11 Transforming Content: Keep the Learning Decision Visible

An interactive page is a delivery format, not evidence that the activity teaches well.

11.1 Start with a task that already works

A paper activity may be effective without becoming an app. Before adding interactivity, identify what the student must decide, what information is supplied and how the response will be checked.

This chapter uses M1 in the pilot pack. Students calculate campaign rates, distinguish an observation from an unsupported profit/causal claim and request useful evidence. That task works on paper. A digital form offers another way to record the same reasoning.

The supplied decision sheet is a small, self-contained HTML example. It has labelled fields, a text-export button and a print route. It does not grade, autosave, submit work to an LMS or send responses to a server. Use fictional information only.

11.2 What to ask AI to help produce

After deciding the activity, a bounded request might be:

Create a self-contained page for this fictional task.
Preserve the supplied facts and learner questions.
Use labelled fields and ordinary keyboard controls.
No external libraries, requests, analytics or login.
Provide a plain-text export and printable alternative.
State clearly whether answers are saved.
Do not add model scoring or invent answer criteria.

A generated file still needs inspection. Do not upload it directly to a live course merely because it opens in your browser.

For readers interested in the development work itself, Build Web, Guide AI covers that route. Here the purpose is evaluating the teaching artifact, not learning to maintain a web application.

11.3 Inspect the supplied example

The M1 sheet displays the complete small dataset and asks for three decisions. It does not hide an answer behind a score or require a model connection.

Test the rate calculation separately: A has a 4% click-through rate; B has 5%. Both have 5% click-to-purchase conversion and 20 purchases. Spend, margin and a comparable experimental design are not supplied. A correct response retains the measurable rate difference while withholding profit and causal claims.

If a form displayed “B wins” merely because its click-through rate is higher, the interaction would teach an unsupported conclusion. Better styling would not repair it.

11.4 Test delivery, saving and access

Open the example using your intended device/browser. Enter fictional responses, export the text and open the saved file. Check that every field is included and that long or non-English text survives. Reloading the page loses the on-page answers; the exported file is the record. There is no automatic import.

Try keyboard-only navigation and the assistive technologies relevant to your students. Check zoom, contrast, labels, mobile layout and printing with long answers. Source-level checks can catch some errors, but are not an accessibility certification or a substitute for target-device testing.

An LMS may block scripts, restrict downloads or display files differently. Use an approved staging area and test its settings. If the form cannot work there, use the printed questions or an institutionally supported form tool under the same data rules.

11.5 From analysis to a deliverable

A model can help format a decision already checked by the learner. For M1, the learner first records the calculations, missing evidence and limits. Only then might a model help turn that decision into a short client memo.

The learner verifies that the memo preserves uncertainty and has not added a profit claim. The final artifact should be accurate as well as clear. Keep the calculation and source note so the decision can be inspected.

For spreadsheets, independently check formulas and totals. For charts, check the underlying values and labels. A file being downloadable does not establish that its analysis is correct.

11.6 Narrow the research claim

A prompt asking a model to identify themes, interpret quotations and write a research memo delegates substantive analysis, not just formatting. Do not describe that as the model merely making a document look professional.

A research activity needs a suitable method, authorised data, traceable coding, source checks and review of alternative interpretations. This chapter does not supply that research curriculum. Use a reviewed synthetic example or appropriate research-methods teaching before asking students to analyse real interview transcripts. The data gate applies throughout.

11.7 Activity: test the response artifact

Allow 15 minutes. Complete M1 using the decision sheet or paper. Check the answer against the pack, then write a short memo preserving the verified rates and limitations.

If using the page, export the responses and check the file. If using paper, ensure another reader can follow the calculation and decision. Both routes are judged on the same reasoning, not the sophistication of the delivery format.

# Transforming Content: Keep the Learning Decision Visible {#sec-transforming} > An interactive page is a delivery format, not evidence that the activity teaches well. ## Start with a task that already works A paper activity may be effective without becoming an app. Before adding interactivity, identify what the student must decide, what information is supplied and how the response will be checked. This chapter uses M1 in the [pilot pack](../appendices/pilot-pack.qmd). Students calculate campaign rates, distinguish an observation from an unsupported profit/causal claim and request useful evidence. That task works on paper. A digital form offers another way to record the same reasoning. The supplied [decision sheet](../tools/decision-sheet.html) is a small, self-contained HTML example. It has labelled fields, a text-export button and a print route. It does not grade, autosave, submit work to an LMS or send responses to a server. Use fictional information only. ## What to ask AI to help produce After deciding the activity, a bounded request might be: ```text Create a self-contained page for this fictional task. Preserve the supplied facts and learner questions. Use labelled fields and ordinary keyboard controls. No external libraries, requests, analytics or login. Provide a plain-text export and printable alternative. State clearly whether answers are saved. Do not add model scoring or invent answer criteria. ``` A generated file still needs inspection. Do not upload it directly to a live course merely because it opens in your browser. For readers interested in the development work itself, *Build Web, Guide AI* covers that route. Here the purpose is evaluating the teaching artifact, not learning to maintain a web application. ## Inspect the supplied example The M1 sheet displays the complete small dataset and asks for three decisions. It does not hide an answer behind a score or require a model connection. Test the rate calculation separately: A has a 4% click-through rate; B has 5%. Both have 5% click-to-purchase conversion and 20 purchases. Spend, margin and a comparable experimental design are not supplied. A correct response retains the measurable rate difference while withholding profit and causal claims. If a form displayed “B wins” merely because its click-through rate is higher, the interaction would teach an unsupported conclusion. Better styling would not repair it. ## Test delivery, saving and access Open the example using your intended device/browser. Enter fictional responses, export the text and open the saved file. Check that every field is included and that long or non-English text survives. Reloading the page loses the on-page answers; the exported file is the record. There is no automatic import. Try keyboard-only navigation and the assistive technologies relevant to your students. Check zoom, contrast, labels, mobile layout and printing with long answers. Source-level checks can catch some errors, but are not an accessibility certification or a substitute for target-device testing. An LMS may block scripts, restrict downloads or display files differently. Use an approved staging area and test its settings. If the form cannot work there, use the printed questions or an institutionally supported form tool under the same data rules. ## From analysis to a deliverable A model can help format a decision already checked by the learner. For M1, the learner first records the calculations, missing evidence and limits. Only then might a model help turn that decision into a short client memo. The learner verifies that the memo preserves uncertainty and has not added a profit claim. The final artifact should be accurate as well as clear. Keep the calculation and source note so the decision can be inspected. For spreadsheets, independently check formulas and totals. For charts, check the underlying values and labels. A file being downloadable does not establish that its analysis is correct. ## Narrow the research claim A prompt asking a model to identify themes, interpret quotations and write a research memo delegates substantive analysis, not just formatting. Do not describe that as the model merely making a document look professional. A research activity needs a suitable method, authorised data, traceable coding, source checks and review of alternative interpretations. This chapter does not supply that research curriculum. Use a reviewed synthetic example or appropriate research-methods teaching before asking students to analyse real interview transcripts. The [data gate](ethics-integrity.qmd) applies throughout. ## Activity: test the response artifact **Allow 15 minutes.** Complete M1 using the decision sheet or paper. Check the answer against the pack, then write a short memo preserving the verified rates and limitations. If using the page, export the responses and check the file. If using paper, ensure another reader can follow the calculation and decision. Both routes are judged on the same reasoning, not the sophistication of the delivery format.