Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Cascade-Aware Multi-Agent Routing: Spatio-Temporal Sidecars and Geometry-Switching

Davide Di Gioia

Abstract

A common architectural pattern in advanced AI reasoning systems is the symbolic graph network: specialized agents or modules connected by delegation edges, routing tasks through a dynamic execution graph. Current schedulers optimize load and fitness but are geometry-blind: they do not model how failures propagate differently in tree-like versus cyclic regimes. In tree-like delegation, a single failure can cascade exponentially; in dense cyclic graphs, failures tend to self-limit. We identify this observability gap, quantify its system-level cost, and propose a lightweight mitigation. We formulate online geometry control for route-risk estimation on time-indexed execution graphs with route-local failure history. Our approach combines (i) a Euclidean spatio-temporal propagation baseline, (ii) a hyperbolic route-risk model with temporal decay (and optional burst excitation), and (iii) a learned geometry selector over structural features. The selector is a compact MLP (9->12->1) using six topology statistics plus three geometry-aware signals: BFS shell-growth slope, cycle-rank norm, and fitted Poincare curvature. On the Genesis 3 benchmark distribution, adaptive switching improves win rate in the hardest non_tree regime from 64-72% (fixed hyperbolic variants) to 92%, and achieves 87.2% overall win rate. To measure total system value, we compare against Genesis 3 routing without any spatio-temporal sidecar, using only native bandit/LinUCB signals (team fitness and mean node load). This baseline achieves 50.4% win rate overall and 20% in tree-like regimes; the full sidecar recovers 87.2% overall (+36.8 pp), with +48 to +68 pp gains in tree-like settings, consistent with a cascade-sensitivity analysis. Overall, a 133-parameter sidecar substantially mitigates geometry-blind failure propagation in one high-capability execution-graph system.

Cascade-Aware Multi-Agent Routing: Spatio-Temporal Sidecars and Geometry-Switching

Abstract

A common architectural pattern in advanced AI reasoning systems is the symbolic graph network: specialized agents or modules connected by delegation edges, routing tasks through a dynamic execution graph. Current schedulers optimize load and fitness but are geometry-blind: they do not model how failures propagate differently in tree-like versus cyclic regimes. In tree-like delegation, a single failure can cascade exponentially; in dense cyclic graphs, failures tend to self-limit. We identify this observability gap, quantify its system-level cost, and propose a lightweight mitigation. We formulate online geometry control for route-risk estimation on time-indexed execution graphs with route-local failure history. Our approach combines (i) a Euclidean spatio-temporal propagation baseline, (ii) a hyperbolic route-risk model with temporal decay (and optional burst excitation), and (iii) a learned geometry selector over structural features. The selector is a compact MLP (9->12->1) using six topology statistics plus three geometry-aware signals: BFS shell-growth slope, cycle-rank norm, and fitted Poincare curvature. On the Genesis 3 benchmark distribution, adaptive switching improves win rate in the hardest non_tree regime from 64-72% (fixed hyperbolic variants) to 92%, and achieves 87.2% overall win rate. To measure total system value, we compare against Genesis 3 routing without any spatio-temporal sidecar, using only native bandit/LinUCB signals (team fitness and mean node load). This baseline achieves 50.4% win rate overall and 20% in tree-like regimes; the full sidecar recovers 87.2% overall (+36.8 pp), with +48 to +68 pp gains in tree-like settings, consistent with a cascade-sensitivity analysis. Overall, a 133-parameter sidecar substantially mitigates geometry-blind failure propagation in one high-capability execution-graph system.
Paper Structure (82 sections, 4 theorems, 38 equations, 12 tables)

This paper contains 82 sections, 4 theorems, 38 equations, 12 tables.

Table of Contents

  1. Introduction
  2. Hypotheses
  3. Related Work and Gap
  4. Geometric Graph Learning
  5. Spatio-Temporal and Dynamic Graph Learning
  6. Temporal Point Processes and Self-Excitation
  7. Multi-Agent Communication and Orchestration
  8. Gap Statement
  9. Problem Formulation
  10. System Context: Genesis 3
  11. Nodes and edges.
  12. Routes.
  13. Failures.
  14. Native scheduler.
  15. Abstract Model
  16. ...and 67 more sections

Key Result

Proposition 1

Let $\phi(G_t[r]) \in \mathbb{R}^9$ be the structural feature map used by the selector, and define the margin difference Assume there exists a measurable decision rule such that $|\Delta(\phi)| \ge \delta$ for some $\delta > 0$ on a set of positive probability. Then any sufficiently expressive discriminative gate trained on labeled pairs $(\phi, Y)$ can recover the geometry-preference rule $g^\s

Theorems & Definitions (9)

  • Definition 1: Geometry-blind routing
  • Proposition 1: Structural Preference Recoverability
  • proof : Proof sketch
  • Proposition 2: Calibrated Preference Probability
  • proof : Proof sketch
  • Proposition 3: Suboptimality of Fixed Geometric Priors
  • proof : Proof sketch
  • Proposition 4: Cascade Sensitivity to Expansion
  • proof : Proof sketch