Skip to main content
Employee Review is the fast, guided, non-technical review layer. It captures human judgment without asking employees to become AI evaluators or treating AI-reviewed platform scores as final human truth.

​
Purpose

The Employee Review layer exists to answer: 
Did Lucia work for a real human in this moment?
It should not ask employees to define training labels, invent ontology terms, or write model-quality explanations.

​
What employees should judge

Employees judge observable experience:
  • Did Lucia understand?
  • Did Lucia help?
  • Did Lucia calm the situation?
  • Did Lucia create risk or confusion?
  • Should a senior reviewer inspect this?
  • Is this useful enough to become reusable learning?
They do not need to know: 
prompt engineering
model behavior
intent taxonomy
adjudication schema
training data theory

​
Current Employee Review fields

understoodNeed
rightNextMove
calmingEffect
riskOrConfusion
seniorReview
reusableLearning
These fields are intentionally simple. They convert employee reaction into structured product signal. The current Quick Review answer options are: 
understoodNeed: yes / mostly / no / not_sure
rightNextMove: yes / mostly / no / not_needed / not_sure
calmingEffect: yes / somewhat / no / not_sure
riskOrConfusion: no / slightly_off / definitely_wrong / not_sure
seniorReview: true / false
reusableLearning: true / false
The UI may show suggested Quick Review selections before the reviewer accepts, overrides, and saves. Suggestions reduce review burden. They do not replace human judgment. AI-reviewed platform runs are useful evidence that the review machinery works. They are not employee review and are not human Lucia-quality approval.

​
Quick Review UX doctrine

Quick Review should feel like a guided flow, not a form. The preferred interaction style is: 
one question
clear choices
save momentum
next question
The UI should reduce:
  • scanning
  • hesitation
  • interpretation burden
  • fear of being wrong
  • freeform reviewer drift

​
Semantic rating sliders

The 1–10 scoring dimensions use semantic confidence bars. These sliders should feel: 
native
calm
premium
low-friction
psychologically guided
They use restrained color progression: 
low score → muted concern
middle score → soft uncertainty
high score → restrained confidence
The color system should guide reviewer intuition without becoming loud, gamified, or dashboard-heavy.

​
Employee rule

Employees should not use their own invented language to train Lucia.
They should use the guided controls and write short notes only when context is needed. The senior review and adjudication layers turn employee signal into canonical learning.

​
Onboarding boundary

Employees should not be onboarded just because the platform readiness gate passed. The evaluator role must remain limited until access, security, and product readiness decisions are final. Read next: Employee Onboarding Gate.