Transparency
Citation Probe
Each Monday at 09:00 UTC, an automated probe asks 30 stable prompts to four production-tier LLMs (120 calls total) and records how many responses cite directory.kvantiq.studio.
Results are committed to this repository as JSON, summarized here, and available for download.
The probe is currently dormant. Runs will resume once paid API access is configured for all four model providers. Until then, this page documents the methodology and remains a transparency commitment for the future.
Methodology
Prompts: 30 prompts spanning 5 dimensions (geographic, modality, category, entity, discovery). The prompt set is version-locked at v1; any change requires a 4-week parallel run before retiring the old version.
Models: Claude Sonnet 4.6 (Anthropic), GPT-4o (OpenAI), Perplexity Sonar Large Online (web-grounded), Gemini 2.0 Pro (Google). 30 prompts × 4 models = 120 calls per run.
Scoring: 2 points if the response contains the URL directory.kvantiq.studio (any path); 1 point if it mentions "Kvantiq Studio" or "Kvantiq Directory" without a URL; 0 otherwise.
Source code: scripts/citation-probe.ts · .github/workflows/citation-probe.yml
Raw history: data/probes/history.json