Thesis
Synix studies how persistent systems adapt over time toward complex goals while maintaining coherence.
The problem
Section titled “The problem”The central question is not just how an agent retrieves the right fact. It is how a system that persists across time organizes information, adapts under changing conditions, learns from experience, and avoids drifting into contradiction, amnesia, or obsolete state.
If a system must operate over long periods, pursue complex goals, revise itself, and coordinate with others, then the organization of information becomes part of the problem — not an implementation detail around an otherwise complete intelligence.
This becomes harder at the collective level. Many important forms of intelligence are structured collectives, not isolated individuals.
Coherence
Section titled “Coherence”Working definition:
- The system learns from experience
- It does not repeat its own mistakes
- It preserves structured, lossy, but recoverable continuity with its own past
- It remains understandable over time as a single evolving system
Contradictions may exist. The goal is not to eliminate them, but to keep the system aligned enough that adaptation does not become fragmentation.
Working hypotheses
Section titled “Working hypotheses”The full invariant set is still under study. This is an active search, not an application of finished theory.
The leading candidates are three structural properties that recur across relevant levels of persistent systems:
- Compression — not everything can remain explicit
- Division — different kinds of information cannot all be governed identically
- Renewal — persistent structures degrade and must sometimes be revised or rebuilt
These are developed formally in the CDR framework.
Synix is not limited to agent memory, RAG, workflow orchestration, or chat memory. Those are application surfaces. The broader question is what coherent long-lived adaptation requires architecturally.
Current stage
Section titled “Current stage”The program is in a research stage. The build system is shipped software that demonstrated some of these ideas. The memory architecture is in design. The Synixolis protocol is a draft specification. The constructive aim is a minimal, implementation-agnostic protocol and reference implementation — but simulation and evaluation come first.