Classifier

The classifier is the reputation model that scores every send before delivery. It runs between your agent.send() call and the actual delivery, reading signal families that capture how the send behaves so the system can protect deliverability without you maintaining the rules.

The four signal families

• Volume curve.Ramp-up rate, time-of-day distribution, batch size, inter-send gap. Sudden spikes against an agent’s normal pattern raise the score.
• Engagement curves. Open rate, reply rate, complaint rate, bounce rate, unsubscribe rate. Post-send signals re-score future sends.
• Content patterns. Merge-tag density, link density, attachment patterns, and other signals correlated with complaint risk.
• Per-agent baseline. Each agent develops a behavioral signature over its first ~200 sends. Deviation from baseline raises the reputation-risk score.

Transparency

Every send response includes classifier_score(0–1, where higher means higher reputation/complaint risk) and top_signals— the highest-weighted features that drove the score. You can see exactly why a send was scored the way it was, and what the sending layer did with it (suppression-at-send, complaint auto-suspend).

If you disagree with a score, request manual review from the dashboard. Reviews complete within an hour during business hours and update the per-agent baseline so future sends score correctly.

See the architecture page for how scoring feeds the sending layer, or the sender reputation glossary entry for how per-agent reputation is built.