---
title: "Investigating a failed AI decision"
description: "How can you separate retrieval, instruction, model, and tool failures?"
status: "coming-soon"
language: "en"
canonical: "https://bulnik.dev/en/articles/topics/T58.10/"
topicId: "T58.10"
---

> Coming soon — this article is being prepared. Below are its question, intended outcome, and place in the story.

## The question

How can you separate retrieval, instruction, model, and tool failures?

## What we will work through

Separate retrieval, instruction, model, and tool failures.

## Before you begin

Previously covered: Regression evaluations; A retrieval evaluation suite.

- [Regression evaluations](/en/articles/topics/T58.03/)
- [A retrieval evaluation suite](/en/articles/topics/T56.10/)