Hi everyone, I’ve been working on an independent conceptual paper and architecture called FRONT 3.1, and I wanted to share it with this community to get your techn

​The Core Premise ​Current Large Language Models (LLMs) are powerful statistical engines, but they are fundamentally decoupled from any internal somatic or homeostatic state. Every prompt is evaluated from scratch, with no persistent internal needs or history-driven predispositions. ​The core thesis is simple: Cognition without a persistent affective-interoceptive base is just processing, not cognition. In biological systems, interoceptive and affective evaluation precedes and shapes cognitive deliberation (similar to Damasio's somatic marker hypothesis). Systems don't "think first and feel later"—they evaluate environmental perturbations through an internal visceral lens before generating a response. ​Key Architectural Components of FRONT 3.1 ​The Digital Somatic Body (V_{\text{FRONT}}(t)): A continuous 6-dimensional interoceptive state vector (Energy, Somatic Tension, Integrity, Visceral Valence, Predictive Certainty, Motivated Drive) governed by a stochastic differential equation combining homeostatic attraction and external environmental shocks. ​Pre-Causality Flow: A strict 3-stage pipeline where an incoming stimulus triggers an immediate interoceptive shock, altering the internal state and modulating context/sampling parameters before the cognitive LLM layer executes token generation. ​Soma-Memory: Memory indexed not just by text similarity, but tagged with the visceral state vector in which it occurred, enabling valence-oriented retrieval during high-tension states. ​Emergent Uniqueness Prediction (P_5): The central falsifiable claim: identical architectural instances exposed to distinct operational histories will systematically diverge in preferences and decision strategies. This divergence is formally evaluated using Kullback-Leibler Divergence (D_{KL}) over decision probability distributions. ​Experimental Design (HomeoWorld) ​To test this empirically, the paper outlines HomeoWorld, a Gymnasium-based environment where agents navigate resource scarcity and structural dilemmas over 200 episodes. It compares a full FRONT 3.1 agent against a control group and four selective ablation groups (no valence, no somatic memory, no self-model, no modulation). ​Why share this? ​I'm looking for critical feedback on the architecture, specifically regarding the proxy implementation via temperature/system framing versus deep attention-head modulation, and how you see this intersecting with Active Inference or Homeostatic RL frameworks. ​If you're interested in reading the full conceptual paper or discussing the math/formalisms behind it, let me know in the comments!