Skip to main content
5.3x speedup over sequential execution

SPOQ

Specialist Orchestrated Queuing

A methodology for multi-agent software engineering with wave-based parallel execution and structured quality gates.

Learn the MethodologyView Paper

Key Innovations

SPOQ addresses coordination overhead, quality control gaps, and limited human oversight through three integrated innovations.

Wave-Based Dispatch
Dependency-aware parallel execution that respects task ordering while maximizing concurrency. Tasks within the same wave execute simultaneously, reducing total completion time.
Dual Validation Gates
Separate quality frameworks for planning (10 metrics, 90+ threshold) and code (10 metrics, 80+ threshold). Both gates must pass before work proceeds, ensuring consistent output quality.
Hierarchical Agent Architecture (HaaA)
Three-tier model selection optimizes cost and capability. Opus handles complex work, Haiku investigates quickly, and Sonnet reviews thoroughly. Each tier serves a distinct purpose.

Three-Tier Agent Hierarchy

Each tier is optimized for its role, balancing capability against cost to maximize value.

Agent roles and their characteristics in the SPOQ system
TierModelRoleTradeoff
WorkerClaude OpusExecute complex implementation tasks requiring deep reasoning and code generationHighest capability, higher cost per token
InvestigatorClaude HaikuRapid codebase exploration and information gathering before task executionFast and economical, limited reasoning depth
ReviewerClaude SonnetValidate completed work against quality metrics with balanced analysisBalanced capability and cost for thorough review

Design Principles

SPOQ is built on principles that distinguish it from prior multi-agent approaches.

  1. Deterministic Quality

    Quality gates use objective, measurable criteria. Validation produces consistent scores regardless of which agent performs the review, eliminating subjective interpretation.

  2. Explicit Dependencies

    All task relationships are declared upfront in structured YAML. The orchestrator builds a directed acyclic graph (DAG) to determine safe parallel execution order.

  3. Atomic Task Boundaries

    Each task represents a single, well-defined unit of work. Clear boundaries enable independent validation and prevent scope creep during execution.

  4. Fail-Fast Validation

    Quality checks run immediately after task completion. Failed validations halt the pipeline early, preventing wasted computation on dependent tasks.

  5. Transparent Provenance

    Every decision, validation score, and output traces back to its source. Complete audit trails enable debugging and continuous methodology improvement.

Pipeline Overview

Four-stage process from epic planning through agent validation

The pipeline consists of five stages: Stage 1 is Epic Planning where goals are defined and decomposed into tasks. Then Epic Validation checks task definitions against a 95 average with 90 minimum threshold. Stage 2 is Agent Execution where parallel agent swarms work on tasks. Then Code Validation ensures quality with a 95 average and 80 minimum threshold. Finally, the Complete stage indicates work is ready for integration. Failed validations trigger revision loops back to the previous stage.