# Dynamic Attentional Context Scoping: Agent-Triggered Focus Sessions for Isolated Per-Agent Steering in Multi-Agent LLM Orchestration

Source: [Dynamic Attentional Context Scoping: Agent-Triggered Focus Sessions for Isolated Per-Agent Steering in Multi-Agent LLM Orchestration](<https://www.sota2.com/research/paper/69d883fb7d0172140ec77999>)

Collection index: [llms.txt](<https://www.sota2.com/research/llms.txt>)

Data updated: 2026-04-11T05:00:04.811Z

Authors: Nickson Patel

Published: 2026-04-09T07:26:42.000Z

## Abstract

Multi-agent LLM orchestration systems suffer from context pollution: when N concurrent agents compete for the orchestrator's context window, each agent's task state, partial outputs, and pending questions contaminate the steering interactions of every other agent, degrading decision quality. We introduce Dynamic Attentional Context Scoping (DACS), a mechanism in which the orchestrator operates in two asymmetric modes. In Registry mode it holds only lightweight per-agent status summaries (\<=200 tokens each), remaining responsive to all agents and the user. When an agent emits a SteeringRequest, the orchestrator enters Focus(a\_i) mode, injecting the full context of agent a\_i while compressing all other agents to their registry entries. Context isolation is agent-triggered, asymmetric, and deterministic: the context window contains exactly F(a\_i) + R\_{-i} during steering, eliminating cross-agent contamination without requiring context compression or retrieval. We evaluate DACS across four experimental phases totalling 200 trials: Phase 1 tests N in {3,5,10} (60 trials); Phase 2 tests agent heterogeneity and adversarial dependencies (60 trials); Phase 3 tests decision density up to D=15 (40 trials); Phase 4 uses autonomous LLM agents for free-form questions (40 trials, Claude Haiku 4.5). Across all 8 synthetic scenarios, DACS achieves 90.0--98.4% steering accuracy versus 21.0--60.0% for a flat-context baseline (p \< 0.0001 throughout), with wrong-agent contamination falling from 28--57% to 0--14% and context efficiency ratios of up to 3.53x. The accuracy advantage grows with N and D; keyword matching is validated by LLM-as-judge across all phases (mean kappa=0.909). DACS outperforms the flat-context baseline by +17.2pp at N=3 (p=0.0023) and +20.4pp at N=5 (p=0.0008) in Phase 4, with the advantage growing with N confirmed by two independent judges.

## Recorded benchmark results

A research result applies to its stated dataset, split, metric and evaluation configuration. Compare like-for-like settings, preserve the metric's unit and direction, and distinguish paper-reported results from independent reproduction. The best result recorded here is not a claim of exhaustive, current global coverage.

- [Multi-agent steering on Synthetic Multi-agent Scenarios](<https://www.sota2.com/research/sota/multi-agent-steering-on-synthetic-multi-agent-scenarios.md>): Multi-agent steering; Synthetic Multi-agent Scenarios; Accuracy: 98.4.

- [Steering Validation on Real Agents LLMAgent, Claude Haiku 4.5 N-scaling Phase 4 (val)](<https://www.sota2.com/research/sota/steering-validation-on-real-agents-llmagent-claude-haiku-4-5-n-scaling-phase-4-val.md>): Steering Validation; Real Agents LLMAgent, Claude Haiku 4.5 N-scaling Phase 4 (val); Haiku Judge Accuracy: 83.7.

- [Steering Validation on Synthetic Stubs Comparison Phase 1](<https://www.sota2.com/research/sota/steering-validation-on-synthetic-stubs-comparison-phase-1.md>): Steering Validation; Synthetic Stubs Comparison Phase 1; Haiku Judge Accuracy: 96.7.

## Original sources and implementation

[Original paper](<https://arxiv.org/abs/2604.07911>)

[PDF](<https://arxiv.org/pdf/2604.07911.pdf>)