Assistant and internal tooling
Typed MCP responses support enterprise assistants and custom workflows without reducing the result to a score.
Enterprise delivery
The index is consumed through typed responses, hiring-system attachments, and a console for the people who own legal and people risk.
Authorization, confidence, evidence state, permitted use, and refusal remain attached to every result.
Primary surface
An assistant, hiring system, or internal tool asks for a read and receives a typed response. Unknown cannot be quietly rounded into zero.
A downstream system cannot read a hidden answer because no answer exists.
Typed response
A placement response includes its confidence band. A refusal carries confidence null next to level null. Permitted use appears on every response.
{
"placement": "RECOGNIZED_CREDIBILITY",
"level": "Recognized Credibility",
"confidence": "High",
"score": null,
"sources_found": 18,
"sources_bound": 12,
"sources_not_read": 3,
"permitted_use": [
"authorized_hiring_read.v1"
],
"receipt": "rcpt_01K9...f12"
}Three enterprise surfaces
The interface may change by context. The contract and boundaries do not.
Typed MCP responses support enterprise assistants and custom workflows without reducing the result to a score.
The read sits on the requisition with authorization, subject confirmation, and the permitted use carried on the record.
Legal, compliance, and talent leadership review evidence state, authorization history, reinvestigation, and refusals.
The result is not complete without the information required to interpret and govern it.
The absence of an answer is structured so a downstream system cannot invent one.
Governed from the first call
Employment use is authorized per engagement and disabled by default, and a documented compliance and reinvestigation process governs every read.
Enterprise engagement
Enterprise access begins with the use case, authorization model, and systems that will consume the result.