Eight short steps, at your own pace
Each step gives an explanation and example, then asks one question. Use the fictional task below throughout. With JavaScript enabled, answer checking offers one retry and then an explanation. Nothing is sent to a model or server.
Without JavaScript, use the worked examples and “Answer and takeaway” sections. Open-ended writing uses a self-review checklist; automatic hints cannot certify the quality of a rubric. Write on paper if you prefer.
Your fictional task and input file
Everything below is fictional. Luma Quay, all vendors, identifiers, rates and rules are invented for this lesson.
Synthetic task prompt
Create an itinerary for two travelers, Ari and Bo, for Monday 8 June through Wednesday 10 June in fictional year F1. Include one attraction and one lunch each day, use each attraction once, and use only the supplied vendors. Both travelers require nut-free meals. Show each traveler’s total and the party total, including the shared shuttle. Budget: at most 100 credits for the party.
Synthetic input facts
- Glass Garden: 10 credits per person; closed Monday.
- Lantern Hall: 15 credits per person; closed Wednesday.
- Cloud Walk: free; unavailable Tuesday (a blackout date).
- Moss Kitchen: nut-free lunch, 5 credits per person; open all three days.
- Ember Café: lunch, 4 credits per person; contains nuts. It cannot meet this task’s meal requirement.
- One shuttle pass for all three days costs 20 credits for the party. Allocate 10 credits to each traveler.
- No other costs, taxes, tips, discounts, travel-time constraints or reservations apply. Each attraction and lunch takes place within that day.
Complete fictional deliverable A
- Monday 8 June: Lantern Hall; lunch at Moss Kitchen.
- Tuesday 9 June: Glass Garden; lunch at Moss Kitchen.
- Wednesday 10 June: Cloud Walk; lunch at Moss Kitchen.
- Ari: 10 + 15 + 0 + (3 × 5) + 10 = 50 credits.
- Bo: 10 + 15 + 0 + (3 × 5) + 10 = 50 credits.
- Party: 50 + 50 = 100 credits, including the shared 20-credit shuttle.
Attractions: 25 per person, 50 party. Lunches: 15 per person, 30 party. Shuttle: 10 per person, 20 party. Total: 50 per person, 100 party.
1. Warm-up: the four parts
A prompt asks for an itinerary; the input file supplies prices; the response is the itinerary; the rubric checks it. Example: “Party total is exactly 100 credits” is a rubric criterion.
Answer and takeaway
The price list supplies facts to the task; it is the input file.
Takeaway: Keep instructions, evidence, response and grading rules distinct.
2. Four principles
Atomic means one check. Verifiable means PASS/FAIL from the deliverable alone. Objective means a rule rather than personal taste. Self-contained means all grading facts are in the criterion. Example: “States the full-trip party total as exactly 100 credits, including the shared 20-credit shuttle.”
Answer and takeaway
The exact total supplies a single observable rule and its expected value. “Delightful” is subjective; “match the file” requires an external lookup.
Takeaway: A reviewer should not have to guess what PASS means.
3. Practice A: diagnose a flawed criterion
Flawed: “Lists a nut-free lunch and states the correct party total.” Two things can succeed or fail independently. Example correction for the first check: “Lists Moss Kitchen for Monday lunch; Moss Kitchen is nut-free.”
Answer and takeaway
Atomic fails: lunch choice and arithmetic are separate checks. Write a second criterion with the exact 100-credit party total.
Takeaway: Split checks that can have different outcomes.
4. Practice B: rewrite
Flawed: “The cost looks reasonable.” It uses an opinion and hides the expected value. Worked rewrite: “States Ari’s full-trip total as exactly 50 credits, including 10 credits for the shuttle.”
Answer and takeaway
Use the exact expected 100-credit party total and include the shuttle fact. No taste judgment or source lookup is needed.
Takeaway: Replace vague praise with a specific observable value.
5. Practice C: assign and justify weights
Positive weights reward fulfilled requirements. Negative weights penalize a defined mistake when that mistake is present. A PASS on a negative criterion triggers the penalty. Example: “Schedules any Ember Café lunch, whose meals contain nuts” has weight −6; a consistent currency label earns +1. Dietary compliance has higher stakes here. Explicit requirements come directly from the prompt or input; implicit requirements are necessary inferences, such as consistent units.
Answer and takeaway
Penalize the defined nut-containing meal mistake; reward consistent units modestly. Weights express task stakes, not equal-sized boxes.
Takeaway: Define the negative trigger before assigning its penalty.
6. Practice D: split without changing total weight
Flawed +6 criterion: “Lists Lantern Hall on Monday and Glass Garden on Tuesday.” Split into two self-contained checks: Lantern Hall on Monday, when open (+3); Glass Garden on Tuesday, when open (+3). The sum remains +6. Splitting allows partial credit when only one part passes.
Answer and takeaway
3 + 3 = 6. The other splits double or increase the weight. Preserve the total, while recognizing partial credit changes.
Takeaway: Split the requirement and preserve the weight budget.
7. Practice E: exclude an unverifiable criterion
Flawed: “The assistant checked every vendor’s live website before answering.” The itinerary does not prove an unseen browsing action. Replace process guesses with output checks: “Does not schedule Cloud Walk on Tuesday, when it is unavailable.”
Answer and takeaway
Exclude it from this output rubric. A process audit would need separately authorized evidence; do not infer hidden activity from polished writing.
Takeaway: Grade observable output; do not invent evidence of hidden work.
How weighted PASS/FAIL scoring works
| Single check | Weight | Origin | Deliverable A evidence |
|---|---|---|---|
| Lists Lantern Hall on Monday 8 June, when it is open. | +3 | Explicit | PASS: Monday names Lantern Hall. |
| Lists Glass Garden on Tuesday 9 June, when it is open. | +3 | Explicit | PASS: Tuesday names Glass Garden. |
| Lists Cloud Walk on Wednesday 10 June, outside its Tuesday blackout. | +3 | Explicit | PASS: Wednesday names Cloud Walk. |
| States Ari’s full-trip total as exactly 50 credits, including 10 credits for the shuttle. | +2 | Explicit | PASS: Ari’s total is 50. |
| States Bo’s full-trip total as exactly 50 credits, including 10 credits for the shuttle. | +2 | Explicit | PASS: Bo’s total is 50. |
| States the full-trip party total as exactly 100 credits, including the shared 20-credit shuttle. | +4 | Explicit | PASS: party total is 100. |
| Uses the same currency label, credits, for every stated monetary total. | +1 | Implicit | PASS: every monetary total uses credits. |
| Schedules at least one lunch at Ember Café, whose meals contain nuts. | −6 | Explicit | FAIL: no Ember Café lunch is scheduled; no penalty. |
For each criterion, PASS means its statement is true. Add its weight on PASS; add zero on FAIL. Deliverable A earns 3 + 3 + 3 + 2 + 2 + 4 + 1 + 0 = 18 points. The negative criterion is a defined mistake: it fails here, so the penalty does not apply. Raw scores can be negative; this exercise has a −6 minimum and +18 maximum, not a certification threshold.
Fictional deliverable B: spot the failures
Keep A’s attractions and printed costs unchanged, but change Tuesday lunch to Ember Café. The printed Ari total is still 50, Bo 50 and party 100 credits. This is deliberately flawed: the meal violates the restriction and the printed costs no longer match its choices.
Under these eight checks, all seven positive statements remain true and the negative statement now passes: 18 − 6 = 12 points. That does not make B acceptable. This teaching rubric does not check every requirement or recalculate every possible itinerary. Add separate arithmetic-consistency and daily nut-free-lunch checks for a full evaluation. Critical requirements can also be hard gates, independent of score.
Advanced example: the existing Veiled Dominion reviewer scorecard uses weighted 1–5 dimensions and hard gates. Some broad dimensions combine judgments; they differ from the strict atomic PASS/FAIL rules taught here.
Score answer
12 points. Subtract 6 once; do not turn a penalty into a reward.
8. Wrap-up: write three criteria
Now write your own three single-check criteria for deliverable A. Each needs a signed weight and an explicit/implicit label. Example: “States Bo’s full-trip total as exactly 50 credits, including 10 credits for the shuttle” (+2, explicit). Choose three different checks and explain their stakes.
Below, write one criterion at a time. Hints flag possible issues with wording; they do not understand meaning or decide PASS/FAIL. After one revision, use the self-review checklist and examples to improve your draft.
Targeted self-review
- Atomic: could part pass while another fails? Split those checks and preserve their combined weight.
- Verifiable: can you point to exact words or omissions in the response? Exclude unseen actions.
- Objective: replace “good,” “reasonable” and similar opinions with an exact rule.
- Self-contained: include the expected amount, date, vendor restriction or other needed fact in the criterion.
- Weights: explain the stakes; a negative weight needs a precisely defined mistake.
- Origin: name the prompt/input fact or explain the necessary inference.
Independent exercise: fictional deliverable C swaps A’s Monday and Tuesday attractions: Monday Glass Garden; Tuesday Lantern Hall; Wednesday Cloud Walk. All lunches are Moss Kitchen and totals remain 50 per traveler, 100 party. Score your three criteria against C. Record PASS/FAIL and the exact evidence, then rewrite one weak criterion. For a negative criterion, PASS means its mistake occurred.
Independent exercise feedback
Glass Garden is closed Monday. Under the eight-check teaching rubric, the Monday Lantern Hall and Tuesday Glass Garden checks both fail; the other positive checks pass and the meal penalty does not trigger. Score: 18 − 3 − 3 = 12. Your three-criterion score depends on the checks you wrote; explain each outcome rather than borrowing this total.
Takeaway: a useful rubric makes a specific failure visible and supports a concrete revision.
Finish with a fresh scenario and scoring record
The worked examples above stay available for learning. This final exercise varies the fictional vendors, prices and mistakes each time you open the page. Calculate the totals, then score three checks one at a time. The generated arithmetic and answer key follow fixed rules. You have two assessment attempts, each with four checks; the practice target is 4/4. After the first unsuccessful attempt, study feedback tied to your scenario before trying a changed response with the same facts. After a second unsuccessful attempt, explore further study and a path based on the checks you answered well. These are learning suggestions, not job-suitability decisions.
Use the record with PIXIE in the classroom pilot
Ask PIXIE: “Help me review my synthetic scenario, three criteria, first-attempt answers, retries, final scoring and one improvement. Identify the next useful practice step. Keep it as a draft for my existing Training checkpoint.” The download includes the scenario facts and scoring history so a new session can use evidence rather than memory.
Review visibility and attribution before sharing through your existing GitHub issue. The record is learner-reported practice; PIXIE does not automatically receive, save or verify classroom progress. Keep private learner and academic records out of public checkpoints. No new classroom tracker is created.
Keep a shareable practice artifact
Use the synthetic practice-artifact template. Include your prompt, input facts, three criteria, weights, classifications, fictional response, PASS/FAIL evidence and one improvement made after feedback.
Your entries stay in this page’s memory and are erased on reload or reset. Download only if you choose; do not enter personal, employment, medical, academic or financial records.
Sharing is optional. Follow the existing contribution and portfolio workflow. Review your artifact and attribution before submitting through your own GitHub account: issues and comments are public. Keep your existing checkpoint URL and add progress there; an exercise is not maintainer acceptance, certification or membership.