# LLM evaluation (evals)

> LLM evaluation scores a language model system's outputs against a fixed set of real test cases, so you can tell whether a change made it better or worse.

- Page: https://methodcrew.com/glossary/llm-evaluation
- Last updated: 2026-10-28
- Publisher: [The Method Crew](https://methodcrew.com/)
- Also known as: LLM testing

## How it works

An eval set is a list of inputs paired with what a good output looks like. In a first version that is usually 50 to 300 real questions, documents or tickets, each with an expected answer, a required source or a pass rule. You run the system over the set and score every result, using the cheapest check you still trust:

- Exact checks: valid JSON, correct field values, the right tool called, a citation present.
- Model-graded checks, where a second model rates relevance or faithfulness to the source against a written rubric. This is often called LLM-as-judge.
- Human review of a sample, which keeps the model grader honest.
- Cost and latency per case, which matter as much as quality once volume grows.

Tools such as promptfoo, Langfuse, LangSmith and Braintrust run these suites and keep the history. A spreadsheet and a short script work for a first pass.

## How businesses use it

Evals let you change a system without guessing. Before you switch models, rewrite a prompt or change chunk sizes in a [RAG](https://methodcrew.com/glossary/retrieval-augmented-generation) system, run the set and compare scores side by side. A cheaper model that scores the same on your cases is an easy decision. One that tops public benchmarks and scores lower on your invoices is a different conversation.

We build the first eval set during the Prove stage of every AI build, from the client's own historical tickets, documents or calls, and hand it over with the system.

## Where it goes wrong

The common mistake is an eval set of easy, invented examples. It passes, and production still fails on the messy cases. Pull cases from real traffic, include the ones that went wrong, and add each new failure as a test.

Model graders have biases of their own, such as favoring longer answers or an answer in a particular position, as [Zheng and colleagues (2023)](https://arxiv.org/abs/2306.05685) showed when they tested LLM-as-judge. Check a sample of their scores by hand every month. Keep the set stable, too: if it changes weekly, scores stop being comparable. Scoring live traffic on an ongoing basis belongs to [LLMOps](https://methodcrew.com/glossary/llmops).

Do not buy tooling before you have a stable system to test. In the first week of a prototype, 20 cases in a spreadsheet are enough.

## Example (hypothetical)

Hypothetical: a software company wants to move its support assistant to a cheaper model. Running 220 past tickets through both versions shows the cheaper model matches on billing questions but cites the wrong help article on about 1 in 8 integration questions, so the team switches only the billing route.

## Related terms

- [LLMOps (AI observability)](https://methodcrew.com/glossary/llmops): LLMOps is the operating discipline for language model systems in production: tracing every call, tracking cost and latency, monitoring quality, and managing prompt and model changes.
- [AI hallucination](https://methodcrew.com/glossary/ai-hallucination): An AI hallucination is output from a language model that is fluent and confident but false, such as an invented citation, policy, figure or product detail.
- [AI guardrails](https://methodcrew.com/glossary/ai-guardrails): AI guardrails are checks placed around a language model that block, rewrite or escalate inputs and outputs that break your rules on safety, accuracy, privacy or scope.
- [Retrieval-augmented generation (RAG)](https://methodcrew.com/glossary/retrieval-augmented-generation): Retrieval-augmented generation is a method where a system first searches your own documents for relevant passages, then gives them to a language model to answer from.
- [Prompt engineering](https://methodcrew.com/glossary/prompt-engineering): Prompt engineering is writing and testing the instructions, examples and context given to a language model so it produces the output a task needs, consistently.

## Services where this matters

- [AI ops monitoring and improvement](https://methodcrew.com/services/ai-ops-monitoring): Quality checks, cost control and model updates for live AI systems. $1,500 to $5,000 per month.
- [RAG and knowledge systems](https://methodcrew.com/services/rag-knowledge-systems): Your documents turned into an assistant that cites its sources. $15,000 to $75,000 per project.
- [Custom AI agents and copilots](https://methodcrew.com/services/custom-ai-agents): Agents that look things up in your CRM or helpdesk and take the next step, with approvals. $15,000 to $90,000 per project.

## Articles that use this term

- [How to build a company knowledge assistant that cites its sources](https://methodcrew.com/insights/build-rag-knowledge-assistant)
- [How to measure the ROI of AI automation, running costs included](https://methodcrew.com/insights/measure-ai-automation-roi)
- [Fractional chief AI officer: what the role covers and what it costs](https://methodcrew.com/insights/fractional-chief-ai-officer)

Full glossary: https://methodcrew.com/glossary