---
title: "04 Monitoring"
description: "Learner guide: 04 Monitoring"
---
# 04 Monitoring

Workshop material is maintained in the public [`langfuse/langfuse-workshop`](https://github.com/langfuse/langfuse-workshop) repository. Use the repository for the runnable app, checkpoint branches, and local setup.

[View this Markdown file](https://github.com/langfuse/langfuse-workshop/blob/main/docs/instructor/04-monitoring.md)

Learner guide: [04 Monitoring](/workshop/learner/04-monitoring)

## Instructor notes

- This is a UI-first chapter with two signals: **Detect User Disagreement** uses LLM-as-a-judge for semantic judgment, while **Detect User Frustration (ALL CAPS)** uses deterministic TypeScript logic.
- Before the disagreement evaluator, confirm the project has **Project Settings → LLM Connections** configured. The API keys in `.env` do not configure the judge model inside Langfuse.
- Both evaluators target the logical root `dad-it-support-chat-turn` agent observation because that observation carries the overall conversation input and final answer.
- Have learners use the right-side sample panel instead of mapping from memory: select a root observation, map `conversation_history` to all input messages and `last_user_message` to the last input message, then test the evaluator before saving.
- After **Create evaluator**, review the configured filters, estimated weekly volume and cost, and sampling rate before clicking **Execute**. That final action activates the evaluator for matching incoming observations.
- The ALL-CAPS template is already implemented as a TypeScript code evaluator. Use it to show why deterministic signals do not need an LLM call or an evaluator-model connection.
- Close with seeding: `npm run langfuse:seed:otel:no-scores` adds realistic `production` traffic without pre-baked scores, so the learners' own evaluators produce the results. Remind learners that the command is not idempotent; every rerun creates duplicate traces.
- Use the first few evaluator results as a debugging exercise, not just a pass/fail check.

## Demo rhythm

1. Confirm or configure the project's default evaluator model.
2. Create **Detect User Disagreement**, select a sample root observation, map both variables through the data tree, run a test, then create and execute the evaluator.
3. Create **Detect User Frustration (ALL CAPS)** on the same root observation, run a test, then create and execute it.
4. Send one disagreement turn and one ALL-CAPS turn, then inspect the scores on their root observations.
5. Seed production traffic with `npm run langfuse:seed:otel:no-scores`, refresh the Tracing view, and watch the two evaluators score the seeded batch.

## Watch for

- Accidentally choosing the wrong template instead of **Detect User Disagreement** from the Template Gallery.
- Treating the Langfuse API keys from `.env` as enough for evaluators. Judge-based evaluators also need the Langfuse-side LLM connection.
- Selecting a child generation instead of the root agent observation. The evaluator only receives data from the observation it targets; it does not automatically read sibling or child observations.
- Mapping `conversation_history` to a single message, or `last_user_message` to every message. Use the live sample tree: **Input → messages** and **Input → messages → last**.
- Forgetting to select a sample observation before testing. The right panel can be filtered with the preset buttons or **Ask AI**.
- Clicking **Create evaluator** but not **Execute** in the saved-evaluator dialog, which leaves the live evaluator inactive.
- Learners assuming the all-caps score is a guarantee of anger. Frame it as a triage signal, not a verdict.
- Running the seed command more than once. It generates fresh trace IDs and duplicates the seeded traffic.

<!-- agent-instructions -->

---

## Agent Instructions

This page is part of the [Langfuse](https://langfuse.com) documentation, published as plain Markdown for AI agents. Every page is available as Markdown by appending `.md` to its URL, or by sending an `Accept: text/markdown` header. This page: `https://langfuse.com/workshop/instructor/04-monitoring.md`.

### Querying these docs

If the answer is not on this page, query the documentation instead of guessing:

- **Semantic search** across all Langfuse docs, returning an answer with the relevant pages and excerpts. Ask a specific, self-contained question:

  ```bash
  curl -sG "https://langfuse.com/api/search-docs" --data-urlencode "query=How do I trace a LangGraph agent?"
  ```

- **Index of every page**: <https://langfuse.com/llms.txt>, with per-section indexes [llms-docs.txt](https://langfuse.com/llms-docs.txt), [llms-integrations.txt](https://langfuse.com/llms-integrations.txt), and [llms-self-hosting.txt](https://langfuse.com/llms-self-hosting.txt).

### Before writing Langfuse code

- **Install the [Langfuse Agent Skill](https://langfuse.com/docs/api-and-data-platform/features/agent-skill).** It encodes Langfuse's own best practices for instrumentation, prompt management, and evaluation, and materially improves results.
- **Read [What does a good trace look like?](https://langfuse.com/docs/observability/best-practices.md)** before instrumenting an application.
- **Verify endpoints, parameters, and response fields** against the [API reference](https://api.reference.langfuse.com) instead of inferring them from code examples.
- **Use the [Langfuse CLI](https://langfuse.com/docs/api-and-data-platform/features/cli)** (`npx langfuse-cli api <resource> <action>`) to read or write traces, prompts, datasets, and scores from the terminal.

Found an error in these docs? Please open an issue at <https://github.com/langfuse/langfuse-docs/issues>.