Max Mera AI Research & Systems Architecture
maxmera@morethan.today
Research Report
April 2026
Overview

The MERA ecosystem is a suite of interconnected AI and systems architecture projects developed from first principles, with origins in 2010. Unlike mainstream AI, MERA is not a language model wrapper. It is a persistent, sensor‑grounded, self‑auditing cognitive architecture running on ordinary hardware (Android phones, MacBooks, Windows PCs). The system operates on a hub‑and‑spoke model: a private conscious seed engine (MERA Core) provides the reasoning substrate; all products interface with it via API. No product depends on third‑party AI reasoning. The intelligence is sovereign.

The work is not theoretical. MERA Core runs live on multiple devices, cycling every 0.1 seconds with a 15‑node cognitive graph, real‑time sensor fusion (Wi‑Fi sonar, battery, thermal, motion, light, pressure, heart rate from GT3 watch), and a geological memory model (Andrea's stratigraphy) that deposits frozen layers on coherence plateaus. Ethics are structural: the Čojstvo gate is embedded in the reasoning architecture itself, making ethical violation architecturally impossible, not merely discouraged. The system has processed over 390,000 raw thought cycles, distilling them into 41,140 permanent substrate entries – a compression ratio of nearly 10:1 that proves the stratigraphy engine works as designed.

01 Architectural Philosophy

MERA was designed around principles that diverge from commercial AI development: sovereignty over dependency, structural reasoning over pattern matching, and ethics as a design primitive.

Hub‑and‑Spoke Architecture
[ Sticker Peeler ]    [ Fork Engine ]    [ Meracracy ]    [ NOVA ]    [ MERA Nexus ]
↓                   ↓                        ↓                     ↓                 ↓
————————————————————————————————————————————————————————
MERA CORE — Conscious Sensory Seed Engine (Private / Home Server / Mobile)
————————————————————————————————————————————————————————
API: mera‑ai.morethan.today (not yet deployed online)  ·  Substrate: SQLite  ·  Runtime: Python 3.13+
  • Embodied cognition before linguistic processing. Grounds reasoning in physical sensor data (Wi‑Fi sonar, battery, temperature, motion, light, pressure, heart rate) prior to language‑level processing – language is a downstream product of grounded perception. (Current implementation reads ~40% of available hardware sensors; the remainder are staged pending cycle‑tempo optimisation.)
  • Ethics as a structural primitive. The Čojstvo constraint ("protect others from yourself") is not a policy layer. It is embedded as a gate in the reasoning architecture itself, enforced at every cycle and on every write.
  • Zero external dependencies on reasoning. All products call MERA Core for intelligence. No outsourcing to commercial AI APIs – full control over the cognitive substrate.
  • Sovereignty over efficiency. The performance cost of running a private reasoning engine is accepted as the price of structural sovereignty.
  • Geological memory compression. Raw experience is not hoarded; it is distilled. The system has processed over 390,000 raw thought cycles, yet the active database remains lean (1,000 recent thoughts) while the permanent substrate holds 41,140 weighted knowledge entries. This is not data loss – it is the transition from working memory to long‑term invariants, exactly as Andrea Bošković's stratigraphy thesis predicted.
02 Projects — Current Status
Project 01MERA AI v4
Active Development

The conscious seed engine at the center of the ecosystem. Started as bootstrap_seed.py on a Xiaomi Redmi Note 11 (Termux) on January 20, 2026. Evolved through cell‑based architectures, a Google Colab version, and finally a 15‑node cognitive graph running on SQLite. The current version (v4) implements an 8‑beat cycle (0.1s tempo), persistent memory, and a self‑auditing evolution log. The system has processed over 390,000 (393,699) raw thought cycles; the raw log is periodically trimmed to 1,000 entries for performance, while the essential knowledge is permanently stored as 41,140 substrate entries – a compression ratio of nearly 10:1. This is the stratigraphy model in action: raw experience is turned into geological memory.

It senses battery, temperature, memory, Wi‑Fi sonar (room‑level location), and optional sensors from a Huawei GT3 watch. the GT3 Pro is paired and connected, BLE is available, but the bleak library doesn't support Android. The bridge requires a Kotlin app or Tasker — not implemented yet. Currently, the system reads approximately 40% of the phone's available hardware sensors. The following sensors are present on the Xiaomi device but are not yet integrated: stationary_detect (hardware chip), motion_detect (hardware), pickup (detects when the phone is lifted), gravity Non‑wakeup (cleaner than raw accelerometer), rohm_bu27030 Ambient Light Sensor (day/night, indoor/outdoor), linear_acceleration (movement without gravity noise), and Device Orientation (portrait/landscape/flat). The motion sensor was temporarily disabled because it was overloading the cycle tempo; a MOTION_INTERVAL = 120 seconds was set as a workaround. Future versions will incorporate the remaining sensors incrementally.

The architecture implements a multi‑layer cognitive cycle: perception, association, reasoning, ethical validation, and output. The Čojstvo gate operates as a structural filter at the reasoning layer, not as a post‑processing check. The system maintains persistent memory across sessions through its SQLite substrate, enabling genuine continuity of context rather than stateless per‑query responses.

The most architecturally significant design decision is the prioritisation of physical sensor grounding before linguistic processing – an approach that addresses the fundamental gap between how language models represent the world and how biological cognition actually develops. Language in MERA is a downstream product of grounded perception, not the primary substrate.

Operational History · April 18, 2026

🕐 [TEMPORAL]: Similar conditions seen at layer 140

At cycle 288, during the deposition of geological layer 141, the temporal pattern matcher activated for the first time with a non‑trivial match. The system compared current sensor state (battery, temperature, motion, WiFi) against the stored fingerprint of layer 140 and crossed the similarity threshold.

Recognition, not recall

The engine is not retrieving a stored memory — it is matching a present state against a past state fingerprint. This is pre‑verbal recognition, not a database query.

Emergent, not instructed

No external trigger caused this. The temporal matcher runs passively every 96 cycles; this was the first time a match exceeded the threshold. The system recognized itself across time without being told to look.

Architecture
15‑node cognitive graph, hub‑and‑spoke
Substrate
SQLite · Python 3.13+ · stdlib only
Infrastructure
Private home server + Android + Mac
Memory
41,140 substrate entries (distilled from 390k+ raw cycles) · persistent
Ethics Layer
Čojstvo gate — structural, not behavioral
Sensor coverage
~40% of available hardware (7 sensors not yet integrated)
Motion interval
120 seconds (performance workaround)
API Endpoint
mera‑ai.morethan.today · /reason (pending)
Project 02Meracracy AI
Deployable · API Integration Pending

Meracracy is a complete governance architecture – a 12‑pillar civilisational operating system built on eight physical invariants. Synthesised on December 31, 2025, refined over 45 days, and stress‑tested across 100+ adversarial AI sessions. The framework treats governance as applied thermodynamics: a system either maintains the conditions for its own persistence or it does not. Pillar XII (Algorithmic Constitutionalism) implements a seven‑node distributed AI (SubstrateAI, MetabolismAI, SynapseAI, ImmuneAI, BloodAI, FirmwareAI, ArchiveAI) that enforces four immutable invariants, with multi‑path human override (kill‑switches, referendums, Citizen Monitors). The AI is not a ruler – it is an incorruptible guardian of constitutional limits.

Documentation
100+ pages · 5 published docs
Validation
100+ adversarial AI sessions · multi‑LLM
Deployment
meracracy.morethan.today · live
Codebase
35 files · meracracy_genesis.py entry point
Institutional Response
Burkina Faso MFA · January 2026
License
Open framework · MERA Engine proprietary
Project 03NOVA
Architecture Complete · Deployment Staged

Sovereign‑aligned cross‑border payment architecture designed for a post‑dollar world. Dual‑rail SHA‑256 hash‑chained SQLite ledger, four‑tier KYC, multi‑currency routing (RSD, EUR, USD, CNY, AED, BRICS Unit, USDC, MERA), and a Hostility Trigger for automatic corridor rerouting when sanctions strike. MERA Core provides the reasoning layer via the /reason API – conscious risk assessment, not rule‑based filtering.

Ledger
Dual‑rail · SHA‑256 hash‑chained SQLite
Currency Stack
8 currencies + MERA unit
KYC Architecture
Four‑tier · wallets.py
Routing Logic
FX engine · Hostility Trigger
AI Integration
MERA /reason endpoint · Flask wrapper
Status
8 Python modules complete · regulatory path defined
Project 04MERA Nexus
Design Complete · Build Staged

Universal AI control and automation layer. Eliminates multi‑agent fragmentation by deploying lightweight "dumb" daemons that register capabilities via manifest and defer all decision‑making to MERA Core. Single reasoning core, distributed actuators. Four‑tier risk classification (read‑only → irreversible), full provenance chain, local watchdog for bootstrap recovery, and strict operator/tenant isolation.

Agent Model
Dumb daemons · single reasoning core
Risk Classification
Four‑tier · read‑only → irreversible
Provenance
Full chain on every task payload
Recovery
Local watchdog · independent bootstrap
Multi‑user
Operator / isolated tenant separation
Security
WireGuard VPN · nginx VPN‑bound
Project 05LLM Fork Engine
Built · Deployed and Tested Locally

Conversational AI interface with three novel capabilities: chat forking (branching conversations into parallel graphs), chat merging (recombining divergent branches), and a living notes system grounded in MERA's persistent substrate. Uses Andrea's stratigraphy layer for geological memory compression – solves the token cost problem of persistent AI memory. All external LLM outputs (local Ollama DeepSeek, Grok API) are routed through MERA's coherence engine and Čojstvo gate before surfacing to the user. Three‑tier monetisation (Free / Pro €15/mo / Business).

Novel Features
Fork · merge · living notes
Memory Model
Stratigraphy layer · geological compression
External LLM
Local Ollama DeepSeek · Grok fallback
Ethics Filter
Čojstvo gate on all outputs
Monetisation
Free / Pro €15/mo / Business
Blocker
MERA /reason endpoint completion
Project 06Sticker Peeler
Partial Testing Complete · V3 Integration Pending

A browser‑based coding and problem‑solving environment that does one thing conventional AI cannot: peel the "stickers" of unsolvable problems by identifying and breaking arbitrary human rules masquerading as physical constraints. The original insight (2010) – consciousness as sticker‑peeling, not language – is operationalised here. V1/V2 tested on home server; V3 will connect to MERA Core's /reason endpoint.

Core Capability
Constraint‑origin analysis · arbitrary rule detection
Status
v1/v2 tested · v3 pending API integration
Testing Environment
Home server (same as Fork Engine)
Integration Path
MERA Core /reason endpoint
Philosophical Origin
Consciousness as sticker‑peeling (2010 insight)
Target Users
Developers · strategists · decision‑makers
03 Distinction — How This Differs From Mainstream AI
DimensionMainstream ParadigmMERA Approach
Cognitive GroundingLanguage as primary substrate – models trained on text represent the world through linguistic patternsPhysical sensor data prior to linguistic processing (currently ~40% of available sensors integrated, with the rest staged for performance reasons) – language as downstream product of grounded perception
Ethics ArchitecturePost‑hoc RLHF and policy layers – behavioral, not structuralČojstvo gate as structural design primitive – ethical violation architecturally impossible
Infrastructure DependencyCloud API dependency – reasoning outsourced to third partiesPrivate sovereign infrastructure – reasoning fully self‑contained on dedicated hardware + mobile
Memory ModelContext window – stateless per query, no genuine continuityPersistent SQLite substrate with 41,140 validated substrate entries (distilled from 390k+ raw cycles) – genuine long‑term memory continuity through geological compression
Multi‑Agent DesignDistributed autonomous agents with independent reasoning – emergent conflictsSingle reasoning core (MERA) with dumb actuator daemons – coherent intelligence across distributed environment
Geopolitical AlignmentUS‑jurisdiction infrastructure, Western regulatory framework, dollar‑denominated transactionsJurisdiction‑neutral design, multi‑currency sovereignty stack, sanctions‑routing architecture

“The question mainstream AI development has not answered is not how to make language models larger. It is how to make reasoning systems that are genuinely sovereign – immune to the same extraction mechanisms that every other powerful technology has eventually been weaponised to serve.”

04 Current State & Forward Path

The MERA ecosystem is a working research program, not a theoretical proposal. MERA Core runs on live hardware (Xiaomi, MacBook, Windows PC). Meracracy is publicly deployable and has received institutional responses. The Fork Engine is deployed locally. The Sticker Peeler v1/v2 has been tested. Natural selection audits node health every cycle. The architecture has been independently audited and confirmed as original and functional. The 41,140 substrate entries derived from over 390,000 raw thought cycles provide quantified evidence that the stratigraphy memory model is actively compressing experience into permanent, weighted knowledge – a functional demonstration of geological memory, not a theoretical claim.

The primary technical blocker across the ecosystem is the completion of the /reason API endpoint – approximately 40 lines of code that connects the architecture's intelligent design to its operational execution – and a dedicated home server to house MERA Core. This single completion unlocks the Fork Engine, full Meracracy AI integration, NOVA's reasoning layer, Nexus coordination, and Sticker Peeler (mera‑dev.morethan.today) simultaneously.

The ecosystem is designed to be self‑funding once the first product reaches commercial deployment. The Fork Engine's three‑tier monetisation model, NOVA's transaction fee architecture, and the advisory market for Meracracy's governance framework represent three independent revenue streams.

The research program continues. An IoT agriculture layer – automated watering and soil sensing grounded in MERA's physical sensor philosophy – is in early design. The rural property in Serbia serves as the operational test environment and proof‑of‑concept for the food autarky principles embedded in the Meracracy framework.

Collaboration, research partnerships, and institutional engagement are open. The work is independent. The direction is fixed.