> ## Documentation Index
> Fetch the complete documentation index at: https://docs.clavata.ai/llms.txt
> Use this file to discover all available pages before exploring further.
# Writing Signals
> How to write signals and rules in the label editor: syntax, conditions, examples, and common mistakes.
This is Moonbounce's label-first experience. If you're looking for the
policy-first experience, switch using the version selector above.
Signals are the words, phrases, and questions you write inside a label to tell
Moonbounce what to look for in content. In the label editor you compose
**rules** from signals and **conditions** (such as `ANY`, `AND`, or exact
match). Together they define when your label should match.
## What is a signal?
A **signal** is a quoted phrase (or question) that names a concept you want the
system to evaluate — for example `"graphic violence"` or `"spammy promotional
language"`.
* Signals are always wrapped in **double quotation marks**.
* A signal can be a single word or multiple words.
* Signals are **not** case-sensitive.
* A **rule** combines one or more signals with optional conditions. If **any**
rule in the label matches, the label matches.
Simple rules can be a single signal on its own line:
```
"graphic violence"
```
That tells the system to evaluate whether the content matches the concept
described by that phrase.
## Rules, conditions, and signal lists
Most labels use **conditions** to combine signals:
| Condition | Purpose |
| ----------------- | --------------------------------------------------------------------------------- |
| `ANY ( … )` | Match if **any** expression inside the parentheses matches |
| `ALL ( … )` | Match only if **every** expression inside matches |
| `NONE ( … )` | Match only if **none** of the expressions inside match |
| `AND` | Both sides must match |
| `OR` | Either side may match |
| `NOT` | Negates the following expression |
| `=( … )` | **Exact match** on literal text (case-insensitive) |
| `~( … )` | **Fuzzy match** on literal text (catches variations and leetspeak) |
| `SENTIMENT ( … )` | Match when content carries the given sentiment (for example threatening language) |
Parentheses group expressions. Separate rules with a **newline** or a **comma**.
### Connecting signals with `ANY` and `AND`
```
ANY ("violence", "gore", "weapons") AND "directed at person"
```
This matches when any of the first three concepts appears together with content
that matches `"directed at person"`.
### Exact and fuzzy literal matches
Use exact match when you need specific terms to appear as written (ignoring
case):
```
=("kill", "murder")
```
Use fuzzy match when you want to catch spelling variants or leetspeak:
```
~("spam")
```
### Sentiment
```
SENTIMENT("threatening", "hostile")
```
### Questions
Append `?` **after** the closing quote to ask a yes/no question about the
content:
```
"is this promotional spam"?
```
## Exceptions with `UNLESS`
Use `UNLESS` to carve out exceptions for the label. The block applies to the
label you are editing:
```
"attacks on personal circumstances"
UNLESS {
"light trash talk"
"fictional context"
}
```
Content that matches an `UNLESS` rule should not cause the label to match, even
when other rules would have matched.
## Comments
Add comments with `#`. Everything after `#` on that line is ignored:
```
# Promotional patterns
ANY ("buy now", "limited time offer")
```
## A fuller example
```
ANY ("violence", "gore", "weapons") AND "directed at person"
=("kill", "murder") OR "threatening language"
UNLESS {
"fictional context"
"educational material"
}
```
## Good vs. weak signals
**Strong** signals describe what you mean in plain language the model can
interpret:
```
"sexually explicit content involving minors"
"unsolicited bulk commercial email"
```
**Weak** signals are vague single words that leave too much room for
interpretation:
```
"bad"
"inappropriate"
```
Prefer specific, testable phrasing — then use your label's dataset to confirm
the behavior you want.
## Common mistakes
| Mistake | Fix |
| ----------------------------------------------------------------- | --------------------------------------------------------------------------------------------------------------------------------------------------- |
| Unquoted text (`spam` instead of `"spam"`) | Wrap every signal in `"double quotes"` |
| Missing separator between rules (`"a" "b"` on one line) | Put each rule on its own line, or separate with a comma |
| Question mark inside quotes (`"is this spam?"`) | Put `?` after the closing quote: `"is this spam"?` |
| Using `[ABOUT]` or other bracket operators | Not supported in the label editor grammar — see [Policy Reference](/policy_reference/syntax/general_syntax_of_policies) for the policy-first syntax |
| `PRIORITY`, `SECTION`, or multiple top-level labels in one editor | Label-first uses a single-label grammar (v3); one label per editor |
| `UNLESS -> "Other Label"` | Use an `UNLESS { … }` block with signals, not a reference to another label |
The editor validates your draft as you work and surfaces syntax errors with line
numbers when something cannot compile.
## Policy-first vs. label-first syntax
The **policy-first** experience supports a broader policy language (including
context operators like `"drugs" [IN REFERENCE TO] "selling"`, `PRIORITY`
chains, and multiple `LABEL` blocks in one policy). That syntax is documented
under **Policy Reference** in the version selector.
The **label-first** editor uses grammar **version 3**: one label body, the
conditions listed on this page, and `UNLESS` blocks — but not context
operators, `PRIORITY`, or `SECTION`.
Full policy language reference (policy-first experience).
Deeper guide to conditions and rule patterns.
## Where to next
The craft → test → publish → integrate loop.
Drafts, published versions, immutability, and test data.
***
## We'd Love to Hear From You
Whether you have a suggestion, feedback, or a bug to report, here are the best ways to get in touch:
* **In the App:** Use the **Feedback button** for direct suggestions.
* **On Slack:** Reach out to the team in your shared channel.
* **With your AM:** Talk to your dedicated account manager.
* **Via Email:** Send a message to [**support@moonbounce.io**](mailto:support@moonbounce.io).
* **Security, availability, or other incidents:** Use the in-app **Feedback button** or email [**support@moonbounce.io**](mailto:support@moonbounce.io). See [Customer Feedback](/feedback/customer_feedback) for what to include.