Model endpoints
8 production model endpoints measured through vendor-hosted APIs, with public-weight availability disclosed only where model-specific.
Method
Every claim traces back to raw model evidence.
Clarivy separates product scope from measurement method. Endpoint coverage, prompt design, raw evidence, QA, and agent-ready files are documented separately.
Direct answer. Clarivy audits production model responses. Each finding must trace back to endpoint, prompt, sampling time, model data, response text, citations, and raw evidence.
Method pillars
Prompt scope
Brand, category, competitor, market, region, and buyer questions are scoped before measurement.
Raw evidence
Each response is indexed with model data, time, prompt, response, citations, and links.
AI-readable delivery
Markdown briefs use stable IDs for findings, evidence, and actions.
QA and limits
The audit measures API surfaces, not every product UI or future ranking outcome.
Agent use
Agents must verify current claims against raw JSON before changing customer systems.
Canonical method pages
Technical methodology
Endpoint matrix, freshness policy, data provenance, and confidence limits.
Agent integrations
Agent Bundle Manifest, adapter rules, and verification rules.
Model weight availability
How public-weight availability is disclosed without claiming API-model reproducibility.