Skip to main content
These recipes are the operator-facing companion to the scoring strategies and contact policies reference. They’re tested patterns — every one is exercised in the platform’s proof bundle and verified against live engine behavior. Each recipe has the same shape: what it solves → why you’d use it → the exact payloads → what to expect in the decision trace. The payloads are deliberately copy-pastable.

​
Available recipes

A/B test with holdout

Split traffic between champion and challenger; reserve a true holdout for causal uplift.

Per-category PRIE split

Route Cards candidates through one scoring profile, Loans through another — same flow.

Atomic channel coupling

Email digests where the hero placement and body placements stand or fall together.

Frequency caps

Cap per-day, per-week, per-category, per-channel — and verify with traces.

Fairness hard-gate

Block publish on disparate impact; configure thresholds + four-eyes override.

​
Reading the traces

Every recipe ends by pointing at the decision trace fields that prove the pattern worked:
  • candidateCount / afterQualification / afterSuppression / afterContactPolicy / finalCount — the funnel.
  • qualificationResults[] — which rules fired and on which candidate.
  • contactPolicyResults[] — which policies suppressed which candidate.
  • scoringResults[] — per-offer score, model type, and (when SHAP is on) per-feature contribution.
  • selectedOffers[] — final ranked + placement-assigned output.
  • channelCoupling[] — for atomic coupling: which channels cascaded.
Pull them via GET /api/v1/decision-traces?customerId={id}&limit=N or open the trace card in Studio → Decision Traces.