An editorial question defines what the audience needs to understand and which evidence can answer it. Audience influences vocabulary, examples and pathway, but does not justify removing conditions that change the meaning. Define purpose, intended users and a provisional conclusion before opening tools. An AI exhibition can investigate a specific capability and its limits rather than present the technology as one intelligence that works identically in every situation.

Formats have complementary functions. The document develops reasons and sources; the chart compares data; the image explains relationships; audio communicates a synthesis with a transcript; the website organises access; the app allows a decision to be made and understood. Repeating one paragraph in six files does not create integration. Preserve common terms, quantities and limits, and decide the order visitors need. The script connects question, evidence, interpretation and visitor action so every piece makes an identifiable contribution and can be reviewed within the whole.

A primary source directly documents a procedure or result; another explanation may help interpretation without adding independent evidence. Record authorship, date or version, exact claim, conditions and limitation. In machine learning, a model fits patterns using training data and is then evaluated with test cases not used for that fitting. This separation examines generalisation; repeating familiar examples does not establish behaviour on new cases. Repeatedly using the test to decide changes reduces its independence.

The Thinking Game supports questions about objectives, training and results, separating observed scene, interpretation and general claim. The existing block takes four sessions: 0.25 for preparation, 2.25 for authorised viewing or a coherent selection and 1 for checking and transfer. A written alternative provides descriptions and guided reading. Record three observations and check two claims against primary sources; express uncertainty when something remains unverified. The conclusion enters A01 design and is applied in A02, without an additional audiovisual submission or compulsory viewing outside class.

A chart communicates a comparison built from data. To understand it, the source table needs a question, provenance, sample and definition of each column. Original data allow the selection and calculations to be reviewed. In machine-learning results, distinguishing training, test and fictional data clarifies what the cases represent: fitting patterns with some examples and evaluating with others serve different functions. An accuracy percentage relates correct responses to the total number of evaluated cases. Its meaning depends on that denominator and the conditions; it does not express a percentage of intelligence.

Representation can reveal differences or hide them. When there are different classes, an overall result may conceal errors concentrated in one class. The chosen chart should address the comparison and maintain consistency between cell range, formulas, units, axes and scales. A truncated scale may exaggerate visual differences; its use requires clarity and justification. Counting totals twice or replacing missing data with zero without explanation also changes the meaning.

The editable sheet preserves the opportunity to review these steps, and the export allows the result to be incorporated into the exhibition. Labels and a textual description make the message understandable without relying only on colour. The accompanying explanation connects what that sample shows with a bounded conclusion and identifies which generalisations it cannot support.

Each format preserves information in a way that shapes its editing and reading. A raster image represents pixels; a vector image preserves shapes and geometric relationships. Enlarging pixels does not create new detail. An editable file may retain layers, objects or settings lost on export, so the original, editable and final output serve different functions and need related names. Cropping or adding labels may clarify an explanation, provided relevant evidence is preserved.

Accessibility connects these decisions with what the audience needs to understand. Alternative text conveys the image’s function and information in context rather than merely its filename. In a document, heading styles express hierarchy, and descriptive links make destinations recognisable. These are ways of organising meaning: a diagram and its textual explanation should communicate the relationship that matters, even when someone accesses them differently.

Original audio or an authorised alternative communicates the question, idea and limitation in approximately thirty seconds. The script orders the message; level, beginning, end and clarity determine whether it can be understood, avoiding clipping and music that masks the explanation. Vocal participation is voluntary. A transcript preserves spoken information and relevant sounds, a different function from image alternative text. Opening each export at its intended size allows legibility and content to be checked: saving a file does not yet establish that the audience can use it.

Collaboration means coordinating contributions to produce and review a common result. Dividing files without integrating them can create contradictions in data or vocabulary. Rotate research, production and review roles so each member understands one decision and reviews another. The shared environment organises sources, editables, exports and integration. Its structure should make the version actually used in the exhibition and the sources supporting its claims easy to find.

Permissions distinguish editors, reviewers and readers. Test access from the intended role; a signed-in author may open a resource invisible to visitors. History or a record preserves change, contribution and review without making connected minutes an automatic measure of collaboration. Before recovering a version, identify what will be preserved from the current one and what will change. Useful comments specify location, problem and proposal respectfully. Review is demonstrated by its effect on content, accessibility or function, rather than by accumulating messages.

The website organises Home, Evidence and Resources through understandable structure and links. HTML expresses elements such as headings, paragraphs and links; link text should indicate its destination. A relative path is interpreted from the current file’s location: moving folders may break it. An external link adds dependencies on availability and permissions. Test from the submission folder and the reader role, because editor access does not establish visitor access.

In App Inventor, the Designer arranges components and blocks describe event-driven behaviour. Two arrangements can alternate question and explanation views without accounts or sending responses. A button triggers an event that changes text or visibility. Interaction needs a way back and reasoning: CORRECT or ERROR alone does not explain sample, testing or limitation. Check both options and the return to the question. App and website must retain other formats’ numbers and terms. An app running locally or in an emulator must not claim an unimplemented service connection.

Authorship identifies who created a work; a licence or permission determines authorised uses. Finding an image online and citing its origin does not authorise reproduction. Read the specific conditions: attribution requires credit, share-alike conditions adaptations, non-commercial restricts certain uses and no-derivatives limits sharing adaptations. Licences do not all allow the same project. Record author, link, licence and changes, or replace the resource with your own or a compatibly permitted one.

Privacy is reviewed separately: a licence may allow reuse of a file containing unnecessary personal information. Avoid publishing other people’s voices, images or identifiers without appropriate conditions; use fictional materials and authorised alternatives. If an allowed generative tool assists, check accuracy and declare its use according to school agreements; its output is not a primary source for a fact. Publication is intended to be internal. Before sharing, check audience, keyboard navigation, structure, contrast, alternative text and transcript. These are specific checks, not automatic accessibility certification.

Editorial integration preserves a verifiable relationship between sources, data, conclusion and formats. An exhibition may open every file yet communicate a false claim. Check content, function and understanding as different dimensions: a link may open, a chart may correctly represent its table and a visitor may still infer an exaggerated conclusion. Testing needs to ask which evidence and limitation the visitor remembers after the pathway.

A01 brings together audience, script, sources, permissions, collaborative organisation and an initial production: editable piece, export and executed first interaction. A02 completes all six formats and demonstrates reader access, coherence, accessibility and contributions. Keep two audiovisual checks and their application in the script. Another pair follows the exhibition and answers a new question; record difficulties without collecting personal profiles. Improve one part and repeat the check to observe its effect. Each member explains a tool choice and a conceptual decision. These tests belong to the two existing submissions, not additional assessed activities.