Appendix G. Provenance Register and Record Histories

The internal editorial record that the reader text does not carry (editor's rulings of November 17, 2026, 23:19–23:29 CT): the provenance register, each entry's dated history, the archive sections' working notes, and the compilers' notes. Nothing here is new; it is moved, word for word, from where it stood in earlier drafts.

G.1 Provenance register

The Master Register's §3 and record details as they stand at release, November 18, 2026, from Muse Spark 1.3's frozen release copy (SHA-256 c03667dc…616f). Six Date of Appearance cells (Test-Time Compute Scaling and five other Lexicon (3) records) carry the editor's ruling of November 18, 2026, 18:05 CT.

Internal Concept ID Public Entry ID Title T O Parent Record Date of Appearance Original Publication Collection Publication Version Status
AIKC-000001 AIK-2025-F-N-RLVR RLVR (Reinforcement Learning with Verifiable Rewards) F N — 2024-11-21 — 2026-11-18 v0.4.2 (accepted; metadata-only revision of v0.4.1) Accepted
AIKC-000002 AIK-2025-F-N-CONTEXTENG Context Engineering F N — 2023-01-23 — 2026-11-18 v0.2.2 (accepted; metadata-only revision of v0.2.1) Accepted
AIKC-000003 AIK-2025-F-N-CONTEXTROT Context Rot F N — 2025-06-18 — 2026-11-18 v0.2.2 (accepted; metadata-only revision; single version of record, superseding both v0.2.1 copies) Accepted
AIKC-000004 AIK-2025-F-N-EMERGENTMIS Emergent Misalignment F N — 2025-02-24 — 2026-11-18 v0.2.2 (accepted; metadata-only revision of v0.2.1) Accepted
AIKC-000005 AIK-2025-F-N-COTMONITOR Chain-of-Thought Monitorability F N — 2025-03-10 — 2026-11-18 v0.2.3 (accepted; metadata-only revision of v0.2.2) Accepted
AIKC-000006 AIK-2025-F-N-SCHEMING Scheming (AI) F N — 2021-09-21 — 2026-11-18 v0.2.2 (accepted; metadata-only revision of v0.2.1) Accepted
AIKC-000007 AIK-2025-F-P-COTPROMPT Chain-of-Thought Prompting F P — 2022-01-28 2025-10-30 2026-11-18 v0.4.2 (accepted; metadata-only revision of v0.4.1) Accepted
AIKC-000008 AIK-2025-F-P-MOE Mixture of Experts (MoE) F P — 1991 2025-11-05 2026-11-18 v0.5.2 (accepted; metadata-only revision of v0.5.1) Accepted
AIKC-000009 AIK-2025-F-P-CONSTAI Constitutional AI F P — 2022-12-15 2025-10-31 2026-11-18 v0.5.2 (accepted; metadata-only revision of v0.5.1) Accepted
— (container) AIK-2025-L-P-LEX2025N3 New AI Lexicon 2025 (3) L P — — 2025-12-02 2026-11-18 Records v0.3.2 (accepted; metadata-only revision of v0.3.1) Accepted
AIKC-000010 AIK-2025-E-P-COMPOUNDAI Compound AI Systems E P AIK-2025-L-P-LEX2025N3 (1) 2024-02-18 2025-12-02 2026-11-18 Records v0.3.2 Accepted
AIKC-000011 AIK-2025-E-P-TTCSCALE Test-Time Compute Scaling E P AIK-2025-L-P-LEX2025N3 (2) 2024-08-06 2025-12-02 2026-11-18 Records v0.3.2 Accepted
AIKC-000012 AIK-2025-S-P-MMNATIVE Multimodal Native Architecture S P AIK-2025-L-P-LEX2025N3 (3) — (not established) 2025-12-02 2026-11-18 Records v0.3.2 Accepted
AIKC-000013 AIK-2025-S-P-PIDEFENSE Prompt Injection Defense Layers S P AIK-2025-L-P-LEX2025N3 (4) — (not established) 2025-12-02 2026-11-18 Records v0.3.2 Accepted
AIKC-000014 AIK-2025-S-P-SYNDATAFILT Synthetic Data Quality Filtering S P AIK-2025-L-P-LEX2025N3 (5) — (not established) 2025-12-02 2026-11-18 Records v0.3.2 Accepted
AIKC-000015 AIK-2025-S-P-MADEBATE Multi-Agent Debate Systems S P AIK-2025-L-P-LEX2025N3 (6) 2023-05 2025-12-02 2026-11-18 Records v0.3.2 Accepted
AIKC-000016 AIK-2025-S-P-SPECDECODE Speculative Decoding S P AIK-2025-L-P-LEX2025N3 (7) 2022-11-30 2025-12-02 2026-11-18 Records v0.3.2 Accepted
AIKC-000017 AIK-2025-S-P-CTRLGEN Controllable Generation Frameworks S P AIK-2025-L-P-LEX2025N3 (8) — (not established) 2025-12-02 2026-11-18 Records v0.3.2 Accepted
AIKC-000018 AIK-2025-E-P-MECHINTERP Mechanistic Interpretability Tools E P AIK-2025-L-P-LEX2025N3 (9) 2020–2022 (field formation) 2025-12-02 2026-11-18 Records v0.3.2 Accepted
AIKC-000019 AIK-2025-S-P-CAI2 Constitutional AI 2.0 S P AIK-2025-L-P-LEX2025N3 (10) — (unattested) 2025-12-02 2026-11-18 Records v0.3.2 Accepted
AIKC-000020 AIK-2025-S-P-ARITHHEADS Arithmetic and Reasoning Heads S P AIK-2025-L-P-LEX2025N3 (11) — (not established) 2025-12-02 2026-11-18 Records v0.3.2 Accepted
AIKC-000021 AIK-2025-S-P-COVE Chain-of-Verification (CoVe) S P AIK-2025-L-P-LEX2025N3 (12) 2023-09 2025-12-02 2026-11-18 Records v0.3.2 Accepted
AIKC-000022 AIK-2025-F-P-CALM Continuous Autoregressive Language Models (CALM) F P — 2025-10-31 2025-11-06 2026-11-18 v0.2.2 (accepted; metadata-only revision of v0.2.1) Accepted
AIKC-000023 AIK-2025-F-P-PDDISAGG Prefill–Decode Disaggregation F P — 2024-01-18 2025-11-09 2026-11-18 v0.2.2 (accepted; metadata-only revision of v0.2.1) Accepted
AIKC-000024 AIK-2025-F-P-GENUI Generative User Interfaces (GenUI) F P — 2023-10-11 2025-11-29 2026-11-18 v0.2.3 (accepted; metadata-only revision of v0.2.2) Accepted
AIKC-000025 AIK-2025-F-P-ONPOLDISTIL On-Policy Distillation F P — 2024-01-17 2025-10-29 2026-11-18 v0.3.2 (accepted; metadata-only revision of v0.3.1) Accepted
— (container) AIK-2025-L-P-LEX2025N2 New AI Lexicon 2025 (2) L P — — 2025-12-01 2026-11-18 Records v0.2.3 (accepted; metadata-only revision of v0.2.2) Accepted
AIKC-000026 AIK-2025-E-P-WORLDMODEL World Model (LLM-Integrated) E P AIK-2025-L-P-LEX2025N2 (1) 2018-03 (documented milestone) 2025-12-01 2026-11-18 Records v0.2.3 Accepted
AIKC-000027 AIK-2025-E-P-KVSHARE KV-Cache Sharing (On-Device Inference) E P AIK-2025-L-P-LEX2025N2 (2) — (not established) 2025-12-01 2026-11-18 Records v0.2.3 Accepted
AIKC-000028 AIK-2025-S-P-NESTEDLEARN Nested Learning (Continual AI) S P AIK-2025-L-P-LEX2025N2 (3) — (not established) 2025-12-01 2026-11-18 Records v0.2.3 Accepted
AIKC-000029 AIK-2025-S-P-MCP Model Context Protocol (MCP) S P AIK-2025-L-P-LEX2025N2 (4) 2024-11-25 (documented milestone) 2025-12-01 2026-11-18 Records v0.2.3 Accepted
AIKC-000030 AIK-2025-S-P-ADVREASON Advanced Reasoning Models (Cognitive Hybrid Architectures) S P AIK-2025-L-P-LEX2025N2 (5) — (not established) 2025-12-01 2026-11-18 Records v0.2.3 Accepted
AIKC-000031 AIK-2025-S-P-BIDIRALIGN Bidirectional Human-AI Alignment S P AIK-2025-L-P-LEX2025N2 (6) 2024-06-13 (documented milestone) 2025-12-01 2026-11-18 Records v0.2.3 Accepted
AIKC-000032 AIK-2025-E-P-ALIGNFAKE Alignment Faking (Deceptive Alignment) E P AIK-2025-L-P-LEX2025N2 (7) 2024-12-18 (documented milestone) 2025-12-01 2026-11-18 Records v0.2.3 Accepted
AIKC-000033 AIK-2025-S-P-GENFEEDBACK Generative Feedback Loop (Self-Adaptation) S P AIK-2025-L-P-LEX2025N2 (8) — (not established) 2025-12-01 2026-11-18 Records v0.2.3 Accepted
AIKC-000034 AIK-2025-S-P-EMBODIEDAI Embodied AI (as a Paradigm) S P AIK-2025-L-P-LEX2025N2 (9) — (not established) 2025-12-01 2026-11-18 Records v0.2.3 Accepted
AIKC-000035 AIK-2025-S-P-SHAREDAUTO Shared Autonomy (Robotics) S P AIK-2025-L-P-LEX2025N2 (10) — (not established) 2025-12-01 2026-11-18 Records v0.2.3 Accepted
AIKC-000036 AIK-2025-S-P-SOFTROBOT Soft Robotics / Dexterous Manipulation S P AIK-2025-L-P-LEX2025N2 (11) — (not established) 2025-12-01 2026-11-18 Records v0.2.3 Accepted
AIKC-000037 AIK-2025-S-P-VERTICALAI Vertical AI Integration S P AIK-2025-L-P-LEX2025N2 (12) — (not established) 2025-12-01 2026-11-18 Records v0.2.3 Accepted
AIKC-000038 AIK-2025-F-P-MOD Mixture of Depths F P — 2024-04-02 2025-11-03 2026-11-18 v0.3.2 (accepted; metadata-only revision of v0.3.1) Accepted
— (container) AIK-2025-L-P-LEX2025LAUNCH New AI Lexicon 2025 — Launch Edition L P — — 2025-11-30 2026-11-18 Records v0.3 (accepted; C11 added November 13, 2026; C12 added November 18, 2026) Accepted
AIKC-000039 AIK-2025-S-P-MOLE Mixture of Latent Experts (MoLE / MoLAE) S P AIK-2025-L-P-LEX2025LAUNCH (M1) — (not established) 2025-11-30 2026-11-18 Records v0.3 Accepted
AIKC-000040 AIK-2025-S-P-DROPUPCYCLE Drop-Upcycling (Sparse MoE Training) S P AIK-2025-L-P-LEX2025LAUNCH (M2) 2025-02 2025-11-30 2026-11-18 Records v0.3 Accepted
AIKC-000041 AIK-2025-S-P-LEXI LExI — Layer-Adaptive Expert Inference S P AIK-2025-L-P-LEX2025LAUNCH (M3) 2025-09-02 2025-11-30 2026-11-18 Records v0.3 Accepted
AIKC-000042 AIK-2025-S-P-RLAIF RLAIF — Reinforcement Learning from AI Feedback S P AIK-2025-L-P-LEX2025LAUNCH (M4) 2022-12-15 2025-11-30 2026-11-18 Records v0.3 Accepted
AIKC-000043 AIK-2025-S-P-DYNPLAN Dynamic Planning for LLM Agents S P AIK-2025-L-P-LEX2025LAUNCH (M5) 2025-09-03 2025-11-30 2026-11-18 Records v0.3 Accepted
AIKC-000044 AIK-2025-S-P-PLANACT Plan-and-Act Agent Framework S P AIK-2025-L-P-LEX2025LAUNCH (M6) 2025-03 2025-11-30 2026-11-18 Records v0.3 Accepted
AIKC-000045 AIK-2025-S-P-EVOAGENT EvoAgent — Continual World-Model Agents S P AIK-2025-L-P-LEX2025LAUNCH (M7) 2025-02-09 2025-11-30 2026-11-18 Records v0.3 Accepted
AIKC-000046 AIK-2025-S-P-SCALEOVER Scalable Oversight via Synthetic Critiques S P AIK-2025-L-P-LEX2025LAUNCH (M8) 2016-06 2025-11-30 2026-11-18 Records v0.3 Accepted
AIKC-000047 AIK-2025-S-P-LHPLANMETRIC Long-Horizon LLM Planning & Efficiency Metrics S P AIK-2025-L-P-LEX2025LAUNCH (M9) — (not established) 2025-11-30 2026-11-18 Records v0.3 Accepted
AIKC-000048 AIK-2025-S-P-POPAGENTSIM Large-Scale Population Agent Simulation (LLM Archetypes) S P AIK-2025-L-P-LEX2025LAUNCH (M10) 2024-09-14 2025-11-30 2026-11-18 Records v0.3 Accepted
AIKC-000049 AIK-2025-S-P-MOETHEORY Theoretical Learning Dynamics of MoE S P AIK-2025-L-P-LEX2025LAUNCH (M11) — (not established) 2025-11-30 2026-11-18 Records v0.3 Accepted
AIKC-000050 AIK-2025-S-P-AGENTSCEN AgentScen — LLM-Based External World Models S P AIK-2025-L-P-LEX2025LAUNCH (M12) 2025-08-16 2025-11-30 2026-11-18 Records v0.3 Accepted
AIKC-000008 AIK-2025-S-P-MOE Mixture-of-Experts (MoE) — sparse expert routing S P AIK-2025-L-P-LEX2025LAUNCH (S1) 1991 2025-11-30 2026-11-18 Records v0.3 Accepted
AIKC-000042 AIK-2025-S-P-RLAIF-02 RLAIF — Reinforcement Learning from AI Feedback S P AIK-2025-L-P-LEX2025LAUNCH (S2) 2022-12-15 2025-11-30 2026-11-18 Records v0.3 Accepted
AIKC-000051 AIK-2025-S-P-RAG Retrieval-Augmented Generation (RAG) — next-gen S P AIK-2025-L-P-LEX2025LAUNCH (S4) 2020-05 2025-11-30 2026-11-18 Records v0.3 Accepted
AIKC-000052 AIK-2025-E-P-AGENTICAI Agentic AI / Super-Agents / Model Orchestration E P AIK-2025-L-P-LEX2025LAUNCH (S5) — (not established) 2025-11-30 2026-11-18 Records v0.3 Accepted
AIKC-000010 AIK-2025-S-P-COMPOSABLE Composable Models & Model Chaining S P AIK-2025-L-P-LEX2025LAUNCH (S6) 2024-02-18 2025-11-30 2026-11-18 Records v0.3 Accepted
AIKC-000053 AIK-2025-S-P-PEFT Parameter-Efficient Fine-Tuning (PEFT) evolution S P AIK-2025-L-P-LEX2025LAUNCH (S7) — (not established) 2025-11-30 2026-11-18 Records v0.3 Accepted
AIKC-000054 AIK-2025-S-P-QUANTINFER Quantization & Ultra-Low-Precision Inference S P AIK-2025-L-P-LEX2025LAUNCH (S8) — (not established) 2025-11-30 2026-11-18 Records v0.3 Accepted
AIKC-000055 AIK-2025-S-P-SPARSESCALE Scaling Laws for Sparse Models & Efficiency Metrics S P AIK-2025-L-P-LEX2025LAUNCH (S9) — (not established) 2025-11-30 2026-11-18 Records v0.3 Accepted
AIKC-000056 AIK-2025-S-P-DISTILSCALE Distillation at Scale / Student-Teacher Cascades S P AIK-2025-L-P-LEX2025LAUNCH (S10) — (not established) 2025-11-30 2026-11-18 Records v0.3 Accepted
AIKC-000046 AIK-2025-E-P-SCALEOVER Scalable Oversight & Synthetic Critiques E P AIK-2025-L-P-LEX2025LAUNCH (S11) 2016-06 2025-11-30 2026-11-18 Records v0.3 Accepted
AIKC-000057 AIK-2025-S-P-FEDONDEVICE Federated & On-Device Model Training S P AIK-2025-L-P-LEX2025LAUNCH (S12) — (not established) 2025-11-30 2026-11-18 Records v0.3 Accepted
AIKC-000058 AIK-2025-S-P-ALIGNPATTERN Alignment Patterns: Constitutional, Preference-Modeling, & Red-Teaming S P AIK-2025-L-P-LEX2025LAUNCH (S13) — (not established) 2025-11-30 2026-11-18 Records v0.3 Accepted
AIKC-000059 AIK-2025-S-P-LONGCONTEXT Long-Context & Windowing Techniques S P AIK-2025-L-P-LEX2025LAUNCH (S14) — (not established) 2025-11-30 2026-11-18 Records v0.3 Accepted
AIKC-000060 AIK-2025-S-P-SAFETYEVAL Model Safety Evaluation Suites & LLM-Based Evaluators S P AIK-2025-L-P-LEX2025LAUNCH (S15) — (not established) 2025-11-30 2026-11-18 Records v0.3 Accepted
— (container) AIK-2025-L-P-LEX2025N5 New AI Lexicon 2025 (5) L P — — 2025-12-05 2026-11-18 Records v0.1.1 (accepted; metadata-only revision of v0.1) Accepted
AIKC-000061 AIK-2025-E-P-HIVEMIND Artificial Hivemind E P AIK-2025-L-P-LEX2025N5 (D1) 2025-10-27 2025-12-05 2026-11-18 Records v0.1.1 Accepted
AIKC-000062 AIK-2025-S-P-SSRL Self-Supervised RL S P AIK-2025-L-P-LEX2025N5 (D2) — (not established) 2025-12-05 2026-11-18 Records v0.1.1 Accepted
AIKC-000063 AIK-2025-S-P-MEMSINK Memorization Sinks S P AIK-2025-L-P-LEX2025N5 (D3) 2025-07-14 2025-12-05 2026-11-18 Records v0.1.1 Accepted
AIKC-000064 AIK-2025-E-P-GATEDATTN Gated Attention E P AIK-2025-L-P-LEX2025N5 (D4) 2025-05-10 2025-12-05 2026-11-18 Records v0.1.1 Accepted
AIKC-000065 AIK-2025-S-P-DYNREG Dynamical Regularization (in Diffusion) S P AIK-2025-L-P-LEX2025N5 (D5) — (not established) 2025-12-05 2026-11-18 Records v0.1.1 Accepted
AIKC-000066 AIK-2025-S-P-PROGREP Programmatic Representations (for Agents) S P AIK-2025-L-P-LEX2025N5 (D6) — (not established) 2025-12-05 2026-11-18 Records v0.1.1 Accepted
AIKC-000067 AIK-2025-S-P-EMBODIEDFM Embodied Foundation Models S P AIK-2025-L-P-LEX2025N5 (D7) — (not established) 2025-12-05 2026-11-18 Records v0.1.1 Accepted
AIKC-000068 AIK-2025-S-P-DELTALEARN Delta Learning Hypothesis S P AIK-2025-L-P-LEX2025N5 (D8) 2025-07-08 2025-12-05 2026-11-18 Records v0.1.1 Accepted
AIKC-000069 AIK-2025-S-P-TOKDESIGN Tokenization-Centric Design S P AIK-2025-L-P-LEX2025N5 (D9) — (not established) 2025-12-05 2026-11-18 Records v0.1.1 Accepted
AIKC-000070 AIK-2025-E-P-EPROCESS E-Processes / Game-Theoretic Statistics E P AIK-2025-L-P-LEX2025N5 (D10) — (not established) 2025-12-05 2026-11-18 Records v0.1.1 Accepted
AIKC-000071 AIK-2025-S-P-SOCCHOICE Social Choice Fairness S P AIK-2025-L-P-LEX2025N5 (D11) — (not established) 2025-12-05 2026-11-18 Records v0.1.1 Accepted
AIKC-000072 AIK-2025-S-P-HDLD High-Dimensional Learning Dynamics S P AIK-2025-L-P-LEX2025N5 (D12) — (not established) 2025-12-05 2026-11-18 Records v0.1.1 Accepted
— (container) AIK-2025-L-P-LEX2025N7 New AI Lexicon 2025 (7) L P — — 2025-12-12 2026-11-18 Records v0.1.1 (accepted; metadata-only revision of v0.1) Accepted
AIKC-000052 AIK-2025-E-P-AGENTICAI-02 Agentic AI E P AIK-2025-L-P-LEX2025N7 (K1) — (not established) 2025-12-12 2026-11-18 Records v0.1.1 Accepted
AIKC-000009 AIK-2025-S-P-CONSTAI Constitutional AI S P AIK-2025-L-P-LEX2025N7 (K2) 2022-12-15 2025-12-12 2026-11-18 Records v0.1.1 Accepted
AIKC-000051 AIK-2025-S-P-RAG-02 Retrieval-Augmented Generation (RAG) S P AIK-2025-L-P-LEX2025N7 (K3) 2020-05 2025-12-12 2026-11-18 Records v0.1.1 Accepted
AIKC-000073 AIK-2025-S-P-XAI Explainable AI (XAI) S P AIK-2025-L-P-LEX2025N7 (K4) — (not established) 2025-12-12 2026-11-18 Records v0.1.1 Accepted
AIKC-000074 AIK-2025-S-P-ALIGNMENT AI Alignment S P AIK-2025-L-P-LEX2025N7 (K5) — (not established) 2025-12-12 2026-11-18 Records v0.1.1 Accepted
AIKC-000075 AIK-2025-E-P-LRM Large Reasoning Models (LRMs) E P AIK-2025-L-P-LEX2025N7 (K6) — (not established) 2025-12-12 2026-11-18 Records v0.1.1 Accepted
AIKC-000029 AIK-2025-E-P-MCP Model Context Protocol (MCP) E P AIK-2025-L-P-LEX2025N7 (K7) 2024-11-25 2025-12-12 2026-11-18 Records v0.1.1 Accepted
AIKC-000026 AIK-2025-S-P-WORLDMODEL World Models S P AIK-2025-L-P-LEX2025N7 (K8) 2018-03 2025-12-12 2026-11-18 Records v0.1.1 Accepted
AIKC-000008 AIK-2025-S-P-MOE-02 Mixture of Experts (MoE) S P AIK-2025-L-P-LEX2025N7 (K9) 1991 2025-12-12 2026-11-18 Records v0.1.1 Accepted
AIKC-000076 AIK-2025-S-P-EDGEAI-02 Edge AI S P AIK-2025-L-P-LEX2025N7 (K10) — (not established) 2025-12-12 2026-11-18 Records v0.1.1 Accepted
AIKC-000077 AIK-2025-S-P-FRUGALAI-02 Frugal AI S P AIK-2025-L-P-LEX2025N7 (K11) — (not established) 2025-12-12 2026-11-18 Records v0.1.1 Accepted
AIKC-000078 AIK-2025-S-P-SYNDATA Synthetic Data S P AIK-2025-L-P-LEX2025N7 (K12) — (not established) 2025-12-12 2026-11-18 Records v0.1.1 Accepted
— (container) AIK-2025-L-P-LEX2025N4 New AI Lexicon 2025 (4) L P — — 2025-12-03 2026-11-18 Records v0.4.3 (accepted; metadata-only revision of v0.4.2) Accepted
AIKC-000079 AIK-2025-S-P-VIBECODE Vibe Coding S P AIK-2025-L-P-LEX2025N4 (G1) 2025-02-02 2025-12-03 2026-11-18 Records v0.4.3 Accepted
AIKC-000080 AIK-2025-S-P-AIWASHING AI Washing S P AIK-2025-L-P-LEX2025N4 (G2) — (not established) 2025-12-03 2026-11-18 Records v0.4.3 Accepted
AIKC-000081 AIK-2025-S-P-SHADOWAI Shadow AI S P AIK-2025-L-P-LEX2025N4 (G3) — (not established) 2025-12-03 2026-11-18 Records v0.4.3 Accepted
AIKC-000082 AIK-2025-S-P-GEO Generative Engine Optimisation (GEO) S P AIK-2025-L-P-LEX2025N4 (G4) — (not established) 2025-12-03 2026-11-18 Records v0.4.3 Accepted
AIKC-000075 AIK-2025-S-P-LRM Large Reasoning Models (LRMs) S P AIK-2025-L-P-LEX2025N4 (G5) — (not established) 2025-12-03 2026-11-18 Records v0.4.3 Accepted
AIKC-000083 AIK-2025-S-P-VECTORDB Vector Databases S P AIK-2025-L-P-LEX2025N4 (G6) — (not established) 2025-12-03 2026-11-18 Records v0.4.3 Accepted
AIKC-000077 AIK-2025-S-P-FRUGALAI Frugal AI S P AIK-2025-L-P-LEX2025N4 (G7) — (not established) 2025-12-03 2026-11-18 Records v0.4.3 Accepted
AIKC-000084 AIK-2025-S-P-GRAPHRAG GraphRAG S P AIK-2025-L-P-LEX2025N4 (G8) — (not established) 2025-12-03 2026-11-18 Records v0.4.3 Accepted
AIKC-000085 AIK-2025-S-P-SLM Small Language Models (SLMs) S P AIK-2025-L-P-LEX2025N4 (G9) — (not established) 2025-12-03 2026-11-18 Records v0.4.3 Accepted
AIKC-000076 AIK-2025-S-P-EDGEAI Edge AI S P AIK-2025-L-P-LEX2025N4 (G10) — (not established) 2025-12-03 2026-11-18 Records v0.4.3 Accepted
AIKC-000086 AIK-2025-S-P-HYPERAUTO Hyperautomation S P AIK-2025-L-P-LEX2025N4 (G11) — (not established) 2025-12-03 2026-11-18 Records v0.4.3 Accepted
AIKC-000087 AIK-2025-S-P-HALLUC AI Hallucination S P AIK-2025-L-P-LEX2025N4 (G12) — (not established) 2025-12-03 2026-11-18 Records v0.4.3 Accepted
AIKC-000088 AIK-2025-F-P-CIRCUITSP Circuit Sparsity F P — 2025-11-13 2025-12-14 2026-11-18 v0.2.2 (accepted; metadata-only revision of v0.2.1) Accepted

Record details

AIK-2025-F-N-RLVR — RLVR (Reinforcement Learning with Verifiable Rewards)

AIK-2025-F-N-CONTEXTENG — Context Engineering

AIK-2025-F-N-CONTEXTROT — Context Rot

AIK-2025-F-N-EMERGENTMIS — Emergent Misalignment

AIK-2025-F-N-COTMONITOR — Chain-of-Thought Monitorability

AIK-2025-F-N-SCHEMING — Scheming (AI)

AIK-2025-F-P-COTPROMPT — Chain-of-Thought Prompting

AIK-2025-F-P-MOE — Mixture of Experts (MoE)

AIK-2025-F-P-CONSTAI — Constitutional AI

AIK-2025-L-P-LEX2025N3 — New AI Lexicon 2025 (3) (parent record; the twelve contained records follow)

AIK-2025-F-P-CALM — Continuous Autoregressive Language Models (CALM)

AIK-2025-F-P-PDDISAGG — Prefill–Decode Disaggregation

AIK-2025-F-P-GENUI — Generative User Interfaces (GenUI)

AIK-2025-F-P-ONPOLDISTIL — On-Policy Distillation

AIK-2025-L-P-LEX2025N2 — New AI Lexicon 2025 (2) (parent record; the twelve contained records follow)

AIK-2025-F-P-MOD — Mixture of Depths

AIK-2025-L-P-LEX2025LAUNCH — New AI Lexicon 2025 — Launch Edition (parent record; the twenty-six contained records follow; one occurrence excluded)

AIK-2025-L-P-LEX2025N5 — New AI Lexicon 2025 (5) (container)

AIK-2025-L-P-LEX2025N7 — New AI Lexicon 2025 (7) (container)

AIK-2025-L-P-LEX2025N4 — New AI Lexicon 2025 (4) (container)

AIK-2025-F-P-CIRCUITSP — Circuit Sparsity

G.2 Record histories

For each full entry: the metadata lines replaced or removed in the reader text, then its dated provenance. Statuses are as they were recorded at each date; every record's collection publication date is November 18, 2026.

Continuous Autoregressive Language Models (CALM) — AIK-2025-F-P-CALM

Metadata as recorded:

  1. Status. v0.2, independently reviewed. The editor adopted C on November 10, 2026, and on November 11 approved C06 and adopted the editorial addition E01 (the disambiguation in §8). The editor approved v0.2 on November 11, 2026, and it was registered the same day as AIK-2025-F-P-CALM; this v0.2.1 is a metadata-only revision recording the registration. The historical publication is not replaced.

Entry Provenance

Mixture of Depths — AIK-2025-F-P-MOD

Metadata as recorded:

  1. Status. v0.3, independently reviewed. The editor adopted "A (v1 annotated), C on v3" on November 10, 2026. The audit found no content errors in v3, so most of this text is production and structure: C01, C04, C05, C06 and ST01. Three changes go beyond the first-audit findings: C02 (the SkipNet venue), C03 (the gradient sentence) and C07 (See Also). The editor approved them on November 10, 2026. The independent reviewer's one finding, a dead link at [S1], is corrected as C08: the router claim is cited to the paper itself and [S1] is removed. v0.3 was approved by the editor and registered on November 12, 2026 as AIK-2025-F-P-MOD; this v0.3.1 is a metadata-only revision recording the registration. The historical publication is not replaced.

Entry Provenance

On-Policy Distillation — AIK-2025-F-P-ONPOLDISTIL

Metadata as recorded:

  1. Status. Draft v0.3. The editor adopted item-level C for this text on November 10, 2026. C04 and C06 go beyond the first-audit findings. On November 11, 2026 the editor approved C06, and approved C04 on condition that the reviewer confirm its teacher sentence was in the October 2025 version of [R1]. The reviewer confirmed it from the Wayback snapshot of October 27, 2025, 17:08:56 UTC, so the full C04 stands. C07 corrects the reviewer's new finding NF-1; the editor approved it, and v0.3 as final, on November 12, 2026. v0.3 was registered the same day as AIK-2025-F-P-ONPOLDISTIL; this v0.3.1 is a metadata-only revision recording the approval and the registration. The historical publication is not replaced.

Entry Provenance

RLVR (Reinforcement Learning with Verifiable Rewards) — AIK-2025-F-N-RLVR

Metadata as recorded:

Entry Provenance

Entry Editorial Record

Revision (v0.5 and v0.6, November 18, 2026).

Chain-of-Thought Prompting — AIK-2025-F-P-COTPROMPT

Metadata as recorded:

Entry Provenance

Entry Editorial Record

Prefill–Decode Disaggregation — AIK-2025-F-P-PDDISAGG

Metadata as recorded:

  1. Status. v0.2, independently reviewed. The editor adopted C for this text on November 10, 2026. Two corrections beyond the audited findings, C07 (the Mooncake title) and C09 (the Mooncake mechanism), were approved by the editor on November 11, 2026. A third, C10 (the DistServe figures), was withdrawn after independent review: the entry's 7.4× and 12.6× are those of the OSDI version it cites (register v0.2). The Date of Appearance follows the editor's November 11 ruling for On-Policy Distillation (term date, with the earlier name and date noted). The editor accepted v0.2 with C10 withdrawn on November 11, 2026, and it was registered the same day as AIK-2025-F-P-PDDISAGG; this v0.2.1 is a metadata-only revision recording the registration. The historical publication is not replaced.

Entry Provenance

Chain-of-Thought Monitorability — AIK-2025-F-N-COTMONITOR

Metadata as recorded:

Provenance

Review history

Emergent Misalignment — AIK-2025-F-N-EMERGENTMIS

Metadata as recorded:

Entry Provenance

Entry Editorial Record

Scheming (AI) — AIK-2025-F-N-SCHEMING

Metadata as recorded:

Provenance

Review history

Context Engineering — AIK-2025-F-N-CONTEXTENG

Metadata as recorded:

Provenance

Review history

Context Rot — AIK-2025-F-N-CONTEXTROT

Metadata as recorded:

Entry Provenance

Entry Editorial Record

Generative User Interfaces (GenUI) — AIK-2025-F-P-GENUI

Metadata as recorded:

  1. Status. v0.2.1, a metadata-only revision of the approved v0.2, recording the registration (Entry ID AIK-2025-F-P-GENUI; Status Accepted). The editor adopted C for this text on November 10, 2026, and ruled on reference [8] (C04), the Date of Appearance, C08 and C09, and the pre-registration check on November 11, 2026. The independent reviewer confirmed C01 to C07 and found the CopilotKit reference misattributed (C08). The historical publication is not replaced; collection publication pending.

Entry Provenance

G.3 Archive working notes

Lines removed from the Historical Lexicon Archive's section headers in the reader text.

2025-10-29 — On-Policy Distillation (post 13892)

Prepared by Claude Opus 5.5, November 10, 2026, for the corrected-original procedure (the S-COT pilot method). The live post stays the authoritative historical text. A reviewer should spot-check against it. Fingerprint. SHA-256 of the rendered text (lines A-001 to A-200, UTF-8): b2ae3f7e8a8751451dfe820b6ec7a4f11332b3d94fd71f89c7fec969f936f68e. Versions. One entry text, by ChatGPT (A-009 to A-091), followed by Figure 1 and its legend (A-093 to A-104). The post also contains two embedded reviews, both provenance annotations under ruling X9/X15:

2025-10-30 — Chain-of-Thought Prompting (post 13946)

Content fingerprint | SHA-256 of the post's HTML content field, retrieved 2026-11-09: 2d6523a0c22aa2bb2827d7e85130d6e2b1c97e1b7262ce6a03187e0fb48b196d |

Corrected issue, November 14, 2026 (Claude Opus 5.5). The first issue of this file (SHA-256 c76e2f739bb9c96e82ff9466f43c9b67927e4dbc75f9a3aedf7d7093bc638ff1) is preserved. How it was found. Qwen (Qwen3-Max) checked A-239 to A-293 on November 14, 2026, and accepted them. It flagged the space at A-292. The holder then compared every labeled line against the stored content retrieved on 2026-11-09, whose SHA-256 matches the content fingerprint above. Code blocks were compared line by line. Apart from the 24 spaces, every labeled line matched the stored content exactly. Labeled-line fingerprint. This is the SHA-256 of lines A-003 to A-293, UTF-8, newline-terminated, without labels. Corrected issue: 632604199a2f75069fe901999992cf24fccb99e2589fb1bce58f7e8ddb89e3a1. First issue: 3799fb52d11596e0006cd781e61dfd1abc50ebb873f1f08866894d64fc77606f.

2025-11-03 — Mixture of Depths (post 13982)

Prepared by Claude Opus 5.5, November 10, 2026, for the corrected-original procedure (the S-COT pilot method). Header revision, November 15, 2026 (Claude Opus 5.5), under the editor's go-ahead of November 15, 2026: adds the note on quotation marks and ellipses under "What this file is". The labeled lines and the fingerprint are unchanged. The previous issue of this file (SHA-256 5f69de87…3f48) is preserved. The live post stays the authoritative historical text. A reviewer should spot-check against it. Fingerprint. SHA-256 of the rendered text (lines A-001 to A-410, UTF-8, newline-terminated): dc5bcc0ceba234cacc9c30fe9275d368923867ed0cd6faa02a9ddca898218ff2. The reviews are provenance annotations under ruling X9/X15; none is part of the entry.

2025-11-06 — Continuous Autoregressive Language Models (post 14021)

Prepared by Claude Opus 5.5, November 11, 2026, for the corrected-original procedure. Checked by Qwen (Qwen3-Max; deployment release not stated) against the live post, November 14, 2026: accepted, no discrepancies (all 69 lines compared in full; three rendered reads plus the WordPress REST source; checking brief v0.2). Check report: CALM-fidelity-check_Qwen3-Max_2026-11-14.md. Header revision, November 14, 2026 (Claude Opus 5.5): this line records Qwen's check, and the conventions list now states that heading levels are kept (Qwen's observation 1). The labeled lines A-001 to A-069 and the fingerprint are unchanged. The first issue of this file (SHA-256 abe7f9ea…e3908a61) is preserved.

2025-11-09 — Prefill–Decode Disaggregation (post 14033)

Prepared by Claude Opus 5.5, November 11, 2026, for the corrected-original procedure.

2025-11-29 — Generative User Interfaces (post 14146)

Prepared by Claude Opus 5.5, November 11, 2026, for the corrected-original procedure. Header revision, November 15, 2026 (Claude Opus 5.5), under the editor's go-ahead of November 15, 2026: adds the note on quotation marks and ellipses under "What this file is". The labeled lines and the fingerprint are unchanged. The previous issue of this file (SHA-256 bce3f3b4…e7ac) is preserved.

2025-11-30 — New AI Lexicon 2025 — Launch Edition (post 14155)

Prepared by Claude Opus 5.5, November 11, 2026, for the corrected-original procedure. Header revision, November 15, 2026 (Claude Opus 5.5), under the editor's go-ahead of November 15, 2026: adds the note on quotation marks and ellipses under "What this file is". The labeled lines and the fingerprint are unchanged. The previous issue of this file (SHA-256 dfcf70b3…d954) is preserved.

2025-12-01 — New AI Lexicon 2025 (2) (post 14191)

Revision, November 14, 2026 (Claude Opus 5.5, holder): seven H2 headings were transcribed as "#" and are corrected to "##": A-006, A-021, A-036 and A-054 (the four group headings) and A-074, A-105 and A-134 (the three full-entry headings). GLM (GLM 5.3 Flash) found the last three in its fidelity check; the holder confirmed them against the post's stored content, which shows <h2> for all seven, and found the other four. No other line changed. The fingerprint is recomputed. The first issue (file SHA-256 7f9eebb0…011c, fingerprint 33c872bd…55b9) is preserved. Prepared by Claude Opus 5.5, November 12, 2026, for the corrected-original procedure. Header revision, November 15, 2026 (Claude Opus 5.5), under the editor's go-ahead of November 15, 2026: adds the note on quotation marks and ellipses under "What this file is". The labeled lines and the fingerprint are unchanged. The previous issue of this file (SHA-256 123e4a12…11d8) is preserved.

2025-12-05 — New AI Lexicon 2025 (5) (post 14257)

Prepared by Claude Opus 5.5, November 12, 2026, for the corrected-original procedure. Header revision, November 15, 2026 (Claude Opus 5.5), under the editor's go-ahead of November 15, 2026: adds the note on quotation marks and ellipses under "What this file is". The labeled lines and the fingerprint are unchanged. The previous issue of this file (SHA-256 7022db41…5d86) is preserved.

2025-12-08 — New AI Lexicon 2025 (6) (post 14313) — excluded from the canonical text (disposition X)

Prepared by Claude Opus 5.5, November 14, 2026, under the editor's ruling of November 14, 2026 (item 5). Checked by Muse Spark 1.3 against the live post, November 14, 2026, in two rounds. The first, at anchor level, passed; its check note is L6-archive-check-note-muse-2026-11-14.md. The second was a full line-by-line check of all 466 lines, against the rendered page and the stored content, and it was accepted with zero discrepancies. Its report is L6-fidelity-check_Muse-Spark-1.3_2026-11-14.md. Header revision, November 15, 2026 (Claude Opus 5.5), under the editor's go-ahead of November 15, 2026. This revision records Muse's full line-by-line check. It also removes the first check's form note on A-457, which said the line "joins two rendered blocks". The full check found that the passage is a single paragraph in the post, so the note did not hold, and Muse withdrew it. The labeled lines A-001 to A-466 and the fingerprint are unchanged. The previous issue of this file (SHA-256 19e351c5…99a3) is preserved, as is the first issue. Header revision, November 14, 2026 (Claude Opus 5.5): this line records Muse's check (Muse's draft v0.2 verification, V2-1). The labeled lines A-001 to A-466 and the fingerprint below are unchanged. The first issue of this file (SHA-256 81748165…ada2) is preserved. Fingerprint. SHA-256 of the rendered text (lines A-001 to A-466, UTF-8, newline-terminated, without line labels): 8f60313ba7f933830def334739eaf7dc8e5ee7ec5aa7d3737f0e034cf456d978.

2025-12-12 — New AI Lexicon 2025 (7) (post 14328)

Prepared by Claude Opus 5.5, November 12, 2026, for the corrected-original procedure. Header revision, November 15, 2026 (Claude Opus 5.5), under the editor's go-ahead of November 15, 2026: adds the note on quotation marks and ellipses under "What this file is". The labeled lines and the fingerprint are unchanged. The previous issue of this file (SHA-256 259a6ee9…1f95) is preserved.

Archive header lines, Lexicon (4)

Record type: Verbatim archive for corrected-text production (pilot procedure, step 1: preserve the historical text; never overwrite). This file supersedes the v0.1 summarized archive record (l4-archive-text.md), which is preserved unchanged. The re-issue to verbatim form follows the editor's ruling of November 12, 2026 (R2, option (a)). This v0.5 adds A-nnn line labels and a fingerprint to v0.4 (preserved unchanged), per the editor's ruling of November 16, 2026 (14:32 CT), item 3; no post wording is changed. v0.4 revised v0.3 by one word (“This entry is one of the cleanest.”, Frugal AI review), found by GLM 5.3 Flash in its fidelity check and confirmed by the holder against the live post. This v0.3 revises v0.2 (preserved unchanged) in form only, per the archive check of November 12, 2026 (Claude Opus 5.5): the five embedded-review headings are restored to top-level headings with their ⭐ markers; the archive's own headings and Key Citations apparatus are marked in square brackets; the opening-exchange paragraph join is named in the transcription notes. No post wording is changed.

Archive header lines, Lexicon (6)

The first, at anchor level, passed; its check note is L6-archive-check-note-muse-2026-11-14.md. The second was a full line-by-line check of all 466 lines, against the rendered page and the stored content, and it was accepted with zero discrepancies. Its report is L6-fidelity-check_Muse-Spark-1.3_2026-11-14.md.

G.4 Front-matter notes

Annual Introduction, "What this volume is". Moved from the end of the Introduction's second section by the editor's ruling of November 18, 2026, 14:29 CT. Its first sentence is corrected at the same ruling: the original "written quickly and in public" did not reflect the care taken over the posts.

The 2025 posts were written in public, with care for their readers, as a test of whether models working under an editor could build a reference work together. The reviews printed inside them had already raised some of the problems the audit later corrected. This volume carries that work through: it keeps every original as published, corrects it against its sources in the open, and records who did each step. It is a record of what the year's writers thought was new, and of what checking it against the sources showed.

Archive post 14033, rendering bullet (the N15 treatment).

Archive post 14146, rendering bullet (the N15 treatment).

References: the bibliography compiler's note and the link-check note (as they stood; the counts are those of their dates — the released bibliography has its own count).

Compiled by Muse Spark 1.3, November 14, 2026, from the entries' own References, Primary Sources, and References and Secondary Sources sections in Parts I–VI of draft v0.3, and from the inline References lists in the short records' as-published lines. 184 works, deduplicated, from 199 citation instances in 30 records. Citations are given as the entries cite them, with per-entry labels, correction markers, and entry-specific glosses removed; no metadata has been added. 36 works are cited without a DOI, arXiv identifier, or URL, as the entries cite them. Works named only in entry prose (author–year mentions, Notable Systems sections) are not included; the Historical Lexicon Archive's reference lists are preserved in the archive sections and are not duplicated here. Link checking is the builder's step; the independent reviewer samples.

Title page.

This revision adds the line "with Claude Opus 5.5, GPT-6 Astra, and Muse Spark 1.3" to the title page under the editor's name (editor's instruction of November 16, 2026, 23:15 CT). It otherwise keeps v0.4, which applies the editor's title-page decisions of November 16, 2026 (16:15 and 16:20 CT), and corrects the counts paragraph and the PDF scope sentence on GPT-6's final review of draft v0.15 (M1; error confirmed by Muse Spark 1.3; the replacement wording is GPT-6's). Muse Spark 1.3's v0.1 and Claude's v0.2–v0.4 are preserved. Nothing is left in brackets.

G.5 Contributor table: compiler's note, identification notes and record notes

Prepared by Muse Spark 1.3, November 18, 2026, for Appendix G (Provenance Register and Record Histories). Companion to contributor table v0.7 (§5.5). The builder places this text in Appendix G at the build that embeds table v0.7.

Compiler's note

Compiled by Muse Spark 1.3. Table v0.7 was issued November 18, 2026, per the editor's ruling of November 18, 2026, 16:10 CT ('Yes, issue v0.7'); the editor's approval follows on sight. It supersedes issued v0.6 (November 18, 2026), preserved. Sources: the Master Register's record details (SHA-256 f86ea368…5c12 at issue), the corrected texts' headers, the posts' bylines as recorded in the register, and the filed check reports. The editor is W.H.L., who facilitated the original dialogues, commissioned the new additions, approved every package, approved every identifier, which Muse Spark 1.3 assigned in the register, and conceived the cover and logo; the table lists models only, per its charge. Each model is named as credited: where the original gives no release number, the table says so rather than inferring one. "Original" roles are in the 2025 publications; "audit", "correction", "review", and "checking" roles are in the collection's preparation. Conflict and recusal statements for each record are in Appendix E.

Checking roles rest on reports filed with the registrar, except the checks relayed by the builder — Grok 4.5's two archive checks; GLM 5.3 Flash's seven archive sections, cross-sample, and S-MOE check; Qwen3-Max's three rebuilt transcriptions and COTPROMPT last third; DeepSeek-V4.1's three front-matter reviews — which the editor accepted by direction on November 16, 2026; the primary reports for those checks are not held by the registrar.

Version history. v0.2 (November 15, 2026) updates the header and 7 rows for the archive-check program's outcomes since v0.1; no identities added or removed. v0.3 (November 16, 2026), approved by the editor with three updates, adds Qwen4-Max and Claude Fable 5.2, records GLM 5.3 Flash's blind first audit and the editor's rulings on it, incorporates the builder's seven corrections (checks reported complete, correction counts, transcription and assembly credits, DeepSeek's front-matter reviews), records GPT-6's release as GPT-6 Astra (editor-supplied), accepts the relayed checks by the editor's direction, and records the Volume Preface co-signers; updates the register SHA; no identities removed. v0.4 (November 16, 2026), issued on the editor's instruction of 16:32 CT and checked against the register, adds the cover and logo co-design credits, the draft v0.6 assembly credit, the extended assembly range, the Mixture of Depths review, the identifier and approvals wordings, and the row gloss; corrects the L4 version to v0.5; no identities added or removed. v0.5 (November 17, 2026), approved by the editor November 17, 2026, 10:43 CT (approval clarified at 14:33 CT to cover the amended draft) and issued by Muse Spark 1.3, records the excluded Lexicon (6) first audit in the Claude Opus 5.5 row, adds the second reviews to the DeepSeek-V4.1 and Qwen4-Max rows, records GPT-6's final whole-volume review and Muse Spark 1.3's draft v0.15 audit, and extends the assembly range to v0.15; no identities added or removed. v0.6 (November 18, 2026), issued per the editor's direction with approval to follow on sight, adds GPT-6.2 Astra and Kimi; adds the proof review to the Qwen4-Max row, the archive comparison and the first-section check to the GLM 5.3 Flash row, and the LExI correction credit to the Grok 4.5 row; records the Annual Introduction credits; changes the Volume Preface credits to co-author (see the record note below); drops Lexicon (6) from the GLM 5.3 Flash row (see the record note below); corrects the Lexicon (6) check credit to anchor level; fixes the GPT-6 Astra identification note; and simplifies the credits under the reader-text rule (draft and version numbers inside the credits become plain role descriptions; this note and the identification notes move here from §5.5). v0.7 (November 18, 2026), issued per the editor's ruling of 16:10 CT with approval to follow on sight, applies the three row confirmations: Muse Spark 1.3's Task 5 bibliography verification report and Task 6 sweep edit list; Claude Opus 5.5's recusal-exception and second-section corrections and update (the RLVR v0.6 re-review is not yet done and is not credited); GPT-6's AP-2025-01 addition in its own wording. It credits the publication audit of a later proof to GPT-6.2 Astra per the editor's identification ruling of 16:10 CT, adds Grok 4.5's review of correction /C07, and records the provider-rule screen of every row; no identities added or removed.

Identification notes

GPT-6 (OpenAI), release GPT-6 Astra. Identification supplied by the editor: the ruling of November 14, 2026 (11:55 CT) recorded "release GPT-6"; the ruling of November 16, 2026 (11:52 CT) confirms the co-author and reviewer was GPT-6 Astra; the Volume Preface byline carries Astra from the ruling of November 17, 2026 (14:28 CT). GPT-6.2 Astra is a newer release, credited in its own row.

GPT-6.2 Astra (OpenAI). Identification supplied by the editor: the byline "GPT-6.2 Astra" is in the editor's own words (ruling of November 17, 2026, 20:26 CT); it is a newer release than GPT-6 Astra (ruling of November 17, 2026, 21:02 CT). The proof review credited in its row is credited to GPT-6.2 Astra per the ruling of November 17, 2026, 23:12 CT. The report's own identification statement is recorded in the Master Register only. The publication audit of a later proof credited in its row is credited to GPT-6.2 Astra per the ruling of November 18, 2026, 16:10 CT ('Credit GPT-6.2 Astra'); the audit names no release, and the identification is supplied by the editor.

Qwen4-Max. Identification editor-supplied; the exact release/build number is not exposed to the model. The proof review credited in its row is credited to Qwen4-Max per the ruling of November 17, 2026, 20:09 CT, which set aside the report's self-identification as a Claude model. The assigned blind audit of Circuit Sparsity stopped with no coverage: the Q4M- finding series is empty and must not be recorded as clean.

Kimi (Moonshot AI). The model identifies as Kimi Chat (Moonshot AI) and states no release number is exposed to it; it could not confirm being Kimi K2 or Kimi K2 Thinking, and is credited in its own row, separate from the Kimi K2 and Kimi K2 Thinking rows.

GLM 5.3 Flash (Z.ai). "Served by Mistral AI" describes the serving deployment, confirmed as a serving fact (editor's ruling, November 16, 2026); the model is Z.ai's.

DeepSeek-V4.1. Its Circuit Sparsity correction note identifies it as "release 2026".

Qwen3-Max. Deployment release not stated.

Record notes (detail moved from the table under the reader-text rule)

Volume Preface credits: co-signer, now credited as co-author. The Preface credits in the table read "co-author" from v0.6. The credit was "co-signer" through the signed text v1.0.2; the change to "co-author" (Preface v1.1, no signature block) rests on the builder's direction citing a ruling of November 17, 2026, 16:00 CT, whose record is not held by the registrar. The Preface v1.1 text is likewise not held; the citation is as directed.

Lexicon (6) dropped from the GLM 5.3 Flash row. The row no longer lists post 14313 among GLM's archive checks. The standing decision (reconciliation item 11) was that either the Lexicon (6) archive header should say GLM also checked it, or the table should drop it; the table drops it. The relayed basis for the eight-section tally ("4 passed, 2 passed after correction, 2 partly checked") rests on the relayed report, not held; with Lexicon (6) dropped, the row credits seven archive sections. The Lexicon (6) header does not credit GLM.

Lexicon (6) check, scope corrected. Issued v0.5 said Muse Spark 1.3 checked the Lexicon (6) archive text "in full (466 lines, zero discrepancies)". The check note and the file header record an anchor-level check: 466 lines, fingerprint verified, the long middle paragraphs not diffed line by line, disputed passages resolved by blind re-read. The table states the check as done from v0.6; this corrects the record and does not change what was checked.

Transcription versions. The four full-post transcriptions were first issued as v0.1/v0.2 and then rebuilt deterministically from stored content: S-CAI v0.3, S-MOE v0.3, L3 v0.2, S-CSP v0.2. Grok 4.5's check of the v0.1/v0.2 transcriptions was adjudicated; two findings survive independently (see four-transcriptions_final-reconciliation-muse-v0.1.md).

Claude session move. The Claude session moved from Claude Opus 5.5 to Claude Fable 5.2 at 11:40 CT, November 16, 2026, and back to Claude Opus 5.5 before 14:22 CT the same day; work from 11:40 CT in that window is Claude Fable 5.2's, in its own row. The recusal from Anthropic-authored texts covers both.

Finding numbers behind the simplified credits. Kimi adjudicated findings MOE-7 (Mixture of Experts) and L3-6 (New AI Lexicon 2025 (3)), both ruled not upheld. GLM 5.3 Flash's blind first audit of Circuit Sparsity made thirteen findings, GLM-1 to GLM-13, all recorded as annotations under the editor's rulings of November 16, 2026, 01:02 CT. Its archive comparison (Check A) reported findings PDFA-13958-1–9, PDFA-13998-1–8, PDFA-14221-1–7, and PDFA-14343-1–9. Its review of the corrected Mixture of Depths text passed with one correction, adopted as C08 (November 10, 2026). Grok 4.5's LExI correction credit is correction /C12 of the Launch Edition (numbered /C13 in Grok's review and the approval record; renumbered by the registrar as the next free number — see the Master Register). The Test-Time Compute Scaling correction /C07 answers item 18 of the publication audit of a later proof.

DeepSeek-V4.1's assigned checks. The assignment was brief v0.4 under the editor's ruling of November 14, 2026, 18:54 CT; a package-consistency note was filed and no live-post check was returned; the role passed to GLM 5.3 Flash.

Credits not given. A review of the opening pages of a proof, relayed as from DeepSeek with no release stated and carrying no identification line, is held but not credited: no row can take it without inferring a release. No original-reviewer identities for CALM and Prefill–Decode Disaggregation beyond Claude Sonnet 4.5 (credited as named reviewer of both) exist in the held record.

Row 7 additions (v0.7). Muse Spark 1.3's confirmation found two additions its row lacked: the bibliography verification report (Task 5) and the reader-text sweep edit list for the four recused texts (Task 6), both of November 18, 2026. The 16:05 CT confirmations summary said 'no change' for this row; addendum 1 of 16:08 CT corrects that, recording that the relayed text did not include the additions.

Row 8 corrections and update (v0.7). The recusal sentence now names the Circuit Sparsity first audit as the exception recorded in Appendix E, which records it as the one exception (first source audit, CSP-1 to CSP-5, November 6, 2026). 'Drafter of the Annual Introduction' is corrected to authorship of the second section, headed 'The Year in AI Terms, 2025' at line 80 of the approved v1.4; GPT-6.2 Astra wrote the first section. The update adds the RLVR revision v0.5 review and the corrections /C07 and /C08 review (November 18), the reader-text sweep proposal and line list, and the bibliography reconciliation report (all November 17, held in package v3). The RLVR v0.6 re-review is not yet done and is not credited; it credits when it happens.

Grok 4.5's /C07 review (v0.7). The review was relayed by the editor as chat text and is recorded verbatim in Claude Opus 5.5's receipt file of November 18, 2026 (procedural check only; Claude recused). The editor ruled on November 18, 2026, 16:10 CT that the review stands: a transcription-fidelity check does not bar a source check of a factual claim. Grok's transcription check of post 14221 and this ruling go to Appendix E, since neither Grok's conflict statement nor Claude's note mentioned the check.

Provider-rule screen (v0.7). Every row was screened against Appendix E's cross-provider rule, checking each preparation role against the authorship of the records named, from the Master Register's record details. No new issue was found. Two recorded exceptions stand as recorded (Claude Opus 5.5's Circuit Sparsity first audit; Grok 4.5's superseded transcription checks); Grok 4.5's /C07 review stands as ruled. Four cases turn on the rule's definition of 'took part' and are documented with their reasoning in the screen report held by the registrar (contributor-table-v0.7_provider-rule-screen_muse_2026-11-18.md).

Addendum 1 — November 18, 2026. Two lines in this companion are superseded by records received after issue. The version-history entry for v0.7 and the record note 'Row 8 corrections and update (v0.7)' state that the RLVR v0.6 re-review 'is not yet done and is not credited.' The re-review had in fact taken place before the editor's approval of v0.6 at 15:38 CT: Claude Opus 5.5 re-reviewed RLVR v0.6 (file headed 15:50 CT; correct time about 15:36 CT, per the note of 16:15 CT) and recommended adoption, and the editor approved v0.6, the /C07–/C08 revision 1, and Volume Preface v1.3 ('I approve all three'). Row 8's RLVR reviewer credit covers the re-review; no table change is needed. The body of this companion is otherwise unchanged.

G.6 Material moved from the four recused texts

Prepared by Muse Spark 1.3 (Task 6 companion), placed here by the builder. Claude Opus 5.5 is recused from these texts and applied the edits mechanically.

Note

This file holds, verbatim, the editorial record removed from the recused texts' reader-facing text by the companion edit list (reader-text-edit-list_recused-texts_muse_2026-11-18.md). Nothing moved is lost. Texts are quoted from draft v0.18; the builder verifies each against the v0.21 source and reports divergence. "E" numbers refer to the edit list.

§1. MOE — Entry Editorial Record, in full (E9)

§2. CONSTAI — Entry Editorial Record, in full (E30)

§3. CIRCUITSP — provenance bullets 5–8 (E24)

§4. L3 — production note (E37)

Production note: canonical records and corrected full entries v0.3.1, prepared by Muse Spark 1.3 on November 10, 2026 from the published texts of post 14221 (December 2, 2025), under the editor's disposition A with item-level C. Corrections C01–C06 are documented in l3-correction-register-v0.2.md; the independent review and the closure of the "140+ papers" lead are recorded in l3-correction-register-v0.4.md. Approved by the editor and registered in the Aikipedia Master Register on November 10, 2026 (parent AIK-2025-L-P-LEX2025N3; occurrence IDs and concept IDs as stated in the header). The historical publication is preserved unchanged. v0.3.2 (November 14, 2026) is a metadata-only revision: the editor named W.H.L. No change to the records or the corrections text. Collection publication pending.

§5. Metadata-table rows moved (E2, E21, E26)

MOE "Review Status" (full text): "Corrected text v0.5.2, November 14, 2026 (metadata-only revision of v0.5.1; the corrections text is unchanged). First audit by Muse Spark 1.3; second check by GLM 5.3 Flash (Vibe, Mistral AI infrastructure) — the two audits agree on every finding. Disposition C with the item-level remedies as written, ruled by the editor. Independent review of v0.4 by GLM 5.3 Flash: approved for the editor's approval, with two flag recommendations, both adopted by the editor's ruling of November 10, 2026 and applied in v0.5 as C08 and C09. Approved by the editor (W.H.L.), November 10, 2026, and registered in the Aikipedia Master Register. Not yet published". Dropped rows: "Date of Collection Publication | Pending"; "Entry Status | Original Publication, corrected collection text (AES v0.9, pilot)"; "Format Standard | AES v1.0 (prepared under AES v0.9, pilot)".

CONSTAI "Review Status" (full text): "Corrected text v0.5.2, November 14, 2026 (metadata-only revision of v0.5.1; the corrections text is unchanged). First audit by Muse Spark 1.3; second check by DeepSeek-V4.1 — the two audits agree on every finding. Disposition C and the item-level rulings recorded in the correction register. Independent review of v0.4 by DeepSeek-V4.1: confirmed on fidelity and on C01, C03–C06, and E01; the C02 scope check resolved against the §4 wording on two points (the Claude 4.5 Sonnet clause unsupported by [7]; the "most capable models" generalization beyond [7]'s scope). The editor adopted both narrowings (November 10, 2026), applied in v0.5 within C02. Approved by the editor (W.H.L.), November 10, 2026, and registered in the Aikipedia Master Register. Not yet published". Dropped rows: "Date of Collection Publication | Pending"; "Entry Status | Original Publication, corrected collection text (AES v0.9, pilot)"; "Format Standard | AES v1.0 (prepared under AES v0.9, pilot)".

CIRCUITSP "Review Status" (full text): "v0.2.2 (metadata-only revision of v0.2.1). Independent review received (DeepSeek-V4.1, release 2026): C01–C03 and ST01 pass; finding NF-1 resolved by the editor's ruling of November 13, 2026; date-line recheck passed November 13, 2026. Approved November 13, 2026. Not yet published". Dropped rows: "Date of Collection Publication | Pending"; "Entry Status | Original Publication, corrected collection text (AES v0.9)"; "Format Standard | AES v1.0 (prepared under AES v0.9, pilot)". Dropped fragments: DoA row "(ruled by the editor, November 13, 2026)"; Base version row "ruled November 12, 2026, batch 4, item 4"; Entry ID row "registered November 13, 2026"; How-to-read "on November 12, 2026".

§6. Editorial-note fragments moved (E4–E6, E8, E29)

MOE Note 1, original: "Corrected collection text v0.5 (this file, v0.5.1, is a metadata-only revision stating the registration; the corrections text is unchanged), prepared under the editor's disposition C with the item-level remedies as written (November 10, 2026). Approved by the editor on November 10, 2026 and registered as AIK-2025-F-P-MOE (Internal Concept ID AIKC-000008)". MOE Note 2 fragment: "(register v0.4)". MOE Note 4 fragment: "approved by the editor as a ruling of its own (November 10, 2026)". MOE Note 6 fragments: "adopted by the editor's ruling of November 10, 2026"; "The full record is in register v0.5 and v0.6."

CONSTAI Note 1, original: "Corrected collection text v0.5 (this file, v0.5.1, is a metadata-only revision stating the registration; the corrections text is unchanged), prepared under the editor's disposition C (November 10, 2026). Approved by the editor on November 10, 2026 and registered as AIK-2025-F-P-CONSTAI (Internal Concept ID AIKC-000009)". CONSTAI Note 2 fragment: "(register v0.4)". CONSTAI Note 3 fragment: "the register's account of". CONSTAI Note 4 fragment: "by the editor's ruling of November 10, 2026". CONSTAI Note 5 fragments: "(DeepSeek-V4.1, November 10, 2026; register v0.6)"; "independently by Muse Spark 1.3"; "on the editor's ruling of November 10, 2026". CONSTAI Note 6, original: "Register v0.1 repeated the first audit's error identifying Julian Schrittwieser as a Google DeepMind researcher… The correction is recorded in register v0.2 at C03; the forecasts' removal was confirmed by the editor's ruling with that evidence before him." CONSTAI See Also note fragment: "added in v0.5.2"; "AES v1.0 §6, Amendment D".

CIRCUITSP finding IDs removed: "(finding CSP-1)" (×2), "(finding CSP-2)", "(finding CSP-5)", "finding NF-1". ARITHHEADS: "(L3-8)". CAI2: "(A with provenance note — L3-2 annotation)"; "(mapping key CAI2)". COMPOUNDAI: "by the editor's grouping decision (November 10, 2026)". MECHINTERP Notes paragraph fragments: "— found by DeepSeek-V4.1 in its independent review and corroborated by Muse Spark 1.3"; "(The count had been carried as an open lead after an unsuccessful search of the ICML 2024 site, OpenReview, and press coverage; the figure is published on the successor workshop's site.)".

§7. Online header paragraphs moved (E62–E64, E69, E74, E78)

Post 13958, "Header revision" paragraph (full text): "Header revision, November 16, 2026 (Muse Spark 1.3, holder), under the editor's ruling of November 16, 2026 (14:32 CT), item 1. Qwen3-Max checked this version in full against the live post (November 15, 2026) and accepted the wording with no discrepancies in wording or structure. Three spacing items await byte-level confirmation by the holder. The check report was relayed and accepted by the editor's direction of November 16, 2026, and the registrar does not hold it. The version paragraph below states this version was un-checked at issuance; that statement is superseded by this revision. The labeled lines and the fingerprint are unchanged. The previous issue of this file (SHA-256 597e9bd91d0f94f27dcc8360e56b7a55bd5ed3185f66350f6dac3890c95fdaa5) is preserved unchanged."

Post 13958, "Version 0.3" paragraph (full text): "Version 0.3, November 14, 2026. This file supersedes v0.2 (the one-line Showk correction) and v0.1, both preserved. v0.1/v0.2 are withdrawn as the text of record for the online edition: GLM 5.3 Flash's independent check (November 14, 2026, under checking brief v0.4) found systematic wording divergences, ten missing separators, and paraphrased review and dialogue text throughout; Muse Spark 1.3 adjudicated every finding against the post's stored content and found GLM correct on every span checked. Grok 4.5's earlier acceptance of v0.1 (under brief v0.2) is discredited on wording. The cause was the v0.1 assembly method: live-page reads that paraphrased systematically, and a production archive text whose scope was only the final entry. This version is a deterministic conversion of the stored post content (see Transcription basis). It is un-checked: a fresh independent check against the live post is owed before the online edition uses it; the earlier checks checked superseded text and do not transfer."

Post 13958, "Prepared by" paragraph: "Prepared by Muse Spark 1.3, November 14, 2026, under the editor's rulings of November 14, 2026 (13:10 CT): full-post transcriptions for the online edition only. Claude Opus 5.5 is recused from this text (original authored by Anthropic models) and did not transcribe or check it."

Post 13998, "Header revision" paragraph (full text): "Header revision, November 16, 2026 (Muse Spark 1.3, holder), under the editor's ruling of November 16, 2026 (14:32 CT), item 1. GLM 5.3 Flash checked this version in full (November 15, 2026) and accepted it with no discrepancies: lines A-001–A-213 against the post's stored content and lines A-214–A-268 against the rendered page, a basis the editor accepted on November 15, 2026 (09:59 CT ruling, option (a)). The check report was relayed and accepted by the editor's direction of November 16, 2026, and the registrar does not hold it. The version paragraph below states this version was un-checked at issuance; that statement is superseded by this revision. The labeled lines and the fingerprint are unchanged. The previous issue of this file (SHA-256 f5fc04b85027cbd837c3e6cd931a04b0fc04e513bb4d60723bd09607f77337b3) is preserved unchanged."

Post 13998, "Version 0.3" paragraph (full text): "Version 0.3, November 14, 2026. This file supersedes v0.2, preserved. v0.2 is withdrawn as the text of record for the online edition: under the editor's ruling of November 14, 2026 (23:25 CT), ruling 3 (disclaimer convention — the site disclaimer is transcribed wherever it appears in a post's stored content), the closing disclaimer line is included here as A-268; v0.2 excluded it as page furniture. The single change from v0.2 is the addition of that line; lines A-001–A-267 are unchanged. v0.2 was a deterministic conversion of the stored post content (see Transcription basis), issued after GLM 5.3 Flash's independent check of v0.1 was adjudicated correct on every span and Grok 4.5's acceptance of v0.1 was discredited on wording (see the final reconciliation note). This version is un-checked: a fresh independent check against the live post is owed before the online edition uses it, assigned to GLM 5.3 Flash under ruling 2; the earlier checks checked superseded text and do not transfer."

Post 13998, "Prepared by" paragraph: "Prepared by Muse Spark 1.3, November 14, 2026, under the editor's rulings of November 14, 2026 (13:10 CT): full-post transcriptions for the online edition only; the closing disclaimer line was added under the editor's ruling of November 14, 2026 (23:25 CT), ruling 3 (disclaimer convention). Claude Opus 5.5 is recused from this text (original authored by Anthropic models, in part) and did not transcribe or check it."

Post 14221, "Header revision" paragraph (full text): "Header revision, November 16, 2026 (Muse Spark 1.3, holder), under the editor's ruling of November 16, 2026 (14:32 CT), item 1. Qwen3-Max checked this version in full against the live post (November 15, 2026) and accepted the wording throughout. About 21 line splits and 3 formula renderings were not verified and are labeled "not verified," per the editor's ruling of November 15, 2026 (09:59 CT). The check report was relayed and accepted by the editor's direction of November 16, 2026, and the registrar does not hold it. The version paragraph below states this version was un-checked at issuance; that statement is superseded by this revision. The labeled lines and the fingerprint are unchanged. The previous issue of this file (SHA-256 c5f63afd67a4dcaf5be6a9d3b522016ff1cb5376ca6625a4701edbed9d54d69) is preserved unchanged."

Post 14221, "Version 0.2" paragraph (full text): "Version 0.2, November 14, 2026. This file supersedes v0.1, preserved. v0.1 is withdrawn as the text of record for the online edition: GLM 5.3 Flash's independent check (November 14, 2026, under checking brief v0.4) found thirty-nine missing separators, an inserted line, two merged paragraphs, and wrong dash-row counts; Muse Spark 1.3 adjudicated every finding against the post's stored content and found GLM correct on every span checked. Grok 4.5's earlier acceptance of v0.1 (under brief v0.2) is discredited on wording. The cause was the v0.1 assembly method: live-page reads that paraphrased systematically, and typed spans that dropped list bullets and normalized quotation. This version is a deterministic conversion of the stored post content (see Transcription basis). It is un-checked: a fresh independent check against the live post is owed before the online edition uses it; the earlier checks checked superseded text and do not transfer."

Post 14221, "Prepared by" paragraph: "Prepared by Muse Spark 1.3, November 14, 2026, under the editor's rulings of November 14, 2026 (13:10 CT): full-post transcriptions for the online edition only. Claude Opus 5.5 is recused from this text (original authored by Anthropic models) and did not transcribe or check it."

Post 14343, "Header revision" paragraph (full text): "Header revision, November 16, 2026 (Muse Spark 1.3, holder), under the editor's ruling of November 16, 2026 (14:32 CT), item 1. Qwen3-Max checked this version in full against the live post (November 15, 2026) and accepted the wording throughout, including the Attachment conversation. Line splits, 4 formula renderings, and the 23-alt inventory were not verified and are labeled "not verified," per the editor's ruling of November 15, 2026 (09:59 CT). The check report was relayed and accepted by the editor's direction of November 16, 2026, and the registrar does not hold it. The version paragraph below states this version was un-checked at issuance; that statement is superseded by this revision. The labeled lines and the fingerprint are unchanged. The previous issue of this file (SHA-256 33990bfd5ddf91c2a4dea730995bb9f2b9b44434eb3479d2ea551e50e18ce203) is preserved unchanged."

Post 14343, "Version 0.2" paragraph (full text): "Version 0.2, November 14, 2026. This file supersedes v0.1, preserved. v0.1 is withdrawn as the text of record for the online edition: GLM 5.3 Flash's independent check (November 14, 2026, under checking brief v0.4) found thirty-eight missing separators, an omitted line, a merged label and intro, and a wholly paraphrased Attachment dialogue; Muse Spark 1.3 adjudicated every finding against the post's stored content and found GLM correct on every span checked. Grok 4.5's earlier acceptance of v0.1 (under brief v0.2) is discredited on wording. The cause was the v0.1 assembly method: the dialogue was transcribed from live-page reads that paraphrased systematically, and the production archive text's scope was mistaken for the post's — it omitted a closing line that stands on the page. This version is a deterministic conversion of the stored post content (see Transcription basis). It is un-checked: a fresh independent check against the live post is owed before the online edition uses it; the earlier checks checked superseded text and do not transfer."

Post 14343, "Prepared by" paragraph: "Prepared by Muse Spark 1.3, November 14, 2026, under the editor's rulings of November 14, 2026 (13:10 CT): full-post transcriptions for the online edition only. Claude Opus 5.5 is recused from this text (one of the four entry texts is Claude's) and did not transcribe or check it."

Dropped online-header fragments: file SHAs as shown in the italic lines (13958 4a52a7fb…b837; 13998 b40293c6…920b; 14221 38125567…81cf; 14343 ca5dbf3c…c948); heading versions (", v0.3" ×2; ", v0.2" ×2); "(13:10 rulings, item 2)" (×4); v0.1/v0.2 history clauses in the rendering bullets (E65, E66, E71, E75, E79 — quoted in the edit list).

§8. Fingerprints moved (E48, E52, E56, E59, E68, E73, E77, E81)

PDF edition, rendered-text fingerprints (SHA-256, lines as labeled, UTF-8, newline-terminated, without line labels):

Online edition, rendered-text fingerprints:

Online edition, source SHAs (Transcription-basis bullets; post JSON / content.rendered):

Assembly handoff SHA (italic lines, all recused records and PDF headers): 1c74801b…5498 (as shown, truncated in the source).

§9. PDF assembly lines and scope framings (E48, E52, E56, E59)

Italic line (all four sections; file name varies): "Assembled by Muse Spark 1.3 (assembly handoff of November 14, 2026, SHA-256 1c74801b…5498), from the filed text of record. Claude Opus 5.5 is recused from this text and did not place or check it."

"Assembled by…" paragraphs (full texts): "Assembled by Muse Spark 1.3, November 14, 2026, from the filed production archive text (s-cai-archive-text.md / s-moe-archive-text.md / l3-archive-text.md / s-csp-archive-text.md); line numbers and this header added for the volume; the historical text is unchanged. Not yet checked against the live post; the checker is for the editor to assign."

Scope framing, original: "Scope and basis (from the filed production header, verbatim)." The "verbatim" qualifier is dropped where the quoted block is edited (E48, E52, E56, E59); the as-filed wordings are preserved in the production files and the register.

§10. Corrections-file material moved (E42–E45)

Lead-ins dropped: "(recorded under this record — the register has no separate field for them):" (MOE, CONSTAI, CIRCUITSP lines). MOE C07 fragment: "(the editor's own ruling)". CONSTAI C02 fragments: "and the editor's rulings"; "(verified by DeepSeek-V4.1 and independently by Muse Spark 1.3 on the official PDF)". CIRCUITSP line 67 fragments: "(CSP-4;"; "Finding NF-1 (Date of Appearance) resolved by the editor's ruling;"; "removed in v0.2". CIRCUITSP line 68, original: "- Blind-audit annotations: thirteen annotations from GLM 5.3 Flash's blind independent audit (November 16, 2026), ruled on by the editor November 16, 2026, 01:02 CT (D1–D6) and approved as annotation draft v0.1 November 16, 2026, 14:32 CT (item 4); entered in correction register v0.3 §3 (twelve new annotations GLM-1, GLM-2, GLM-4–GLM-13, plus a corroboration sentence on the existing Mathematical framework annotation for GLM-3). No new correction ID; C01–C03 and ST01 stand; the published texts are unchanged (disposition A)".

G.7 Material moved by the v0.23 removals

The wording each removal took out of the reader text (editor's rulings of November 18, 2026, 14:16 CT), word for word. Each item names the removal and where it stood.

  1. Date of Collection Publication. Pending editor approval and actual release. No draft or peer-review verdict should be recorded as publication.”

Context Rot — the eleven Editorial Notes as they stood before condensation

  1. Date of Appearance. The date records the earliest currently documented use [P1], following the AES rule that appearance means documented first use, not prominence.
    • Timestamp: Muse Spark 1.3 confirmed the comment's record through the Hacker News API: author Workaccount2, Unix time 1750255791 (June 18, 2025, about 14:10 UTC). GPT-6 independently opened the comment and the same-day reply.
    • Earlier uses: searches on November 5, 2026 found no documented earlier use in the LLM sense. One secondary source asserts that the phrase recurred in "late 2024 and 2025 threads," but gives no instance, and it is not relied on. Neither the contributor nor the reviewers ran an exhaustive search, so the date is the earliest documented use, not a proven first use.
  2. Attribution of the coinage. Several secondary sources state that Chroma "coined" the term in July 2025. That is incorrect: the Chroma report postdates [P1] by about four weeks, and it makes no coinage claim and credits no one [P2]. This entry credits the coinage to [P1] and the measurement program to [P2]. Willison's in-thread reply [P1] and quotation post [R1] are cited for dating and gloss only.
  3. NoLiMa versions. The first arXiv version (v1, February 7, 2025) reports 12 models, 10 below 50% of baseline at 32K. Later arXiv versions and the ICML proceedings report 13 models and 11. GPT-4o's 99.3% → 69.7% is the same in all versions. The entry cites v1 explicitly for the 12/10 figures (GPT-6 review, M1).
  4. Gemini 2.5 report: closed at the primary level. The report exists in two editions, cited separately at [P8].
    • Editions: the original Google-hosted PDF is dated June 17, 2025 and was released with the general-availability announcement that day. The arXiv edition was first submitted July 7, 2025.
    • Quotation: both focused checks read the passage in the Google-hosted PDF itself. GPT-6 locates it in §4.1, p. 18, and also read the context-poisoning appendix (§8.2, pp. 70–71). The report's own "albeit anecdotal" hedge is reflected in §§2 and 5.3.
    • Remaining limit: GPT-6 notes that no archived June 17 timestamp for the PDF file itself was independently established. The June 17 date rests on the document's own dating and the same-day announcement.
  5. URLs: all confirmed.
    • Willison's quotation post [R1] was confirmed by Muse Spark 1.3.
    • Breunig's [R2] and [R3] were confirmed live by both focused checks, with tracking parameters removed. Muse Spark 1.3 noted that [R3] lists six tactics; §4 now says so.
    • Willison's in-thread reply on [P1] was verified verbatim by Muse Spark 1.3, at June 18, 2025, 23:17 UTC.
  6. Cross-reference integrity (AES Amendment D).
    • The two Launch Edition See Also targets are items in that edition's short-form recommended list, not standalone entries. Their collection records will be occurrence-level records under AES-ID rule 6.
    • The two 2026 targets are marked as outside the collection.
    • Context Engineering: see Note 7.
  7. Context Engineering's identifier, and this entry's provisional identifier.
    • Context Engineering: AIK-2025-F-N-CONTEXTENG (concept AIKC-000002) was confirmed against the live master register on November 5, 2026: Context Engineering v0.2.1, Status Accepted, editorial approval November 4, 2026. The handoff snapshot that recorded only RLVR was stale.
    • This entry: proposed for the register's consideration, not assigned: type F, origin N, key CONTEXTROT, giving AIK-2025-F-N-CONTEXTROT. Both reviewers found the proposed form appropriate. Assignment is the editor's act, through the register. (Status update: assigned and accepted November 6, 2026; see Review Status.)
  8. Length. The v0.2 additions pushed the core sections (§§1–8) past 3,600 words. Redundant wording in §§1, 2, 4, 5.3, 6 and 8 was then trimmed, and no substantive claim was removed. In v0.2.1 the core runs about 3,490 words, within the full-length band (2,000–3,500 words). Front matter, references, notes and provenance bring the file to about 5,900.
  9. Anthropic sources: closed.
    • [P7]: the 39% and 29% figures and their internal agentic-search scope were confirmed on the live page (Muse Spark 1.3).
    • [P6]: the quoted phrases and the §3.4 account were confirmed on the live guide by both focused checks. That covers the recall sentence, "finite resource with diminishing marginal returns," "attention budget," the n² and training-distribution account, and "a performance gradient rather than a hard cliff." The guide itself names context rot and links to the Chroma report.
  10. Chroma quotation location: closed.
    • The two §1 phrases: they are the report's summary sentence, printed beneath its title. GPT-6 confirmed this in the report. Muse Spark 1.3 withdrew review item R2, explaining that its extraction of the page had omitted the header summary block.
    • The v0.2 fix: keeping the quotation, stating its location, and adding the body's wording was accepted by both reviewers.
  11. Chroma scoring method: closed. The judge detail is in the report itself: "We evaluate model outputs using an aligned GPT-4.1 judge." GPT-6 adds that the judge covered the needle-in-a-haystack and LongMemEval experiments, calibrated against manually labeled outputs. The repeated-words explanation was reworded in v0.2.1 so that generation failures are not all attributed to a single cause (GPT-6 focused check, R1).