SPOQ Glossary
AI Agent Orchestration Terminology
- Epic
- High-level goal with a set of atomic tasks, dependencies, and success criteria. An epic is the top-level unit of work in SPOQ, decomposed into a dependency graph of tasks that are dispatched across parallel waves.
- Task
- 1-4 hours of focused work with one clear deliverable. Tasks are the atomic units of execution in SPOQ -- each has explicit acceptance criteria, dependency declarations, and a complexity estimate.
- Wave
- Group of tasks with no inter-dependencies that can execute in parallel. Waves are computed by topologically sorting the task dependency graph. All tasks within a wave run simultaneously, and the next wave begins only after the current wave completes.
- Phase
- Synonym for wave; the field name used in task YAML definitions. Each task declares its phase number, which determines which wave it belongs to during parallel dispatch.
- Dependency Graph (DAG)
- Directed acyclic graph of task prerequisites that determines execution order. SPOQ builds a DAG from task dependency declarations and topologically sorts it into parallel execution waves. Cycles are rejected during validation.
- Planning Validation
- Pre-execution quality gate scoring 10 metrics with a 95 average and 90 per-metric minimum threshold. This gate evaluates epic quality before any agents begin work, catching structural problems early. See Methodology →
- Agent Validation
- Post-execution quality gate scoring 10 metrics with a 95 average and 80 per-metric minimum threshold. This gate evaluates delivered code after agent execution, ensuring quality before integration. See Methodology →
- HaaA (Human-as-an-Agent)
- Bidirectional human-AI collaboration paradigm where the developer participates as a first-class agent. The human shapes task decomposition, validates plans, and can be consulted by AI agents during execution -- rather than being excluded from the process. See Methodology →
- Quality Amplification
- The synergistic effect of combining human judgment with AI execution. Human judgment improves task decomposition and plan quality, while AI agents scale the human's productivity across parallel execution waves.
- Worker Agent
- Opus-tier agent that executes implementation tasks. Workers handle the most complex coding, reasoning, and generation work. They operate at the highest capability tier, trading cost for deep reasoning ability.
- Reviewer Agent
- Sonnet-tier agent for quality assurance scoring. Reviewers evaluate completed work against the 10 code quality metrics, providing balanced capability at moderate cost for thorough assessment.
- Investigator Agent
- Haiku-tier agent for build failure triage and codebase exploration. Investigators handle lightweight tasks like error diagnosis and information gathering, optimized for speed and economy.
- Wave-Based Dispatch
- Topological sorting of a directed acyclic graph (DAG) into parallel execution waves for maximum throughput. Independent tasks run simultaneously within each wave, while dependency ordering is maintained between waves.
- Dual Validation Gates
- Two-gate quality system: planning validation before execution and agent validation after. Each gate scores against 10 structured metrics with explicit pass/fail thresholds, ensuring quality at both the planning and implementation stages.
- Critical Path
- Longest dependency chain in the task DAG, determining the minimum possible completion time. Even with unlimited parallelism, total execution time cannot be shorter than the sum of tasks along the critical path.
- Speedup
- Ratio of sequential to parallel execution time. For example, a 5.3x speedup means parallel wave-based dispatch completed work 5.3 times faster than sequential single-agent execution. SPOQ achieved speedups from 1.3x to 5.3x across case studies.