducible and the outputs agree.

Engine	Public benchmark stance	Required context
Gold1	Prime-count benchmarking can be compared with primecount when the exact interval and validation agree.	CPU, threads, versions, L/H, rounds, count agreement
Red2	Prime-sieve measurements are reported only with workload and reference implementation context.	CPU, threads, limit, warm-up, reference, exact output
Blue3	Search throughput alone is not a primality proof.	Candidate rule, range, filters, primality verification path
Violet4	Research-model output is not treated as experimental physics validation.	Model version, assumptions, fixtures, numerical checks

EVIDENCE RULE

The public site does not automatically inherit every development benchmark as a commercial claim.

A faster wrong answer does not win.

Comparisons must use the same range or task definition and an appropriate reference.

CPU, operating system, compiler/runtime, threads and software versions are part of the result.

Logs and structured results should survive beyond the headline.