Engagement
How an assessment runs
A HOP engagement is evidence review against a defined decision class, conducted under the protocol’s independence and separation-of-duties rules. It is bounded, it is scoped before it starts, and it ends in a determination built to be read by someone outside your organization. Engagements are fixed fee, scoped to one decision class, and quoted in writing before any engagement letter. Reassessment is scoped to what changed rather than repriced from scratch.
Scoping call
You describe one decision class. We tell you plainly whether HOP applies, what evidence would be needed, and what we would expect to find. If it does not apply, we say so on the call.
Engagement letter and scope
The decision class is fixed in writing, along with the period under review, the evidence to be produced, and the independence terms. Scope is fixed before the assessment begins, and any change to it is agreed in writing before work continues.
Evidence production
You produce what exists: how the decision is made, who was empowered to change it, and what survives afterward.
Assessment against the four conditions
Each condition is tested against the evidence produced, separating control conditions from evidence records. Findings are capped by the evidence. Assertions without evidence are recorded as assertions.
Determination and Decision Record
You receive one of three findings, the reasoning behind it, the specific gaps if any, and a frozen, versioned Decision Record that supports the finding and can be reconstructed years later.
Reassessment, when you are ready
If the finding identifies gaps, you close them however you choose and with whomever you choose. We reassess. We do not implement, and we do not recommend a vendor to implement.
Scoping
What counts as one decision class
HOP is scored against the level at which liability attaches, which is narrower than a system and wider than a single case. A decision class is a repeatable determination, made the same way, with the same human role and the same consequence.
- A decision class
- Prior-authorization denials for a named benefit category. Adverse action on a defined credit product. First-party claim denials above a stated threshold.
- Not a decision class
- "Our use of AI." An entire claims platform. A department. Those are systems and organizations, and they are assessed with different instruments.
- Period under review
- Fixed in the engagement letter. A determination speaks to the decision class as it operated during that period, under the model and workflow versions in effect.
- Change triggers
- A model change, a workflow change, a role change or a volume change can invalidate a determination. Material-change reassessment is scoped against what actually moved.
Deliverables
What you receive
Four artefacts, each of which is designed to survive being handed to somebody who is motivated to break it.
HOP Assessment
The engagement itself. Evidence review and testing against the four conditions for one decision class.
HOP Determination
The finding, the reasoning, and the specific gaps. Stated plainly, never as a maturity score.
Decision Record
A frozen, versioned evidence package. Same version, same evidence, same rationale, years later.
Continuous Decision Assurance
Keeps a determination current as the model, agent or process changes. Not dashboard access.
Questions
The hard questions first
If your question is not here, it is a good first question for the scoping call.
Does a HOP assessment create a discoverable document?
This is the first question most General Counsel ask, and it deserves a direct answer. A HOP assessment produces a structured record under a defined engagement scope. Whether and how privilege attaches depends on how the engagement is directed and scoped, which is settled with your counsel before the assessment begins, not afterwards.
What is not in question is the alternative. The unstructured AI exhaust already sitting in your systems is discoverable today, was not designed to help you, and will be read by someone else's expert. A structured record produced under a designed posture is a different exhibit from a pile of logs.
Who is this for?
The person who has to answer for a consequential decision once it is challenged. General Counsel, Chief Risk Officer, Chief Audit Executive, CFO, and compliance or model-risk leadership at regulated enterprises.
An independent finding is evidence an internal review cannot supply, which is what makes it useful when remediation has to be funded or a decision has to be defended.
Do you need access to our model or our AI vendor?
No. HOP assesses human oversight of a decision, not the AI system. We need documentation of the workflow, the human role, what was shown to the reviewer, and the records that survive. We do not need model weights, training data or vendor cooperation.
How is this different from ISO 42001, NIST AI RMF or an AI governance programme?
Those assess the AI system and the organization's management of it. They are the right instruments for those questions.
HOP assesses one thing they do not: whether the human control around a specific decision held, and whether it can be proven. A company can hold a governance certification and still be unable to prove that a qualified person governed the decision now being litigated.
What if the determination is Insufficient?
Then you know before someone else finds out, which is the entire point of commissioning it. The determination names the specific condition that failed and whether it failed as a control or as an evidence gap. Those need different remedies.
We do not publish findings. The determination is your document, produced under your engagement scope, and how it is handled from there is governed by the posture you and your counsel set at the start.
Will you help us fix what you find?
No, and that is deliberate. PromptID does not build, configure or remediate the controls it assesses, and does not refer you to a partner who will. You close the gaps however you choose. We reassess.
An assessor who also sells the remediation is not independent, and a determination from one is worth less to the referee you are eventually going to show it to.
What is in a Decision Record?
The evidence relied on, its provenance, the method version in effect, the assessor, the reasoning against each of the four conditions, and the determination. It is frozen and versioned so that the basis of the finding cannot be edited after the fact without the edit being obvious.
We are US-only. Does the EU AI Act matter to us?
Not necessarily, and the case does not rest on it. California AB 316 and the CJEU's SCHUFA ruling are in force now, and the sector rules that already govern your decisions — utilization review, adverse action, claims handling — carry human duties that predate any AI statute.
The EU AI Act matters if you operate in Europe or sell into it. It is a second driver on a known clock, not the reason to act.
What happens when our model or workflow changes?
A determination speaks to the decision class as it operated during a defined period, under stated model and workflow versions. Material change can invalidate it. Continuous Decision Assurance and material-change reassessment are scoped against what actually moved rather than repeating the whole assessment.
Start with one decision
Describe a single decision class and we will tell you whether HOP applies, what evidence would be needed, and what we would expect to find. Twenty minutes, no charge, no obligation.