Reproducibility requirements
Workload and environment, software versions, configurations, repeated runs, sample definition, raw-data provenance and collection date.
Benchmarks
Measurements you can inspect before you use them.
Benchmark results only make sense with the workload, environment, collection method and limitations attached.
Published benchmarks
No verified APPNEURAL benchmark results are published.
Previous draft metrics and sample counts have been removed from public discovery because no reproducible source evidence was supplied.
Interpretation requirements
Units, axes, comparison fairness, uncertainty, tested scope, limitations and a distinction between observation and causal inference.
Related
Read architecture guidance instead.
Illustrative patterns can help frame a design conversation, but are not benchmark results or production proof.
Enterprise Intelligence
Need a benchmark for your workload?
A useful comparison begins with a defined task, environment, baseline and acceptance method.
Scope and availability confirmed individually · Written terms define the engagement