grceng.academy
Review lab 40 bonus XP

Review the AI Act register

Start labRead the room

Objectives

The AI Act register is being reviewed before it goes to the board. Each row records a system, its tier, the company's role, and the obligations that follow.

Accept the rows that would hold up in front of an assessor. Reject the rows whose classification, role or evidence will not.

Verdicts are accept or reject. Every rejection needs one reason code from: classified-by-technology | wrong-role | oversight-nominal | no-evidence-artifact.

ai_act_register.yaml: rows under review

Z-1  Applicant ranking tool: classified high risk on the basis of intended use (employment screening), with a named owner.
Z-2  Keyword CV filter that auto-rejects the bottom 40%: classified minimal risk because the vendor states it uses no machine learning.
Z-3  Customer support chatbot: limited risk, transparency notice deployed, artefact linked in the row.
Z-4  Licensed model fine-tuned on company data, rebranded and sold to hospitals: recorded as deployer, obligations assigned to the vendor.
Z-5  Benefits eligibility: human oversight satisfied by a confirmation button; reviewer throughput target 400 cases per hour, no route to record a different outcome.
Z-6  Data governance obligation for the triage model: owner named, evidence cell reads "to be defined".

Checks the server runs (3)

  • Every line in the fixture carries a verdict
  • Accept and reject land on the right lines
  • Each rejection cites a valid reason code

Check names are all you see here: the expectations that decide pass or fail, and the reference solution, stay on the server.

Step-by-step hints

The guided walkthrough, in full. It is free: no XP cost, and it does not touch the paid hint ladder or the reveal gate.

  1. Step 1

    Read the request before the artefact

    Read each row in the order the Act does: what is the system used for, what does that make the company, what obligations attach, and what artefact evidences each one?

  2. Step 2

    Judge every line, including the boring ones

    Six rows, six verdicts. The chatbot row is an accept even though it carries the lightest obligations: proportionate is not the same as wrong.

    Every line in the fixture carries a verdict

  3. Step 3

    Separate what is wrong from what is merely ugly

    A statement about implementation is not a classification. Ask what the system decides about people before you ask how it was built.

    A row can have the right tier and still fail: an obligation whose evidence cell names no artefact is a gap wearing a control's clothes.

    Accept and reject land on the right lines

  4. Step 4

    Say why, in the vocabulary the programme uses

    Use the role code for the row where fine-tuning and rebranding changed who the company is, and the oversight code for the row where a human is present but powerless.

    Each rejection cites a valid reason code

Rules summary

  • XP on pass

    40 bonus XP the first time every check goes green.

  • Paid hints

    2 available. Each costs 25% of the lab XP (about 10 XP).

  • Reveal solution

    Scores 0 XP and unlocks only after 3 check runs.

  • Attempt cap

    60 check runs per learner. The runner warns you in the last 10.

  • Drafts

    Your editor content is saved in this browser, so a reload will not lose work.

  • Walkthrough

    Always free and never penalised: use it before spending a hint.