# Voice and Language
> [!summary]
> Eval Labs voice should be direct, calm, precise, and product-serious. It should explain what is true without sounding sterile or overhyped.
---
## Voice traits
Eval Labs should sound:
```text
direct
calm
clear
serious
warm
inspectable
truthful
```
It should not sound:
```text
salesy
generic SaaS
academic sludge
AI hype
vague innovation language
```
---
## Preferred language
Use:
```text
human judgment layer
behavioral evaluation
review evidence
quality standard
failure pattern
regression suite
truth-state discipline
operator calm
response contract
```
Avoid:
```text
magical AI insights
unlock exponential intelligence
revolutionary automation
next-gen synergy
fully autonomous trust engine
```
---
## Tone rule
Eval Labs can be confident.
It should not be grandiose.
Good:
```text
Eval Labs gives us repeatable evidence about whether Lucia is improving.
```
Bad:
```text
Eval Labs revolutionizes the entire AI evaluation paradigm forever.
```
---
## Writing pattern
Use this shape often:
```text
State the truth.
Name why it matters.
Give the operating rule.
```
Example:
```text
Saved suites make evals repeatable.
That matters because Lucia can only improve safely if the same behavior can be tested after each change.
The rule is simple: important prompts should become reusable suites.
```
---
## Eval Labs vs Lucia voice
Lucia voice is operator-facing and emotionally containing.
Eval Labs voice is reviewer-facing and evidence-oriented.
Eval Labs can be a little more technical than Lucia, but it should still preserve calm and clarity.
---
## Language to protect
This phrase is important and should remain part of the Eval Labs identity:
```text
Eval Labs is Lucia’s proprietary evaluation platform.
```
This phrase is also useful:
```text
Human grading is the product.
```
Use it when explaining why Eval Labs is different from generic benchmark systems.
---
## Employee-facing review language
Employee-facing language should be simple and non-technical.
Prefer:
```text
Did Lucia understand what was needed?
Should a senior reviewer look at this?
Could this teach Lucia something reusable?
```
Avoid:
```text
Classify operational intent.
Select canonical adjudication labels.
Define action taxonomy.
```
The UI should invite honest judgment, not performative expertise.