// 003
Guardrails
The constraints built into the product, and why each one is not configurable.
These are properties of the system rather than settings. They cannot be turned off from an admin panel, because a guardrail that can be disabled is a default.
A person makes every decision
No code path advances, declines, or hires a candidate without a person acting. The system produces recommendations and drafts; it does not send them.
Protected characteristics are excluded
Race, ethnicity, religion, gender, sexual orientation, disability, age, and nationality where not legally relevant are excluded from matching. The system does not attempt to infer any of them, and photographs are not analysed.
Names and photographs can be hidden during assessment so that the first read of a candidate is of their evidence.
No inference about people from style
YoHe does not perform facial analysis, emotion recognition, or personality prediction, and does not infer traits from writing style. Findings are about stated experience, not inferred character.
AI output is labelled
Anything generated by the model is marked as generated, remains editable, and can be reverted to the original. Edited text is attributed to the person who edited it.
YoHe does not claim to remove bias from hiring. It narrows what the system is allowed to consider, and makes the reasoning visible enough that a biased conclusion can be seen and challenged.