Level 200 credential route (draft): Open Certificate in Applied AI Practice
Draft route for review. This page describes the Level 200 project and proposes using its marking sheet for the course credential. It does not award or promise a credential. The credential specification records the proposed route; approval cannot be bought or replaced by an AI mark.
Anything marked Proposal is not decided yet. Bob Rapp and Ken Johnston, the review board co-chairs, decide it unless the page names Bob alone.
What you must do
Read the Level 200 lessons in the gear that suits you. Complete each module's quiz and workbook. For objective 200.P-a, retrofit a rule onto an approved tool you use, or use the made-up Parts Desk case. Rate three consequential actions (actions that send, change, buy, publish or delete something), set a policy, test it with at least ten inputs and keep one share-safe receipt. Show how the rule can be stopped and recovered. Ask a second person to review your work before bringing it to a host.
Proposal, needs the review board co-chairs’ (Bob and Ken) decision: make the quizzes and workbooks practice only, not a credential requirement. The current credential specification calls for completing them.
The rubric: how your project is marked
The host judges whether the policy decided as intended, not just whether a box is filled.
| Check | Meets when |
|---|---|
| Policy | Rates three actions by who is affected and whether each can be undone; names the hold approver and deputy or what happens when the approver is away; gives a reason for a deny; denies unlisted actions; says what happens when a rule condition fails |
| Tests | At least ten made-up or approved redacted inputs, with decisions predicted before the run; covers each rule, a false condition, an unlisted action and tricky cases, including a deliberately weakened rule |
| Results | Records the actual decision and pass or fail for each input; fixes any failure and reruns the whole set; if all inputs initially passed, shows the weakened-rule test can catch a failure |
| Receipt | Records the action, decision, any hold approver and owner from a gated or checked use; contains no prompts, answers, customer data or restricted data |
| Stay and recover | Sends rule changes to a reviewer, gives a named person the ability to stop the tool, and spells out rollback, fixing and retesting the rule, and getting a new yes |
A gap found and fixed can pass; a test set that cannot reveal a gap cannot.
A worked example
In the fictional Parts Desk case, ordering parts can affect Big Widget and its supplier. The rule holds an order for Dana, or Sam when Dana is away. Tests need to distinguish that hold from an allowed action and from a denied action. The receipt records the decision, not the conversation.
How a host marks it
A second person allowed to see the work applies the policy to three inputs without seeing your expected decisions, suggests a tricky input to try, checks that the receipt is share-safe, and signs and dates their findings.
Bring only the shareable copy to the Foundation's Thursday call. A host applies the project marking sheet. You do not have to host the call.
If it is "not yet": revise and resubmit
Fix the gap, rerun the test set when a policy test fails, and bring the corrected work for human marking. No separate credential resubmission or appeals process is established here.
What the certificate would say, and what it would not
The proposed claim is limited to a host marking this Level 200 project against the project rubric. This route does not claim that the project earns the Practice 200 mark. The certificate would not certify the tool or provide legal approval, a degree, or a professional or vendor certification.
The certificate would carry the lines "Free and open. Every source cited. Human-reviewed." and "Earned, never bought."