# Administrator knowledge check

Ten questions on access, automation and governing AI on top of your metadata. You need 8 correct to pass. Unlimited retakes, with an explanation for every answer.

Question 1

1

Why assign access to teams rather than to individual users?

Access follows organizational structure, so a joiner or leaver is one change in one placeIndividual assignment is not permittedTeam assignment grants broader permissionsIt is the only way to use roles

Question 2

2

Which part of the access lifecycle carries the most risk and is most often left undesigned?

ProvisioningTeam creationPassword policyDeprovisioning and periodic review

Question 3

3

What is the relationship between a policy and a role?

They are interchangeable termsA role is a rule, a policy bundles rolesA policy is a rule, a role bundles policies so they can be assignedPolicies apply to teams, roles apply to users

Question 4

4

Why does starting broad and trimming permissions later tend to fail?

The platform does not allow permissions to be removedIt never gets trimmed, because nobody wants to remove access that might be load bearingBroad permissions cannot be auditedIt causes ingestion failures

Question 5

5

Separation of duties requires that a proposer is not the approver. Where does that property come from?

From splitting the relevant policies across different rolesIt is enforced automatically by the permission modelFrom assigning every user a unique roleFrom disabling self service entirely

Question 6

6

A persona hides the administrative surface from business analysts. Are those features secured?

Yes, hidden features cannot be accessedNo, a persona shapes what is surfaced while a role decides what is permittedYes, provided the persona is applied to a teamOnly if the analysts have no role assigned

Question 7

7

Lineage is missing for a set of tables, though ingestion is clearly working. What is the useful first step?

Identify which agent owns lineage and check its recent runsRecreate the connectorRe-run the full ingestionRaise a support ticket immediately

Question 8

8

How should agent run frequency be decided?

Run everything as often as the platform allowsMatch the cadence to how fast the underlying thing changes, since schema and volume move at different ratesRun everything nightlyFrequency has no cost implication

Question 9

9

You are designing a new AI agent. What most determines whether it can be trusted and debugged?

How often it runsThe size of the model behind itThe set of tools and actions you grant it, and how narrowly you scope themThe number of users who can invoke it

Question 10

10

What is the failure mode of pointing an AI agent at poorly governed metadata?

The agent refuses to answerIngestion stopsThe agent runs more slowlyThe agent answers fluently and confidently from stale or mislabeled metadata

Check my answers0 of 10 answered