Operational Drift

Definition

Operational drift is the gradual divergence of an engineering system’s actual state from its governed operational intent. Unlike context drift, which occurs within a single AI-assisted session, operational drift accumulates across many sessions, many decisions, and many engineering contexts over time.

A system is in a state of operational drift when its implementation no longer reflects the governing decisions, architectural constraints, and operational standards that were established for it — even if those decisions and standards were never formally rescinded.

How Operational Drift Accumulates

Operational drift is rarely caused by a single decision. It accumulates through a pattern of individually small divergences that compound over time:

• A dependency is updated outside the approved version constraint, “just for this sprint”
• An architectural boundary is crossed in one service, then normalized across others
• A governance-required review is skipped under time pressure, once, then routinely
• A new engineer joins and makes decisions based on the current implementation, not the original governed intent

Each individual divergence may be minor. The compound effect is a system whose actual operational state has drifted significantly from its governed intent — with no record of when, how, or why the divergence occurred.

The Traceability Problem

What makes operational drift particularly damaging is the absence of a traceable record. If each divergence had been documented — captured as an append-only record in the operational corpus — the drift would be visible. Engineers could review the lineage, understand the history, and make informed decisions about whether to accept or correct each divergence.

Without an append-only corpus, the divergences are invisible. The system is in an unknown state relative to its original governed intent.

Prevention vs. Detection

Operational drift can be managed at two points:

Prevention: governance-first workflows with active Rules and Roles constrain the scope within which drift can occur during any given session.

Detection: convergence validation at release gates surfaces accumulated drift before it reaches production.

Yanzi addresses both. The append-only corpus makes accumulated drift visible. Certification checks confirm that drift has not exceeded governance boundaries at release time.