Software Developer, Digital Artist, and Builder of Things That Learn

I build client-side autonomous entities, digital grimoires, and interactive simulations that run entirely in browser-based and device-native environments.

🌐 Experience My Active Projects: ardorlyceum.itch.io

An immersive narrative simulation that models human consciousness, memory, and reality as a command-line operating system.

• Interactive Terminal: Run BIOS of Being on Itch.io
• The Consciousness Operating System Manual: A 100-page privileged manual (DLC) featuring kernel decryption keys and archetype installation codes. Archived in the San Diego Central Library's permanent collection.
• Master Registry: Lyric Database: Database Uplink
• BIOS_OS: The Sonification Cycle: Listen to the 24 Tracks
• Keygentia AI Taxonomy Engine: keygentia.netlify.app

A sci-fi living illustration published on Steam. Not a game. A microscope interface you actually look through, finding patterns and shapes that reveal the lifeforms that live on the surface of your skin.

• Steam Store: Integument on Steam
• DLC Expansion: Integument - Database Gates

An experimental browser-based digital art piece exploring how an artificial entity learns and visualizes its own mind.

• Live App & Devlog: Run SUKOSHI on Itch.io

v7 - Self-Directed Capability Development

ARMINTA is a Python-based autonomous causal discovery agent running continuously on Linux. It does not passively monitor the OS. It actively intervenes, measures outcomes, and builds a grounded causal model of your specific hardware from scratch. Every edge in the model is earned through real actions and empirical observation.

Key v7 Features:

• Wish Pipeline (W1-W4) - self-directed capability development triggered during SELF_ASSESS. W1 detects causal dead zones and situation gaps in the learned world model. W2 searches for procurement candidates from existing system utilities and the action registry. W3 runs shadow staging for ~2000 steps with no execution - observe only, with routing validation gates. W4 grades over 5000 steps and returns WIN/TIE/LOSE verdicts. W4b generates implementation code from her own source via AST on WIN, backs up the current source, and appends staged actions for human review. 24 wins to date, 17 active wishes.
• Full v6 foundation retained: HobbyCore - voluntary external engagement layer. Fires during DREAM cycles when emotional state is receptive. Samples four probe domains (public network latency, local hardware sensors via sysfs, system load index, and solar/daylight context) using intensity-weighted domain interest. 923 total fetches, 72 novel observable edges discovered.
• EarlyOOM Observation Node - earlyoom_ct as an observational-only SCM metric. All action -> earlyoom_ct causal edges are poison-listed at write time. The agent learns preconditions that precede OOM kills.
• Circadian Memory Look-Ahead - _check_circadian_memory() fires compact_memory during predicted idle lulls before historically high-RAM hours. Log prefix [CIRC-MEM].
• Full v5 foundation retained: PriorityShift (focus-aware dynamic process priority, RL-learned nice delta), SelfTuner + ActionProposer + SandboxRunner (self-expanding action space), zRAM-aware memory management, battery-aware action gating, and the complete v4 cognitive hierarchy (Temporal Causal Graph, DDQN CMC, MosaicCore, LexicalCore, WebLearner, SomaticConfidenceModel, etc.).

Live Stats (pushed directly from the running agent): Live Agent Dashboard

Full architecture, cognitive hierarchy (updated Mermaid diagram), version lineage, and detailed documentation are in the repo.

Status: Active development at v7.

• NoSight
• Drawmation

Maintainer: Jason German (mematron)