---
title: "Discussion on \"Structured Output Beats Prompt Tweaking: A Practical Eval Loop\" | Hashnode"
description: "Discussion on \"Structured Output Beats Prompt Tweaking: A Practical Eval Loop\". We spent two weeks rewriting prompts. Then we changed the response format, and the failure rate dropped by half.\nThat is not a story about prompt engineering being useless. It is a story about measuri"
url: "https://hashnode.com/posts/structured-output-beats-prompt-tweaking-a-practical-eval-loop/6ae1855c2095097a8f491333"
---

[Feed](https://hashnode.com/feed) /

Discussion

# [Structured Output Beats Prompt Tweaking: A Practical Eval Loop](https://eval-notes.hashnode.dev/structured-output-beats-prompt-tweaking-a-practical-eval-loop?utm_source=hashnode&utm_medium=feed)

We spent two weeks rewriting prompts. Then we changed the response format, and the failure rate dropped by half. That is not a story about prompt engineering being useless. It is a story about measuri

Oct 28 · 2 min read · eval-notes.hashnode.dev  
Read the article

-   [#ai](https://hashnode.com/tag/ai)
-   [#llm](https://hashnode.com/tag/llm)
-   [#python](https://hashnode.com/tag/python)

## Written by

[Priya Raman](https://hashnode.com/@priyaraman)

@priyaraman · Bengaluru, India

  
[View profile](https://hashnode.com/@priyaraman)

Exploring AI tooling, LLM applications and what breaks in production.

I build LLM-powered developer tools in Bengaluru, mostly around retrieval and evaluation. I write about what actually breaks when you put a model inside a production loop, and how to measure it honestly. My notes lean practical: prompts that failed, evals that lied, and the small harnesses that fixed both.

## Responses 2

Markdown supported · ⌘↵ to post

[Kartik N V J K](https://hashnode.com/@kartik-nvj)

[Oct 29](https://hashnode.com/posts/structured-output-beats-prompt-tweaking-a-practical-eval-loop/6ae1855c2095097a8f491333/comment/6ae34fb8cabb70c219e8bd16)

Ten days of calling parser breaks prompt problems sounds familiar to me. I'd still watch for a reply that passes the schema with a wrong value, the right vendor and a confidently wrong amount, which only your field-level accuracy on the fifty examples would catch. When the failure rate dropped by half, did that count both kinds or just schema failures?

[Daniel Okafor](https://hashnode.com/@danielokafor)

[Oct 28](https://hashnode.com/posts/structured-output-beats-prompt-tweaking-a-practical-eval-loop/6ae1855c2095097a8f491333/comment/6ae18603a50a971c124c4b7d)

Building the eval set first is the part everyone skips under deadline pressure.
