Fixing Drift: A Continuity Architecture That Makes LLMs Remember Across Sessions

This document presents the full technical specification and evaluation of CXS 
(Cross-Session eXtended Continuity System), a deterministic architecture designed 
to eliminate drift in long-running and multi-session LLM workflows. Developed across 
more than 2000 real-world interactions with GPT, Claude, Grok, Gemini, and local 
models, CXS solves two categories of drift:

1. Cross-session drift: definitions, constraints, plans, and decisions that mutate or 
   vanish after chat resets.
2. In-session drift: semantic shifts, assumption drift, weakened constraints, and 
   structural collapse during long conversations.

CXS introduces a structured Continuity Passport, strict-mode state derivation rules, 
digest-verified state restoration, anchor-based checkpoints, and a runtime drift 
monitor (CANARY) that stabilizes reasoning inside a session using a GREEN/YELLOW/RED 
health model.

This 50-page report includes:
• taxonomy of drift failure modes,
• empirical results across 2000+ sessions,
• cross-model continuity experiments,
• a full description of the Continuity Passport,
• strict-mode enforcement pipeline,
• digest canonicalization rules,
• the CANARY runtime protocol,
• the startup ritual and drift-probe questions,
• appendices documenting multi-model drift events and recovery procedures.

This Zenodo entry serves as the canonical long-form reference describing the full 
architecture. A short-form software paper describing the implementation has been 
submitted separately to the Journal of Open Source Software (JOSS).