What we do, what we hold, and what is recorded.
This page exists so that a procurement, security or legal reviewer can find the answers without asking. Where the honest answer depends on the arrangement, it says so rather than implying more.
Where the humans are.
Sign-off is required at four points, and it is held by your own people. The platform does not release unreviewed work, and Serinyx does not rule on the substance of your simulation.
Mandate approval
A named person on your side accepts the matter as framed, its decomposition, the criteria for an answer and the scope. No research begins before this.
Intake ruling
After the brief has been challenged, a named person rules on whether it proceeds, proceeds reframed, or is refused.
Challenge resolution
Where the system’s disagreement cannot be resolved on the evidence, a named reviewer rules. The ruling is recorded with the dissenting position alongside it.
Release sign-off
A named person signs off the output before it leaves the platform.
Each decision is attributed, and the person who made it can be asked about it afterwards.
Client data is isolated per organisation.
Work carried out for one organisation does not inform another’s output, in either direction. Connections are drawn within your own material and against the public record, never across the private material of different organisations. This is a design decision rather than a configuration option, and it applies to every client without exception.
Where data is held.
Data residency options are available and are agreed as part of the licence. We describe the specific arrangement for your organisation in writing before any data is transferred. We do not publish a general residency claim, because the honest answer depends on the arrangement.
Your data is not used to train models.
Material you provide is used to run your own decision simulations and for nothing else. It is not used to train or fine-tune any model, ours or a third party’s, and it does not inform the output of any other organisation’s work. Where a licence requires a specific contractual form of this commitment, we will put it in writing.
What is retained, and what you can reconstruct.
For every decision simulation: the brief as originally pre-registered, every revision with its stated reason, the decomposed questions, the sources relied on per claim, the confidence grades, the disagreements and how they were ruled, the work attributed per unit, and every sign-off with the name attached.
The purpose is specific. When the decision is challenged, whether by a board, a regulator, an auditor or a counterparty, the reasoning can be reconstructed by someone who was not there.
If you are assessing this under an AI governance framework.
We do not hold certifications and we do not claim compliance on your behalf, because compliance attaches to your deployment rather than to our platform. What we can do is tell you precisely where to look.
If your framework asks who exercises human oversight, the answer is the four named gates above, all held by your own people, and the platform does not release unreviewed work. If it asks what is logged, the answer is the pre-registered brief, every revision with its stated reason, the sources relied on per claim, the grades, the disagreements and how each was ruled. If it asks whether the output can be explained to someone who was not present, that is the property the audit record exists to provide. If it asks about training on your material, the answer is that we do not, under any arrangement.
Where your assessment needs any of this in a specific contractual form, ask and we will put it in writing.
If something is wrong, tell us.
If you believe a finding the platform produced is incorrect, or you have a security or data-protection concern, write to us and a named person will respond. We will tell you what we found and what we changed. A finding that came out wrong is a defect in the process, and we would rather know.
Ask us a hard question