Context Drift
Context Drift describes the loss, staleness, or misweighting of project context as agent work moves across sessions, tools, and summaries.
Context Drift is the loss of fit between the context an agent uses and the actual state of the work. It can happen when summaries compress away reasoning, old rules remain active after requirements change, abandoned paths re-enter the task, or important constraints are present but not weighted correctly. Context Drift is not simply a shortage of tokens. It is a failure to keep context valid, scoped, and connected to current lifecycle state.
The problem it names
The problem it names is that agent systems often treat context as a pile of relevant text. Real project context is more structured than that. It includes decisions, rejected options, active constraints, role boundaries, evidence expectations, unresolved risks, and acceptance criteria. When those pieces are flattened into notes or summaries, the agent may read more material while understanding less about what still matters now.
Why existing approaches are not enough
Retrieval, memory, and long-context models help agents access more information, but access is not the same as authority. A retrieved note may be outdated. A rule file may conflict with a newer decision. A summary may preserve a conclusion while losing the trade-off that made the conclusion valid. Without lifecycle semantics, the system has no reliable way to distinguish current context from stale context, background context, or nonbinding history.
How it relates to AI Agent Lifecycle
AI Agent Lifecycle treats context as a stateful layer that must be loaded, updated, invalidated, and evidenced as work moves forward. Context Drift is closely tied to Intent Drift: once context stops representing the real project state, intent becomes easier to reinterpret incorrectly. Cognitive OS is the runtime path for this concern, while Evidence Chain and MPLP show why context needs traceable links to plans, confirmations, and outcomes.
White paper source trace
Context Drift is adjacent to GAIC through lifecycle continuity and evidence integrity; R3K-0 did not assign a direct chapter/table/MRO anchor for this route.
The concept supports MRO and ALCS reasoning by naming how active work context can detach from accepted responsibility.
A workflow can keep running while its active assumptions no longer match the original intent, constraints, or review state.
This source trace is author-analytical. It is not legal advice, certification, legal compliance proof, regulator approval, vendor ranking, procurement guidance, or a claim that MPLP is required.
Evidence route
The evidence route for Context Drift starts in the origin essay's discussion of summaries, rule files, and stale project state. It continues through Cognitive OS as the runtime path: the place where context has to become active state, not background reading. Validation Lab matters because context validity eventually has to be checked against evidence, especially when the same work resumes across sessions or agents.