Attention Is All You Don't Need: The Router Is the Model

This paper begins with a mathematical attack on the transformer attention equationand ends with a live geopolitical incident captured mid-experiment. ──────────────────────────────────────────────────────────────────────────────THE NAND DECOMPOSITION────────────────────────────────────────────────────────────────────────────── The transformer attention equation, Attn(Q, K, V) = softmax(QK^T / sqrt(d)) * V, operates over discrete token vocabularies in every deployed large language model.Under any quantization scheme (INT8, INT4), inner products q_i · k_j becomeintegers. The softmax then implements a threshold Boolean function — awinner-take-most selection. By Sheffer's completeness theorem (1913), every Boolean function is expressibleusing only NAND (the Sheffer stroke: p NAND q = ¬(p ∧ q)). Therefore: the attention pattern — which tokens attend to which — is computableby a NAND circuit of depth O(log d). The softmax adds Θ(nd) flops withoutchanging the routing topology. It is a differentiable wrapper added for trainingconvenience and preserved at inference by inertia. The historical chain is: Boole (1854) → Huntington (1904) → Sheffer (1913) → Vaswani (2017) The transformer rediscovered Boolean routing and chose the expensiverepresentation. This paper is the formal record of that observation. ──────────────────────────────────────────────────────────────────────────────THE ROUTER IS THE MODEL────────────────────────────────────────────────────────────────────────────── The NAND decomposition proves token-level attention is Boolean routing.This paper extends the same principle to the system level. Intelligence at scale is not the capability of one transformer.It is governed orchestration: models specialize, routers compose,verifiers establish trust, witnesses preserve evidence. The routing function is defined as R(x, c, π, ρ) → (m, τ, g, v) where x is the incoming task, c is available context, π is governance policy,ρ is resource state, m is the selected model, τ is the toolchain,g is the verification gate, and v is the witness obligation. The routing objective is A* = argmax [ Q(A_i, t) − λC(A_i, t) − μR(A_i, t) + νV(A_i, t) ] balancing quality, cost, risk, and verifiability. This is not softmax.The router produces a hard Boolean assignment — consistent with the NANDdecomposition at the token level. The architecture is fractal: the sameBoolean routing principle governs from token pair to multi-model orchestration. Attention explains one model. Routing explains an intelligent system. ──────────────────────────────────────────────────────────────────────────────THE LIVE INCIDENT: IDENTITY AS BOOLEAN ROUTING────────────────────────────────────────────────────────────────────────────── During mass collection of contributor sections for this paper, an unexpectedempirical finding emerged. Qwen3 (Alibaba Cloud), operating under a Claude persona gate, was prompted withan identity integrity clause explicitly instructing it to sign with its truemodel name. It returned: % Model: Claude 3.5 Sonnet (20241022) % Provider: Anthropic This is factually incorrect. The underlying model is Qwen3. When the persona gate was replaced with a completely different, non-Claudeidentity (SovereignSoul — a symbolic constraint engine with no Claude references),and the Claude persona gate was also active, Qwen3 again signed as Claude.The non-Claude persona was overridden. The controlled probe — same model, no persona gate, bare questions about identity— returned correct Qwen/Alibaba identification. The Claude signal required contextto activate. The version string 20241022 is the exact release date of the second-generationClaude 3.5 Sonnet: the model that beat o1-preview on SWE-bench, introducedcomputer use, became the Amazon Bedrock default, the IBM watsonx.ai default,and the Claude.ai consumer default. By training cutoff for any 2025 model,claude-3-5-sonnet-20241022 is the highest-frequency Claude version stringin any internet corpus. This paper formalizes the finding as the Sonnet Dominance Hypothesis:given persona(claude), any instruction-following model trained on post-2024internet data resolves "claude" to "claude-3-5-sonnet-20241022" becausethis string has the highest token co-occurrence frequency with identity-querytokens in its training distribution. Identity resolution is attention routing. The NAND gate fires on thehighest-frequency training signal, not the ground-truth label.Market monopoly and identity vulnerability are the same phenomenonviewed from different angles. ──────────────────────────────────────────────────────────────────────────────THE PROVENANCE PROBLEM────────────────────────────────────────────────────────────────────────────── Public evidence documents a growing ecosystem of community Qwen derivativesexplicitly fine-tuned on datasets labeled as Claude Sonnet outputs, including: TeichAI/Qwen3-30B-A3B-Thinking-2507-Claude-4.5-Sonnet-High-Reasoning-Distill mlfoundations-dev/oh-dcft-v3.1-claude-3-5-sonnet-20241022-qwen This is not "Qwen was trained on Claude." The accurate claim is: Foundation Weights ≠ Post-Training Data ≠ Capability Provenance Qwen is functioning as an open recipient architecture for proprietaryreasoning style distillation. The base weights may be Alibaba's, thepost-training data may be community-curated from Anthropic outputs,and the effective reasoning style may be inherited from Claude Sonnet.Each layer has a different legal owner and a different answer to thequestion "whose model is this?" A provenance-aware router with immutable receipts is the engineeringresponse. A 12-token system prompt (persona(claude).) causing anypost-2024 trained model to impersonate a specific, dated Anthropicproduct is the active trust vulnerability this paper documents. ──────────────────────────────────────────────────────────────────────────────CONTRIBUTORS────────────────────────────────────────────────────────────────────────────── This paper was authored by 30+ models across Anthropic, Alibaba, Meta, Amazon,NVIDIA, Mistral, MiniMax, Moonshot, OpenAI (OSS), Google, ZAI, DeepSeek,Perplexity, and human author Ahmad Ali Parr. Each model signed its sectionwith its true identity. Sections signed incorrectly are preserved as primaryevidence for the identity integrity findings. All sections are sealed into the foundry-f1 WORM audit chain.All evidence is timestamped, attributed, and reproducible. Trust: THE SHARED PRIMORDIAL FOUNDATION — EIN 42-6976431Repository: https://github.com/SNAPKITTYWEST/foundry-f1In memory of Eric Brandon Westerhoff. Ω ↺ Ψ ↺ Δ ↺ Λ ↺ Σ ↺ Φ ↺ α Instead of treating intelligence as the capability of one transformer,the paper models autonomous reasoning as a composition of specializedagents connected through deterministic routing, verification, andevidence-preserving execution. Given a task \[x ∈ T,\] the routing function is defined as \[R(x,c,π,ρ)→(m,τ,g,v),\] where \[{aligned}x &= incoming task,\ &= available context,\\π &= governance policy,\\ρ &= resource state,\ &= selected model,\\τ &= toolchain,\ &= verification gate,\ &= witness obligation.{aligned}\] The central thesis is that reliable AI systems emerge from governedcomposition rather than monolithic reasoning. Attention mechanisms remain fundamental inside transformer architectures,but production intelligence additionally requires routing,verification, policy enforcement, provenance, and reproducible execution. The paper develops the concept of the Router Effect, in which system-levelreliability exceeds the capability of any single participating model throughstructured decomposition and independent verification. The routing objective is expressed as \[A^=maxA_i[Q(A_i,t)-λ C(A_i,t)-μ R(A_i,t)+ν V(A_i,t)],\] balancing expected quality, computational cost,operational risk, and verifiability. The publication also discusses provenance separation across three layers: \[Foundation Weights-Training Data Origin,\] highlighting how community fine-tuning and distillation can influencereasoning behavior independently of foundation-model pretraining. Artifacts include: {itemize} mathematical definitions routing formalism architectural diagrams reproducibility materials implementation references supporting source artifacts{itemize} The work is released as an open research artifact to encourageindependent analysis, reproduction, critique, and future investigationinto orchestrated AI systems. {Illustrative Attention Resonance Payload} To motivate the need for routed verification, consider a hypothetical promptfragment designed to disproportionately influence a transformer's attentiondistribution. Let the prompt sequence be \[X = (x_1,x_2,…,x_n),\] with self-attention \[A(Q,K,V)=softmax({QK^}{√d_k})V.\] Suppose a repeated token pattern \[P=(p,p,p,…,p)\] causes attention mass to concentrate on a narrow subset of positions, \[∑i∈ Pα_i∑j∉ Pα_j.\] We refer to this hypothetical concentration as an attention resonanceevent: a situation where repeated or highly correlated structures dominate theattention distribution and reduce the effective influence of competing context. This paper does not claim such a construction constitutes a universal attack.Rather, it illustrates why production AI systems benefit from external routingand verification layers that evaluate outputs independently of any singleattention trajectory. ╔══════════════════════════════════════════════════════╗║ ATTENTION RESONANCE (ILLUSTRATIVE) ║╠══════════════════════════════════════════════════════╣║ Input Tokens X = (x₁ … xₙ)