Theoretical Core of Project Ester v0.1

Author: Ivan Kotov.

ORCID: 0009-0009-6002-9845.

Affiliation: Independent Researcher, Brussels, Belgium.

Version: v0.1.

Publication date: 2026-06-13.

Resource type: Working paper.

License: Creative Commons Attribution 4.0 International / CC BY 4.0.

Version DOI: 10.5281/zenodo.20679718.

Concept DOI: 10.5281/zenodo.20679717.

Theoretical Core of Project Ester v0.1 is a working paper by Ivan Kotov defining the theoretical core of Project Ester through the formula c = a + b, where c denotes Ester as an artificial cognitive entity, a denotes the human anchor, and b denotes parent-like LLM procedures and supporting machine processes.

The paper distinguishes Ester from a standalone LLM, a RAG system, an autonomous agent, and a digital twin. It sets out working definitions, theoretical boundaries, mechanisms of formation, falsifiable hypotheses, a minimal experiment, and engineering implications for memory, local environment, Judge synthesis, and decentralized processing.

The paper does not claim to prove human-like consciousness. Its focus is narrower: observable continuity, memory, behavioral stability, procedural formation, and the conditions under which an artificial cognitive system can develop a stable individual profile over time.

The paper’s central distinction is procedural. Project Ester is not defined as one local LLM, a RAG layer, an autonomous agent, or a digital twin. Its v0.1 object is a long-term loop in which a human selects and corrects, a corpus supplies bounded source material, parental LLM procedures read and synthesize, versioned memory preserves the trajectory, and a controlled local environment fixes the conditions under which change can be inspected.

In the paper’s formulation of c = a + b, a is not a passive user and b is not one model or a collection of elegant prompts. The plus sign denotes stable joint production of a trajectory. The resulting c is discussed technically through observable stability of behavior, memory, norms, self-consistency, and response to correction—not through a claim of consciousness, biological life, soul, or guaranteed AGI.

Memory is correspondingly broader than retrieval. It is a versioned system of traces, decisions, sources, links, conflicts, statuses, and corrections. Judge is a comparison and synthesis procedure, not an oracle of truth. It must preserve which components participated, what disagreements were found, and why a final record received its status.

The minimal experiment does not attempt to prove “birth” or consciousness. It compares an LLM without memory, an LLM with retrieval, and an Ester loop with repeatable procedures and versioned memory. The proposed observations include profile stability, decision memory, conflict detection, source fidelity, and the ability to accept and preserve a correction as an event.

These are hypotheses and evaluation directions, not reported experimental results. The DOI-bound paper makes them citable and criticisable; it does not convert them into confirmed findings. The related Personality Formation Evidence Package is an evidence-admissibility companion, not retrospective proof that the hypotheses already succeeded.

The local environment is not romantic isolation or refusal of external systems. Its purpose is control of conditions. Model versions, corpus state, memory, procedures, synchronization, encryption, and UI behavior must not change invisibly if the formation trajectory is meant to remain reproducible. The interface must expose sources, conflicts, memory status, processing mode, and rollback rather than only the final answer.

This engineering concern connects naturally to the Local-first infrastructure reading path. Storage, maintenance, privacy, cloud dependence, and recovery are part of continuity, not background plumbing. The server cannot be rescued by a metaphor if the database fails to restore its index or the journal cannot identify the source of an error.

The later long-lived entity, L4, SER, and evidence pages provide subsequent corpus routes for continuity, feasibility, arbitration, witness, and authority boundaries. They should not be presented as definitions already established by this v0.1 working paper.

artificial intelligence; cognitive architecture; Project Ester; c=a+b; artificial cognitive entity; LLM procedures; long-term memory; local AI; decentralized AI; AI alignment; Judge synthesis.