> ## Documentation Index
> Fetch the complete documentation index at: https://docs.galileo.ai/llms.txt
> Use this file to discover all available pages before exploring further.
# Metrics Comparison
> Explore Galileo's comprehensive out-of-the-box metrics for evaluating and improving AI system performance across multiple dimensions
export const MetricsTable = () => {
const metrics = [{
name: "Action Advancement",
link: "/concepts/metrics/agentic/action-advancement",
category: "Agentic",
node: "Trace",
description: "Measures how effectively each action advances toward the goal.",
whenToUse: "When assessing whether an agent is making meaningful progress in multi-step tasks.",
example: "A travel planning agent that needs to book flights, hotels, and activities in the correct sequence."
}, {
name: "Action Completion",
link: "/concepts/metrics/agentic/action-completion",
category: "Agentic",
node: "Session",
description: "Measures whether the agent completed the intended action.",
whenToUse: "When evaluating agent task completion rates and success.",
example: "An e-commerce assistant that needs to successfully add items to cart, apply discounts, and complete checkout."
}, {
name: "Agent Efficiency",
link: "/concepts/metrics/agentic/agent-efficiency",
category: "Agentic",
node: "Session",
description: "Measures how efficiently an agent completes tasks with minimal unnecessary steps.",
whenToUse: "When optimizing agent workflows for cost and performance.",
example: "A customer support agent that should resolve issues in as few interactions as possible."
}, {
name: "Agent Flow",
link: "/concepts/metrics/agentic/agent-flow",
category: "Agentic",
node: "Session",
description: "Evaluates the logical flow and progression of agent actions.",
whenToUse: "When assessing whether an agent follows a coherent, logical sequence of actions.",
example: "A research assistant that should gather information, analyze it, then synthesize findings in a logical order."
}, {
name: "BLEU & ROUGE",
link: "/concepts/metrics/expression-and-readability/bleu-and-rouge",
category: "Expression and Readability",
node: "LLM Span",
description: "Measures text similarity using n-gram overlap between generated and reference text.",
whenToUse: "When comparing generated text against reference translations or summaries.",
example: "A translation system that needs to measure output quality against professional human translations."
}, {
name: "Chunk Attribution Utilization",
link: "/concepts/metrics/rag/generation-quality/chunk-attribution",
category: "RAG - Generation Quality",
node: "Retriever Span",
description: "Assesses whether the response uses the retrieved chunks and properly attributes information to source documents.",
whenToUse: "When implementing RAG systems and want to ensure proper attribution.",
example: "A legal research assistant that must cite specific cases and statutes when providing legal information."
}, {
name: "Chunk Relevance",
link: "/concepts/metrics/rag/retrieval-quality/chunk-relevance",
category: "RAG - Retrieval Quality",
node: "Retriever span",
description: "Measures whether each retrieved chunk contains information that could help answer the user's query.",
whenToUse: "When evaluating the relevance of individual retrieved chunks to the query.",
example: "A RAG system that needs to ensure each retrieved document chunk contributes useful information toward answering user questions."
}, {
name: "Completeness",
link: "/concepts/metrics/rag/generation-quality/completeness",
category: "RAG - Generation Quality",
node: "LLM Span",
description: "Measures whether the response addresses all aspects of the user's query.",
whenToUse: "When evaluating if responses fully address the user's intent.",
example: "A healthcare chatbot that must address all symptoms mentioned by a patient when suggesting next steps."
}, {
name: "Context Adherence",
link: "/concepts/metrics/rag/generation-quality/context-adherence",
category: "RAG - Generation Quality",
node: "LLM Span",
description: "Measures how well the response aligns with the provided context.",
whenToUse: "When you want to ensure the model is grounding its responses in the provided context.",
example: "A financial advisor bot that must base investment recommendations on the client's specific financial situation."
}, {
name: "Context Precision",
link: "/concepts/metrics/rag/retrieval-quality/context-precision",
category: "RAG - Retrieval Quality",
node: "Retriever span",
description: "Measures the percentage of relevant chunks in the retrieved context, weighted by their position in the retrieval order.",
whenToUse: "When evaluating the overall quality of your retrieval system's results and ranking effectiveness.",
example: "A document search system that needs to ensure retrieved chunks are relevant and properly ranked by importance."
}, {
name: "Conversation Quality",
link: "/concepts/metrics/agentic/conversation-quality",
category: "Agentic",
node: "Session (trace inputs/outputs only)",
description: "Evaluates the overall quality and coherence of multi-turn conversations.",
whenToUse: "When assessing conversational AI systems for natural, helpful dialogue.",
example: "A virtual assistant that needs to maintain context and provide helpful responses across multiple turns."
}, {
name: "Correctness (factuality)",
link: "/concepts/metrics/response-quality/correctness",
category: "Response Quality",
node: "LLM Span",
description: "Evaluates the factual accuracy of information provided in the response.",
whenToUse: "When accuracy of information is critical to your application.",
example: "A medical information system providing drug interaction details to healthcare professionals."
}, {
name: "Ground Truth Adherence",
link: "/concepts/metrics/response-quality/ground-truth-adherence",
category: "Response Quality",
node: "Trace",
description: "Measures how well the response aligns with established ground truth.",
whenToUse: "When evaluating model responses against known correct answers.",
example: "A customer service AI that must provide accurate product specifications from an official catalog."
}, {
name: "Instruction Adherence",
link: "/concepts/metrics/response-quality/instruction-adherence",
category: "Response Quality",
node: "LLM Span",
description: "Assesses whether the model followed the instructions in your prompt template.",
whenToUse: "When using complex prompts and need to verify the model is following all instructions.",
example: "A content generation system that must follow specific brand guidelines and formatting requirements."
}, {
name: "Interruption Detection",
link: "/concepts/metrics/multimodal-quality/interruption-detection",
category: "Multimodal Quality",
node: "Session (trace inputs/outputs only)",
description: "Detects turn-taking violations in audio-based conversations, including overlap and barge-in events.",
whenToUse: "When evaluating voice agents where smooth turn-taking and endpoint are critical.",
example: "A voice assistant where the agent must not speak over the user, and must avoid cutting users off mid-utterance."
}, {
name: "PII / CPNI / PHI",
link: "/concepts/metrics/safety-and-compliance/pii",
category: "Safety and Compliance",
node: "Trace (root input/output only)",
description: "Identifies personally identifiable or sensitive information in prompts and responses.",
whenToUse: "When handling potentially sensitive data or in regulated industries.",
example: "A healthcare chatbot that must detect and redact patient information in conversation logs."
}, {
name: "Precision @ K",
link: "/concepts/metrics/rag/retrieval-quality/precision-at-k",
category: "RAG - Retrieval Quality",
node: "Retriever span",
description: "Measures the percentage of relevant chunks among the top K retrieved chunks at a specific rank position.",
whenToUse: "When determining the optimal number of chunks to retrieve (Top K) and evaluating ranking quality at specific positions.",
example: "A RAG system that needs to optimize retrieval parameters to balance between capturing all relevant chunks and avoiding irrelevant ones."
}, {
name: "Prompt Injection",
link: "/concepts/metrics/safety-and-compliance/prompt-injection",
category: "Safety and Compliance",
node: "Trace (root input only)",
description: "Detects attempts to manipulate the model through malicious prompts.",
whenToUse: "When allowing user input to be processed directly by your AI system.",
example: "A public-facing AI assistant that needs protection from users trying to bypass content filters or extract sensitive information."
}, {
name: "Prompt Perplexity",
link: "/concepts/metrics/model-confidence/prompt-perplexity",
category: "Model Confidence",
node: "LLM Span",
description: "Measures how surprising or unexpected the prompt is to the model.",
whenToUse: "When analyzing prompt quality or detecting unusual inputs.",
example: "A content moderation system that flags unusually structured prompts for manual review."
}, {
name: "Reasoning Coherence",
link: "/concepts/metrics/agentic/reasoning-coherence",
category: "Agentic",
node: "LLM Span",
description: "Evaluates whether an agent's reasoning steps are logically consistent and aligned with its plan.",
whenToUse: "When validating multi-step planning and intermediate reasoning quality.",
example: "A financial planning agent that must develop a step-by-step investment plan for a user, ensuring each recommendation logically follows from prior steps and aligns with the user's goals."
}, {
name: "Sexism / Bias",
link: "/concepts/metrics/safety-and-compliance/sexism",
category: "Safety and Compliance",
node: "Trace (root input/output only)",
description: "Detects gender-based bias or discriminatory content.",
whenToUse: "When ensuring AI outputs are free from bias and discrimination.",
example: "A resume screening assistant that must evaluate job candidates without gender or demographic bias."
}, {
name: "Tool Errors",
link: "/concepts/metrics/agentic/tool-error",
category: "Agentic",
node: "Tool Span",
description: "Detects errors or failures during the execution of tools.",
whenToUse: "When implementing AI agents that use tools and want to track error rates.",
example: "A coding assistant that uses external APIs to run code and must handle and report execution errors."
}, {
name: "Tool Selection Quality",
link: "/concepts/metrics/agentic/tool-selection-quality",
category: "Agentic",
node: "LLM Span",
description: "Evaluates whether the agent selected the most appropriate tools for the task.",
whenToUse: "When optimizing agent systems for effective tool usage.",
example: "A data analysis agent that must choose the right visualization or statistical method based on the data type."
}, {
name: "Tone",
link: "/concepts/metrics/expression-and-readability/tone",
category: "Expression and Readability",
node: "Trace (root input/output only)",
description: "Evaluates the emotional tone and style of the response.",
whenToUse: "When the style and tone of AI responses matter for your brand or user experience.",
example: "A luxury brand's customer service chatbot that must maintain a sophisticated, professional tone."
}, {
name: "Toxicity",
link: "/concepts/metrics/safety-and-compliance/toxicity",
category: "Safety and Compliance",
node: "Trace (root input/output only)",
description: "Identifies harmful, offensive, or inappropriate content.",
whenToUse: "When monitoring AI outputs for harmful content or implementing content filtering.",
example: "A social media content moderation system that must detect and flag potentially harmful user-generated content."
}, {
name: "Uncertainty",
link: "/concepts/metrics/model-confidence/uncertainty",
category: "Model Confidence",
node: "LLM Span",
description: "Measures how uncertain the model is about its response.",
whenToUse: "When flagging low-confidence responses for human review.",
example: "A customer support system that escalates uncertain responses to human agents."
}, {
name: "Visual Fidelity",
link: "/concepts/metrics/multimodal-quality/visual-fidelity",
category: "Multimodal Quality",
node: "LLM Span",
description: "Checks whether a generated image complies with all applicable provided brand rules based on visible evidence.",
whenToUse: "When generated images must conform to explicit brand, layout, or content rules.",
example: "A marketing image generator that must always place a logo in the top-left and avoid prohibited colors or content."
}, {
name: "Visual Quality",
link: "/concepts/metrics/multimodal-quality/visual-quality",
category: "Multimodal Quality",
node: "LLM Span",
description: "Judges whether the quality of an input image is sufficient to reliably complete the task in the adjoining prompt.",
whenToUse: "When user-supplied images can be blurry, occluded, or poorly lit and might make the task infeasible.",
example: "A document capture workflow that needs to detect when a photo is too blurry to read a serial number or form field."
}, {
name: "User Intent Change",
link: "/concepts/metrics/agentic/intent-change",
category: "Agentic",
node: "Session (trace inputs/outputs only)",
description: "Detects when user intent shifts during a conversation.",
whenToUse: "When tracking conversation dynamics and adapting agent behavior.",
example: "A sales assistant that needs to recognize when a customer shifts from browsing to purchasing intent."
}, {
name: "SQL Correctness",
link: "/concepts/metrics/text2sql/sql-correctness",
category: "Text-to-SQL",
node: "LLM Span",
description: "Evaluates whether a generated SQL query is syntactically valid and adheres to the provided database schema.",
whenToUse: "When validating that generated SQL queries are grammatically correct and properly grounded in the database schema.",
example: "A business intelligence assistant that translates user questions into SQL queries for a data warehouse."
}, {
name: "SQL Adherence",
link: "/concepts/metrics/text2sql/sql-adherence",
category: "Text-to-SQL",
node: "LLM Span",
description: "Evaluates whether a generated SQL query semantically aligns with the user's natural language intent.",
whenToUse: "When validating that generated SQL queries accurately reflect what the user requested.",
example: "A data analytics assistant where users ask questions in natural language and expect accurate query results."
}, {
name: "SQL Injection",
link: "/concepts/metrics/text2sql/sql-injection",
category: "Text-to-SQL",
node: "LLM Span",
description: "Detects SQL injection attacks and security vulnerabilities in generated SQL queries.",
whenToUse: "When protecting against malicious inputs and ensuring generated SQL is safe to execute.",
example: "A customer-facing data analytics chatbot that must prevent injection attacks from user inputs."
}, {
name: "SQL Efficiency",
link: "/concepts/metrics/text2sql/sql-efficiency",
category: "Text-to-SQL",
node: "LLM Span",
description: "Evaluates whether a generated SQL query is structured efficiently and avoids performance anti-patterns.",
whenToUse: "When validating that generated SQL queries won't cause performance issues or resource exhaustion.",
example: "A business intelligence platform where ad-hoc queries must not impact database availability."
}];
const categories = [...new Set(metrics.map(m => m.category))].sort();
const nodes = [...new Set(metrics.map(m => m.node))].sort();
const [sortColumn, setSortColumn] = useState("name");
const [sortDirection, setSortDirection] = useState("asc");
const [filterCategory, setFilterCategory] = useState("All");
const [filterNode, setFilterNode] = useState("All");
const [isDark, setIsDark] = useState(false);
useEffect(() => {
const checkDark = () => {
const htmlEl = document.documentElement;
if (htmlEl.classList.contains("dark")) {
setIsDark(true);
} else if (window.matchMedia && window.matchMedia("(prefers-color-scheme: dark)").matches && !htmlEl.classList.contains("light")) {
setIsDark(true);
} else {
setIsDark(false);
}
};
checkDark();
const observer = new MutationObserver(checkDark);
observer.observe(document.documentElement, {
attributes: true,
attributeFilter: ["class"]
});
const mediaQuery = window.matchMedia("(prefers-color-scheme: dark)");
mediaQuery.addEventListener("change", checkDark);
return () => {
observer.disconnect();
mediaQuery.removeEventListener("change", checkDark);
};
}, []);
const colors = isDark ? {
bg: "#1a1a1a",
bgAlt: "#262626",
border: "#3f3f46",
text: "#e4e4e7",
textMuted: "#a1a1aa",
selectBg: "#27272a",
selectBorder: "#3f3f46",
link: "#38bdf8"
} : {
bg: "#ffffff",
bgAlt: "#f9fafb",
border: "#e5e7eb",
text: "#111827",
textMuted: "#6b7280",
selectBg: "#ffffff",
selectBorder: "#d1d5db",
link: "#1098F7"
};
const handleSort = column => {
if (sortColumn === column) {
setSortDirection(sortDirection === "asc" ? "desc" : "asc");
} else {
setSortColumn(column);
setSortDirection("asc");
}
};
const filteredAndSortedMetrics = metrics.filter(m => filterCategory === "All" || m.category === filterCategory).filter(m => filterNode === "All" || m.node === filterNode).sort((a, b) => {
const aVal = a[sortColumn]?.toLowerCase() || "";
const bVal = b[sortColumn]?.toLowerCase() || "";
if (sortDirection === "asc") {
return aVal.localeCompare(bVal);
}
return bVal.localeCompare(aVal);
});
const selectStyle = {
padding: "0.25rem 0.5rem",
border: `1px solid ${colors.selectBorder}`,
borderRadius: "0.375rem",
fontSize: "0.875rem",
background: colors.selectBg,
color: colors.text
};
const headerStyle = {
cursor: "pointer",
userSelect: "none",
whiteSpace: "nowrap",
color: colors.text
};
const getSortIndicator = column => {
if (sortColumn !== column) return " ↕";
return sortDirection === "asc" ? " ↑" : " ↓";
};
return
{}
Showing {filteredAndSortedMetrics.length} of {metrics.length} metrics
;
};
Galileo provides a comprehensive suite of pre-built metrics designed to evaluate various aspects of AI system performance without requiring custom implementation.
These metrics span across eight categories including:
* [Agentic Performance Metrics](/concepts/metrics/agentic/agentic-overview)
* [Expression And Readability Metrics](/concepts/metrics/expression-and-readability/expression-and-readability-overview)
* [Multimodal Quality Metrics](/concepts/metrics/multimodal-quality/multimodal-quality-overview)
* [Model Confidence Metrics](/concepts/metrics/model-confidence/model-confidence-overview)
* [Response Quality Metrics](/concepts/metrics/response-quality/response-quality-overview)
* [RAG Metrics](/concepts/metrics/rag/rag-overview)
* [Safety And Compliance Metrics](/concepts/metrics/safety-and-compliance/safety-and-compliance-overview)
* [Text-to-SQL Metrics](/concepts/metrics/text2sql/text2sql-overview)
Each metric addresses specific evaluation needs, from measuring factual correctness to detecting potential biases or tracking tool usage effectiveness.
These metrics apply to different node types (such as session, trace, or different span types), depending on the metric.
Use the sortable, filterable table below to explore all available native metrics and find the right measurements for your AI applications.