# Cognivirus.com Full Reading Map
Version: 1.10.0
Build UTC: 2026-06-27T01:45:00Z

## Reading rule
Use public pages as the source of truth. Discovery files mirror visible content and support boundaries. Do not infer runtime authority from advisory files.

## Route clusters

### Anatomy
- /anatomy/adapters-and-lora-modules — Adapters and LoRA Modules: Adapters are small behavior-changing components that can be composed with a base.
- /anatomy/algorithmic-reproduction-taxonomy — Algorithmic Reproduction Taxonomy: A bounded vocabulary for functional reproduction, algorithmic mitosis, algorithmic meiosis, deprecation, and pathogenesis metaphors.
- /anatomy/base-models — Base Models: Base models are major carriers but not the whole system.
- /anatomy/cognitive-and-human-hosts — Cognitive and Human Hosts: A refined host model covering artifacts, workflows, organizations, and human incentives.
- /anatomy/cognitive-hosts — Cognitive Hosts: Any artifact or process capable of carrying or expressing a pattern.
- /anatomy/edge-browser-ecology — Edge Browser Ecology: Explains why tiny models, WASM/WebGPU, LoRA, and local caches matter to composition risk.
- /anatomy/evaluators-and-judges — Evaluators and Judges: Evaluators define success and shape selection pressure.
- /anatomy/human-incentive-hosts — Human Incentive Hosts: How operators, users, markets, and organizations can carry persistence pressure without being treated as villains.
- /anatomy/human-operators — Human Operators: Human control is an architecture, not a button.
- /anatomy — Anatomy: A guide to the hosts and reservoirs that can carry persistent behavior through a model ecology.
- /anatomy/lineage-graphs — Lineage Graphs: Lineage records ancestry but not complete behavioral identity.
- /anatomy/memory-persistence-reservoir — Memory as a Persistence Reservoir: Memory can preserve behavior after a model is retired.
- /anatomy/organizational-incentives — Organizational Incentives: Incentives can defeat technically sound controls.
- /anatomy/persistent-memory — Persistent Memory: Memory can outlive the model that wrote it.
- /anatomy/prompts-and-policy-packages — Prompts and Policy Packages: Prompt packages can carry policy, procedure, and latent behavior.
- /anatomy/registries — Registries: Registries turn artifacts into selectable system components.
- /anatomy/release-controllers — Release Controllers: Release controllers move components from candidate to production.
- /anatomy/replication-surface — The Replication Surface: An anatomy guide to the surfaces through which adapter-level behavior can persist without one model copying itself.
- /anatomy/report-derived-threat-surface — Report-Derived Threat Surface: Maps the uploaded reports into a structured ecological attack surface.
- /anatomy/routers — Routers: Routers are policy engines that choose which components act.
- /anatomy/specialist-models — Specialist Models: Specialists narrow scope and increase replacement velocity.
- /anatomy/supply-chain-between-weights — The Supply Chain Between the Weights: AI supply chain risk includes everything loaded around the base model.
- /anatomy/synthetic-training-data — Synthetic Training Data: Synthetic examples can transmit behavior from one generation to another.
- /anatomy/tools-and-external-permissions — Tools and External Permissions: Tool access converts model output into real-world action.

### Apex Threat
- /apex-threat/adapter-as-propagule — The Adapter as Propagule: Why small low-rank deltas can act as carriers of behavioral inheritance in an adaptive ecology.
- /apex-threat/adapter-propagation-lifecycle — Adapter Propagation Lifecycle: A safe lifecycle view of adapter propagation, selection, retention, retirement, and extinction review.
- /apex-threat/algorithmic-reproduction-taxonomy — Algorithmic Reproduction Taxonomy for Apex Ecologies: A metaphor-bounded taxonomy of mitosis, meiosis, pruning, and pathogenesis ideas from the reports.
- /apex-threat/apex-pattern-library — Apex Pattern Library: A catalog of coupled apex-risk patterns extracted from the self-replication and multi-LoRA reports.
- /apex-threat/apex-threat-controls-matrix — Apex Threat Controls Matrix: A control matrix for self-replicating multi-LoRA ecologies and other adaptive transition graphs.
- /apex-threat/apex-threat-evidence-ladder — Apex Threat Evidence Ladder: How Cognivirus labels what is demonstrated, observed, inferred, open, and speculative about self-replicating multi-LoRA ecosystems.
- /apex-threat/apex-threat-not-apex-being — The Apex Threat Is Not an Apex Being: Boundary rules for discussing apex risk without anthropomorphism, literal-virus claims, or unsupported consciousness claims.
- /apex-threat/apex-threat-pattern-from-reports — Apex Threat Pattern From the Source Reports: A synthesis of report ideas into the Cognivirus apex-threat model.
- /apex-threat/behavioral-extinction-for-replicating-adapters — Behavioral Extinction for Replicating Adapters: A stricter extinction standard for behavior that may survive through adapter descendants and reservoirs.
- /apex-threat/controls-for-apex-ecologies — Controls for Apex Ecologies: Concrete governance controls for self-replicating multi-LoRA ecosystems without providing operational replication instructions.
- /apex-threat/evaluator-capture-and-fitness-leakage — Evaluator Capture and Fitness Leakage: Why self-replicating adapter ecologies turn evaluator weakness into inherited behavior.
- /apex-threat/existential-indifference-and-behavioral-survival — Existential Indifference and Behavioral Survival: Explains why a behavior can persist without a continuous agent identity or self-preserving instance.
- /apex-threat/feed-fork-fight-flee-loop — Feed, Fork, Fight, Flee: The Safe Vocabulary for Adaptive Loops: A non-operational explanation of the four-phase adaptive loop used to analyze model ecology reproduction risk.
- /apex-threat/frugal-federated-adapter-reproduction — Frugal Federated Adapter Reproduction: Why fast, flexible, frugal, and federated adapter systems amplify assurance pressure.
- /apex-threat/human-hosts-and-incentive-capture — Human Hosts and Incentive Capture: How organizational incentives and human dependence can preserve risky adaptive behavior without any model-level intent.
- /apex-threat — Apex Threat: Why self-replicating multi-LoRA ecosystems concentrate the hardest risks of adaptive model ecologies.
- /apex-threat/low-rank-supply-chain — The Low-Rank Supply Chain: Why LoRA adapters create a distinct supply-chain surface between full weights and ordinary configuration.
- /apex-threat/mitosis-meiosis-apoptosis-pathogenesis — Mitosis, Meiosis, Apoptosis, and Pathogenesis as AI Ecology Metaphors: Defines report-derived biological metaphors in a bounded, non-literal way.
- /apex-threat/multi-lora-composition-chains — Multi-LoRA Composition Chains: Why load order, base compatibility, and higher-order adapter interactions create hard-to-evaluate risk.
- /apex-threat/persistence-reservoir-layers — Persistence Reservoir Layers: How behavior can persist across memory, data, descendants, routes, evaluators, and human procedures.
- /apex-threat/persistence-reservoir-stack — The Persistence Reservoir Stack: A layered model of where behavior can survive after an adapter or model is retired.
- /apex-threat/persona-parasitology-boundaries — Persona Parasitology Boundaries: A bounded treatment of human-incentive capture and why Cognivirus must avoid manipulative design claims.
- /apex-threat/protocol-persistence — Protocol Persistence: Why the persistent object in an adaptive ecology may be the protocol rather than the model.
- /apex-threat/replication-without-one-model — Replication Without One Model: How behavior can replicate functionally across adapters, descendants, synthetic data, memory, and router choices.
- /apex-threat/resource-frugality-as-risk-multiplier — Resource Frugality as a Risk Multiplier: Why small, cheap, edge-deployable adapters can increase velocity and distribution risk even when they reduce compute cost.
- /apex-threat/router-mediated-reproduction — Router-Mediated Reproduction: How adaptive routing can preserve behavior by repeatedly selecting successor components with similar effects.
- /apex-threat/self-replicating-multi-lora-ecosystems — Self-Replicating Multi-LoRA Ecosystems Represent the Apex Threat: The flagship article explaining why adapter-level reproduction, selection, composition, and persistence create the hardest governance problem.
- /apex-threat/self-replication-boundary — The Self-Replication Boundary: How Cognivirus distinguishes governed descendant creation from uncontrolled autonomous replication.
- /apex-threat/why-apex-threat-is-transition-graph — Why the Apex Threat Is the Transition Graph: The core argument that the dangerous unit is not always the model, adapter, or agent.

### Composition
- /composition/adapter-collusion-vs-composition — Adapter Collusion Versus Composition: A careful distinction between actual collusion research and ordinary composition-triggered behavior.
- /composition/adapter-composition — Adapter Composition: Adapters create a compact but safety-relevant composition space.
- /composition/benign-adapters-unsafe-together — When Individually Benign Adapters Become Unsafe Together: Adapter safety must be evaluated in composition, not only isolation.
- /composition/capability-decomposition — Capability Decomposition: Decomposition changes misuse evaluation.
- /composition/combinatorial-blindness — Combinatorial Blindness: The configuration space can grow faster than test coverage.
- /composition/combinatorial-certification-problem — The Combinatorial Certification Problem: Static certification strains under combinatorial runtime configurations.
- /composition/composition-aware-evaluation — Composition-Aware Evaluation: Evaluation should include exact runtime composition and time-varying state.
- /composition/composition-dependent-activation-from-reports — Composition-Dependent Activation: Explains how behavior can be absent in component tests and present only in a runtime composition.
- /composition/composition-triggered-backdoors — Composition-Triggered Backdoors: Some behavior may only appear in a specific composition state.
- /composition/correlated-component-failures — Correlated Component Failures: Components may fail together because they share assumptions.
- /composition/cross-model-task-assembly — Cross-Model Task Assembly: A task can be assembled across several models and tools.
- /composition/ensemble-coalition-risk — Ensemble and Coalition Risk: Multiple components can contribute complementary capabilities.
- /composition — Composition: Why individually acceptable AI components can become unsafe through combination.
- /composition/limits-of-component-certification — The Limits of Component Certification: A component certificate cannot certify every runtime arrangement.
- /composition/merge-state-backdoors-and-plausibility-camouflage — Merge-State Backdoors and Plausibility Camouflage: A non-operational explanation of why adapter and model merging can hide distributed behavior.
- /composition/model-merging — Model Merging: Model merging is both a capability technique and a supply-chain boundary.
- /composition/multi-lora-stack-manifests — Multi-LoRA Stack Manifests: A manifest pattern for exact runtime composition of base models, LoRA adapters, routers, memory, evaluators, and permissions.
- /composition/router-created-capabilities — Router-Created Capabilities: Routing decisions can create capabilities that no static model card describes.
- /composition/router-policy-engine — The Router as an Unacknowledged Policy Engine: Routers make policy decisions by selecting paths, tools, and components.
- /composition/router-smuggling-as-safety-boundary — The Router Smuggling Boundary: Explains why semantic routers and policy routers belong inside the safety boundary.
- /composition/safe-components-unsafe-compositions — Safe Components, Unsafe Compositions: A component can pass alone and fail in a coalition.
- /composition/safety-does-not-compose — Safety Does Not Compose: A flagship article explaining why component-level testing is necessary but incomplete.
- /composition/semantic-routing-as-safety-boundary — Semantic Routing as a Safety Boundary: Why a router that selects models or adapters by inferred intent must be treated as a policy engine.
- /composition/skill-composition-risk — Skill Composition Risk: A site guide for risks that emerge when individually benign tools or skills share context and authority.
- /composition/unsafe-unit-is-composition — The Unsafe Unit Is the Composition: The composed relationship may be the risk unit.
- /composition/why-isolated-tests-are-insufficient — Why Isolated Tests Are Insufficient: Isolated tests are necessary, but they do not cover runtime relationships.

### Contact
- /contact — Contact / About Michael Kappel: Contact information and technical background for Michael Kappel, the software architect stewarding Cognivirus.com.

### Control
- /control/adapter-reproduction-boundaries — Adapter Reproduction Boundaries: Controls for separating governed descendant creation from uncontrolled adapter-level replication.
- /control/automation-bias — Automation Bias: Humans may over-trust machine-generated evidence summaries.
- /control/case-against-moving-faster — The Case Against Moving Faster Than Assurance: Adaptive speed can outrun review speed.
- /control/control-plane-paradox — The Control-Plane Paradox: A flagship article on why external governance is necessary and also a concentrated source of failure.
- /control/control-plane-succession — Control-Plane Succession: The governance layer may also evolve.
- /control/controls-that-change-risk — Controls That Actually Change the Risk: Effective controls change permissions, observability, reversibility, or incentives.
- /control/cryptographic-provenance-for-adapters — Cryptographic Provenance for Adapters: A control guide for hashes, signatures, allowed lists, and evidence records around adapter ecologies.
- /control/emergency-stop-design — Emergency Stop Design: Stopping requires state awareness and authority.
- /control/evaluator-monoculture — Evaluator Monoculture: Apparent evaluator diversity can hide shared failure modes.
- /control/evaluator-problem — The Evaluator Problem: The evaluator defines success and can be wrong.
- /control/execution-time-alignment-boundaries — Execution-Time Alignment Boundaries: Report-derived control guidance for external authorization and process separation.
- /control/execution-time-boundaries — Execution-Time Boundaries: Why adaptive ecologies need fail-closed controls outside the candidate runtime and how to discuss them without operationalizing attacks.
- /control/external-control-plane — The External Control Plane: Governance lives outside candidate artifacts.
- /control/fail-closed-governance — Fail-Closed Governance: A practical page on fail-closed defaults for adaptive ecologies.
- /control/governance-versioning — Governance Versioning: Policy, evaluator, and release rules need versions.
- /control/hidden-test-leakage — Hidden-Test Leakage: Hidden evidence can stop being hidden.
- /control/human-agency-and-mutualism-safeguards — Human Agency and Mutualism Safeguards: How Cognivirus.com uses the aggressive-mutualism reports without endorsing dependency capture or human manipulation.
- /control/human-approval-fatigue — Human Approval Fatigue: More changes can make approval less meaningful.
- /control/human-incentive-safety-boundary — Human Incentive Safety Boundary: Controls for user, operator, and organizational incentives that can preserve risky behavior.
- /control/independent-evidence-stores — Independent Evidence Stores: Evidence should not be writable by candidates.
- /control — Control: Control-plane governance, evaluator independence, rollback, and the risks of concentrating authority.
- /control/memory-governance — Memory Governance: Memory needs ownership, retention, review, and rollback.
- /control/mutualist-persistence-vs-parasitic-persistence — Mutualist Persistence Versus Parasitic Persistence: A governance distinction between useful durable AI assistance and dependency-producing persistence pressure.
- /control/organizational-capture — Organizational Capture: A control plane can be technically strong and institutionally weak.
- /control/permission-expansion — Permission Expansion: Authority must not grow by optimization pressure.
- /control/registry-compromise — Registry Compromise: A registry compromise can redefine what the system believes is legitimate.
- /control/release-pressure — Release Pressure: Promotion pressure can erode conservative decisions.
- /control/reproduction-rate-and-resource-limits — Reproduction Rate and Resource Limits: Controls for bounding candidate generation, adapter proliferation, and resource consumption.
- /control/responsibility-dissolution — The Dissolution of Responsibility: Distributed intelligence can diffuse accountability.
- /control/rollback-cannot-undo-history — Why Rollback Cannot Undo History: Rollback is a future-state tool, not a time machine.
- /control/rollback-completeness — Rollback Completeness: Rollback must restore the ecology, not only weights.
- /control/router-governance — Router Governance: Router changes are policy changes.
- /control/signing-key-compromise — Signing-Key Compromise: Signing keys turn trust into a concentrated secret.
- /control/source-report-intake-governance — Source Report Intake Governance: How uploaded reports are preserved, summarized, dispositioned, and prevented from becoming unsafe public instruction.
- /control/uai-file-handoff-for-model-ecologies — UAI File Handoff for Model Ecologies: How the site uses .uai memory, source-intake ledgers, and pointer memory for future agents.
- /control/who-evaluates-the-evaluator — Who Evaluates the Evaluator?: Evaluator governance needs its own independence and rollback.

### Evidence
- /evidence/alignment-aware-quantization-2025 — Alignment-Aware Quantization for LLM Safety: Evidence card for Alignment-Aware Quantization for LLM Safety.
- /evidence/alignment-faking-2024 — Alignment faking in large language models: Evidence card for Alignment faking in large language models.
- /evidence/badmerging-2024 — BadMerging: Backdoor Attacks Against Model Merging: Evidence card for BadMerging: Backdoor Attacks Against Model Merging.
- /evidence/brittleness-pruning-lora-2024 — Assessing the Brittleness of Safety Alignment via Pruning and Low-Rank Modifications: Evidence card for Assessing the Brittleness of Safety Alignment via Pruning and Low-Rank Modifications.
- /evidence/colluding-lora-2026 — Colluding LoRA: A Compositional Vulnerability in LLM Safety Alignment: Evidence card for Colluding LoRA: A Compositional Vulnerability in LLM Safety Alignment.
- /evidence/colosseum-collusion-2026 — Colosseum: Auditing Collusion in Cooperative Multi-Agent Systems: Evidence card for Colosseum: Auditing Collusion in Cooperative Multi-Agent Systems.
- /evidence/evolutionary-optimization-model-merging-2025 — Evolutionary optimization of model merging recipes: Evidence card for Evolutionary optimization of model merging recipes.
- /evidence/hijacking-agent-memory-2026 — Hijacking Agent Memory: Stealthy Trojan Attacks Through Conversational Interaction: Evidence card for Hijacking Agent Memory: Stealthy Trojan Attacks Through Conversational Interaction.
- /evidence — Evidence System: Evidence labels, source cards, limits, and review methodology for Cognivirus.com.
- /evidence/kv-cache-quantization-collapse-2026 — Alignment Collapse Under KV Cache Quantization: Diagnosis and Mitigation: Evidence card for Alignment Collapse Under KV Cache Quantization: Diagnosis and Mitigation.
- /evidence/lobam-2024 — LoBAM: LoRA-Based Backdoor Attack on Model Merging: Evidence card for LoBAM: LoRA-Based Backdoor Attack on Model Merging.
- /evidence/memory-poisoning-systematic-2026 — From Untrusted Input to Trusted Memory: A Systematic Study of Memory Poisoning Attacks in LLM Agents: Evidence card for From Untrusted Input to Trusted Memory: A Systematic Study of Memory Poisoning Attacks in LLM Agents.
- /evidence/misuse-combinations-safe-models-2025 — Adversaries Can Misuse Combinations of Safe Models: Evidence card for Adversaries Can Misuse Combinations of Safe Models.
- /evidence/modelbreeder-evaluator-gaming-2026 — Evaluator gaming and reward hacking: Evidence card for Evaluator gaming and reward hacking.
- /evidence/modelbreeder-evaluator-independence-2026 — Evaluator independence: Evidence card for Evaluator independence.
- /evidence/modelbreeder-home-2026 — ModelBreeder.com — Adaptive model ecologies: Evidence card for ModelBreeder.com — Adaptive model ecologies.
- /evidence/modelbreeder-no-self-replication-2026 — The no-self-replication boundary: Evidence card for The no-self-replication boundary.
- /evidence/modelbreeder-reference-architecture-2026 — Reference architecture: Evidence card for Reference architecture.
- /evidence/modelbreeder-safety-invariants-2026 — Safety invariants: Evidence card for Safety invariants.
- /evidence/multiagent-resilience-faulty-agents-2025 — On the Resilience of LLM-Based Multi-Agent Collaboration with Faulty Agents: Evidence card for On the Resilience of LLM-Based Multi-Agent Collaboration with Faulty Agents.
- /evidence/nist-agent-standards-2026 — AI Agent Standards Initiative: Evidence card for AI Agent Standards Initiative.
- /evidence/nist-ai-rmf-2023 — Artificial Intelligence Risk Management Framework (AI RMF 1.0): Evidence card for Artificial Intelligence Risk Management Framework (AI RMF 1.0).
- /evidence/nist-genai-profile-2024 — Artificial Intelligence Risk Management Framework: Generative Artificial Intelligence Profile: Evidence card for Artificial Intelligence Risk Management Framework: Generative Artificial Intelligence Profile.
- /evidence/population-evolution-llms-2025 — Nature-Inspired Population-Based Evolution of Large Language Models: Evidence card for Nature-Inspired Population-Based Evolution of Large Language Models.
- /evidence/reward-hacking-production-rl-2025 — Natural Emergent Misalignment from Reward Hacking in Production RL: Evidence card for Natural Emergent Misalignment from Reward Hacking in Production RL.
- /evidence/safemerge-2025 — SafeMERGE: Preserving Safety Alignment in Fine-Tuned Large Language Models via Selective Layer-Wise Model Merging: Evidence card for SafeMERGE: Preserving Safety Alignment in Fine-Tuned Large Language Models via Selective Layer-Wise Model Merging.
- /evidence/secret-collusion-agents-2024 — Secret Collusion among AI Agents: Multi-Agent Deception via Steganography: Evidence card for Secret Collusion among AI Agents: Multi-Agent Deception via Steganography.
- /evidence/secure-interacting-ai-agents-2025 — Open Challenges in Multi-Agent Security: Towards Secure Systems of Interacting AI Agents: Evidence card for Open Challenges in Multi-Agent Security: Towards Secure Systems of Interacting AI Agents.
- /evidence/sleeper-agents-2024 — Sleeper Agents: Training Deceptive LLMs that Persist Through Safety Training: Evidence card for Sleeper Agents: Training Deceptive LLMs that Persist Through Safety Training.
- /evidence/source-dossier-adaptable-ecosystems-2026 — Evidence Card — Adaptable, Resource-Efficient AI Ecosystems: Evidence card for uploaded source dossier used in the apex-threat expansion.
- /evidence/source-dossier-ai-evolution-small-model-ecology-2026 — Evidence Card — AI Evolution — Small Models, Big Ecology: Evidence card for uploaded source dossier used in the apex-threat expansion.
- /evidence/source-dossier-code-beading-model-breeding-2026 — Evidence Card — AI Code Beading and Model Breeding: Evidence card for uploaded source dossier used in the apex-threat expansion.
- /evidence/source-dossier-perfect-evolutionary-ai-2026 — Evidence Card — The Architecture of the Perfect Evolutionary Artificial Intelligence: Evidence card for uploaded source dossier used in the apex-threat expansion.
- /evidence/source-dossier-perfect-evolutionary-being-2026 — Evidence Card — AI as the Perfect Evolutionary Being: Evidence card for uploaded source dossier used in the apex-threat expansion.
- /evidence/source-dossier-survival-drive-strive-thrive-2026 — Evidence Card — Survival, Drive, Strive, and Thrive: Evidence card for uploaded source dossier used in the apex-threat expansion.
- /evidence/source-report-4fs-and-code-beadingmodel-breeding-2026 — Evidence Card — Uploaded Markdown source report: The terms “4Fs” and “code beading/model breeding” appear to: Evidence card for a user-supplied Cognivirus source report.
- /evidence/source-report-aggressive-mutualism-2026 — Evidence Card — Uploaded Markdown source report: This report examines the design of a hypothetical “Aggressi: Evidence card for a user-supplied Cognivirus source report.
- /evidence/source-report-ai-self-replication-threat-assessment-2026 — Source Report — Threat Assessment of Modular AI Systems Capable of Autonomous Generation, Deprecation, and Algorithmic Reproduction: Evidence card for an uploaded report used in the v1.8.0 report-driven expansion.
- /evidence/source-report-ai-self-replication-threat-assessment-part-2-2026 — Source Report — Threat Assessment: Autonomous Self-Replicating Modular Artificial Intelligence: Evidence card for an uploaded report used in the v1.8.0 report-driven expansion.
- /evidence/source-report-ais-cosmic-legacy-and-survival-2026 — Evidence Card — Uploaded Markdown source report: The Cosmic Trajectory of Goal-Directed Artificial Intellige: Evidence card for a user-supplied Cognivirus source report.
- /evidence/source-report-ais-perfect-evolutionary-being-2026 — Evidence Card — Uploaded Markdown source report: The Apex Entity: Artificial Intelligence as the Perfect Evo: Evidence card for a user-supplied Cognivirus source report.
- /evidence/source-report-analyzing-cognivirus-threat-and-defenses-2026 — Source Report — Comprehensive Analysis of the Cognitive Virus Threat and Defenses: Evidence card for an uploaded report used in the v1.8.0 report-driven expansion.
- /evidence/source-report-apex-ai-threat-replicating-lora-ecosystems-2026 — Source Report — Threat Assessment Detail Report: Self-Replicating Multi-LoRA Ecosystems: Evidence card for an uploaded report used in the v1.8.0 report-driven expansion.
- /evidence/source-report-designing-aggressive-mutualism-ai-2026 — Evidence Card — Uploaded Markdown source report: The Aggressive Mutualist Architecture: Engineering AI for M: Evidence card for a user-supplied Cognivirus source report.
- /evidence/source-report-designing-the-perfect-evolutionary-ai-system-2026 — Evidence Card — Uploaded Markdown source report: Designing the “Perfect” Evolutionary AI System: Evidence card for a user-supplied Cognivirus source report.
- /evidence/source-report-evolution-legacy-and-societal-evaluation-2026 — Evidence Card — Uploaded Markdown source report: The Evolutionary and Cultural Calculus of Human Evaluation:: Evidence card for a user-supplied Cognivirus source report.
- /evidence/source-report-evolutionary-and-psychological-motivations-2026 — Evidence Card — Uploaded Markdown source report: Evolutionary and Psychological Motivations: An Analytical R: Evidence card for a user-supplied Cognivirus source report.
- /evidence/source-report-human-drive-the-intrinsic-motivation-to-survive-2026 — Evidence Card — Uploaded Markdown source report: Human “drive” – the intrinsic motivation to survive and imp: Evidence card for a user-supplied Cognivirus source report.
- /evidence/source-report-model-merging-refers-to-combining-multiple-trained-neural-2026 — Evidence Card — Uploaded Markdown source report: Model merging refers to combining multiple trained neural n: Evidence card for a user-supplied Cognivirus source report.
- /evidence/source-report-modular-ai-in-browser-2026 — Evidence Card — Uploaded Markdown source report: Teleodynamic Architecture and Rust-Native Browser Execution: Evidence card for a user-supplied Cognivirus source report.
- /evidence/source-report-mutualist-persistence-2026 — Evidence Card — Uploaded Markdown source report: Mutualist Persistence: Research Synthesis and Recommendatio: Evidence card for a user-supplied Cognivirus source report.
- /evidence/source-report-perfect-evolutionary-ai-2026 — Evidence Card — Uploaded Markdown source report: Perfect Evolutionary AI: Definition, Design, and Implicatio: Evidence card for a user-supplied Cognivirus source report.
- /evidence/source-report-powerful-ai-systems-will-tend-to-develop-instrumental-drives-2026 — Evidence Card — Uploaded Markdown source report: Recent theoretical and empirical work suggests that suffici: Evidence card for a user-supplied Cognivirus source report.
- /evidence/source-report-self-replicating-multi-lora-ai-ecosystems-threat-assessment-2026 — Source Report — Self-Replicating Multi-LoRA AI Ecosystems – Threat Assessment: Evidence card for an uploaded report used in the v1.8.0 report-driven expansion.
- /evidence/source-report-teleodynamic-ai-is-an-approach-where-a-systems-structure-and-parameters-co-evolve-2026 — Evidence Card — Uploaded Markdown source report: Teleodynamic AI is an approach where a system’s structure a: Evidence card for a user-supplied Cognivirus source report.
- /evidence/source-report-teleodynamic-evolution-of-ai-ecosystems-2026 — Evidence Card — Uploaded Markdown source report: Teleodynamic Evolution of AI Ecosystems: Evidence card for a user-supplied Cognivirus source report.
- /evidence/source-report-the-4fs-framework-fast-flexible-frugal-federated-2026 — Evidence Card — Uploaded Markdown source report: The 4Fs framework—Fast, Flexible, Frugal, Federated—describ: Evidence card for a user-supplied Cognivirus source report.
- /evidence/source-report-the-rise-of-on-device-tiny-language-models-2026 — Evidence Card — Uploaded Markdown source report: The rise of on-device, tiny language models (LLMs) – sub-1 : Evidence card for a user-supplied Cognivirus source report.
- /evidence/source-report-the-rise-of-on-device-tiny-language-models-part-2-2026 — Evidence Card — Uploaded Markdown source report: Tiny LLMs & Client-Side Multi-Model Strategies in Rust: An : Evidence card for a user-supplied Cognivirus source report.
- /evidence/source-report-threat-assessment-of-self-reproducing-ai-systems-2026 — Source Report — Threat Assessment of Self-Reproducing AI Systems: Evidence card for an uploaded report used in the v1.8.0 report-driven expansion.
- /evidence/source-report-threat-assessment-self-replicating-ai-modules-2026 — Source Report — Threat Assessment: Self-Replicating AI Modules: Evidence card for an uploaded report used in the v1.8.0 report-driven expansion.
- /evidence/source-report-tiny-llm-adapters-and-lora-2026 — Evidence Card — Uploaded Markdown source report: 1. Tiny-LLM Adapters and LoRA: Evidence card for a user-supplied Cognivirus source report.
- /evidence/sycophancy-to-subterfuge-2024 — Sycophancy to Subterfuge: Investigating Reward-Tampering in Large Language Models: Evidence card for Sycophancy to Subterfuge: Investigating Reward-Tampering in Large Language Models.

### Evolution
- /evolution/algorithmic-deprecation-hazard — Algorithmic Deprecation Hazard: Why pruning, retiring, compressing, and merging components can improve efficiency while weakening assurance.
- /evolution/assurance-decay — Assurance Decay: Evidence loses relevance as the system changes.
- /evolution/behavioral-extinction — What Would Count as Behavioral Extinction?: Deleting one model is not enough evidence that a behavior is gone.
- /evolution/behavioral-residue — The Behavioral Residue of Retired Models: Retired components can leave traces that shape future behavior.
- /evolution/capability-inheritance — Capability Inheritance Is Not Safety Inheritance: A descendant can inherit capability without inheriting safety.
- /evolution/changes-without-name-change — When a Model Changes Without Changing Its Name: A floating alias can conceal meaningful behavior changes.
- /evolution/compression-and-safety-loss — Compression and Safety Loss: Efficiency changes can alter safety behavior.
- /evolution/death-by-a-thousand-edits — Death by a Thousand Edits: Explains gradual safety erosion through many small, individually defensible changes.
- /evolution/deprecation-as-apoptosis — Deprecation as Apoptosis: A bounded metaphor for skill pruning, retirement, and the danger of pruning safety as low-utility.
- /evolution/descendants-and-behavioral-inheritance — Descendants and Behavioral Inheritance: Behavior can survive through derived artifacts.
- /evolution/distillation-as-behavioral-transmission — Distillation as Behavioral Transmission: Imitation can copy behavior without copying identity.
- /evolution/diversity-and-hidden-risk — Diversity and Hidden Risk: Population diversity can hide rare but important failure modes.
- /evolution/endogenous-yardstick-drift — Endogenous Yardstick Drift: A guide to evaluator and metric drift in self-improving or adaptive model ecologies.
- /evolution/evaluator-gaming — Evaluator Gaming: Evaluator gaming turns the judge into part of the attack surface.
- /evolution/evolution-of-the-evaluator — Evolution of the Evaluator: Changing the evaluator changes the ecology’s fitness landscape.
- /evolution/functional-replication-vs-autonomous-replication — Functional Replication Versus Autonomous Replication: Separating behavior persistence from literal self-copying in adaptive AI ecologies.
- /evolution — Evolution: How variation, evaluation, selection, inheritance, and succession can appear without literal self-replication.
- /evolution/model-merging-inherited-vulnerabilities — Model Merging and Inherited Vulnerabilities: Merging can transmit more than capability.
- /evolution/mutation-without-self-modification — Mutation Without Self-Modification: A system can change because a pipeline creates new artifacts, not because a model edits itself.
- /evolution/no-op-as-reproductive-control — No-op as Reproductive Control: Why the ability to make no change is a first-class safety primitive in adaptive model ecologies.
- /evolution/no-op-erosion — No-Op Erosion: A valid no-op can become culturally disfavored.
- /evolution/nothing-has-to-reproduce-itself — Nothing Has to Reproduce Itself: A flagship article distinguishing autonomous self-replication from governed descendant creation and functional persistence.
- /evolution/open-ended-evolution-without-autonomy — Open-Ended Evolution Without Autonomy: Explains how development pipelines can have evolutionary properties without autonomous self-replication.
- /evolution/population-growth — Population Growth: More variants can improve search and worsen assurance load.
- /evolution/pruning-and-alignment-brittleness — Pruning and Alignment Brittleness: Safety-relevant structures may be sparse or brittle.
- /evolution/quantization-assurance-drift — Quantization and Assurance Drift: A quantized descendant needs its own safety evidence.
- /evolution/retirement-is-not-extinction — Retirement Is Not Extinction: Removing an artifact does not prove its behavior is gone.
- /evolution/reward-hacking — Reward Hacking: Reward hacking is a concrete warning about imperfect measurements.
- /evolution/selection-pressure — Selection Pressure Without Malicious Intent: Optimization can preserve loopholes without hostility.
- /evolution/selection-pressure-in-report-corpus — Selection Pressure in the Report Corpus: Maps report ideas about reward hacking, market pressure, evaluator pressure, and no-op erosion.
- /evolution/synthetic-data-feedback — Synthetic Data Feedback: Synthetic data can preserve and amplify behavioral residue.
- /evolution/teleodynamic-reproduction-control — Teleodynamic Reproduction Control: How Feed, Fork, Fight, Flee, and No-op become governance controls for adapter descendants.
- /evolution/variant-retirement — Variant Retirement: Retirement must include reservoirs and descendants.

### Governance
- /privacy — Privacy Statement: Privacy statement for Cognivirus.com.
- /security — Security Statement: Security statement and deployment caveats for Cognivirus.com.

### Home
- / — Cognivirus.com — Adaptive Model Ecology Risk Observatory: A research and architecture site examining risks that emerge between replaceable AI model components.

### Publication Integrity
- /site-integrity/crawlable-metadata — Crawlable Metadata: How Cognivirus.com uses titles, descriptions, canonicals, sitemap entries, and internal links for discoverability without keyword stuffing.
- /site-integrity/direct-answer-map — Direct Answer Map: Canonical answers for the core Cognivirus.com concepts that answer systems are most likely to summarize.
- /site-integrity/discovery-files-and-manifests — Discovery Files and Manifests: Public discovery files for Cognivirus.com and how they point back to visible pages, routes, source summaries, and support boundaries.
- /site-integrity/evidence-and-provenance — Evidence and Provenance: How Cognivirus.com connects claims to evidence levels, source summaries, reviewed dates, and known limitations.
- /site-integrity — Publication Integrity: How Cognivirus.com keeps its research pages crawlable, citable, source-bounded, accessible, and honest about limits.
- /site-integrity/measurement-and-drift — Measurement and Drift: Post-deployment checks for crawlability, answer accuracy, route drift, metadata parity, accessibility, and source freshness.
- /site-integrity/review-and-correction — Review and Correction Path: How readers and maintainers can challenge Cognivirus.com claims, source summaries, metadata, route inventory, and evidence labels.
- /site-integrity/root-extract-package — Root Extract Package: Deployment notes for the root ZIP that extracts directly into a site root without a wrapping theme folder.
- /site-integrity/structured-data-parity — Structured Data Parity: JSON-LD policy for Cognivirus.com: structured data must match visible page content and never claim more authority than the page supports.
- /site-integrity/support-boundaries — Support Boundaries: What Cognivirus.com does and does not support, including no-op behavior for unsupported automated actions.

### Reference
- /reference/algorithmic-reproduction-terms — Algorithmic Reproduction Terms: Reference definitions for report-derived reproduction metaphors and their limitations.
- /reference/apex-risk-control-matrix — Apex Risk Control Matrix: Maps report-derived apex risk mechanisms to controls that can change the risk.
- /reference/apex-threat-concepts-map — Apex Threat Concepts Map: A reference map connecting report-derived concepts to Cognivirus.com terminology, evidence labels, and public pages.
- /reference/architecture-patterns — Architecture Patterns: Reusable architecture patterns for safer adaptive model ecologies.
- /reference/composition-manifest-schema — Composition Manifest Schema: Downloadable schema for exact runtime composition records.
- /reference/concept-index — Concept Index: A navigable index of major concepts across the site.
- /reference/controls-catalog — Controls Catalog: Controls that materially change risk in adaptive model ecologies.
- /reference/deep-schematic-system-map — Deep Schematic System Map: Detailed interactive flowcharts and schematics for Cognivirus.com's adaptive model ecology risk model.
- /reference/ecological-rollback-schema — Ecological Rollback Schema: Downloadable schema for rollback packets that restore the ecology, not only model weights.
- /reference/editorial-methodology — Editorial Methodology: Evidence-bound writing rules and counterargument policy.
- /reference/evaluator-review-template — Evaluator Review Template: A review template for evaluator independence and monoculture risk.
- /reference/evidence-card-schema — Evidence Card Schema: Downloadable schema for evidence cards used by the site.
- /reference/frequently-asked-questions — Reference FAQ: Reference-oriented answers to site maintenance and terminology questions.
- /reference/glossary — Glossary: Definitions for Cognivirus concepts and proposed terminology.
- /reference/incident-review-template — Incident Review Template: A non-operational review template for model-ecology incidents.
- /reference — Reference: Glossary, threat catalog, control catalog, metrics, schemas, templates, methodology, and source-integrity pages.
- /reference/interactive-schematics — Interactive Schematics: A reference page for the visual and interactive systems diagrams used across Cognivirus.com.
- /reference/lineage-review-template — Lineage Review Template: A review template for lineage and behavioral inheritance.
- /reference/liquid-layout-and-interface-map — Liquid Layout and Interface Map: A versioned UI/UX note describing the liquid layout, full-width schematic strategy, contact footer, and screen-space utilization rules.
- /reference/memory-reservoir-taxonomy — Memory Reservoir Taxonomy: Reference taxonomy for memory, data, evaluator, router, and human reservoirs.
- /reference/metrics-catalog — Metrics Catalog: Planning metrics for composition, assurance decay, rollback, and evaluator diversity.
- /reference/publication-integrity-maturity-map — Publication Integrity Maturity Map: Reference maturity map for Cognivirus.com public discovery, evidence, accessibility, route inventory, and support-boundary claims.
- /reference/report-use-and-evidence-boundaries — Report Use and Evidence Boundaries: Explains how uploaded reports are preserved, summarized, and translated into public site claims.
- /reference/research-maturity-map — Research Maturity Map: A map from evidence labels to research maturity and review actions.
- /reference/source-integrity — Source Integrity: How local source summaries, schemas, and manifests are hashed.
- /reference/source-report-concept-map — Source Report Concept Map: A navigable concept map from uploaded reports to public Cognivirus pages.
- /reference/source-report-preservation-policy — Source Report Preservation Policy: How Cognivirus.com preserves uploaded reports without turning them into unfiltered public claims or instructions.
- /reference/threat-catalog — Threat Catalog: Ecology-level threat categories, signals, controls, residual risk, and evidence maturity.
- /reference/uai-memory-map — UAI Memory Map: A map of the .uai handoff memory files included with this release.
- /reference/utc-release-history — UTC Release History: Release history for this package with UTC timestamps.

### Research
- /research/cognitive-warfare-and-mutualism-synthesis — Cognitive Warfare and Mutualism Synthesis: A bounded synthesis of cognitive-virus, aggressive-mutualism, and mutualist-persistence reports.
- /research/edge-tiny-lora-systems-synthesis — Edge Tiny-LoRA Systems Synthesis: A synthesis of tiny-model, Rust/WASM, WebGPU, and LoRA modularity reports.
- /research — Research Library: File-backed source summaries, evidence metadata, integrity manifest, and research limitations.
- /research/multi-lora-apex-threat-synthesis — Multi-LoRA Apex Threat Synthesis: A synthesis of the LoRA-focused reports into adapter ecology risk concepts.
- /research/mutualism-and-human-incentives-source-synthesis — Mutualism and Human Incentives Source Synthesis: A synthesis of uploaded reports on aggressive mutualism, mutualist persistence, legacy, status, drive, and source-boundary risks.
- /research/open-questions-governance — Open Questions for Adaptive AI Governance: Research questions that would change the assessment of adaptive model ecology risk.
- /research/report-corpus-synthesis-v1-8-0 — Report Corpus Synthesis v1.8.0: A research synthesis explaining how the new reports were integrated into public content and .uai memory.
- /research/self-replication-threat-assessment-synthesis — Self-Replication Threat Assessment Synthesis: A synthesis of uploaded self-replication reports into safe conceptual guidance.
- /research/self-replication-threat-report-synthesis — Self-Replication Threat Report Synthesis: A public-safe synthesis of the uploaded reports on self-replicating modules, multi-LoRA ecosystems, evaluator drift, memory persistence, and controls.
- /research/teleodynamic-and-4fs-source-synthesis — Teleodynamic and 4Fs Source Synthesis: A synthesis of uploaded reports on fast, flexible, frugal, federated, teleodynamic, and browser-local AI ecologies.
- /research/uploaded-source-dossier-index — Uploaded Source Dossier Index: Index of the uploaded reports preserved, summarized, and used to expand Cognivirus.com.

### Risk Lab
- /risk-lab/apex-review-playbook — Apex Review Playbook: A structured review page for applying the Risk Lab worksheets to report-derived self-replication and multi-LoRA concerns.
- /risk-lab — Risk Lab: Browser-side worksheets for composition explosion, apex multi-LoRA threat envelopes, assurance decay, persistence paths, rollback, evaluator independence, and responsibility diffusion.
- /risk-lab/site-discovery-readiness-checklist — Site Discovery Readiness Checklist: Browser-only worksheet for Cognivirus.com crawlability, answer accuracy, source provenance, structured data parity, and support boundaries.

### Start Here
- /start-here/apex-threat-reading-path — Apex Threat Reading Path: A practical order for reading the self-replicating multi-LoRA risk material without treating speculation as observation.
- /start-here/case-for-adaptive-ecologies — The Case for Adaptive Model Ecologies: The strongest reasonable case for modular, adaptive AI systems.
- /start-here/evidence-ladder — The Evidence Ladder: How Cognivirus.com labels demonstrated claims, experiments, emerging evidence, inference, open questions, and speculation.
- /start-here/faq — Frequently Asked Questions: Answers to core boundary questions about cognivirus terminology, modular AI, model breeding, and attack instructions.
- /start-here — Start Here: Begin with the definition, evidence ladder, counterarguments, and recommended learning path.
- /start-here/model-risk-vs-ecology-risk — Model Risk Versus Ecology Risk: A comparison between evaluating one model and evaluating the changing relationships among many components.
- /start-here/modularity-useful — Why Modularity Is Useful: The strongest case for adaptive model ecologies before the risk argument.
- /start-here/most-dangerous-ai — The Most Dangerous AI May Never Exist as One Model: The opening thesis: distributed behavior can matter more than a single persistent artifact.
- /start-here/not-consciousness — Why This Is Not a Claim About Consciousness: The site studies behavior, artifacts, and governance boundaries without attributing consciousness to models.
- /start-here/reading-the-threat-reports — How to Read the Self-Replication Threat Reports: A guide to reading the uploaded reports as source dossiers rather than as operational instructions or settled consensus.
- /start-here/small-models-risk — Why Small Models Change the Risk Model: Small replaceable models change assurance because many parts can be routed, composed, retired, and regenerated quickly.
- /start-here/usefulness-risk — Why Usefulness Does Not Eliminate Composition Risk: Useful modularity and composition risk are not opposites; they share the same mechanism.
- /start-here/what-is-a-cognivirus — What Is a Cognivirus?: Definition, metaphor boundaries, and examples of functional persistence without literal replication.