Knowledge Base

How evidence-backed recommendation rules work

Updated August 5, 2026

Recommended Next Actions are created by deterministic rules, not by invented report copy. Each action names the source pack, rule key, rule version, owner, category, severity, benefit, and the exact measured evidence that matched.

What users see

Open a saved report and find Recommended Next Actions. Rule and Evidence identifies the versioned rule and values that caused the recommendation. A missing measurement does not become zero and cannot trigger a numeric rule.

What administrators manage

Administrators open A&B Platform > ERP Rules. A source pack stores the approved document text, source name, source version, URL when available, and a server-calculated SHA-256 hash. Rules are saved as append-only versions with a trigger, evidence paths, category, severity, action, reason, owner, and benefit.

  1. Create a draft source pack from the complete approved source.
  2. Add draft rules using allowlisted JSON conditions.
  3. Back-test the pack against saved report packages.
  4. Approve each validated rule.
  5. Approve the pack to make it primary.
  6. Retire a rule or pack without deleting its history.

ERP 3.0 source gate

The A&B Measured Evidence Baseline keeps current reports useful but is not represented as ERP 3.0. ERP 3.0 becomes active only after the authoritative document is imported, its rules are reviewed and back-tested, and an administrator explicitly approves the pack. This prevents draft or guessed rules from reaching client reports.

Security and history

Only administrators can create, approve, retire, or back-test packs and rules. Every mutation requires a WordPress nonce and is written to the existing activity log. Rule conditions use allowlisted operators and bounded data paths; they never execute PHP, SQL, JavaScript, or uploaded code.