The Transfer
A prototype earns a transfer to the main stage when two things are true: you can show it changes behavior, and you know exactly how it ships. Here is how Greenroom would measure success, and what it takes to move it from this demo to production.
How we’d know it works
Was it worth opening?
A lightweight useful / not-useful on each answer.
Did the idea land?
Whether the person can restate the next step in their words.
Did they actually do it?
The did-it-land signal on the first cue. This is the one Greenroom captures today.
Did anything change?
Peer and manager-observed change, aggregated over time across a team.
The bet is simple: a static library can prove L1, that someone opened a page. Greenroom is built to prove L3, that the behavior happened. That is the difference between content delivery and behavior change.
From this demo to production
A governed store the Talent team owns
The seeded library becomes a content store the team authors into, behind a real vector index. Calendar and Slack supply the triggers.
Seconds and cents per answer
~10s per answer today, a few cents a session, with a cached system prefix and capped output. Streaming and batching tighten it further at scale.
The Reviews become a release gate
The labeled eval set runs on every library or prompt change. Ship only if retrieval, refusal, and groundedness hold. Add human-rated sampling.
Watch groundedness and refusals live
Production dashboards on groundedness, refusal rate, and retrieval drift, with alerts when the curve moves.
Pilot one team, measure the L3 lift, expand
Start with a single team, prove a behavior-change lift, then widen. The Slack and calendar surfaces ride existing tools, so adoption is low-friction.
Build the judgment, buy the commodities
Build the orchestration and the evals (the real IP). Buy the model and embeddings, integrate the vector store, Slack, calendar, and HRIS.
Governance, before launch
The same posture as House Rules, made formal: a privacy and fairness review with Legal and Infosec, consented and aggregated outcome data, and a governed content store the Talent team owns. The measurement serves the person’s growth, never surveillance.
Greenroom is a portfolio prototype. The outcomes framework and production plan are how a real Talent AI feature would move from demo to deployment, written to be acted on, not just admired.