Why replayability, not explainability, is the AI governance standard for agentic systems →
// The Decision Integrity Chain™ (DIC™)
The Decision Integrity Chain™ maps how an institutional decision travels from purpose to execution and back. It gives everyone accountable for the decision one shared map — so they can see where a decision sits, which layer is drifting, and which join has broken.
Institutions already govern the components. Model risk is covered by Basel and supervisory model-risk guidance; AI systems by the NIST AI Risk Management Framework and the EU AI Act; operational resilience by MAS, HKMA and Basel requirements; conduct by the FCA and its peers. Each governs a part. The Decision Integrity Chain™ governs what sits between them — whether the institution’s intent survived the journey from board mandate to executed decision, and whether anyone can still prove it did.
// The chain
Purpose → Strategy → Intent → Rules → Judgment → Decision → Outcome → Feedback
Each layer has an owner. Control is usually lost not inside a layer but at the joins between them — where intent, authority and accountability begin to separate.
What the institution exists to protect or achieve.
The choice about how to pursue that purpose.
The authorised outcome and its boundaries.
Intent translated into policy, thresholds and code.
Where a person, model or agent interprets information or exercises discretion.
The act itself — the decision that is actually made.
What follows from the decision.
Whether the chain learns — and returns to someone who can correct it.
// Who it's for
The Decision Integrity Chain™ — and the book When Decisions Break — is written for the people who answer for institutional decisions and the systems that now make them: boards, executives, risk and compliance leaders, technologists and engineers, and regulators. Each sees a different part of the chain. DIC™ gives them one language for the whole of it.
Real decisions do not move neatly from left to right. They loop back. An outcome may expose a rule that was wrong from the beginning. A judgment may reveal that the original intent was unclear. Feedback may force a change in strategy — effectively sending the whole chain into reverse.
Used this way, DIC™ locates the break. In one credit case, intent was never encoded into rules. A healthcare scheduling case breaks between outcome and feedback. A clinical scoring case breaks at judgment, because authority was delegated without a clear boundary. In another, purpose called for fair customer outcomes while the rules optimised margin — a contradiction that continued for twenty-two months before an inquiry exposed it. The cases differ; the map stays the same.
// Common questions
The Decision Integrity Chain™ (DIC™) maps the eight layers an institutional decision moves through: Purpose, Strategy, Intent, Rules, Judgment, Decision, Outcome and Feedback. It gives boards, risk teams and engineers one shared map to see where a decision sits, which layer is drifting and which join has broken.
Purpose (what the institution exists to protect or achieve), Strategy (how it chooses to do it), Intent (the authorised outcome and its boundaries), Rules (intent translated into policy, thresholds and code), Judgment (where a person, model or agent interprets or exercises discretion), Decision (the act itself), Outcome (what follows) and Feedback (whether the chain learns).
The Decision Integrity Chain™ was created by Deepak Aggarwal as part of Decision Engineering™, and is developed across his SSRN research and his book, When Decisions Break.
Every term used across this work — replayability, the Decision Integrity Chain™, the Fiduciary Gap™, decision drift — is defined once in the glossary.