AI-assisted vocational lessons and demonstrations

Learn in your browser · Chapter 01 of 08

A practical lesson succeeds when learners can perform and explain the intended task. A neat worksheet or correctly packed example may conceal guessing, copying or excessive help. Begin by specifying the performance you want to observe, then decide how AI can assist your preparation.

Merefield Adult Skills Centre is fictional. Its tutors are preparing an introductory workshop using pre-cut rectangular cards, shallow trays, blunt rulers and invented order cards. Learners inspect and pack sample kits. There are no powered tools, electrical connections, blades or real customer records. The setting still needs ordinary local checks for accessible seating, clear walkways and suitable materials. This teaching example does not establish an occupational standard or authorise work with hazardous equipment.

The course-authored procedure is revision R3. Kit A requires two blue cards and one cream card. Kit B requires one blue card and two cream cards. Every card must have a recorded width from 48 through 52 millimetres, including both endpoints. These invented requirements provide a stable reference against which a tutor can check explanations, demonstrations and learner reasoning.

R3 applies HOLD first when the revision is missing or is not R3, the order type is unknown, a card is unreadable or torn, a width is missing or non-numeric, or the quantity cannot be established. Only complete, readable R3 records proceed to the matching checks. Wrong colour quantities or any width outside the inclusive range produce REWORK. Otherwise the result is ACCEPT. A torn card measuring 47 mm therefore produces HOLD, not REWORK: the earlier condition takes precedence.

Turn that procedure into four observable outcomes. A learner identifies the order type and revision; establishes whether the information and cards are usable; compares the quantities and widths with the appropriate requirements; and explains the resulting disposition. Each outcome needs an action you can see or hear. Replace 'understands quality' with 'identifies the mismatch and points to the relevant R3 rule'.

Specify the conditions of performance as carefully as the outcome. For the first attempt, the learner may use the reference sheet and ask procedural questions. During a later independent check, the tutor presents a different invented order and records any assistance. Independence does not mean forbidding legitimate reference use. It means stating what support is allowed and distinguishing the learner's decision from a decision supplied by someone else.

A finished product cannot reveal every important step. An apparently correct Kit A could have been copied from a neighbour without checking its revision. Conversely, a learner who correctly holds an unreadable order has demonstrated useful judgement even though no kit is completed. Build the observation sheet around decisions, evidence and explanations, rather than rewarding speed or the number of trays marked ACCEPT.

Before introducing AI, write a plain lesson brief: audience, four outcomes, R3 reference, available materials, permitted assistance and expected evidence. An AI tool can suggest alternative wording or practice cases from that brief. It must not quietly change the procedure or manufacture evidence of learner performance. The Department for Education's Generative artificial intelligence (AI) in education guidance, updated 12 August 2025 and applicable to England, retains professional responsibility for the accuracy and appropriateness of generated material.

Learn in your browser · Chapter 02 of 08

A useful teaching explanation makes a procedure easier to understand without changing its meaning. Treat AI-generated text as a proposed adaptation. Check every requirement, decision and example against the reference before it becomes part of a demonstration or learner resource.

Create a compact reference map for R3. One column identifies the requirement, another records its exact teaching meaning, and a third shows where it appears in the lesson. Keep the original course-authored procedure alongside the map. A summary is a derivative teaching aid, not a replacement authority. If the tutor changes a rule deliberately, that requires an explicit new procedure revision and corresponding changes to examples and checks.

Separate three kinds of content. Requirements define what must happen: Kit B needs one blue and two cream cards. Explanations describe why the teaching task uses a rule: checking the order type prevents applying the wrong quantity pattern. Presentation choices describe how to teach it: place a Kit A example beside a Kit B example. AI may propose explanations and presentation choices, but none of them may silently redefine the requirements.

The order of decisions carries meaning. A width of 47 mm normally leads to REWORK after the information checks have passed. If the same record has a missing revision, the result is HOLD. A draft that tells learners to measure first and immediately select REWORK can therefore teach the wrong outcome. Preserve the HOLD-first logic even when rearranging paragraphs to improve readability.

Use contrasting records to test the explanation. Record One is a complete R3 Kit A with two blue cards, one cream card and widths of 48, 50 and 52 mm: ACCEPT. Record Two is otherwise identical but has a width of 53 mm: REWORK. Record Three has that same width and an unreadable card: HOLD. These examples distinguish a complete mismatch from information that cannot support the ordinary comparison.

Give an AI assistant a bounded preparation request: Using only the supplied R3 procedure, draft a learner explanation and three practice cases. Preserve the decision order, quantities and inclusive limits. List any uncertainty separately. Return each proposed case with its inputs, result and rule-based explanation. Supplying a source and an output structure makes checking easier; it does not make the output reliable by itself.

Role: resource-drafting assistant for a fictional adult tutor.
Reference: supplied R3 procedure only.
Task: propose an explanation and three contrasting practice cases.
Constraints: preserve counts, inclusive 48..52 and HOLD precedence;
no real learner data, equipment procedures or invented requirements.
Process: map each proposed claim to R3; list unsupported requests.
Verification: tutor independently traces complete and mixed-fault cases.
Output: case inputs | proposed result | supporting rule | uncertainty.
Failure: omit unsupported instructions and ask for a reference decision.

An average request such as make an engaging workshop leaves the task, permitted sources and failure conditions unspecified. The structured request makes the proposal inspectable. It still needs a real review of both the instructions and their examples. Do not treat a model declaring its own answer verified as independent evidence.

Review generated material in two passes. First check meaning: revision, order type, counts, numeric widths, boundary values and precedence. Then check teaching usefulness: language, unnecessary detail, ambiguity and answer leakage. Recalculate every example yourself. Ask a colleague to trace a small set of cases using only the proposed learner sheet. If they need an unwritten assumption, improve the sheet before the lesson.

Keep a correction record that is small enough to use. Record the draft statement, the relevant R3 requirement, the change and the reviewer. For example, replace 'between 48 and 52, excluding the endpoints' with '48 through 52 mm inclusive'. Do not retain real learner submissions merely to demonstrate that a prompt worked. This preparation task needs only invented records and material the centre is permitted to use.

The ASQA Assessment Practice Guide under the 2025 Standards requires relevant NVR registered training organisations to review assessment tools before use and align them with the training product. Its examples warn against generic, uncontextualised tools. That supports checking and contextualising resources; it does not mean this fictional R3 exercise is an accredited training product or that a reviewed prompt satisfies Australian assessment requirements.

Illustration description: This original illustration is for adult tutor planning, not a photograph or an observed class. The reference states Kit A needs two blue cards and one cream, Kit B one blue and two cream. Every width must be 48 through 52 mm inclusive. Three panels all state R3, Kit A and quantity 3. Panel 1 has blue 48 mm, blue 50 mm and cream 52 mm, all readable and intact: ACCEPT. Panel 2 changes blue 50 to 53 mm while all information remains readable and intact: REWORK. Panel 3 retains those 53 mm inputs but marks one card unreadable: HOLD first. The unavailable information must be resolved before ordinary comparisons. The final line reminds the tutor that a correctly held order may show reasoning, while a finished kit alone does not reveal the learner decision or assistance.

Learn in your browser · Chapter 03 of 08

A demonstration combines physical actions, spoken explanation and learner attention. Those elements can support one another or compete. Design what learners will watch, what they will hear and when they will respond, then rehearse the sequence from their viewing position.

Begin with the end task in view. Show an invented order card and the sample tray, then explain what the learner will eventually decide. Avoid revealing the result before learners know what evidence matters. For the first demonstration, choose a complete R3 Kit A with matching quantities and widths of 48, 50 and 52 mm. A straightforward case lets the tutor establish the routine before adding exceptions.

Prepare the physical layout deliberately. Keep the reference sheet, order card, unchecked cards and checked cards in distinct positions. Choose a viewing angle from which the relevant markings can be seen. A tutor's hands can hide the feature being discussed, and a projected close-up may reverse or crop useful information. Rehearse the actual view rather than assuming that a description in the lesson plan makes the demonstration visible.

Use a four-column script: show, say, ask and check. Under show, write the action or item that should be visible. Under say, write the short explanation that accompanies it. Under ask, place a question at a point where the learner can reason from available evidence. Under check, specify the response or action that would justify continuing. The columns prevent a polished monologue from becoming the entire lesson.

Demonstrate the HOLD check before the matching checks. Point to R3, identify the known order type and establish that the cards and measurements are usable. Explain that a known mismatch differs from missing information. Then compare quantities and widths. Pause at 48 mm and ask whether the lower boundary is included. The learner should connect the answer to inclusive, not simply repeat that this tray is expected to pass.

After the straightforward case, change one relevant feature. Keep the order as Kit A but replace a recorded width of 50 mm with 53 mm. Learners should explain REWORK using the changed value. Next present a separate torn-card case and establish HOLD. Do not physically perform a misleading sequence and hope that a verbal warning will undo it. Label deliberately incorrect examples before discussing the error.

Separate watching from recording when both demand the same attention. Give learners a short observation question before a close demonstration, then provide a pause to make notes afterwards. If a learner needs a different view or pace, arrange that without treating the request as evidence of low ability. A printed sequence can support recall, but it should not obscure the particular judgement the learner is being asked to make.

AI can help propose shorter explanations, observation questions or a timing draft. It cannot observe whether the actual room arrangement blocks a learner's view or whether the tutor has demonstrated the wrong card. Record rehearsal findings from direct observation. HSE's Training and competence guidance links appropriate instruction, demonstration and supervision to the actual work context; it does not support treating an AI script as proof that practical instruction is adequate.

Plan a complete lesson rather than filling every minute with presentation. Chapter 8 provides a fifty-minute starting schedule covering orientation, demonstration, supported practice, fresh attempts and feedback. It is a planning example, not a required ratio. Adjust it using the task, access needs and observed progress, while preserving time for learners to practise.

Illustration description: The original teacher-planning illustration has four panels in reading order. SHOW places the R3 order upright with cards and markings visible. SAY uses 48 through 52 millimetres including both endpoints. ASK poses what changes if a card becomes unreadable, without printing its answer in that panel. CHECK requests the reason before the tutor supplies a disposition. A cream band says same input, different access, no answer supplied, with readable print, structured text and spoken description. Three bottom columns distinguish product, explanation and observation: a completed tray does not establish who decided; explanations record reasoning and prompts; observations record actual actions and keep gaps unobserved. On a fresh case, ask for the learner decision before further prompts or feedback. Formal competence decisions and child-facing tutor use remain separate. This is a planning diagram for adults, not evidence that anyone learned, a complete access audit or an approved qualification assessment.

Learn in your browser · Chapter 04 of 08

A useful question distinguishes between plausible ways of thinking. It does more than invite the class to agree with the tutor. Design examples that reveal a particular misunderstanding, then decide what you will teach or ask next when that misunderstanding appears.

Start with a misconception map for the sample-kit task. Possible misunderstandings include treating a boundary value as a failure, using the Kit A quantities for every order, assuming any visible defect must mean REWORK, or accepting a familiar-looking order with an old revision. These are hypotheses for teaching design, not diagnoses of real learners. An incorrect answer can also result from an unclear question, inaccessible presentation or an ordinary slip.

Construct a small sequence of examples that changes one concept at a time. Case A is a complete R3 Kit A with two blue cards, one cream card and widths of 48, 50 and 52 mm: ACCEPT. Case B changes only the order type to Kit B: REWORK, because the quantities no longer match. Case C restores Kit A but changes one width to 53 mm: REWORK. Ask learners to identify the changed evidence, not merely name the outcome.

Now introduce precedence. Case D has the same dimensional mismatch as Case C, but also contains a torn card: HOLD. Case E is otherwise matching but uses revision R2: HOLD. These contrasts test whether learners are applying the decision structure or simply counting defects. A learner who says HOLD because it is worse has not yet explained the rule. Ask which condition is present and where it appears in R3.

Use worked examples to expose reasoning without turning every activity into copying. First show one complete explanation. Next give a partially completed explanation with the decisive comparison missing. Finally present a fresh record with no answer scaffold. Keep the permitted reference sheet available where appropriate. Removing all support at once may test memory for wording rather than the ability to use the procedure.

Write questions that require a discriminating response. Does everyone understand provides little evidence. A complete Kit A includes a card exactly 52 mm wide; which R3 phrase determines whether that width is acceptable focuses the response. Follow it with a changed case: what changes if that card is torn? The pair tests the boundary and then precedence without pretending that one correct answer proves complete understanding.

Build plausible distractors from the misconception map. For the torn 53 mm card, REWORK is plausible because the learner may notice the width but miss precedence. ACCEPT is less plausible, unless the question specifically tests ignoring the defect. Avoid making the correct option conspicuously longer, more qualified or more formal than every alternative. Option shuffling cannot remove a wording cue that remains attached to the correct answer.

An AI assistant can propose variants, but test both the facts and the question's educational purpose. Check that every supplied value is complete, only the intended feature changes, the keyed answer follows R3 and the explanation does not introduce a new rule. Reject questions that can be answered from grammatical clues or that require information not supplied. Keep a short note naming the misconception each retained question is intended to probe.

Use responses to choose a teaching action, not to attach a label to the person. If a learner confuses quantities, compare Kit A and Kit B with cards in view. If they miss precedence, trace the two stages using the mixed-fault example. After feedback, use a fresh case to see whether the explanation transfers. For formal qualification assessment, applicable JCQ AI guidance and awarding-body requirements distinguish supported learning from the candidate's own demonstrated attainment.

Learn in your browser · Chapter 05 of 08

Change how learners receive instructions and show their reasoning when appropriate. Preserve the capability being taught and check that an adjustment does not quietly perform the task for them.

Merefield's original kit sheet uses small grey type, colour alone and a photograph of a ruler. Tutor Noor first states the purpose: learners interpret the approved order, check the sample and explain a disposition. The purpose is not to read tiny print or guess what a photograph shows. An accessible version can remove those barriers while retaining the checking decision. This distinction must be reconsidered for each real training product; our fictional paper exercise is not an adjustment policy for a qualification.

Begin with an access conversation through the institution's appropriate process. Ask what would make the material usable, not for a diagnosis to paste into a chatbot. Keep individual support records in their authorised system. For lesson design, an invented requirement such as 'provide a readable text description of the example' is sufficient. Do not ask AI to classify a learner's disability, predict effort or decide entitlement to an adjustment.

Prepare the base reference as structured, selectable text. Use explicit headings and a consistent reading order: revision, order type, required counts, measurements, then decision. State millimetres beside values. Explain that 48 and 52 are included. Put the HOLD conditions before the REWORK checks so a spoken version does not reverse the rule. Keep the same information in the print sheet, digital text and spoken script, with the R3 identifier on each.

Colour is part of this invented kit specification, but colour alone should not carry the teaching instructions. Each physical sample can also carry the readable word blue or cream and a distinct tactile identifier agreed for the exercise. The label describes the sample; it does not say ACCEPT or REWORK. If matching colour unaided were genuinely the assessed occupational skill, changing that requirement would require the relevant assessment decision. Here we are teaching inspection reasoning, not certifying colour discrimination.

A magnified view or a spoken reading of an already recorded width may make a reasoning example accessible. It is not evidence that the learner independently measured the object. Noor therefore splits the record into 'interprets recorded measurements' and 'demonstrates the specified measuring task'. She uses the first in this course's paper simulation and refers any real practical adjustment to the appropriate assessor. This prevents a supportive adaptation from turning into an unsupported claim of practical competence.

For digital resources, W3C WCAG 2.2 supplies useful criteria, including keyboard operation, text alternatives, readable contrast and reflow. Check the actual page at increased text size and with the keyboard. A diagram needs a text explanation of its relationships, not just an image filename. A narrated demonstration needs accurate captions and a usable transcript. These checks do not establish complete WCAG conformance or resolve physical access to a workshop.

Scroll sideways to see every column.

Worked example: a digital card reads R3, kit A, blue widths 48 and 52, cream width 50, all readable and intact. The accessible learner version states every input in text and asks for a disposition with reasons. ACCEPT remains the correct result. A second version supplies the sentence 'all conditions are met' before the learner responds. That sentence performs a material part of the task and is removed. The tutor keeps a separate answer sheet.

An AI assistant may draft a plain-language version using only that invented card and the R3 rules. The tutor then compares every numeral, condition and decision word with the reference. Readability is not a reason to replace '48 through 52 inclusive' with 'about 50'. Ask a colleague to trace the accessible version without the source beside it and identify anything missing. Record the reviewed version, not a generic promise that AI makes material accessible.

Learn in your browser · Chapter 06 of 08

A copied success tells you less than a fresh attempt with an explanation. Plan how support changes over the lesson and how you will respond to different kinds of error.

Noor has shown the R3 checking sequence. Now she must find out whether learners can use it. She distinguishes three phases: a guided attempt where prompts are expected, a supported retry after feedback, and a fresh attempt where the learner makes the decision before receiving help. These are lesson-design choices for this fictional adult workshop, not a universal timetable or a mandated supervision ratio.

Prepare two sets of cases so the independent task is not the tutor's example with a different name. Set one includes a complete kit B with blue 50 and cream 49 and 51: ACCEPT. Set two includes kit A with blue 48 and 52.2 and cream 50: REWORK once the record checks pass. A third case has a torn order card and width 53: HOLD first. Learners must explain the reason for the precedence, not merely memorise three output words.

During the guided attempt, the tutor can ask which reference applies and what must be checked before measuring. Make that assistance visible in the notes. Do not later describe a prompted answer as independent. The physical demonstration should use safe pre-cut cards and a blunt ruler. The optional code uses recorded widths and cannot observe measurement, posture, attention or understanding.

Consider three fictional responses. Lee puts a kit A with one blue and two cream samples in ACCEPT: the record is complete but the count comparison is wrong, so it needs REWORK. Amira puts width 48 in REWORK because she thinks the lower boundary is excluded. Tomas sees an unreadable order and asks the tutor for a readable reference before deciding. These require different teaching moves. Lee needs an explicit count-to-order comparison; Amira needs a boundary example; Tomas has used an appropriate stop.

Feedback should identify the observed action, its consequence within the exercise and the next attempt. To Amira: 'You excluded 48. R3 includes both endpoints. Place 47.9, 48 and 48.1 against the rule and explain where the decision changes.' This is more actionable than 'be more accurate'. Then provide a fresh upper-boundary example without the answer. If the learner now applies 52 correctly, record that new evidence and the support that preceded it.

A slip and a misconception can look similar in the finished tray. Ask a neutral question before diagnosing: 'Show how you compared this order with the samples.' A correct explanation followed by a transposed count may need a checking habit; repeated exclusion of both endpoints suggests an issue with the meaning of inclusive. This is a tentative teaching interpretation, not a psychological assessment. Keep the response open to new evidence.

AI can propose feedback wording from wholly invented responses. It should not watch a live learner and automatically declare readiness. A useful constrained request includes the reference, the observed fictional action, the desired next attempt and a ban on invented learner characteristics. The tutor checks that the suggestion neither changes R3 nor humiliates the learner. Feedback must allow a question or disagreement and provide a route back to the actual reference.

Scroll sideways to see every column.

The tutor's observation sheet needs space for the case ID, reference revision, actual action, exact prompt if any, learner explanation and next step. Use invented labels in the course workbook. Real learner records belong in the institution's approved system with applicable access and retention controls. A shorthand score without the underlying observation makes later feedback and review harder.

Learn in your browser · Chapter 07 of 08

Use observations, explanations and products for the capabilities they can actually support. Keep teaching feedback, a knowledge quiz and formal qualification decisions visibly separate.

A tidy finished kit is evidence about that kit. It does not show who chose the samples, whether another learner supplied the answer or whether the learner can explain a HOLD. A polished written reflection is likewise not proof of the practical sequence. The tutor should decide what each capability requires before collecting evidence, then record the conditions and support under which it was produced.

For the fictional workshop, Noor's evidence map has four rows: establish the current instruction; compare quantities and measurements; follow and explain the check sequence; and report a discrepancy without guessing. Each row names an observable action and a question. The record includes 'not yet observed' as a legitimate state. Missing evidence should create a further learning or assessment opportunity, not an invented pass or an automatic failure of the person.

This course is informal adult professional learning about teaching design. A real regulated qualification has its own training product, assessment requirements, permitted assistance, authenticity arrangements, assessor responsibilities and appeal process. Check those with the actual awarding organisation or regulator. ASQA's Assessment Practice Guide is relevant to NVR registered training organisations in its stated Australian scope. Its evidence principles do not convert this course's invented rubric into an approved qualification instrument.

For assessments covered by JCQ arrangements, use the applicable current guidance and qualification instructions. An acknowledgement of AI use does not demonstrate a candidate's own achievement. Where a centre permits AI-assisted marking under JCQ arrangements, the responsible human assessor must examine the complete submission and decide its mark independently of the AI output. Sampling an AI-generated mark or approving it without that review is insufficient. The tutor must distinguish what assistance was allowed in preparation from what the learner needs to demonstrate in the assessment itself. The boundaries cannot be guessed from a generic AI policy.

Scroll sideways to see every column.

Worked judgement: Sam's preparation included permitted AI drafting. In the fresh case, the reference was torn and Sam chose ACCEPT without requesting a readable copy. Noor records the case and action, asks Sam to explain the choice and identifies a gap in applying the stop condition. She does not use the fluent reflection to cancel the observed problem. The next learning step is a guided precedence comparison followed by a fresh independent case. Any formal outcome would require the actual assessment process.

An observation rubric can use explicit descriptors: identifies R3 before checking; compares required with observed counts; explains the inclusive limits; applies HOLD before REWORK; and communicates what must be resolved. Avoid labels such as excellent attitude or seems confident as substitutes for evidence. Where the activity did not reveal a descriptor, record it as unobserved. Preserve the learner's opportunity to explain a disputed record through the institution's proper route.

Do not feed real student submissions, voices, photographs or support records into an unapproved model to save marking time. Data ownership and copyright questions are separate from an individual's willingness to try a tool. Sovereign control involves who can access, retain, reuse or change the system and who retains the decision. Use the institution's authorised processes; the course's synthetic example gives no permission to upload real material.

The Mickai knowledge assessment accompanying this course draws twenty questions from a mapped bank. Strictly above 80 percent is required: sixteen correct is 80 percent and fails; seventeen is 85 percent and passes. Written exercises, the teaching pack and peer microteach are ungraded practice. A named certificate records the course knowledge result and issuance. It does not confer a vocational qualification, professional authorisation, verified identity or proof that an organisation complies with law.

Learn in your browser · Chapter 08 of 08

A lesson pack is ready for a bounded trial when its instructions, examples, access routes and evidence checks agree. Rehearsal should reveal where the teaching fails, not just confirm that the slides look finished.

Bring the previous chapters together into a fifty-minute adult workshop at fictional Merefield. The output is a teachable pack, not a chatbot transcript. Include the R3 reference, performance objectives, task analysis, demonstration script, learner materials, separate teacher answers, practice cases, observation sheet and a revision log. The time plan is an initial design assumption. Revise it from the rehearsal rather than claiming that every group learns at the same speed.

Scroll sideways to see every column.

In a peer microteach, one colleague plays the learner and another observes the instruction, not the personality of the tutor. Give the observer three questions: could the learner see the comparison; did each instruction preserve R3; and did the question reveal the intended reasoning? Avoid filming by default. If recording is proposed in a real institution, its permission, access and retention arrangements need separate attention. This fictional course needs no recording or learner upload.

Worked rehearsal: the observer notes that Noor says 'larger than 48' while the printed sheet says '48 through 52 inclusive'. Two peers then reject the lower-boundary sample. The diagnosis is an inconsistent spoken rule, not poor learner motivation. Noor corrects the script, highlights both endpoints in the teacher notes and rehearses the boundary explanation again. A fresh kit B containing 52 is used to see whether the revised instruction transfers beyond the original example.

A second observation is that the learner cannot see the order while Noor holds a sample over it. Noor places the order upright beside the tray, then pauses before moving the sample. This is a specific design repair tied to an observed barrier. The revision log records version, issue, evidence, change, owner and recheck result. A box saying 'human reviewed' without those details is not equivalent evidence.

Estimate preparation effort honestly. In an invented comparison, a template-only pack takes 35 minutes to adapt, 15 to check and 20 to rehearse: 70 minutes. An AI-assisted draft takes 12 minutes to generate, 28 to correct, 15 to adapt access materials and 20 to rehearse: 75 minutes. The AI route is five minutes longer in this example. It might still offer useful variant ideas, but the timing does not support a speed claim. These numbers are a worked calculation, not a Mickai performance study.

The optional downloadable checker makes one part of the reference testable. Read its README, inspect the supplied synthetic cases and run the standard-library tests in an approved local Python environment. No model or network is needed. The paper table and expected decisions provide the same reasoning exercise without code. Change one field at a time, including 48, 52 and an invalid reference, and explain the result before checking the key. A passing test suite says nothing about an individual learner's practical attainment.

Choose the relevant jurisdiction pathway before adapting the pack for a real course. England's DfE material, the selected European Commission educator guidance and Australia's RTO guidance have different scopes. Identify the actual institution, learner age, qualification, regulator and permitted technology use. Where the reference is only an overview, say so. A general AI-literacy discussion or locally hosted model does not supply legal clearance. Under-18 use and safeguarding remain outside this adult prototype.

Mickai's Trust Agent approach keeps the educator responsible for teaching and the institution responsible for its systems. A useful institutional request might be an approved resource-drafting workflow with bounded sources, controlled access and retained human review. That is a design requirement to validate in the actual organisation, not a claim that this course includes a deployed service or makes all AI safe. Keep confidential source collections and real learner data out of these public exercises.