brianletort.ai
Back to Research
Preprint

Paged Context Memory: Runtime Systems for Evidence Blocks, Locality, Speculation, Linking, and Policy Preservation

D. Brian Letort, Ph.D.

Digital Realty

April 12, 2026

DOI: 10.5281/zenodo.19546800

Program hub covering one foundational theory paper followed by a three-paper trilogy on IR, runtime lifecycle, and precision-aware optimization.

Summary

Paper 2 extends the trilogy from semantics into runtime behavior. It argues that compiled context should live like managed memory: evidence blocks enter working sets, locality governs movement, speculative execution prepares likely futures, context GC removes dead state, and provenance remains attached through lifecycle transitions.

Why This Matters

Even a strong IR is not enough if compiled context decays once it enters execution. Enterprise systems need locality, lifecycle management, provenance, and policy preservation at runtime so context remains legible, efficient, and governable after compilation.

Key Contributions

  1. A runtime model based on evidence blocks, working sets, and locality-aware movement
  2. Paging-style discipline for hot, warm, and cold context transitions
  3. Speculation, branch prediction, and linker semantics for context assembly
  4. Context garbage collection and provenance as a causality graph
  5. Policy-preserving runtime transitions that keep governance attached to context state

Who Should Read This

  • Architects designing memory layers for agent or RAG systems
  • Engineers concerned with provenance, lifecycle, and runtime cost control
  • Researchers exploring locality, speculation, and semantic memory models
  • Operators who need context systems to stay explainable after execution begins

Related Writing

What This Points To Next

  • Runtime schedulers for locality-aware context promotion and eviction
  • Speculative prefetch strategies for agent workflows and long-running sessions
  • Reference implementations of context GC and linker passes in MemoryOS
  • Operational metrics for policy-preserving runtime transitions