causal-credit-review

Review causal evidence for agent memory prioritization by evaluating causal ledger data against an evidence-tier rubric to determine if association evidence warrants promote, demote, or reroute recommendations. The process applies confounder analysis and degraded-state constraints to ensure safe priority changes.

The evidence-tier rubric is a strict T0–T5 classification system that determines what priority actions are allowed based on sample size and confounder control. It gates recommendations so that only sufficiently controlled, verified evidence can trigger memory priority changes.

Validate confounder control by running a confounder checklist against your causal ledger evidence before any promote decision. The review process stops with specific gate codes if evidence is missing, unverified, underpowered, uncontrolled, or unquarantined.

No, causal inference decisions for memory rerouting require explicit JSON outputs from verified `ee causal ... --json` commands or a verified evidence bundle artifact. Direct database scraping or inference from raw transcripts is forbidden to prevent outsourced causal judgment.

Causal credit assignments are limited by evidence-tier rules, confounder control requirements, and degraded-state constraints. The review stops with specific gate codes when evidence is missing, unverified, underpowered, or unquarantined, preventing unsafe priority changes.

Yes, causal credit review handles redacted and degraded-state evidence by validating provenance status and applying specific safety controls. It checks redaction and provenance status as part of the confounder checklist before allowing any promote, demote, or reroute action.