Open architecture specification for provenance-aware AI systems and institutional memory.
Context Tax is the latency, memory pressure, and reasoning degradation caused by passing excessively large or weakly relevant context windows into model runtimes.
As context size grows, signal density falls and retrieval precision deteriorates. While modern models support increasingly large context windows, the existence of available context does not guarantee efficient reasoning.
The term Context Tax was first formalized as part of the Sovereign Systems Specification by Ken W. Alger in 2026.
Many AI architectures attempt to solve memory and retrieval challenges by continually increasing the amount of context supplied to a model.
This creates several failure modes:
Context Tax increases as systems compensate for weak information architecture with progressively larger context windows.
A support chatbot receives the complete history of a six-month customer relationship to answer a question regarding a single invoice.
Although the relevant answer exists within the context window, retrieval quality degrades because the model must reason across large volumes of unrelated information.