> ## Documentation Index
> Fetch the complete documentation index at: https://docs.fibonax.com/llms.txt
> Use this file to discover all available pages before exploring further.

# Snippet quality

> Every snippet gets a score from 0 to 10 for how clear, complete and trustworthy it reads, with a concrete fix for each weak point.

A snippet is sent again and again, so a small mistake in it is sent again and again too. The editor scores each snippet as you write and tells you what to fix.

## See the score

Open a snippet. Under the body, next to the word count, **Quality** shows the score for the language you are editing, like `Quality 7.8`. Click it to open the window.

The window shows:

* The score out of 10, and one word for it: **Excellent** (8 or more), **Good** (5 or more) or **Needs improvement**.
* One row per check, ticked when it passes. Click a row to see why it scored that way and an example of the fix.

The score updates as you type. It is worked out on your computer: your text is not sent anywhere for it.

## What is checked

The checks follow Google's questions for helpful, trustworthy writing, adapted to a text you send to people.

| Group | What it checks |
| - | - |
| **Clarity** | Short sentences, specific words instead of vague ones like "stuff" or "etc.", and clean wording: no repeated words, shouting in capitals or doubled punctuation |
| **Completeness** | A message opens with a greeting and ends with a sign-off, tells the reader what to do next, and has enough detail to act on |
| **Value and trust** | Concrete details such as numbers, dates, names, links or fields, and a calm tone without pressure or exaggeration |
| **Title** | The name says what the snippet is, without exaggerating |
| **Care** | Nothing left unfinished, like `TBD`, `XXX`, `[name]` or a part of the snippet that is not closed |

Fields and formulas in curly braces, like `{name}`, count as details the reader gets, never as gaps. The automatic greeting field counts as a greeting.

A text that is not written to someone, like opening hours or an address, is not asked for a greeting.

### Short snippets

A short snippet is often short on purpose: an address, a signature, "Thanks, talk soon". Under 12 words, a snippet is judged on clarity, its title and anything left unfinished, and never marked low for its length.

### What the score cannot tell

The checks read the words, not the meaning. They cannot tell whether a price, a date or a fact is right, so a high score never means the content is correct. Read it through before you share it.

When another snippet says much the same thing, the window says so. Keeping one and updating it is easier than keeping two in step. This note does not change the score.

## Snippets that need work

In the snippet list, a snippet that scores 4 or less wears an amber wrench. Point at it to see the score; for a snippet in several languages it names the weakest one. Open the snippet and click **Quality** to see what to fix.

A snippet with a broken part wears a red wrench instead, and a snippet among your most used wears a trophy. A snippet shows one of these at a time.

## Improve with AI

> [!NOTE]
>
> Improve with AI is part of Fibonax Pro.

In the Quality window, click **Improve with AI** for written suggestions from an AI that reads the meaning as well as the words. You get up to five, each naming the problem and a fix, with new wording in the snippet's own language. A short summary says what works and what does not.

* The suggestions never change the score or your snippet. Make the changes you agree with, then **Save**.
* The AI does not add facts. Where a fix needs a detail it cannot know, like a date, it leaves a gap or a field for you.
* Your text is sent to Claude, by Anthropic, to write the suggestions. Nothing is stored.
* Change the text and click **Check again** for a fresh set.


This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.