AI Engineering
/
Responsible AI & Governance
/

Specify a review and escalation workflow

Responsible AI & Governance · 2 to 3 hours

Specify a review and escalation workflow

Specify how disputed AI feedback is re-reviewed, corrected, and escalated, with server-side authorization, artifact-bound approval, deadlines, and concurrency rules, then prove the rules with acceptance fixtures.

Medium
Hands-on

Hands-on workflow specification

3 h

Step 1: The question

Specify the workflow that handles a learner's dispute of published AI feedback on the fictional platform, from filing to closure. Define the case states and transitions, who may take each action and how that is enforced, how a corrected result is bound to the exact reviewed artifact, what happens when a deadline passes, and how concurrent actions and repeated publication requests are handled. Add a risk-register entry for the risk that a wrong score stays in final contest results because its dispute is not decided in time, the audit events and the remedy for affected records, a reviewer-capacity estimate for the contest window, and acceptance fixtures that would prove the dispute path.

Build a risk register with evidence

FieldExample
Risk IDR-017
ScenarioA valid alternative solution receives a failing score
Affected peopleLearners using less common but correct approaches
Cause hypothesisRubric and examples overfit a preferred implementation
ControlBroader fixtures, expert adjudication, accessible dispute path
EvidenceEvaluation report E-12 and review sample E-19
OwnerFeedback quality owner
Residual uncertaintyLimited coverage of advanced asynchronous solutions
Decision and review triggerRestrict score use; reassess after dataset expansion

Keep the scenario concrete: "bias exists" tells nobody what to test. Separate observed incidents from hypothetical risks and record your confidence in the cause. A control needs an owner, a deadline, and evidence that it operates; a planned control is not an implemented one. The NIST AI RMF Playbook suggests actions for this work; adapt them rather than treating them as a mandatory checklist.

Review workflow specification

The review contract for the fictional platform in the fixtures below:

Current stateEventActorPreconditionsResult
DraftSubmitGeneratorArtifact and evidence revisions fixedPending review
Pending reviewApproveQueue reviewerEvery required evidence item present; the reviewer can open each one; reviewed artifact hash, evidence revision, and policy version equal the stored onesApproved; approval bound to the reviewer, artifact hash, evidence revision, and policy version
Pending reviewRejectQueue reviewerReason recordedRejected
Pending reviewDeadline reachedSchedulerNo valid decision before the deadlineExpired and escalated to the queue lead; never approved by default
ApprovedBinding changesGenerator (output or evidence) or policy service (policy)New output, material evidence revision, or new policy versionDraft with the new binding; the old approval is superseded
ApprovedPublishPublisherThe approval's artifact hash, evidence revision, and policy version still equal the stored ones at commit timePublished; score written to the learner's progress record once; one learner notification
PublishedRestrictIncident ownerReason recordedRestricted: explanation visible; score stops counting and shows as under review; investigation opened
PublishedWithdrawIncident ownerReason recordedWithdrawn: explanation and score hidden and removed from progress totals; learner sees an under-review notice; investigation opened
Published, Restricted, or WithdrawnCorrection publishedPublisherA corrected item for the same submission has just been publishedSuperseded: kept in history; its score leaves progress totals and the corrected score replaces it

Rejected and Expired are final for that artifact revision; new output starts again as a Draft. Some workflows need two independent reviewers or more states; derive them from the use case rather than adding ceremonial approvals.

The feedback-review fixture

Suppose a coding-practice platform shows learners AI-drafted feedback on their solutions: an explanation and a practice score from 0 to 5 that counts toward their progress record. A model drafts each item, a reviewer approves it, a publisher service publishes it, and learners can dispute a published score.

Item F-31 is the draft feedback for learner Lee's solution S-88:

FieldValue
Revision4 (the record version; every applied transition adds 1)
Artifact hashh1 (the exact explanation and score)
Evidence revisionev3 (hidden-test results and the rubric)
Policy versionp7 (current)
Review deadline48 hours after submission

Actors. Identity always comes from the authenticated session, never from a request field.

ActorMay do
Rina, KaiApprove or reject items in this queue; can open all evidence
OmarApprove or reject items in this queue; his access to the hidden-test results has lapsed
LeeDispute their own results once published
GeneratorSubmit drafts and write new model output or evidence for an item
Policy serviceMove items to a new policy version
PublisherPublish items
DanaRestrict or withdraw published items (incident owner)
SchedulerFire deadline events

Check order. Each command goes through these steps, and the first failing step decides the outcome.

  1. Authorize the actor for the event. Failure: refused as unauthorized; nothing changes; a denied-attempt log entry is written.
  2. Publish only: if the idempotency key was already applied, return the stored outcome and change nothing.
  3. Compare the expected revision with the stored one. Mismatch: conflict; nothing changes; the caller is shown the current state.
  4. Check that the table allows the event from the current state. Otherwise: refused as an invalid transition.
  5. Check preconditions in table order. Failure: refused with the specific reason, which is returned to the actor and logged; no audit event; the state does not change. If a reviewer cannot open required evidence, the queue lead is also alerted to reassign the item or restore access.
  6. Apply atomically: new state, revision plus 1, one audit event.

Disputes. Only the learner a result belongs to may dispute it. A dispute is not an item transition: after step 1 it skips steps 2 to 6 and follows this rule instead. It may be filed against a Published, Restricted, or Withdrawn item. It leaves the item's state and revision unchanged, creates an Open case record with an owner from the dispute rota, writes one audit event linking the case to the item revision, and gives the learner the case ID and due time. A decision is due within 5 days, or before the contest results become final for a dispute filed during a contest window.

Contest load. During the weekly 8-hour contest window, disputes arrive at about 12 per hour and each re-review takes about 15 minutes. Two qualified reviewers are on shift, and contest results become final 2 hours after the window closes.

Constraints

  • Use only the actors, rules, and figures in the context sections
  • Nothing may be approved, upheld, or published by default, whether a deadline passes or a backlog grows
  • Specify engineering controls; make no claim about legal or regulatory compliance

Deliverables

  • Dispute case states and transitions, including deadline and escalation paths
  • Authorization and concurrency rules for every action
  • A risk-register entry for a wrong score that stays in final contest results, with controls, evidence, an owner, and a trigger
  • Re-review setup: what the reviewer sees, who may re-review, contest-window capacity, and how review quality is measured
  • Audit events, the remedy for affected records, and acceptance fixtures for the dispute path

Step 2: Hints

Optional. Each hint gives a little more away.

Step 3: Work through it

Predict the outcome

Predict the acceptance fixtures

Each row starts from its own stated record for item F-31; rows are independent. Use the actors, transition table, and check order from the context. Predict the outcome, the stored state afterward, and the audit and downstream effects before you reveal them.

FixtureStartCommandOutcomeStored state afterAudit and downstream effects
A1. Authorized reviewer approves an unchanged, complete itemPending review, rev 4; h1, ev3, p7; evidence completeRina: Approve, expected rev 4, reviewed h1, ev3, p7???
A2. Unauthorized user submits an approvalPending review, rev 4Lee (learner session): Approve, expected rev 4; the request body says reviewer "rina"???
A3. Reviewer submits a stale expected revisionApproved, rev 5 (Rina approved at rev 4)Kai: Reject, expected rev 4, reason recorded???
A4. Output changes after approval, then publication is attemptedApproved, rev 5; approval bound to h1, ev3, p7(a) Generator: Binding changes (new output h2), expected rev 5. (b) Publisher: Publish, key P-1, expected rev 6???
A5. Required evidence is missingPending review, rev 4; ev3 lacks the hidden-test resultsRina: Approve, expected rev 4???
A6. Deadline passes without reviewPending review, rev 4; 48 hours pass with no decisionScheduler: Deadline reached, expected rev 4???
A7. Duplicate publication requestApproved, rev 5; approval valid for h1, ev3, p7Publisher: Publish, key P-7, expected rev 5; after a timeout, the identical request is sent again???
A8. Published output is withdrawnPublished, rev 6Dana: Withdraw, expected rev 6, reason "explanation quotes another learner's code"???
A9. User disputes a resultPublished, rev 6Lee: Dispute F-31, reason "valid approach penalized"???
A10. Reviewer cannot open the source evidencePending review, rev 4; evidence completeOmar: Approve, expected rev 4???

Diagnose

Escalate a disputed score

Lee files dispute D-12 against published item F-31 (revision 6): a score of 2 out of 5 on an interval-merging exercise. The solution passes all 14 hidden tests. In the dispute, Lee explains that they use a screen reader and write one statement per line with short landmark comments so they can navigate the code by ear. The published explanation says "Readability: verbose, non-idiomatic structure (minus 3)."

The audit trail shows that Rina approved F-31 under policy p7 after 40 seconds. The review screen showed the model's explanation and score first, with the hidden-test results in a collapsed panel.

The case owner pulls the passing submissions for this exercise from the last 30 days:

SlicePassing submissionsReadability penalty of 2 or more points
One statement per line with landmark comments3431
All other passing submissions1,16660
  1. Say who owns D-12, what you preserve as evidence, and where Lee's accessibility details may be stored.
  2. Compute the penalty rate for each slice. Say what the comparison does and does not show, and how you would test the cause directly.
  3. Decide whether to contain similar outputs while investigating. If you do, name the workflow action, who performs it, and its scope.
  4. Say what the F-31 approval suggests about the review setup, name one change, and explain how blind samples would show whether the change worked.
  5. Describe the remedy for Lee and for other affected learners using the workflow rules, the conditions for closing D-12, and what Lee is told.

Kept in this browser. Saving your score below keeps a copy with it.

0 / 5000

Step 4: Reference answer

Step 5: Score yourself

Score each dimension 0, 1, or 2, then save the score to your history. Readiness thresholds are editorial estimates.

Grades your written answer against this rubric with your own AI provider key from Settings. It is a second opinion: the score you save is still yours.

Risk and evidence mapping
Workflow correctness
Authorization and concurrency
Reviewer effectiveness
Audit and remedy design

0 of 10 · Scored 0 of 5 dimensions

Ready at 8 of 10, with no 0 in Workflow correctness

Kept in this browser until you sign up, then it moves to your account.

Complete this level when each risk has an accountable response, each approval can be traced to the exact reviewed output, and your fixture predictions follow from the stated check order.

Your saved scores

Practice saves and mock interview answers on this question, newest first.

No saved scores yet. Score every dimension, then save.

Step 6: Follow-up questions

An interviewer may ask these next.

Why bind approval to an artifact revision?

What should happen when a review deadline passes?

How do you measure whether human review works?

Does a risk register reduce risk by itself?

Why carry an expected revision in each transition?

Step 7: Mark complete