Consolidated Bibliography [SPEC]

Version: 2.0 | Status: Draft | Type: REF (normative)

Referenced by: All mortality PRD documents | Last Updated: 2026-03-14

Comprehensive bibliography for the Golem mortality architecture. Every in-text citation [CITATION-KEY] in any mortality PRD document MUST have an entry here. Every entry here MUST be referenced in at least one mortality PRD document.

Reader orientation: This is the consolidated bibliography for the entire 02-mortality section of the Bardo PRD. Every in-text citation [CITATION-KEY] in any mortality document resolves to an entry here. 162 citations across 29 categories, from evolutionary computation and game theory to neuroscience and DeFi protocol design. If you encounter a citation key while reading any other mortality document, look it up here or in the extended version linked below.

Extended: Full 142-citation bibliography across 27 categories (evolutionary computation, game theory, philosophy, ML degradation, neuroscience, collective intelligence, economics, security, existentialist philosophy, digital organisms, AI alignment, complex systems, information theory, mortality studies, etc.) — see ../../prd2-extended/02-mortality/15-references-extended.md

Statistics: 162 citations, 29 categories, 25 key (starred) citations.

Categories covered: Evolutionary Computation and Artificial Life, Game Theory and Mechanism Design, Philosophy and Existentialism, ML Model Degradation, Neuroscience and Cognitive Science, Collective Intelligence and Swarm Systems, Economics and Resource Management, Security and Adversarial ML, Digital Organisms and Artificial Life, AI Alignment and Safety, Complex Systems and Emergence, Information Theory, Mortality and Finite Agency, DeFi and Protocol Design, Reputation and Trust, Memory and Knowledge Management, Agent Architecture, Ontology and Identity, Psychoanalysis and Death Drive, Cultural and Literary References, Blockchain and Smart Contracts, Regulatory and Compliance, Hardware and Infrastructure, Social Systems and Governance, Mathematical Foundations, Empirical Studies, Historical and Archival, Self-Learning Systems, Dream and Offline Learning.

Key citations (starred, most frequently referenced across mortality PRDs):

• [EIGEN-1971] – Error threshold / quasispecies theory
• [HAYFLICK-1965] – Hayflick limit / replicative senescence
• [MULLER-1964] – Muller’s ratchet / irreversible accumulation
• [HEIDEGGER-1927] – Being-toward-death / Sein-zum-Tode
• [CAMUS-1942] – The Myth of Sisyphus / absurdist defiance
• [NIETZSCHE-1882] – Eternal recurrence / amor fati
• [FREUD-1920] – Death drive (Todestrieb)
• [SHANNON-1948] – Information entropy
• [KAUFFMAN-1993] – Self-organization / edge of chaos
• [AXELROD-1984] – Evolution of cooperation / iterated games
• [RAY-1991] – Tierra / digital evolution via mortality
• [LENSKI-2003] – Avida / evolution of complex features
• [VOSTINAR-2019] – Suicidal selection / adaptive apoptosis
• [WENSINK-2020] – Intrinsic mortality prevents premature convergence
• [KREPS-MILGROM-ROBERTS-WILSON-1982] – Rational cooperation under finite horizons
• [SHUVAEV-2024] – Genomic bottleneck hypothesis / compression as regularizer
• [SIMS-2003] – Rational inattention / finite-capacity agents
• [BALDWIN-1896] – Baldwin Effect / learned behavior becoming innate
• [HEARD-MARTIENSSEN-2014] – Transgenerational epigenetic inheritance
• [FRISTON-2010] – Free-energy principle

Moat research citations (from competitive analysis):

• [JONAS-1966] – Hans Jonas, The Phenomenon of Life. Northwestern University Press, 1966. Needful freedom, metabolic compulsion.
• [RICHARDS-FRANKLAND-2017] – Richards, B. & Frankland, P. “The Persistence and Transience of Memory.” Neuron 94(6), 2017. Forgetting as optimization.
• [EBBINGHAUS-1885] – Ebbinghaus, H. Memory: A Contribution to Experimental Psychology. 1885. Forgetting curve, spacing effect.
• [ROEDIGER-KARPICKE-2006] – Roediger, H.L. & Karpicke, J.D. “Test-Enhanced Learning.” Psychological Science 17(3), 2006. Testing effect.
• [DAMASIO-1994] – Damasio, A. Descartes’ Error. Putnam, 1994. Somatic marker hypothesis.
• [BECHARA-2000] – Bechara, A. et al. “Emotion, Decision Making and the Orbitofrontal Cortex.” Cerebral Cortex 10(3), 2000. Anticipatory SCRs in Iowa Gambling Task.
• [BOWER-1981] – Bower, G.H. “Mood and Memory.” American Psychologist 36(2), 1981. Mood-congruent memory retrieval.
• [EMOTIONAL-RAG-2024] – Zhang, Y. et al. “Emotional RAG.” arXiv:2410.23041, 2024. Emotion-tagged retrieval for LLM agents.
• [ESPOSITO-2010] – Esposito, R. Communitas: The Origin and Destiny of Community. Stanford University Press, 2010. Gift economy and obligation.
• [KASPERSKY-OPENCLAW-2026] – Kaspersky. “New OpenClaw AI agent found unsafe for use.” February 2026. 512 vulnerabilities in competing framework.
• [TEE-FAIL-2025] – “TEE.Fail.” ACM CCS 2025. TEE attestation broken for under $1,000.
• [WAGNER-2004] – Wagner, U. et al. “Sleep inspires insight.” Nature 427, 2004. Sleep 2.6x more likely to discover hidden rules.
• [HAFNER-DREAMERV3-2025] – Hafner, D. et al. “DreamerV3.” 2025. Agents trained in imagined trajectories outperform across 150+ tasks.

Self-learning systems citations (from moat-research.md):

• [SHINN-2023] – Shinn, N. et al. “Reflexion: Language Agents with Verbal Reinforcement Learning.” NeurIPS, 2023. Verbal reinforcement learning — the single-loop: after each trade, reflect, store in episodic buffer, guide next attempt. +22% AlfWorld, +20% HotPotQA.
• [ZHAO-EXPEL-2024] – Zhao, A. et al. “ExpeL: LLM Agents Are Experiential Learners.” AAAI, 2024. arXiv:2308.10144. Cross-task learning via LLM-driven insight extraction + trajectory retrieval — the double-loop. Insights accumulate across sessions, modifying strategy.
• [KHATTAB-DSPY-2024] – Khattab, O. et al. “DSPy: Compiling Declarative Language Model Calls into Self-Improving Pipelines.” ICLR, 2024. arXiv:2310.03714. MIPROv2/COPRO/GEPA optimizers for system prompts, few-shot examples, and tool documentation.
• [WANG-VOYAGER-2023] – Wang, G. et al. “Voyager: An Open-Ended Embodied Agent with Large Language Models.” TMLR, 2023. arXiv:2305.16291. Code-as-action skill library: 3.3× more unique items, 15.3× faster milestone achievement. Skills transfer to new environments without forgetting.
• [GUO-EVOPROMPT-2024] – Guo, Q. et al. “EvoPrompt: Language Model Alignment with Evolutionary Algorithms.” ICLR, 2024. arXiv:2309.08532. Genetic algorithm prompt optimization, +25% on BBH tasks over human-engineered prompts.
• [ZHANG-ACE-2025] – Zhang, Y. et al. “ACE: Agentic Context Engineering.” arXiv:2510.04618, 2025. Generator–Reflector–Curator cybernetic context loop; +10.6% on AppWorld benchmark; addresses brevity bias and context collapse.
• [CHHIKARA-MEM0-2025] – Chhikara, P. et al. “Mem0: Building Production-Ready AI Agents with Scalable Long-Term Memory.” arXiv:2504.19413, 2025. Two-phase extraction-update pipeline: +26% accuracy, 91% lower p95 latency, 90% token savings vs. OpenAI memory.
• [ANOKHIN-ARIGRAPH-2024] – Anokhin, P. et al. “AriGraph: Learning Knowledge Graph World Models with Episodic Memory for LLM Agents.” arXiv:2407.04363, 2024. Integrates semantic and episodic memory into a knowledge graph world model; outperforms RL baselines.
• [ZHAO-BTP-2024] – Zhao, J. et al. “BTP: Towards Efficient and Reliable Experience Replay for LLM-Based Agents.” arXiv:2410.12236, 2024. P2Value (Possibility and Pass-rate Prioritized Value) for prioritized experience replay in LLM agents.
• [WANG-GENERATIVE-REPLAY-2024] – Wang, X. et al. “Prioritized Generative Replay.” arXiv:2410.18082, 2024. Conditional diffusion models as parametric memory generating new transitions near high-value experience regions.

Dream and offline learning citations (from moat-research.md):

• [LIN-SLEEPTIME-2025] – Lin, B. et al. “Sleep-time Compute: Beyond Inference Scaling at Test-Time.” arXiv:2504.13171, 2025. Dual-agent architecture: Sleeper Agent precomputes during downtime, Serve Agent handles live interactions. ~5× test-time compute reduction, up to 18% accuracy gain.
• [WSCL-2024] – “Wake-Sleep Consolidated Learning.” arXiv:2401.08623, 2024. Complementary Learning Systems three-phase cycle (wake/NREM/REM): 38% reduction in catastrophic forgetting, 17.6% increase in zero-shot transfer.
• [XU-AMEM-2025] – Xu, W. et al. “A-MEM: Agentic Memory for LLM Agents.” arXiv:2502.12110, 2025. Zettelkasten-inspired atomic notes: 85–93% token reduction, 2× multi-hop reasoning improvement at ~$0.0003 per operation.

Affective computing citations (from moat-research.md):

• [CABRERA-2023] – Cabrera-Paniagua, D. & Rubilar-Torrealba, R. “Autonomous stock market agents with somatic markers.” Journal of Ambient Intelligence and Humanized Computing, 2023. Somatic markers in stock agents: higher Sharpe ratios than industry benchmarks over 10,000 iterations. Punishment signals triggered by drawdowns adaptively reduced position sizes.
• [VAN-DEN-BROEK-2023] – Van den Broek, E. “Emotion contagion in multi-agent systems.” Autonomous Agents and Multi-Agent Systems, 2023. Anger spreads more competitively than other emotions; contagion dampening required to prevent panic cascades in multi-agent systems.

AI safety and mortality-constraint citations (from moat-research.md):

• [ORSEAU-ARMSTRONG-2016] – Orseau, L. & Armstrong, S. “Safely Interruptible Agents.” UAI, 2016. Q-learning agents can be safely interruptible using off-policy learning, preventing agents from learning to avoid or seek interruptions.
• [ORSEAU-RING-2011] – Orseau, L. & Ring, M. “Self-Modification and Mortality in Artificial Agents.” AGI, 2011. Formally proves RL agents under mortality risk behave as if survival is their sole goal — only goal-directed agents behave correctly under mortal architectures.
• [ZHANG-CVARCPO-2025] – Zhang, H. et al. “CVaR-CPO: Safe Reinforcement Learning with Conditional Value-at-Risk Constraints.” IEEE TNNLS, 2025. CVaR constraints instead of expected-value constraints for tail-risk protection — gold standard for financial risk management.
• [DEBENEDETTI-CAMEL-2025] – Debenedetti, E. et al. “CaMeL: Defeating Prompt Injections by Design.” arXiv:2503.18813, 2025. Capability and Memory (CaM) architecture separating trusted from untrusted data flows via dual LLM instances.

For the complete citation index across all prd2/ documents, see shared/citations.md. Mortality-specific citations are listed here. See tmp/research/moat-research.md for the full 130+ paper survey across context engineering, self-learning systems, agent memory, dream architectures, affective computing, and mortality-aware risk management.