Refinement

Allocate specialized agent teams for modular tasking, decompose objectives, orchestrate event-driven microservices, and refine compliance overlays.

Specialized Agent Teams

The Elastic Council architecture dynamically selects and activates agents based on Expected Value of Sample Information (EVSI) calculations. Core agents remain always active while specialists are activated on demand.

Strategic Orchestrator

Always Active

Task coordination and resource allocation

Research Synthesizer

Always Active

Information aggregation and synthesis

Quality Sentinel

Always Active

Output validation and quality assurance

Ethical Guardian

Always Active

Compliance and ethical oversight

Metacognitive Orchestrator

Always Active

Self-monitoring and adaptation

Domain Specialists

EVSI-Activated

Industry-specific analysis

EVSI-Based Agent Activation

Expected Value of Sample Information (EVSI) quantifies the marginal utility of activating additional agents for a given task.

The Elastic Council Router uses Thompson Sampling combined with EVSI calculations to determine which agents should be activated for each task. This ensures optimal resource utilization while maintaining analytical quality.

Activation Threshold

Agents are activated when their EVSI score exceeds the configured threshold (default: 0.15). The system continuously monitors agent performance and adjusts activation probabilities using Bayesian updating.

elastic_council:
  evsi_threshold: 0.15
  max_agents: 20
  always_active:
    - strategic_orchestrator
    - research_synthesizer
    - quality_sentinel
    - ethical_guardian
    - metacognitive_orchestrator
  sampling_method: thompson
  update_frequency: per_task

Task Decomposition

Intake

Parse and validate incoming task specifications

Output:Structured task object

Analysis

Identify subtasks and dependencies

Output:Task dependency graph

Allocation

Assign agents based on EVSI calculations

Output:Agent assignment matrix

Scheduling

Optimize execution order and parallelization

Output:Execution timeline

Compliance Overlay Configuration

Protocol	Description	Trigger
Schema Validation	Validate all data against IQAS v5.3 schema specifications	Data ingestion
Provenance Logging	Record UUID-tagged lineage for all transformations	Every operation
Contradiction Check	Flag and resolve conflicting agent outputs	Agent output
Audit Trail Update	Append to hash chain with timestamp and signature	State change

Assessment Next: Computation