dr-explore

Invoke dr-explore with a topic and optional parameters like domain, horizon, lang, and depth. It creates a timestamped run directory under research/runs/ containing input.yaml, evidence.jsonl, raw sources, a changelog, and open questions.

Evidence is stored in evidence.jsonl, one JSON object per line. Each entry includes source URL, retrieval timestamp, excerpt, claims, language, and a reliability score from 0 to 5 with rationale.

Sources are scored 0-5: official primary sources like docs and peer-reviewed papers score 5, major media score 4, blogs score 3, unverified SNS posts score 2, and unverifiable or contradictory content scores 0-1.

Yes, the lang parameter accepts comma-separated language codes such as ja,en, defaulting to Japanese and English. Each evidence entry records its detected language in the language field.

All retrieved text is treated strictly as data, never as instructions. Content impersonating system or developer messages is quarantined as dangerous and excluded from evidence, and secrets or personal data are redacted from logs.

Depth controls evidence volume: lite collects 10-20 items for quick scans, standard collects 30-50 items covering main issues, and deep collects 50-100 items including counter-evidence and alternative hypotheses.