CDIP v3.1 PERIMETER_AUX Ricci Flow on Knowledge Graphs v0.2-RC Specification

This document specifies PERIMETER_AUX/RICCI_FLOW as a strict NI-2 observational sidecar layer for CDIP v3.*, implementing discrete Ollivier–Ricci curvature on a read-only knowledge graph snapshot and a virtual (shadow) Ricci flow on shadow edge weights. The layer detects early signs of cognitive rigidity (high positive curvature / contraction dominance) and bottleneck expansion (negative curvature regions) without modifying any sealed upstream gate sequences (CORE/OPA/G-τ .4/Telomere/TAO-FQC/MDIT-CS/Fisher–Finsler).Key distinction from MDIT-CS: MDIT-CS monitors connectivity collapse via minimum cut invariants (e.g., κG). PERIMETER_AUX/RICCI_FLOW monitors geometric singularity signa- tures and structural stiffness via curvature evolution on the knowledge graph topology. Joint observation detects the regime “connected but rigid”.This is an alerts-only layer: all outcomes are confined to PERIMETER_AUX/RICCI_FLOW and SHALL NOT be lifted into TAO/FQC verdicts.