1. Which modules are in the evaluation, by maturity label?
A finished answer is a list the room can point at later. It places Platform Core and GenAI Prompt DLP under available for this POC, places AI usage observability under coming next if that module is even in the conversation, and places every other crew direction under design-partner roadmap. It says which of those rows will be exercised, and it says that a roadmap conversation ends in notes.
Ask. Which modules are in this evaluation, and what maturity label is attached to each one?
Why. A scope line that only says the platform will be read afterwards as a promise covering the whole roadmap.
Weak answer. We will look at everything on the website and decide later what counted.
2. Which test cases are approved before the first run?
The finished answer names the synthetic fixtures, the person who approved them, and the date of that approval. It states that live operational text and records from a real desk stay out of the prompt path. It says where the approved bytes are kept, so a later run can be compared with the same text.
Ask. Which test cases are approved before the first run, and who approved them?
Why. Text that nobody approved is how material from outside the fixture set enters a prompt path.
Weak answer. We will paste whatever is handy on the day and tidy the story afterwards.
3. Who may see a verdict, and who may unwrap evidence?
Write two lists, not one. The first names the roles that may read a verdict. The second names the enrolled officers whose keys may take part in a reveal, and it records that custody in this POC is a software key the development profile holds. A role grant by itself unwraps nothing, because the ceremony checks enrolled key material.
Ask. Who may see a verdict in this evaluation, and who may take part in unwrapping evidence?
Why. The envelope does no work if every person in the room holds a copy of the software key that opens it.
Weak answer. The project team can open anything it needs to.
4. What is the rollback path if a policy snapshot is wrong?
The finished answer names the prior published version, the account that drafts the rollback, and the different account that approves it. A security-admin account drafts it, and a second security-admin account or a security-officer account approves it. The active pointer moves only after that approval, and the published rules stay as they were.
Ask. If a published policy snapshot is wrong, what action restores the previous version, and who may perform it?
Why. The supported recovery is to move the active pointer, and editing a published version in place is a different change with a weaker record.
Weak answer. We will correct the rule directly in the database and move on.
5. Which stand-ins are accepted for this round?
List each stand-in in words a later reader can match to the screen, and write beside it the claim it cannot support. The usual set is local password accounts, software officer keys, a software machine CA, a simulated endpoint fleet, an emulated acknowledgement of audit checkpoints, and a synthetic log source. Accepting the set is a decision, recorded before the run.
Ask. Which stand-ins does this round accept, and what does each one leave unproved?
Why. Local accounts, software keys, and a software CA change the strength of a result even when the screen resembles the intended control.
Weak answer. Ignore the labels and treat the build as production.
6. Where will the appliance sit, and what may it reach?
The finished answer is a one-page sketch of the host, the addresses it binds, who may reach it, and what it may reach. On that sketch, three statements stay separate: the appliance is designed for air-gapped operation, the build has been shown to complete with no route out, and egress from a running appliance is still something this round must measure if anyone intends to report it.
Ask. Where will the appliance sit during this evaluation, and which network paths is it allowed to open?
Why. Air-gap language without a sketch cannot be failed or passed.
Weak answer. It is isolated because the description says air-gap-capable.
7. What would count as a failed test?
Write the miss conditions before the first run, in observable terms. A view of effective policy that disagrees with enforcement fails the round. So does a stand-in shown without its label, a latency or quality figure this POC has not measured, or text from outside the approved fixtures on the prompt path. The team records the miss and stops the claim there.
Ask. What observable result will count as a failed test in this round?
Why. A test with no miss condition can be narrated as a success after the fact.
Weak answer. We will know a success when we see one.
8. Who signs the record of limits at the end?
Name the roles, on the evaluating side and on the product side, who will sign the written limits with the results. The page they sign covers the stand-ins that were accepted, the measurements that were not taken, and the modules that stayed on the roadmap. That page is kept with the outcome.
Ask. Which roles will sign the written record of limits when the evaluation ends?
Why. The disclosed gaps belong in the result, and a summary that keeps only the highlights has dropped half the evidence.
Weak answer. We will write up the highlights and leave the limits in the working notes.
How to read this
| Read it as | Which means | Do not read it as |
|---|---|---|
| Scope | Named modules and their maturity labels | “The platform” |
| Test text | Approved before the first run | Whatever is handy to paste |
| A failed test | Defined before the run starts | A feeling at the end |