House Rules
A Talent AI tool earns trust by what it refuses to do. These are the rules built into Greenroom from the start, not bolted on after. They are the same privacy-first, governed-content posture a real People Development feature would need before it ships.
It stays inside the governed library.
Greenroom answers only from Strand's curated People Development content. It is instructed to use the retrieved sources and nothing else, so it cannot wander into ungoverned or off-brand advice the Talent team never sanctioned.
It refuses rather than guesses.
When the best retrieval similarity falls below 0.44, Greenroom declines before any model call and says the library does not cover the question. Saying "I don't have that" is a feature, not a failure.
It cites only what it actually retrieved.
A server-side check confirms every cited source was in the retrieved set. Anything outside it is flagged on the run sheet, not trusted. The model does not get to grade its own grounding.
It collects no personal data.
No login, no PII, no individual tracking. The 'did it land?' signal is stored in your own browser for this demo. In production it would be consented and reported in aggregate, never used to surveil or rate a person.
It never rates or decides about people.
Greenroom coaches behavior. It does not score employees, feed performance reviews, or make decisions about individuals. The measurement it captures is whether a behavior happened, framed for the person's own growth.
It hands off to a person when a person is needed.
Some development needs a human: a mentor conversation, a coach, a manager. Greenroom points there rather than pretending to replace it.
It is built to be inspected.
Every answer exposes its model, retrieval, and groundedness on the run sheet, and the whole system is graded in The Reviews. Before production, this routes to Legal and Infosec for a privacy and fairness review, the same way a real Talent AI feature would.
What Greenroom will not do
- Answer outside Strand's governed library
- Store or request personal or sensitive data
- Rate, rank, or make decisions about individuals
- Feed performance reviews or surveillance
- Invent advice when the library is silent
In production these rules become contracts: a governed content store the Talent team owns, consented and aggregated outcome data, and a privacy and fairness review with Legal and Infosec before launch.