|
| 1 | +# SIARE Glossary |
| 2 | + |
| 3 | +Definitions of key terms used throughout the SIARE documentation. |
| 4 | + |
| 5 | +--- |
| 6 | + |
| 7 | +## Core Concepts |
| 8 | + |
| 9 | +### SOP (Standard Operating Procedure) |
| 10 | +In SIARE, an SOP is a complete pipeline configuration that defines how agents collaborate. Represented as `ProcessConfig` in code. An SOP includes roles, graph structure, prompts, and constraints. |
| 11 | + |
| 12 | +**Also called:** Pipeline, ProcessConfig |
| 13 | + |
| 14 | +### Role |
| 15 | +An individual agent within a pipeline. Each role has a model, tools, prompts, and defined inputs/outputs. In user-facing documentation, roles may be called "agents" for clarity. |
| 16 | + |
| 17 | +**Code:** `RoleConfig` |
| 18 | +**User-facing:** Agent |
| 19 | + |
| 20 | +### Director |
| 21 | +The AI service that analyzes pipeline performance and proposes improvements. Acts as the "brain" of the evolution system - both diagnosing weaknesses and suggesting mutations. |
| 22 | + |
| 23 | +**Code:** `DirectorService` |
| 24 | + |
| 25 | +### Gene Pool |
| 26 | +Storage for SOP versions and their performance history. Tracks ancestry (parent-child relationships), computes Pareto frontiers, and maintains diversity via QD Grid. |
| 27 | + |
| 28 | +**Code:** `GenePool` |
| 29 | + |
| 30 | +### Mutation |
| 31 | +A targeted change to an SOP designed to improve performance. SIARE supports 7 mutation types: PROMPT_CHANGE, PARAM_TWEAK, ADD_ROLE, REMOVE_ROLE, REWIRE_GRAPH, CROSSOVER, META_PROMPT_CHANGE. |
| 32 | + |
| 33 | +**Code:** `MutationType`, `MutationProposal` |
| 34 | + |
| 35 | +--- |
| 36 | + |
| 37 | +## Evolution Terminology |
| 38 | + |
| 39 | +### Evolution Job |
| 40 | +A complete evolution run with defined budget, constraints, and phases. Orchestrated by the `EvolutionScheduler`. |
| 41 | + |
| 42 | +### Generation |
| 43 | +One iteration of the evolution loop: execute → evaluate → diagnose → mutate. |
| 44 | + |
| 45 | +### Pareto Frontier |
| 46 | +The set of SOPs where no single SOP dominates all others across all metrics. Represents optimal trade-offs (e.g., high accuracy vs. low cost). |
| 47 | + |
| 48 | +### QD Grid (Quality-Diversity Grid) |
| 49 | +A data structure that maintains diverse solutions across behavioral dimensions. Prevents evolution from converging to a single local optimum. |
| 50 | + |
| 51 | +### Fitness |
| 52 | +The overall performance score of an SOP, typically a weighted combination of multiple metrics. |
| 53 | + |
| 54 | +--- |
| 55 | + |
| 56 | +## Metrics & Evaluation |
| 57 | + |
| 58 | +### EvaluationVector |
| 59 | +A collection of metric scores for a single SOP execution. Contains individual metric values plus an aggregate score. |
| 60 | + |
| 61 | +**Code:** `EvaluationVector` |
| 62 | + |
| 63 | +### LLM Judge |
| 64 | +A metric that uses an LLM to evaluate quality. Common for subjective measures like "answer quality" or "relevance." |
| 65 | + |
| 66 | +**Code:** `MetricType.LLM_JUDGE` |
| 67 | + |
| 68 | +### Programmatic Metric |
| 69 | +A metric computed by code (not LLM). Examples: latency, cost, word count. |
| 70 | + |
| 71 | +**Code:** `MetricType.PROGRAMMATIC` |
| 72 | + |
| 73 | +### Runtime Metric |
| 74 | +A metric that measures execution characteristics like latency or API cost. |
| 75 | + |
| 76 | +**Code:** `MetricType.RUNTIME` |
| 77 | + |
| 78 | +--- |
| 79 | + |
| 80 | +## Pipeline Structure |
| 81 | + |
| 82 | +### Graph |
| 83 | +The DAG (Directed Acyclic Graph) that defines execution order and data flow between roles. |
| 84 | + |
| 85 | +**Code:** `GraphEdge` |
| 86 | + |
| 87 | +### Conditional Edge |
| 88 | +A graph edge that only executes if a condition is met. Enables dynamic routing based on intermediate outputs. |
| 89 | + |
| 90 | +**Code:** `GraphEdge.condition` |
| 91 | + |
| 92 | +### PromptGenome |
| 93 | +The collection of prompts for all roles in an SOP. The "DNA" that gets mutated during evolution. |
| 94 | + |
| 95 | +**Code:** `PromptGenome` |
| 96 | + |
| 97 | +### PromptConstraints |
| 98 | +Rules that limit what evolution can change in a prompt. Protects critical instructions from modification. |
| 99 | + |
| 100 | +**Code:** `PromptConstraints` |
| 101 | + |
| 102 | +--- |
| 103 | + |
| 104 | +## Tools & Adapters |
| 105 | + |
| 106 | +### Tool Adapter |
| 107 | +A class that wraps external functionality (APIs, databases, search engines) for use by agents. |
| 108 | + |
| 109 | +**Base class:** `ToolAdapter` |
| 110 | + |
| 111 | +### Tool Registry |
| 112 | +A global registry of available tool adapters. Adapters register via the `@register_adapter` decorator. |
| 113 | + |
| 114 | +**Code:** `ToolRegistry` |
| 115 | + |
| 116 | +--- |
| 117 | + |
| 118 | +## Prompt Evolution |
| 119 | + |
| 120 | +### TextGrad |
| 121 | +A prompt optimization strategy that uses LLM-generated critiques as "gradients" to improve prompts. |
| 122 | + |
| 123 | +**Code:** `TextGradStrategy` |
| 124 | + |
| 125 | +### EvoPrompt |
| 126 | +A prompt optimization strategy based on evolutionary algorithms (genetic algorithms or differential evolution). |
| 127 | + |
| 128 | +**Code:** `EvoPromptStrategy` |
| 129 | + |
| 130 | +### MetaPrompt |
| 131 | +A prompt optimization strategy that uses LLM meta-analysis to identify patterns and propose improvements. |
| 132 | + |
| 133 | +**Code:** `MetaPromptStrategy` |
| 134 | + |
| 135 | +--- |
| 136 | + |
| 137 | +## API & Infrastructure |
| 138 | + |
| 139 | +### Execution Trace |
| 140 | +A complete record of one pipeline run, including all role inputs/outputs, timing, and costs. |
| 141 | + |
| 142 | +**Code:** `ExecutionTrace` |
| 143 | + |
| 144 | +### ConfigStore |
| 145 | +Persistent storage for SOPs, prompts, and configurations. |
| 146 | + |
| 147 | +**Code:** `ConfigStore` |
| 148 | + |
| 149 | +### Circuit Breaker |
| 150 | +Fault tolerance mechanism that prevents repeated failures from overwhelming the system. |
| 151 | + |
| 152 | +**Code:** `CircuitBreaker` |
| 153 | + |
| 154 | +--- |
| 155 | + |
| 156 | +## Terminology Mapping |
| 157 | + |
| 158 | +| User-Facing Term | Code Term | Description | |
| 159 | +|------------------|-----------|-------------| |
| 160 | +| Agent | Role | An individual component in the pipeline | |
| 161 | +| Pipeline | SOP / ProcessConfig | The complete configuration | |
| 162 | +| Evolution | Mutation + Selection | The optimization process | |
| 163 | +| Score | EvaluationVector | Performance metrics | |
| 164 | +| Memory | Context | Information passed between agents | |
| 165 | + |
| 166 | +--- |
| 167 | + |
| 168 | +## See Also |
| 169 | + |
| 170 | +- [Architecture](architecture/SYSTEM_ARCHITECTURE.md) — System design |
| 171 | +- [Data Models](architecture/DATA_MODELS.md) — Complete model reference |
| 172 | +- [Mutation Operators](reference/mutation-operators.md) — All 7 mutation types |
0 commit comments