From Automation to Agency: Observations from a Local-First, Conversational Personal Systems Lab

This project documents research on reinforcement memory for stateless AI systems, conducted within an independent personal systems lab. Current version: v0.3 (March 15, 2026) — "From Hypothesis to Observed Result" The Ayala Sigil protocol enables persistent, cross-session memory for AI systems using symbolic anchors, structured logging, and temperature-based lifecycle management — without modifying model weights. Version 0.3 reports empirical results from 20 days of continuous operation across eight distinct AI model families. Key findings: - Bootstrap confidence improved from 0.62 to 1.0 over 20 operational days - 8 model families adopted the logging protocol, including autonomous adoption (milestone x=4.4) - Local compaction engine (Qwen 2.5 3B) produced 167 capsule artifacts autonomously - No vector database, no embeddings, no cloud dependency — runs on SQLite FTS5 + MCP tools Root files: - Ayala_2026_Sigil_Protocol_v0.3.md — compiled paper with empirical data - Ayala_2025_From_Automation_to_Agency.pdf — original v0.1 paper (provenance) - README.md — project overview and citation Dashboard: https://wanatux.net/sigil/