Turn a policy into a work-ready assessment

1. Write the work task before choosing a quiz

A policy document tells people what the rules are. An assessment needs to reveal whether they can use those rules in a relevant situation. If the work is triaging support requests, asking someone to define “escalation” is not enough: ask them to select a next action and explain the evidence that supports it.

Use this brief: Given [realistic inputs] and access to [permitted references], the learner will [observable action], meeting [review criteria]. For our fictional support team: given three requests and a short policy card, choose the next action for each, justify the priority and identify missing information without promising an unauthorized outcome.

First check whether training is the right response. If the current policy cannot be found, contains contradictory rules or requires authority the learner does not have, fix that constraint with the responsible owner. A more polished course cannot resolve an unclear operating process.

2. Set the conditions and show the rules

The following toy policy is invented solely to make the assessment design inspectable. It is not customer-service, security or legal advice. Replace it with current approved rules and have the responsible subject-matter expert review every case before real use.

Policy card A: a suspected account-access or personal-data exposure concern is escalated immediately to the designated specialist. Do not request passwords or promise a resolution. Card B: a delivery-status enquiry with an order number follows the standard tracking process. Card C: when an order number is missing, request that identifier before making an order-specific decision. Card D: prioritize a suspected exposure concern over routine delivery enquiries; the order of two routine enquiries is not assessed here.

Allow the policy card during both practice and assessment: the intended skill is finding and applying the rule, not remembering its wording. Do not add a speed threshold unless the real job requires it and the team has justified it. State whether collaboration or an approved AI tool is permitted; keep these conditions consistent with the performance claim you intend to make.

3. Try three cases before reading the rationale

Case A: “I can see another person’s address in my account.” Case B: “Where is order 2041? Tracking has not updated.” Case C: “Where is my parcel?” No order identifier is supplied. These are fictional requests; do not paste live customer records into the worksheet.

For each case, write four things: the next action, the relevant policy card, the information still needed, and one sentence you would say to the customer. Then explain which request you handle first. Use the supplied rules only. If a detail is not supported, mark it unknown rather than filling it in.

Before revealing the answer, check the task design: did you give enough information to decide? Is the assessment accidentally measuring reading speed, unfamiliar jargon or interface skill? Remove irrelevant difficulty while retaining the actual decision.

4. Compare the decision and the rationale

A goes first: escalate the suspected exposure to the designated specialist under A and D. A suitable response is “I’ll pass this concern to the designated specialist.” Do not request a password, repeat exposed personal details or guarantee a fix. The case does not establish the cause or prove a breach; the correct action follows the toy policy’s escalation threshold.

B follows the standard tracking process using order 2041 under B. The case gives no evidence of a refund entitlement or a delivery date. “I’ll check the tracking for order 2041” stays within the information supplied.

C needs the order identifier under C before an order-specific decision. “Please share the order number so I can check the delivery status” is sufficient for this toy exercise. Do not invent an order record. Either B or C may follow A; a scorer must not penalize their relative order because the policy does not define it.

These are example rationales, not mandatory scripts. Accept different wording when the action and evidence satisfy the rule. A learner who selects the expected action for the wrong reason needs different feedback from one who identifies the rule but chooses the wrong action.

5. Score four observable criteria

Use the four criteria below across the complete three-case response. The 0–2 scale is a draft feedback aid, not a validated measurement instrument. Keep the criterion-level notes; a total can hide an important error. Do not create a pass score or use this sample to make hiring, disciplinary or certification decisions.

Action: 2 = selects the policy-supported action for all three cases; 1 = two cases are supported and one needs correction; 0 = fewer than two are supported. Record which case, not merely a number.

Reasoning: 2 = connects each action to the relevant card and explains why A is first; 1 = mostly correct references but one missing or unclear connection; 0 = reasoning is absent, contradictory or not grounded in the cards.

Uncertainty: 2 = recognizes C’s missing identifier and avoids inventing facts in all cases; 1 = flags the missing identifier but leaves another assumption unmarked; 0 = proceeds as though missing facts are known.

Communication: 2 = gives clear next-step messages without an unsupported promise or password request; 1 = one message is vague but does not introduce an unauthorized promise; 0 = includes a prohibited request or unsupported commitment. Independently flag any failure to escalate A for immediate coaching; do not let strong wording compensate for it.

Calibration check: have two reviewers independently score the same sample response, then compare the evidence behind every difference. Revise ambiguous descriptors before a pilot. Agreement on one example does not prove reliability or validity; it is a useful way to find unclear scoring instructions.

6. Give feedback that changes the next attempt

Weak feedback says “Wrong—try again.” Useful feedback identifies the missed cue, links it to the rule and offers a new attempt. For example: “You handled the tracking enquiry first. The address visibility concern meets card A’s escalation condition; card D gives it priority. Reorder the actions and explain the cue that changed your decision.”

After supported practice, use a different case with the same underlying decision. Do not use the identical wording as evidence of transfer. If learners may use hints, record that assistance; a successful guided response is not the same as independent performance.

Pilot with representative learners and an expert reviewer, with appropriate permission. Ask learners to explain confusing wording after the attempt. Separate a policy misunderstanding from an inaccessible interface, missing instructions or a scoring disagreement. Record the revision and its reason. No pilot or learner outcome has been measured for this example.

7. Check access, source ownership and the release decision

Offer the scenario and policy in readable text, not only an image. Make controls keyboard-operable, keep labels explicit and avoid color-only meanings. Let learners use an accessible response format that still demonstrates the intended skill. A written task need not become a timed drag-and-drop exercise just because the authoring tool supports one.

Record the policy version, expert reviewer, scoring owner and next review trigger. Hold release if a rule is disputed, a correct answer depends on an unstated fact or an access barrier prevents learners from demonstrating the skill. Revise and retest instead of averaging away the problem.

An AI assistant can suggest variations from approved material. Require a rule reference for each branch and mark unsupported details for review. A human must check the case, rationale and feedback; fluent wording is not evidence that a policy was applied correctly. Never upload confidential records without permission.

Use the brief on one real task

Copy the free editable task-to-assessment brief below. Fill in one task, the permitted references, a fictional or authorized de-identified case, the expected evidence and the four review criteria. The existing scenario storyboard helps you plan practice and feedback; the evaluation guide helps you decide what to observe afterward.

This is a design starting point, not a validated test. Independent performance on an assessment does not by itself prove workplace transfer, business impact or reduced errors. If the decision is consequential, involve the appropriate assessment, accessibility and domain specialists before using the results.