A product comparison is valid only within a declared scope. ARuntime does not calculate a universal score.
Required scope
- Runtime category and layer
- Product/runtime versions and verification date UTC
- Deployment target and operating model
- Model formats, models, hardware, and precision
- Workload and service objectives
- Security, governance, and data boundary
- Required capabilities and non-goals
- Authoritative source for each material field
Performance evidence
Require model and version, hardware/topology, runtime version/configuration, precision/quantization, input/output lengths, concurrency, batch behavior, cache state, warm/cold state, metrics, quality check, and measurement method. Do not compare vendor figures with independent tests as equivalent evidence.
Capability evidence
Use official documentation or repository/release material for product-specific claims. Distinguish built-in, plugin, integration, preview, experimental, and roadmap. A link to a marketing page does not establish operational semantics.
Excluded dimensions
State what the comparison does not cover—support quality, total operations cost, security review, regional availability, licensing interpretation, quality for other models, or future releases. An omitted field is not a negative score.
Publication rules
- No universal ranking.
- No unsupported compatibility or adoption claims.
- Show unknown and unverified values.
- Record review date and version assumptions.
- Provide correction route and change history.