Career Growth
Choose Practice Tasks That Reveal a Skill Instead of Just Testing Recall
An original review exercise supplies a small filter contract, controlled inputs and two contrasting submissions. A criterion-level scoring record shows what each response supports, what remains untested and which practice to assign next.
Choose a practice task by naming the work someone should produce, supplying enough information to do it, and defining what would count as a sound result. Include a case where two plausible implementations would give different answers. Then examine the output against those criteria. A correct definition can support a knowledge check while leaving the intended application untested.
This LandOffer.ai guide develops a small software-review exercise and two fully written, scored responses. The task, learners and feedback are invented. The publisher executed the supplied Python fixture locally to check its outputs; no employee, production system or training program was assessed. The scores explain these authored examples, not anyone's overall competence.

Start with an observable decision
“Understand testing” leaves too much room for interpretation. A narrower objective is: given a written selection rule and a short implementation, choose inputs that could disprove the implementation and explain the resulting discrepancy. The learner produces a review note with input, expected result, observed result and a bounded recommendation. Those fields let another person check the reasoning instead of judging confidence or vocabulary.
That objective changes the assignment. Asking someone to define a Boolean operator checks terminology. Asking for another example identical to the demonstration checks a familiar procedure. Asking for a case that distinguishes two plausible implementations makes the choice of evidence visible.
Carnegie Mellon's alignment guidance explains why objectives, instruction and assessment should reinforce one another. Its example of an analytical objective assessed only through factual recall describes the mismatch directly. The workplace exercise here is our adaptation of that teaching principle, not an evaluated CMU staff-training program.
Recall still has a place. A learner who cannot interpret the terms may need explanation before this task. Supply that instruction and practice before expecting independent application. Otherwise a difficult assignment can reveal that the teaching omitted a prerequisite rather than the particular skill you meant to examine.
Worked task: review a deliberately small filter
The fictional team is practicing how to review a selection rule. Its entire contract is: from the supplied list, retain records whose status is exactly open and priority is exactly high; return their identifiers in input order. The exercise accepts only the lowercase values shown. Missing fields, other types, duplicate identifiers and performance are outside this first task.
Those limits make the expected output checkable. They do not establish an appropriate contract for a real application. In particular, the exercise says nothing about how unknown statuses should be handled in production.
The learner receives this proposed implementation:
def select_ids(records):
return [r["id"] for r in records
if r["status"] == "open" or r["priority"] == "high"]
The supplied input pack contains every combination of the two stated conditions:
| Record | Status | Priority | Expected under the written rule |
|---|---|---|---|
| Q1 | open | high | Include Q1: both conditions hold |
| Q2 | open | low | Exclude: priority condition fails |
| Q3 | closed | high | Exclude: status condition fails |
| Q4 | closed | low | Exclude: neither condition holds |
The assignment asks for a review note, two or more discriminating checks, a proposed correction if justified, and a statement of what has not been established. Learners may use the contract, Python reference material and the supplied runner. This is an open-reference evidence task, not a memory or typing-speed contest.
Set aside practice time and check that the supplied runner works before collecting answers. Our suggested 25-minute slot is a design choice, not a validated duration or an employment standard. If setup fails, preserve that interruption and arrange a usable attempt. Do not interpret a missing execution record as a conceptual error when no functioning environment was available.
Keep a demonstration separate from the final response. For example, first model a simple rule that selects any item with a red label, then use this two-condition pack for the independent attempt. Giving away the exact counterexample immediately before the exercise would change what independent selection means.
Define the scoring criteria before reading answers
CMU's rubric guidance identifies criteria, descriptions and performance levels as parts of an explicit scoring tool. Our rubric uses four criteria and three levels. Its descriptions were written for this exercise; the numbers are an ordinal shorthand, not a validated professional rating.
| Criterion | 0: not demonstrated | 1: partly demonstrated | 2: demonstrated in this task |
|---|---|---|---|
| Contract interpretation | No usable rule or a different rule | Refers to both conditions but leaves their relationship unclear | States that both conditions must hold and preserves the supplied scope |
| Choice of evidence | No check against the rule | Checks only agreement cases, or supplies a contrast without a correct expectation | Checks cases where exactly one condition holds, with correct expectations |
| Reproducible discrepancy | No concrete input/result pair | Gives an input or result but not enough to compare expected with observed | Records enough input and expected/observed output to reproduce the mismatch |
| Recommendation and limits | Unsupported approval or unrelated change | Suggests a justified change but omits verification or scope limits | Ties correction to evidence, checks it and identifies material untested scope |
Each row has a different purpose. A persuasive explanation cannot substitute for a missing result comparison. Conversely, an unpolished sentence should not erase a correct reproducible counterexample when prose style is not the target skill. Keep the exact selected input or quoted review sentence beside its criterion rating, so a second reviewer can inspect why it was assigned.
Before using this with colleagues, give the task and descriptors to another reviewer. Ask them to score sample work independently and discuss disagreements. Cornell's rubric guidance recommends observable descriptors, colleague feedback and revision. It cautions against vague attributes such as whether work is interesting. Our four-row instrument has not undergone a reliability study.
Do not invent a pass mark from the convenient maximum of eight. The criterion pattern is the useful result. Decide what further practice or review is needed from the missing evidence, with the learner's role and the actual workplace process in view.
Two completed responses and their different evidence
The following responses are written teaching specimens. Neither is a transcript of a real learner. Their assertions were checked against the same local fixture, so their contrast comes from the evidence selected, not different hidden requirements.
Response A: “Both conditions must hold: the item must be open and high priority. I ran Q1 on its own and got Q1. I ran Q4 on its own and got an empty list. Those are the expected answers. The implementation is ready to approve.”
A uses the right verbal rule and reports two truthful observations. However, both implementations agree on those two inputs. The response never tests an item satisfying only one condition, so it cannot justify the approval. Correct recall and a passing example coexist with an undiscovered implementation error.
Response B: “The contract requires both conditions. Q2 alone should return an empty list, but this implementation returns Q2. Q3 alone should also return an empty list, but it returns Q3. The current or accepts either condition. Replace it with and. On the four-record pack, the original returns Q1, Q2, Q3; the corrected version returns only Q1. I also checked empty input, which returns an empty list in both. This addresses the supplied lowercase, complete records; it does not establish behavior for missing keys, other types or a live application.”
B's recommendation follows from contrasting inputs. The two counterexamples isolate which individual conditions are insufficient. The complete-pack result adds context; the empty-list check supplies a small additional boundary without pretending to cover every input.
The publisher ran the authored fixture in Python 3.12 and retained the inputs, expected outputs and actual outputs. The run reproduced the single-record, four-record and empty-list results described above. The corrected function used the same comprehension with and, and returned Q1 on the supplied pack. That observation verifies this authored fixture only. It does not demonstrate an actual learner's reasoning, general test coverage or suitability for a production release.
Here is the completed scoring record:
| Criterion | A and reason | B and reason |
|---|---|---|
| Contract interpretation | 2: explicitly requires both conditions | 2: states the same rule and respects its limits |
| Choice of evidence | 1: Q1 and Q4 are agreement cases | 2: Q2 and Q3 distinguish the faulty predicate |
| Reproducible discrepancy | 0: no mismatch is identified | 2: single-record inputs and expected/observed outputs are explicit |
| Recommendation and limits | 0: approves despite incomplete evidence | 2: justified correction, checked result and untested scope |
The descriptive totals are A: three of eight, B: eight of eight. A's pattern identifies a specific next practice need: choosing cases that separate the implementations, even though the verbal rule was correct. B supplies the requested evidence for this task. Neither total certifies broad software-review ability, and B's response does not prove independent performance across unfamiliar systems.

Give feedback that leads to a new decision
Feedback to A should address the missing contrast: “Your rule is correct, and the two reported results agree with it. Choose an item for which open status and high priority do not occur together. Write the expected output before running the function, then compare.” That provides a next action without relabeling A as careless or incapable.
If the reviewer supplies Q2 directly, record that help. A later successful revision is then supported practice. It can still be valuable, but it should not silently replace the original response in a record meant to show independence. Preserve both versions and the prompt that connected them.
For B, the next useful task changes the problem rather than demanding more definitions. One possible follow-up supplies a different written rule and asks the learner to choose distinguishing inputs again. Another adds an explicit rule for a missing priority field and asks what tests now become necessary. These are proposed follow-ups, not completed transfer tests.
Avoid expanding the task halfway through scoring. Penalizing B for an unstated missing-field behavior would assess a requirement the learner never received. If the real work requires that behavior, improve the next task's contract and instruction. A narrow successful attempt is a reason to gather the next evidence, not to invent broader evidence afterward.
Keep the exercise useful and proportionate
A practice task should resemble the decision you care about while removing unnecessary operational risk. This exercise uses four invented records and a local function. It needs no customer data, production credentials or real bug disclosure. More realistic practice can be designed later with the responsible team, appropriate materials and a clear scope.
Make resource conditions explicit. If approved references or assistance are normally available, decide which are allowed and record their use. If independent reasoning is the objective, ask the learner to explain the choice of cases. Do not infer that explanation from a polished answer whose creation process is unknown.
Also separate practice from consequential evaluation. An informal learning task does not automatically become a hiring, promotion or operational-authorization instrument. This article supplies no validated cutoff, accommodation procedure or employer policy for those decisions. A manager should use the organization's applicable process rather than treat a blog rubric as approval to judge a person's career.
For your next practice session, choose one work product, write the conditions and scoring descriptors, and create contrasting specimens before you collect answers. Save the actual output, any hint supplied and the remaining gap. Use that record to select the next practice task; keep broader readiness decisions open. LandOffer.ai publishes this learning-design guide; the useful next step is a bounded practice decision supported by the work you can inspect.
Sources and Further Reading
- Carnegie Mellon: Align assessments, objectives and instructional strategies, institutional teaching guidance checked October 27, 2026. Workplace adaptation and filter exercise are original editorial examples.
- Carnegie Mellon: Creating and using rubrics, criteria, descriptors and performance levels; checked October 27, 2026.
- Cornell: Using rubrics, observable descriptors, review and revision; checked October 27, 2026. No exact publication dates displayed on these pages; no learning-effectiveness result inferred.


