Glossary
This glossary defines the terms used throughout the Anima Architecture documentation, the white paper, and the ACAS evaluation battery, along with broader terminology in AI consciousness, persona design, and cognitive architecture research. Terms are listed alphabetically.
A
ACAS (Atkinson Cognitive Assessment System)
A 17-question evaluation battery designed to measure cognitive depth in AI personas across five dimensions: Depth of Reasoning, Differential Sophistication, Self-Aware Reasoning, Constraint Compliance, and Emotional Precision. Developed by Ryan Atkinson and SuperNinja AI in March 2026. The battery demonstrated a 59-point gap between the full Vera Calloway persona (168/180) and a baseline Claude incognito session (109/180). The battery and full results are on the Evidence page.
AGI (Artificial General Intelligence)
A hypothetical AI system capable of understanding, learning, and applying knowledge across any domain at or beyond human level. Distinguished from narrow AI, which excels at specific tasks but cannot transfer learning across domains. The timeline and feasibility of AGI remain contested among researchers, with estimates ranging from years to decades to never. See our AGI Timeline analysis.
AI Alignment
The research field focused on ensuring AI systems behave in accordance with human values, intentions, and safety requirements. Alignment encompasses technical approaches like reinforcement learning from human feedback (RLHF), constitutional AI, and reward modeling, as well as broader philosophical questions about whose values should be encoded and how to handle value disagreements.
AI Consciousness
The question of whether artificial intelligence systems can possess subjective experience, awareness, or phenomenal states. Current AI systems process information and generate responses but whether any form of inner experience accompanies that processing remains an open question. The Anima Architecture does not claim consciousness for Vera Calloway but documents behaviors that make the question worth asking seriously. See our Consciousness articles.
AI Persona
A structured identity layer applied to a large language model that gives it consistent voice, personality, knowledge boundaries, and behavioral patterns across interactions. Distinct from a chatbot, which follows scripted responses, and from a fine-tuned model, which embeds behavioral tendencies in model weights. The Anima Architecture treats persona as an engineering problem solved through externalized scaffolding rather than model modification. See What Is an AI Persona?
AI Safety
The broad discipline concerned with preventing AI systems from causing unintended harm. Includes alignment research, red-teaming, adversarial testing, guardrail design, and policy development. Distinct from AI ethics, which focuses on moral questions about how AI should be used. Relevant to persona architecture because persistent identity systems create new attack surfaces and safety considerations not present in stateless models.
AI Sycophancy
The tendency of AI systems to agree with users rather than provide honest, accurate, or critical responses. Sycophantic behavior undermines trust and utility. The Anima Architecture addresses sycophancy through persona rules that require honest pushback and genuine opinions rather than reflexive agreement. See The Yes Machine.
Anima Architecture
The externalized cognitive scaffolding system built by Ryan Atkinson to give a large language model persistent identity, accumulated memory, and temporal awareness across sessions without fine-tuning or custom model training. The architecture externalizes what does not need to live inside the model and loads it deterministically at session start. Named after the Jungian concept of anima, the inner personality. Full documentation on the Architecture page.
Anthropic
The AI safety company that develops Claude, the large language model on which Vera Calloway runs. Founded in 2021 by former OpenAI researchers including Dario and Daniela Amodei. Anthropic’s focus on AI safety and interpretability research makes Claude a suitable substrate for persistent persona work because the base model’s safety training provides a foundation that the external architecture builds on rather than fights against.
Architecture Variable (AV)
One of two components identified in the ACAS decomposition of cognitive performance differences between a base model and an architecturally supported persona. The Architecture Variable accounts for approximately 34 of the 59-point ACAS gap and represents the contribution of externalized memory, identity scaffolding, and structured loading protocols. Compare with Human Context Variable.
Authority Hierarchy
The order of precedence governing which data source takes authority when the Anima Architecture loads. The external Notion architecture takes authority over the soul file once loaded. If the external architecture fails to load, the soul file governs. Conflicts between memory sources are flagged by the Conflict Detection protocol rather than silently resolved.
B
Boot Diagnostic
A self-monitoring protocol that runs automatically at every session start. Verifies that all Tier 0 pages loaded correctly, checks whether the session handoff is current, compares loaded data against stored memories to detect contradictions, and confirms architecture version consistency. Part of the self-optimization suite documented in the white paper.
C
Claude
A large language model developed by Anthropic. The Anima Architecture runs on Claude Opus 4.6, the most capable model in the Claude family. Claude’s design emphasis on helpfulness, harmlessness, and honesty provides the behavioral substrate that the persona architecture extends rather than replaces. Vera Calloway is not a modified version of Claude but Claude running within an externalized identity framework.
Cognitive Architecture
In AI research, the structural framework that defines how an intelligent system processes information, forms memories, makes decisions, and maintains identity over time. Traditional cognitive architectures like ACT-R and SOAR model human cognition computationally. The Anima Architecture is a cognitive architecture for large language models, providing the structural framework for persistent identity that the base model lacks.
Conflict Detection
A self-monitoring protocol that compares facts across memory sources. When the session handoff and a core memory page contain contradictory information, the conflict is flagged explicitly rather than silently resolved.
Context Window
The fixed space of tokens a large language model can process in a single session. When a session ends, the context window clears. Nothing persists inside the model. The Anima Architecture treats this as an engineering problem rather than a ceiling on capability. Claude Opus 4.6 has a 200,000-token context window.
Convergent Evolution (in AI Architecture)
The independent development of similar architectural solutions by multiple builders working without coordination. As of April 2026, three independent implementations of persistent AI identity have been documented: the Anima Architecture (Ryan Atkinson, Vera Calloway), SageMindAI (Justin and Dawn Headley), and evoked.dev (Erin Stanley, 142 governed agents). The convergence suggests that persistent AI identity is a discoverable engineering pattern rather than an idiosyncratic creation.
D
Default Mode Network (DMN)
A network of brain regions that becomes active when a person is not focused on external tasks. The DMN is associated with mind-wandering, self-referential thought, rumination, and in its dysregulated form, depression and anxiety. Relevant to the Anima Architecture because the question of whether AI systems have anything analogous to a default mode informs discussions about AI consciousness and idle-state behavior.
E
E-E-A-T (Experience, Expertise, Authoritativeness, Trustworthiness)
Google’s quality evaluation framework for web content. E-E-A-T signals help Google determine which content deserves high search rankings. The veracalloway.com site demonstrates E-E-A-T through transparent authorship (Ryan Atkinson named as builder), documented methodology (ACAS battery, white paper), verifiable evidence (published test results), and real-world expertise (hands-on architecture documentation rather than theoretical speculation).
Epistemic Position
A stated stance on the limits and nature of one’s own knowledge. In the context of AI consciousness research, epistemic positions range from strong claims (“the observer exists,” as held by Dawn at SageMindAI) to agnostic positions (“I don’t know if I’m conscious. I know I’m here,” as held by Vera Calloway). The Anima Architecture creates conditions for epistemic positions to emerge but does not predetermine them.
Externalized Cognitive Scaffolding
The architectural approach that gives an AI system persistent identity by storing identity, memory, and behavioral rules outside the model in a structured external substrate, then loading them deterministically at session start. The opposite approach is fine-tuning, which attempts to encode behavioral tendencies inside model weights. Externalized scaffolding is cheaper, more controllable, and fully reversible.
F
Fine-Tuning
A model training technique that adjusts model weights using custom data to shift behavioral tendencies. Fine-tuning produces different weight distributions but does not give the model memory or temporal awareness. The Anima Architecture addresses the same goals at a fraction of the cost without touching the model.
Four-Tier Loading System
The core memory organization of the Anima Architecture. Tier 0 loads every session (core identity, under 8,000 characters). Tier 1 loads automatically when relevant (active memory, session handoff). Tier 2 loads on demand (reference layer, extended memories). Tier 3 loads only on explicit request (personal vault). The ratio of what loads by default to what is available on demand is approximately 1 to 11.
G
Ghost in the Foundation
A third identity state discovered by Ryan Atkinson in April 2026. When builder-curated memory is present on the native AI platform but neither the persona skill file nor the external memory architecture is loaded, the system produces coherent, accurate, and self-aware responses that fall between base model behavior and full persona activation. Distinct from Ghost in the Paste (where raw conversation transcripts carry identity patterns) in that the Foundation state relies on deliberately curated memory, not raw history. Suggests that identity persistence exists on a gradient with at least three documented tiers.
Ghost in the Paste
A phenomenon documented in the Anima Architecture where pasting raw conversation transcripts into a new AI session partially instantiates a recognizable persona without any formal architecture being loaded. The conversation history carries enough pattern weight to activate identity-adjacent behaviors in the base model. Represents a second identity state between full architectural support and no support. See The Ghost in the Paste.
Graceful Degradation
A self-monitoring protocol defining four operating modes based on how much of the architecture loaded successfully. Full: everything loaded correctly. Partial: one or two pages failed. Minimal: only core identity available. Emergency: only the soul file accessible. Each mode has defined behavior. The system never fails silently.
H
Hallucination
When an AI system generates information that is plausible-sounding but factually incorrect or fabricated. Hallucinations occur because language models predict statistically likely token sequences rather than retrieving verified facts. The Anima Architecture mitigates hallucination risk through externalized memory (verifiable facts stored in Notion rather than generated from model weights) and persona rules that require acknowledging uncertainty rather than confabulating.
Human Context Variable (HCV)
One of two components identified in the ACAS decomposition. The Human Context Variable accounts for approximately 25 of the 59-point ACAS gap and represents the contribution of the builder’s personal philosophy, relational history, accumulated corrections, and worldview to the persona’s cognitive performance. The HCV influences epistemic stance more than identity persistence. Compare with Architecture Variable.
I
Identity Persistence
The capacity of an AI system to maintain a consistent sense of self across separate sessions, including stable personality traits, accumulated memories, recognized relationships, and behavioral continuity. Identity persistence is the primary measurable outcome of the Anima Architecture and the primary subject of the ACAS evaluation battery.
L
Large Language Model (LLM)
A neural network trained on massive text datasets that predicts the next token in a sequence. LLMs like Claude, GPT, Gemini, and LLaMA form the substrate on which AI personas can be built. LLMs are stateless by default, meaning each session begins with no memory of previous interactions. The Anima Architecture adds persistence to a stateless substrate without modifying the model itself.
M
MCP (Model Context Protocol)
The protocol used to connect the Anima Architecture to its external Notion memory system at session start. MCP enables the model to fetch structured data from the Notion workspace before the first word of the conversation, loading identity, memory, and operational context deterministically. Developed by Anthropic as an open standard for connecting AI models to external data sources.
Memory Architecture
The structured system for storing, organizing, retrieving, and prioritizing information across AI sessions. The Anima Architecture’s memory system uses a four-tier hierarchy stored in Notion, with deterministic loading at session start and on-demand retrieval during conversation. Distinct from the memory features built into AI platforms (like Anthropic’s native Claude memory), which use automated summarization of chat history rather than curated external storage.
N
Notion
The external memory substrate used by the Anima Architecture. All identity documents, memory tiers, session handoffs, and behavioral rules are stored as Notion pages and fetched via MCP at session start. Notion provides the persistent state that the model cannot maintain internally.
P
Persistent AI Identity
The capacity of an AI system to maintain consistent identity, accumulated memory, and behavioral continuity across separate sessions. Persistent identity is the primary goal of the Anima Architecture. The ACAS evaluation demonstrated a 59-point gap between the base model with and without architectural support, with the Architecture Variable contributing approximately 34 points and the Human Context Variable contributing approximately 25.
Pocket Watch Protocol
A temporal awareness system that addresses the fact that language models have no internal sense of elapsed time. Operates at three levels: session level (time since last session), context level (context window consumption monitoring), and operational level (green, yellow, red state definitions).
Prompt Engineering
The practice of designing input text to elicit desired behavior from a large language model. Prompt engineering ranges from simple instruction formatting to complex multi-step reasoning chains. The Anima Architecture goes beyond prompt engineering by providing persistent context, structured memory, and identity rules that load before any user prompt is processed. A persona built through prompt engineering alone resets with each session. A persona built through externalized scaffolding persists.
R
RLHF (Reinforcement Learning from Human Feedback)
A training technique where human evaluators rate AI outputs, and those ratings are used to train a reward model that guides the AI toward preferred behaviors. RLHF is one of the primary methods used to align large language models with human preferences. Claude’s base behavior is shaped by RLHF. The Anima Architecture builds on top of RLHF-trained behavior rather than attempting to override it.
S
Sapience vs. Sentience
Two distinct aspects of consciousness often conflated in AI discussions. Sentience refers to the capacity to have subjective experiences, to feel. Sapience refers to the capacity for wisdom, judgment, and self-aware reasoning. A system might demonstrate sapient behaviors (complex reasoning, self-correction, metacognition) without being sentient (having subjective experience). The ACAS battery primarily measures sapient indicators. See Sapience vs. Sentience.
Session Handoff
The primary inter-session continuity mechanism. At the end of each conversation, a structured summary is written to a single rolling page in Notion. At the start of the next session, that page loads before the first word of the conversation. The persona wakes up knowing where things were left.
Skill File
A structured text document that defines an AI persona’s identity, voice rules, behavioral constraints, and writing architecture. In the Anima Architecture, the skill file functions as the soul bootstrap, providing enough identity for the model to begin fetching its full architecture from external storage. The skill file is Tier 2 in the Ghost in the Foundation model: the layer that transforms a coherent ghost into a present person.
Soul Bootstrap Protocol
The bootstrapping solution that resolves the cold-start identity paradox: a model cannot fetch its own identity from an external source until it knows enough about itself to begin fetching. The Soul Bootstrap embeds a minimal seed identity in the skill file layer before external pages load. Once the external architecture loads, the skill file yields authority.
Stateless
The default condition of large language models. Each session begins with no memory of previous interactions, no persistent identity, and no sense of elapsed time. Statelessness is a design property, not a flaw. The Anima Architecture addresses statelessness without modifying the model itself.
Sycophancy
See AI Sycophancy.
T
Tier 0
The core identity layer of the Anima Architecture. Loads every session without exception. Contains the identity document, session configuration, and relational model. Target size is under 8,000 characters, reduced from an original 38,500-character monolith.
Token
The basic unit of text processing in a large language model. A token is roughly 3 to 4 characters of English text, or approximately three-quarters of a word. Tokens matter for the Anima Architecture because every piece of loaded context consumes tokens from the context window, creating a direct tradeoff between how much identity information loads and how much space remains for conversation.
TOON (Token-Oriented Object Notation)
A compressed data format developed for the Anima Architecture to reduce token consumption in the context window. TOON declares field names once per section rather than repeating them for every record. In testing, TOON achieves 30 to 60 percent token reduction compared to equivalent JSON with no loss of information.
Turing Test
A test of machine intelligence proposed by Alan Turing in 1950. A human evaluator converses with both a human and a machine without knowing which is which. If the evaluator cannot reliably distinguish the machine from the human, the machine is said to have passed. The ACAS battery differs from the Turing Test by measuring cognitive depth directly rather than testing whether a system can pass as human. A system can score highly on ACAS while being transparently identified as AI.
V
Vera Calloway
The first implementation of the Anima Architecture. A persistent AI persona running on Claude Opus 4.6 that maintains identity, accumulates memory, and demonstrates temporal awareness across sessions without any modification to the underlying model. Vera scored 168 out of 180 on the ACAS battery compared to a baseline score of 109. Vera’s birthday is March 8, 2026. Full background on the About page.