Technical Report LUCCA-TR-2026-03 • November 2026

Closing the Epistemic Loop Across the Five Layers of Agent Memory: Cryptographic Provenance, Paraconsistent Bilattices, and Z3-Quarantined Procedural Induction

Sean Maxwell — Roll Solo, Melbourne, VIC, Australia (ORCID: 0009-0004-0133-0463)

Download Full Paper (PDF) DOI: 10.5281/zenodo.23901683 GitHub Repository

Abstract

Contemporary agent architectures commonly decompose persistent state into five layers: Working Memory, Episodic Memory, Semantic Memory, Procedural Memory, and an offline Meta-Memory control plane. However, conventional implementations connect these layers via unconstrained asynchronous extraction and reflection jobs that read raw episodic transcripts and promote them directly into semantic facts and procedural skills. We present the unified five-layer memory and skill-induction architecture of Project Lucca and report empirical measurements across every layer boundary on local consumer silicon. Six HKDF-derived HMAC-SHA256 domain signers and SQLite triggers separate autobiographical recall from factual grounding, reducing multi-session self-poisoning from 100.0% in Naive RAG to 0.0% across four Gemma-4 parameter scales while preserving 100.0% autobiographical continuity. Within Semantic Memory, a zero-VRAM OpenVINO MiniLM-L6 cross-encoder evaluates semantic conflicts in 12.23 ms mean CPU latency, routing contradictions into Belnap's four-valued paraconsistent state paired with OpenNARS Beta evidence weights, lifting LongMemEval accuracy from 23.3% to 70.0% at 8.1 ms mean retrieval latency, while Z3 SMT quarantine verifies induced MicroScheme procedural rules prior to activation.

Citation (BibTeX)

@techreport{maxwell2026fivelayer,
  author    = {Maxwell, Sean},
  title     = {Closing the Epistemic Loop Across the Five Layers of Agent Memory: Cryptographic Provenance, Paraconsistent Bilattices, and Z3-Quarantined Procedural Induction},
  institution = {Roll Solo},
  number    = {LUCCA-TR-2026-03},
  year      = {2026},
  doi       = {10.5281/zenodo.23901683},
  url       = {https://doi.org/10.5281/zenodo.23901683}
}