Update branding to JEV InvariantOne-v1 and add agent routing & decision tags

README.md

@@ -7,15 +7,21 @@ base_model:
 pipeline_tag: text-classification
 library_name: invariantone
 tags:
+- jev
+- agent-routing
+- tool-selection
+- agentic-ai
+- order-bias
+- decision-making
 - decision-model
 - permutation-equivariance
 - lora
 - qwen
 ---
 
-# Model Card: InvariantOne-v1
+# Model Card: JEV InvariantOne-v1
 
-**Release Version:** `InvariantOne-v1`  
+**Release Version:** `InvariantOne-v1` (JEV InvariantOne)  
 **Date:** September 24, 2026  
 **License:** Apache-2.0  
 **Base Backbone:** `Qwen/Qwen3.5-4B-Base` (Revision `1001bb4d826a52d1f399e183466143f4da7b741b`, Native BF16)  
@@ -23,11 +29,11 @@ tags:
 
 ---
 
-## Try InvariantOne
+## Try JEV InvariantOne
 
 [Interactive Demo](https://huggingface.co/spaces/TanmaySah/InvariantOne-Demo) | [GitHub](https://github.com/tradertanmay/InvariantOne)
 
-Try InvariantOne-v1 directly in the browser on structured multi-candidate decision problems.
+Try JEV InvariantOne-v1 directly in the browser on structured multi-candidate decision problems and AI agent routing.
 
 ---
 
@@ -63,11 +69,11 @@ print(result.probabilities)
 
 ---
 
-## 1. Model Overview
+## 1. Model Overview: JEV InvariantOne
 
-`InvariantOne-v1` is an order-equivariant probabilistic decision model designed for multi-candidate decision-making under complex operational constraints and relational rules. 
+`JEV InvariantOne-v1` is an order-equivariant probabilistic decision model designed for multi-candidate decision-making, AI agent routing, and tool selection under complex operational constraints and relational rules. 
 
-Standard autoregressive language models suffer from candidate-order and position sensitivity in joint multi-option evaluation. `InvariantOne-v1` addresses this structural sensitivity through a decoupled comparative architecture:
+Standard autoregressive language models suffer from candidate-order and position sensitivity in joint multi-option evaluation. `JEV InvariantOne-v1` addresses this structural sensitivity through a decoupled comparative architecture:
 1. **Candidate-Independent Scoring:** Each candidate action is scored independently conditioned on context, eliminating candidate order sensitivity and cross-option attention artifacts.
 2. **Analytic Distributional Equivariance:** Option-span pooled representations pass through a lightweight direct comparative projection head, producing scalar scores normalized via calibrated softmax. The architecture is mathematically permutation-equivariant by construction in real arithmetic.
 3. **Upper-Layer Relational Pre-Adaptation:** Parameter-efficient adaptation (LoRA on layers 28–31) trained across a diverse relational operator universe (N_op=16) instills strong, transferable relational reasoning capabilities into the comparative readout.