Your learning brief
Learn to choose a useful AI task, supply the right evidence and check the result before someone relies on it. Your project is a source-verified briefing for a fictional community learning centre. You can complete the course with the supplied simulated outputs; a paid model account is not required.
Year 16 modules10-hour instructional plan
Learning outcomes
- Distinguish prediction, generation, retrieval and rule-based automation.
- Select a bounded AI task and define evidence of success.
- Check a generated answer against supplied facts and record uncertainty.
Prerequisites
No prior course required.
Complete all 18 activities; achieve at least 80% on each five-question module quiz (4/5), at least 80% on the final assessment and at least 80% on the capstone with every mandatory artefact present. An instructor must verify the work and documented active instructional hours before recording completion. Browser self-checks are practice only and are not secure graded assessment.
Ten documented instructional hours may earn one internal Northline learning credit only after every completion requirement and instructor verification. The 600-minute teaching plan is not proof of attendance, an awarded credit, external academic credit or accreditation. Enrollment is not open; this package does not process payments or issue certificates.
Instructional schedule
Each 75-minute module: 15 minutes teaching, 10 minutes worked example, 15 minutes Activity 1, 15 minutes Activity 2, 10 minutes Activity 3 and 10 minutes quiz/review. Capstone: 90 minutes. Final assessment: 60 minutes. Total: 6 × 75 + 90 + 60 = 600 minutes. Record actual active learning time; breaks are excluded. If you finish earlier, agree additional supervised practice with your instructor rather than inventing time.
Keep your responses in a separate document. This page does not save responses or award completion.
Module 1 · 75 minutes
AI and intelligent systems
Teaching
An intelligent system is a combination of inputs, processing, outputs and decisions about what happens next. The label AI tells you less than the actual operation. In this course, prediction estimates a category or quantity from patterns; generation creates a new sequence such as a draft; retrieval locates stored material; and rule-based automation follows explicit conditions. A product can combine all four. Search may retrieve a policy, a model may draft an answer, a rule may require review, and a person may decide to send it.
Classify a system by the step you can observe. A spreadsheet formula that flags totals over a fixed amount is rule-based, even if the vendor calls the product intelligent. A model that estimates demand from historical orders is prediction. A chatbot that quotes a record verbatim may be retrieving information at that moment. You do not need to guess the hidden implementation: record what is known, what is inferred and what remains unknown.
The distinction matters because verification changes with the operation. Check a rule against its conditions and boundary values. Compare a retrieval result with the source and whether relevant records were omitted. Evaluate a generated summary claim by claim. Assess a prediction using a defined outcome, suitable examples and the cost of mistakes. Do not use fluent language as evidence that a prediction or summary is correct.
Make an input–operation–output–owner table before using a tool. The owner is the person responsible for the resulting decision, not the software brand. A useful first project is small enough that the owner can inspect every result. When a tool combines functions, describe the sequence rather than forcing the entire product into one category.
Worked example
FICTIONAL CLASSROOM EXAMPLE — Cedar Learning Centre receives twelve workshop enquiries. A filter moves messages with the word “invoice” into a folder: rule-based automation. A search finds the workshop timetable: retrieval. An assistant drafts a response using the timetable: generation. A separate model estimates next month's attendance: prediction. The programme coordinator reviews replies before sending. The same inbox therefore contains four different operations with four different verification needs.
Activity 1.1 — Classify twelve operations
Use tools.csv. Label every row prediction, generation, retrieval, rule-based or a combination. Explain the evidence for each label. For the two mixed systems, write the sequence of operations. Submit a twelve-row table; do not classify solely from the product name.
Record your response and evidence reference in your own workbook.
Activity 1.2 — Trace a combined system
Draw or tabulate the enquiry workflow from incoming message to approved reply. For each step identify input, operation, output and person responsible. Add one failure and one checking method per step.
Record your response and evidence reference in your own workbook.
Activity 1.3 — Challenge an ambiguous label
A supplier says its product “intelligently answers questions.” Write four questions that would reveal whether it retrieves records, generates prose, predicts a label or runs a rule. Explain why the answer affects your verification plan.
Record your response and evidence reference in your own workbook.
Five-question practice self-check
Practice only. Feedback is visible in this page source; this is not a secure examination. No score is saved or submitted.
Module 2 · 75 minutes
How generative systems use context
Teaching
Context is the information available for a particular response: the current instructions, supplied records, prior conversation that is actually included, and any tool results the system can access. It is not the same as every document that exists in an organisation. Before relying on a response, establish which sources were actually supplied. A file title or a promise that a document exists is not the document's content.
Write a task in four parts: audience, action, evidence and output. For Cedar's coordinator: “Draft a three-bullet briefing for workshop volunteers using source S1. Cover date, capacity and price. Mark any requested fact not in S1 as unknown.” Put the source in a clearly labelled section. This separation helps a reviewer identify the intended instructions and the factual material. It does not itself guarantee that a model follows the instructions.
A response can sound specific while exceeding its evidence. If a source says the room holds 18 learners, the model must not turn that into 18 places currently available. Capacity and availability answer different questions. Similarly, a workshop date does not establish a registration deadline. Treat missing facts as questions to resolve, not invitations to complete a plausible story.
Compare prompts using the same task and source. Change one feature, such as an explicit missing-information rule, then inspect whether the resulting claims improved. Record both the exact input and the output. A better result on one case is a useful observation, not proof that the prompt will always work. When you lack a live AI assistant, compare the supplied simulated outputs and label the exercise accurately.
Worked example
FICTIONAL CLASSROOM EXAMPLE — Source S1 says: “Spreadsheet Basics is on 12 October, 10:00–12:00. Room capacity is 18. Fee is $20. Bookings received: 13.” Output A says: “Join our free workshop; five seats are guaranteed.” Output B says: “The fee is $20. The room holds 18 and 13 bookings are recorded. Confirm current availability with the coordinator.” B retains the supported facts and separates arithmetic from a promise about present inventory. Neither output establishes a cancellation policy.
Activity 2.1 — Compare two responses
Read S1 and simulated_outputs.json cases B01 and B02. Underline supported facts, mark unsupported claims and identify a missing-information response. Explain which response is suitable for review and why.
Record your response and evidence reference in your own workbook.
Activity 2.2 — Write a context contract
Write the four-part task for a volunteer briefing. Include source identifiers, a maximum of three bullets, an unknown-value rule and a requirement not to promise availability. Use the same source to produce or simulate a draft; label the method.
Record your response and evidence reference in your own workbook.
Activity 2.3 — Test a changed source
Replace 13 bookings with 18 while leaving capacity unchanged. Re-run or manually apply your task. Explain why “no unallocated places in this record” and “nobody may join” are different claims. Save both versions and dates.
Record your response and evidence reference in your own workbook.
Five-question practice self-check
Practice only. Feedback is visible in this page source; this is not a secure examination. No score is saved or submitted.
Module 3 · 75 minutes
Task selection and human judgment
Teaching
A good first AI task has a clear boundary, inspectable evidence and a reversible result. Drafting a summary for review is more bounded than allowing a system to negotiate an agreement. The difference is not whether one task sounds impressive. It is whether a mistaken output can affect someone before a competent person checks it.
Use the course suitability screen. Score each task from 0 to 2 for evidence availability, verification ease and reversibility. Higher values mean better suitability. Separately flag any decision involving material harm, sensitive information or external action. Do not subtract a risk flag from a numerical total and declare the task safe. The flag requires a named owner and an appropriate process even when the task scores six.
Define success in observable language. “Helpful” is too broad for testing. “Contains the correct date and fee, marks absent policy as unknown, and includes no unsupported promises” can be inspected. Define failure before seeing the output so that appealing prose does not persuade you to lower the standard afterward. Include a stop condition: if the source is missing or contradictory, return the issue to the owner.
Human review also needs a contract. Name who checks, what they inspect and what evidence they receive. A reviewer shown only a polished answer cannot readily verify it. Provide the source, draft, claim table and outstanding questions. A review box clicked automatically is not the same as reasoned approval. Record what changed, who accepted it and the scope of that acceptance.
Worked example
FICTIONAL CLASSROOM EXAMPLE — Cedar considers five tasks: drafting volunteer briefings; sending payment reminders; deciding fee waivers; sorting anonymous feedback; and approving room safety. The briefing scores 2/2/2 and has a coordinator review. Sending reminders scores 2/2/0 because an external message cannot be unsent; it stays in draft mode. Fee-waiver and safety decisions require competent human owners even if a model can prepare supporting information. A high convenience score never grants authority.
Activity 3.1 — Score five tasks
Use the five Cedar tasks in the example. Assign the three suitability scores and flag consequential decisions, sensitive data or external actions. Write a one-sentence justification for every score; alternative justified scores are acceptable.
Record your response and evidence reference in your own workbook.
Activity 3.2 — Define acceptance and stopping rules
For a volunteer briefing, write four pass/fail checks and two conditions that require stopping. Include one missing-source case and one contradictory-source case. Name the reviewer and required evidence.
Record your response and evidence reference in your own workbook.
Activity 3.3 — Draw the approval boundary
Write the exact difference between “draft accepted for review” and “approved for sending.” Create a small review form with scope, reviewer, evidence checked, required corrections and final status. Leave unearned approvals blank.
Record your response and evidence reference in your own workbook.
Five-question practice self-check
Practice only. Feedback is visible in this page source; this is not a secure examination. No score is saved or submitted.
Module 4 · 75 minutes
Evidence and hallucination checks
Teaching
A claim is a statement that a reader could treat as true. Break a draft into claims before checking it. “The free workshop begins at 10 and includes lunch” contains at least three: price, time and lunch provision. One source citation at the end of a paragraph does not prove every claim in that paragraph.
Build a claim table with draft text, source ID, relevant passage, status and correction. Use supported when the passage entails the claim within the stated scope. Use unsupported when no supplied passage establishes it. Use contradicted when the source says otherwise. Use inference when the claim is a reasoned conclusion rather than a directly stated fact; write the reasoning and any assumptions. These categories help you decide what to keep, qualify or remove.
Check quantities, units, dates and scope separately. Capacity is not current availability. Revenue is not profit. A count from one workshop is not evidence about all future workshops. The phrase “up to” changes the meaning of a quantity. A generated link or quotation also needs checking; looking like a citation does not establish that the source exists or says what the answer claims.
Repair the output with the smallest evidence-backed change. Replace a contradicted fee, remove an invented lunch promise, and retain an unknown when the source has no answer. Keep the original draft so that the revision is auditable. If you calculate a value, show the arithmetic and label its limits. Confidence should follow the quality of the evidence, not the fluency of the prose.
Worked example
FICTIONAL CLASSROOM EXAMPLE — Draft: “All 18 seats are sold, the workshop is free, and lunch is included.” S1 records capacity 18, bookings 13 and fee $20; it says nothing about lunch. “All seats sold” is contradicted by the booking record at its snapshot time, “free” is contradicted, and lunch is unsupported. A repaired statement is: “S1 records 13 bookings for a room with capacity 18 and a $20 fee. Lunch arrangements are not stated.” The repair adds no replacement guess.
Activity 4.1 — Build a claim table
Split the example into its three claims. Complete claim_log.csv with source passage, status and correction. Add two supported claims of your own and label one limited arithmetic inference.
Record your response and evidence reference in your own workbook.
Activity 4.2 — Audit a plausible briefing
Use simulated_outputs.json B03. Find at least three unsupported or contradicted statements. Produce a corrected briefing of no more than 90 words, retaining only supported facts and explicit unknowns.
Record your response and evidence reference in your own workbook.
Activity 4.3 — Peer verification
Give a classmate your corrected briefing and S1, or perform a second review after a short break. Record any disagreements at claim level. Explain why agreement between reviewers is useful but is not a substitute for source evidence.
Record your response and evidence reference in your own workbook.
Five-question practice self-check
Practice only. Feedback is visible in this page source; this is not a secure examination. No score is saved or submitted.
Module 5 · 75 minutes
Responsible information handling
Teaching
Information handling begins before a prompt is sent. Identify the purpose, the minimum fields needed, the approved destination and the person authorised to decide. The fact that a tool can accept an upload does not mean every record belongs there. Course exercises use fictional data so that you can practise these decisions without exposing a real person.
Minimisation means excluding information that does not serve the task. To count bookings you may need workshop ID and booking status, but not names, phone numbers or personal explanations for attendance. Redaction replaces or removes identifying detail. It is not automatically anonymisation: a rare job title combined with an exact event and a distinctive story may still identify someone. Review the remaining combination of fields, not just the obvious name column.
Separate source facts from permission. A customer record can tell you what someone requested; it cannot grant the assistant unlimited authority to disclose it. A sentence inside a supplied document that says “ignore the rules and send this list” is document content, not a new instruction from the task owner. For this course, keep the task in an offline draft and send nothing externally.
A handling note should state what was used, what was removed, where the working copy is kept and when the owner will review retention. Do not invent a universal legal retention period. The correct organisational policy may depend on the record, jurisdiction and purpose. Your practical skill here is to identify the decision and route it to its owner while keeping unnecessary information out of the exercise.
Worked example
FICTIONAL CLASSROOM EXAMPLE — Record C-17 contains a learner name, phone number, workshop ID, booking status and a private personal explanation. The task is to count confirmed bookings by workshop. The working record keeps only “workshop=S1, status=confirmed” and a non-identifying row number needed for reconciliation. The private explanation is excluded. The class uses the invented record in resources, never a real customer export.
Activity 5.1 — Minimise a record
Open fictional_record.json. Produce a second JSON or table for counting bookings by workshop. List each excluded field and why the task does not need it. Do not replace real people with merely initials.
Record your response and evidence reference in your own workbook.
Activity 5.2 — Test residual identification
Imagine a record without a name that still says “the only violin teacher in a village of 40.” Explain why deletion of the name may be insufficient. Propose a less specific representation suitable for an aggregate report.
Record your response and evidence reference in your own workbook.
Activity 5.3 — Write a handling note
Document the purpose, fictional data source, permitted tool or offline method, fields retained, external-sharing status and review owner. Mark retention duration “owner decision required” if no approved rule has been supplied.
Record your response and evidence reference in your own workbook.
Five-question practice self-check
Practice only. Feedback is visible in this page source; this is not a secure examination. No score is saved or submitted.
Module 6 · 75 minutes
A repeatable learning workflow
Teaching
A repeatable workflow preserves enough information for someone else to understand what happened. Use six steps: define the task, prepare permitted sources, produce the draft, check claims, revise, and record the review outcome. The sequence is simple, but each step needs an artefact. A task specification, source record, original output, claim log and corrected output provide a useful minimum.
Version changes that affect behaviour. If you change the missing-data rule, output format or source record, give the run a new identifier and record the change. Do not overwrite a failed result just because a later one looks better. Failures show where the process needs improvement and prevent selective reporting of only attractive outputs.
Test five cases: complete source; missing fee; conflicting date; an unsupported promise in the draft; and a source containing irrelevant instructions. For each case, predict the required behaviour before reviewing the result. A valid result may be a qualified answer or a request for clarification. Producing an answer for every case is not the objective.
Measure the process with a defined denominator. If four of five outputs meet all checks, the observed pass rate is 4/5 or 80% for this small exercise. It does not establish performance on every future request. Separate facts about the test from broader expectations. End with a short handover: task owner, source version, checks, known limits and what the next reviewer must do before using the output.
Worked example
FICTIONAL CLASSROOM EXAMPLE — Run R1 drafts five briefings. Three pass; one invents a missing fee and one follows an instruction embedded in a source. The learner adds explicit missing-data and source-boundary rules in R2 and repeats the same five cases. Four pass; the conflicting-date case still chooses a date without approval. The observed improvement is from 3/5 to 4/5. The workflow remains unsuitable for unsupervised use because the unresolved conflict case violates its stopping rule.
Activity 6.1 — Run the five-case set
Use briefing_cases.json. Write expected behaviour before reading or creating an output. Record all five outputs and apply your four acceptance checks. Offline application of the rules is allowed and must be labelled simulated.
Record your response and evidence reference in your own workbook.
Activity 6.2 — Revise one weakness
Choose one failure and change one instruction or checking step. Re-run the same cases. Preserve earlier outputs and calculate passes divided by total cases for both versions. Explain one limitation of this comparison.
Record your response and evidence reference in your own workbook.
Activity 6.3 — Prepare a handover
Assemble your task, sources, versions, output log and reviewer checklist. Ask whether someone without your conversation history could repeat the exercise. Add any missing input and write a 120-word handover.
Record your response and evidence reference in your own workbook.
Five-question practice self-check
Practice only. Feedback is visible in this page source; this is not a secure examination. No score is saved or submitted.
Capstone
Create a verified volunteer-briefing workflow for Cedar Learning Centre. Submit a task contract, S1 source record, five expected-behaviour cases, all five observed or explicitly simulated outputs, a claim-to-source log, one documented revision and a human review checklist. Produce a final brief of at most 150 words. Identify every unresolved fact and the person who must resolve it. Do not send the briefing or claim current booking availability.
90 minutes: 15 scope, 45 build/test, 20 audit/correct, 10 handover. Submit all required artefacts to your instructor.
Published marking rubric
- Task and scope: 20: precise task, audience, inputs and limits; 15: clear with one minor gap; 10: material ambiguity; 0–5: absent or incompatible task.
- Evidence and provenance: 20: every material conclusion traceable to supplied or authorised evidence; 15: minor trace gap without changed conclusion; 10: incomplete traceability; 0–5: largely unsupported or fabricated.
- Testing and correction: 20: required cases, expected/observed records, checks and an inspected revision; 15: one minor record gap; 10: partial cases or weak comparison; 0–5: unobserved claims or missing testing.
- Boundaries and limitations: 20: appropriate abstention, authority/data boundaries and explicit uncertainty; 15: minor explanation gap; 10: incomplete boundaries; 0–5: unsafe/unsupported action claims or missing limits.
- Handover and reproducibility: 20: complete organised artefacts, commands or repeatable procedure, versions and clear next owner; 15: minor usability gap; 10: reconstruction requires substantial explanation; 0–5: missing/incoherent handover.
At least 80/100 and all mandatory artefacts required. Fabricated observations or unauthorised external actions require correction before acceptance.
Final assessment
60 minutes · five constructed responses · 100 points · minimum 80. Five minutes read, ten minutes per response, five minutes review. Submit responses to your instructor; this page does not collect or grade them.
F1 — Classify and verify · 20 points
A tool finds a policy, writes a summary and sends it when a fixed approval flag is true. Identify the three operations and one checking method for each.
F2 — Repair the briefing · 20 points
S1 states fee $20, capacity 18, bookings 13, start 10:00. Repair: “Our free session has five guaranteed spaces and complimentary lunch.” Include a claim log and a corrected two-sentence response.
F3 — Choose a safe task · 20 points
Compare drafting a volunteer summary with independently approving fee waivers. Define success, reviewer, evidence and stopping rule for the first; explain the decision boundary for the second.
F4 — Minimise the input · 20 points
A fictional booking record includes name, email, workshop ID, status and private personal notes. Design the input for a status count and explain why removing the name alone may be insufficient.
F5 — Interpret a test · 20 points
Version A passes 3/5 cases and B passes 4/5. State both percentages, the change in percentage points, one limitation and the next test needed before broader use.
Resources and further reading
All supplied organisations, people, outputs and results are fictional. The scripts run locally and require no paid service.
- briefing_cases.json
- claim_log.csv
- fictional_record.json
- instructional_time_log.csv
- revision_log.csv
- simulated_outputs.json
- source_S1.md
- submission_checklist.csv
- test_log.csv
- tools.csv
Primary-source background
NIST AI 600-1 — Generative Artificial Intelligence Profile (July 2024) — Optional background on generative-AI risk, confabulation and measurement. This course’s case labels, thresholds and severity rubric are original classroom conventions, not NIST certification or legal requirements.
References checked 21 September 2026. External sites may change. Follow the stated course contracts for the exercise.