Updated
Updated · O'Reilly Media · Nov 17
Decision Lineage Tracks 1 AI Finding Across 4 Systems to Stop Stale Decisions
Updated
Updated · O'Reilly Media · Nov 17

Decision Lineage Tracks 1 AI Finding Across 4 Systems to Stop Stale Decisions

2 articles · Updated · O'Reilly Media · Nov 17

Summary

  • A proposed “decision lineage” model would link an AI-generated assertion to every decision, report and retrieval artifact that reused it, so a later correction can trigger targeted review instead of leaving stale conclusions active.
  • In the supplier example, a certificate finding corrected at 11:00am still left a dashboard and program report showing the supplier cleared at 2:00pm because systems updated the source record but not the decisions already based on it.
  • The proposal separates an agent’s output from business authority: findings enter as proposals, gain approval only through a deterministic workflow, and carry a dependency record tying evidence, scope, dependent decisions and derived artifacts together.
  • That record would let systems recompute dashboards, regenerate reports, refresh search indexes and halt pending workflows automatically, while routing already executed decisions back to human owners rather than reversing them on their own.
  • The article argues the pattern is most valuable for high-consequence uses such as approvals, payments, safety and compliance, and suggests measuring it by decision coverage, time to impact and stale-evidence leakage.

Insights

If an AI error is corrected today, how many hidden business decisions are still secretly running on its initial mistake?
When a machine hallucination becomes enterprise ground truth, who is legally responsible for the downstream chaos it leaves behind?