🧠 L1.5 Signal Console
</> sourcegrounded · gemini-2.5
Machine intuition for healthcare agents — signals computed before the agent, fed in as labeled facts (never raw text in the context window).
L1 Truth (dbt/warehouse) → L1.25 Features → L1.5 Signals → L2 Agent reads labels → L3 Human decides
L1 · Truth (facts)
L1.5 · Signals — feature in → labeled signal out
L2 · Agent decision live
…
Truth says
Signal says
L3 · Human:
Signal methods — what powers each label (honest: only Cluster is trained ML)
A senior signal layer uses the simplest method that clears the bar — K-Means where clustering genuinely helps, statistical / rule-based / IR scoring where that's more reliable and auditable than a black box. Every signal is pre-computed, labeled, and evaluated.