Instructional Visual/Slide Design Pipeline

When a clean slide hides a broken export

A six-slide architecture lesson contains a dense diagram, two screenshots, a comparison table, and speaker notes carrying important explanation. The source editor looks orderly, but the PDF export flattens reading order and one screenshot has no alternative description. Reviewing only the source would miss a real release defect.

This lesson builds Instructional Slide & Visual Production Contract v1. Every slide is bound to an outcome and claims, content model, visual semantics, alt/description, notes, layout, reading order, density, contrast, export state, review state, revision, and provenance. The strongest positive state is not automatic release; when the evidence is complete the ceiling is VISUAL_PRODUCTION_REVIEW_REQUIRED. When source/export evidence or presenter/self-paced semantics diverge, the pipeline must fail closed.

Lesson route and ownership boundary

MKN-012 owns the Instructional Visual/Slide Design Pipeline. MKN-007 is the frozen hard prerequisite and MKN-011 is the soft prerequisite. MKN-011 owns Accessible Multimedia Learning UX and MKN-013 owns Ethical Gamification; this lesson does not absorb either outcome.

The Foundation route traces a small deck from outcome to export QA. Repair fixes only the failed evidence dimension and reruns the affected checks. Experienced learners must explain why a visually polished slide or a PDF that merely opens does not prove instructional completeness.

Checkpoints: V1 purpose/claim trace, V2 semantic alternative plus reading order, V3 transfer from presenter-led deck to self-paced HTML without treating the baseline as an oracle.

Mental model: purpose → content model → visual semantics → source QA → export QA → review

The pipeline is:

learning outcome + claims → slide content model → visual choice → semantics + notes + alternatives → layout/readability QA → export → export accessibility QA → review/revise/release

Keep three boundaries explicit. First, the content model is separate from layout; a template change must not destroy claim structure. Second, visual semantics are separate from appearance; a screenshot or diagram is instructional only when its role and alternative are explicit. Third, source QA is separate from export QA because reading order, text extraction, and alternatives can be lost during conversion.

If any boundary lacks decisive evidence, the pipeline returns a typed HOLD or REVIEW_REQUIRED state.

Foundation contract: nineteen evidence dimensions

The frozen dimensions are deck_id, lesson_outcome_ref, slide_id, slide_purpose, claim_ids, content_model_ref, visual_type, visual_semantics, alt_or_description, speaker_note_ref, layout_template, reading_order, text_density, contrast_state, export_format, export_accessibility_state, review_state, revision, and provenance.

Minimum invariants: each slide has a purpose and claim trace; decorative and instructional visuals are not equivalent; instructional visuals need semantics and a usable alternative; essential meaning cannot remain only in speaker notes unless the delivery context deliberately supports it; reading order is checked in source and export; readability/contrast are evidence rather than reviewer taste; export accessibility is measured on the exported artifact; review binds to an exact revision; provenance links source/export/review; and a delivery-mode change requires recomputation.

Deep explanation: instructional semantics are not rendering

W3C WAI presentation guidance — FACT. WAI recommends consistent slide design, limited and readable slide text, describing relevant visual information, and providing materials in accessible formats. It does not prescribe MAKNOW slide templates or density thresholds.

WCAG 2.2 — FACT. WCAG 2.2 provides technology-independent, testable success criteria for web content. When a slide experience is delivered as HTML, these criteria can inform accessibility QA; WCAG does not decide instructional sequencing or whether a diagram improves learning.

Pipeline separation — INFERENCE. Because semantic meaning can be lost in conversion/export, source review alone is insufficient. Source and exported artifacts need independent evidence.

MAKNOW release contract — DESIGN. Visual grammar, density limits, templates, review state, and release gates are local, versioned design authority.

Worked examples: good source, broken export, and self-paced transfer

Baseline. A six-slide deck has a reasonable source structure. The architecture diagram has semantics and the notes carry presenter context, but PDF reading order is flattened and the second screenshot lacks an alternative description. The state is HOLD_EXPORT_A11Y_REPAIR_REQUIRED.

Wrong approach. The designer removes speaker notes while keeping deliberately sparse slides, even though part of the explanation existed only in the notes. The state is HOLD_SPEAKER_NOTE_DEPENDENCY until the missing meaning is restored.

Transfer. The same sequence becomes self-paced HTML with no live presenter. Information formerly carried by narration/notes must move into visible explanation, expandable notes, transcript, or assistive alternatives. The baseline release state is not an oracle; the transfer is independently recomputed and at best reaches SELF_PACED_REVIEW_REQUIRED.

Failure modes that must fail closed

Fail closed when slide purpose or claim trace is missing; an instructional diagram is replaced by an unlabeled screenshot; semantics exist but alt/description is absent; a sparse slide depends on removed notes; PDF reading order diverges from the source; contrast is FAIL or UNKNOWN; export accessibility is inferred from the source editor; review points to an older revision; provenance does not connect source/export/review; or self-paced transfer copies the presenter-led result.

A HOLD is a valid outcome. Release is not owed to a visually clean artifact.

Alternatives and trade-offs: diagram, screenshot, table, or text-first

A diagram represents relations and flow well, but semantic structure can become trapped in shape/color. A screenshot is fast and realistic, but has weak semantic structure and becomes stale quickly. A table is strong for comparison and audit, but can be dense on small screens or live slides. Text-first is searchable and often robust for accessibility, but may communicate some relationships more slowly.

The right question is not “which visual looks best?” It is “which representation serves the learning outcome, which information must be visible, what alternative is required, and what risks does the target export introduce?” Removing a decorative visual or splitting a dense slide can be the better instructional decision.

Experienced bridge: treat the slide as a view over a versioned content model

In a professional pipeline the slide file is not the sole source of truth. Treat it as a view over a versioned content model: claim and semantic nodes are independent of coordinates; each visual asset has an identifier and purpose; notes and alternatives are sibling evidence; layout is a presentation concern; each export has a revision/hash; review evaluates source and export for the same revision; and accessibility repair should not require rewriting claim structure.

This separation allows PDF, HTML, and presenter modes to share a content model without pretending that their affordances are identical.

Guided and independent practice

Guided. Open the baseline fixture and label which fields come from the source and which come from the export. Repair only the reading-order fault and rerun the gate; then repair the missing alternative description separately. Do not hide multiple defects behind a vague “fix accessibility” step.

Independent. Add a new comparison-table slide. Define its purpose, claims, reading order, density, and export state. Then create a mutation that changes only revision while leaving review_state untouched; the system should detect stale review.

Experienced. Explain which checks can be automated and which—such as instructional purpose and adequacy of the semantic alternative—still require human judgment.

Verification evidence: what must be reproducible?

The package must prove without hidden reasoning that the frozen Bible hash is preserved, all 13 semantic roles exist in both locales, the three-row table covers all 19 dimensions, the lab is deterministic, source/export failures have typed states, freshness claims reference real source IDs, and a reviewer can reproduce the state from visible fixtures.

Machine validation is not proof of pedagogical quality, but it exposes structural drift. The final package therefore combines content, table, lab, source ledger, acceptance record, critical hash lock, and manifest.

Assessment and transfer: presenter-led to self-paced HTML

Assessment has four layers: retrieval checks the lesson boundary, application builds the baseline contract, diagnosis repairs a shortcut, and transfer changes the delivery mode.

In transfer the presenter disappears. speaker_note_ref can no longer represent essential hidden information; required meaning must become visible content or an accessible alternative. The HTML target also requires its own accessibility QA. Copying the baseline state produces HOLD_BASELINE_ORACLE_REUSE.

Passing means explaining which evidence changed because the delivery context changed, not merely exporting the same deck again.

References, freshness, and authority boundaries

The primary external sources are W3C WAI Accessible Presentations and WCAG 2.2. Rechecked on 2026-09-13, WAI continues to recommend consistent design, limited/readable slide content, describing relevant visual information, and providing accessible materials. WCAG 2.2 remains a W3C Recommendation with technology-independent, testable criteria for web content.

These sources define FACT boundaries. Visual grammar, density thresholds, templates, review workflow, and release states remain MAKNOW DESIGN authority. If W3C guidance or target-platform semantics materially change, freshness review creates a new package version; the accepted Golden is never mutated in place.