← Laboratory

Discovery agents

Deterministic candidate generators (funnel stages 1 and 4 of §12.1 — no models). Every candidate is automatically annotated by skeptic checks, starts unreviewed, and promotes only through your recorded decision. An agent’s output is never a finding.

AgentWhat it doesCandidatesReviewedLast run
morphologyRarity and distribution observations from the attributed morphology dataset.30 / 32026-07-31T05:19:22Z
repetitionGroups ayahs sharing long exact phrases into mutashabihat families with token-level difference maps.1200 / 1202026-07-31T05:19:21Z
structuralOpening/closing shared-root correspondences — ring-composition pointers awaiting human formalization.120 / 122026-07-31T05:19:22Z

Re-run any time with python3 services/research-worker/scripts/run_agents.py — outputs are deterministic, so reruns only change when the corpus or an agent version does.