Grade C — exploratory
Small sample, single model, one-time observation, or incomplete evidence.
How it works
An answer is stochastic. One run is not a measurement. Everything below exists so a result can be traced back to the evidence that produced it.
Confidence is calculated, not asserted
Grade B — directional
Adequate platform and prompt coverage, with documented gaps.
Grade A — decision-grade
Multi-model, multi-run, geography-controlled, current evidence, human validation, reproducibility checks, and documented corrections.
An operator may lower a grade with rationale, and may not raise it above system eligibility.
Not built yetGrade eligibility is calculated from explicit conditions in slice 1C. Until then every displayed result is labelled Grade C by construction, not by calculation.Share of Answer is reported in parts
Inclusion rate, recommendation rate, citation rate, accuracy rate, and competitive-win rate are reported separately. They are never silently collapsed into one visibility number.
Engine coverage, as it actually stands
The sampling regime calls for three engines. One adapter is built. Stating the other two as live would be the exact overclaim this page exists to prevent, so here is the real state of each.
Answers diverge sharply between engines — a company present in one can be absent from another for the same question. Coverage is therefore a limitation on every result until all three run.
- Perplexity SonarAdapter built and proven against a real call: 20 citations parsed from one response on 2026-08-03, cost reported by the provider at $0.005670. That observation is what the sample assessment is built from. It runs in the Syntax workspace, and is not yet wired into this application.
- Not built yet
- Google GeminiGrounding metadata carries the citations. No adapter exists in any repository.
- Not built yet
- OpenAI ResponsesA proxy for the ChatGPT consumer answer, not a reproduction of it. No adapter exists in any repository.
- Not built yet
What we do not claim
No deterministic control over model outputs. No manufactured reviews, testimonials, analyst opinions, or awards. Findings are an observed association within a stated window and sample, never a causal promise.
Not built yetThe correction policy is published; the submission and adjudication surface lands in slice 1D.