Route-based accessibility QA for interactive film: a matrix for input, media, and state
Build a risk-based accessibility test matrix that combines representative routes, input methods, media alternatives, five locales, and state recovery without an unmanageable case count.
Introduction
A branching work cannot run each accessibility check once, and it cannot test the full Cartesian product of every variable. Use a risk-driven route matrix: cover every unique rule, then add the main, shortest, resume, no-input, and preference-switch paths.
Interactive film accessibility cannot be reduced to a subtitle check before launch. The choice interface, playback controls, branch state, and media content form one complete journey. If any layer cannot be perceived, operated, or understood, a viewer can lose the means to continue. WCAG 2.2 defines the principles of perceivable, operable, understandable, and robust web content. This article translates relevant guidance into a production method for interactive film. It does not claim that following one article proves conformance; teams still need to confirm platform, market, and user-testing requirements.
Write an experience promise that can be accepted or rejected
The Letter from Fog Harbor has twelve nodes, three input methods, five locales, and several preferences. Every combination explodes quickly, while a Chinese mouse-only main route misses remote focus, Japanese wrapping, description permissions, and old-save migration.
The examples use a fictional short film, The Letter from Fog Harbor. A viewer investigates a swapped letter at night and chooses among a warehouse, a waiting room, and a lighthouse. All people, routes, test records, and asset IDs are original teaching material. They are not real user research, completed test results, or automatic DramaFork features.
Start with one experience promise and one failure condition. The promise names who must complete which task, on what device, and under what condition. The failure condition identifies the point at which release must stop. “A keyboard-only viewer can complete every choice” is testable; “keyboard friendly” is not. Deduplicate by node rule, not number of endings. Nodes sharing a choice component can use a representative, but unique timing, dark scenes, convergence, backtracking, and endings each need a case. Record locale, input, preference, entry state, expected feedback, and exit state.
Connect requirements, implementation, and evidence
| Field | Question | Evidence |
|---|---|---|
| Viewer task | What must the viewer perceive, decide, or control? | A concise task and context |
| Interaction rule | Does the default and alternative path reach the same result? | Input, focus, timing, or presentation rule |
| Branch effect | Do settings, backtracking, and re-entry preserve story state? | Entry, exit, and retained-state record |
| Acceptance | Who tests which device, and where does a failure return? | Route ID, action log, and owner |
The core set includes keyboard main route, remote shortest route, switch-scan no-input route, reduced-motion lighthouse route, described warehouse route, five-locale text boundary route, and old-save resume route. After a high-risk fix, retest affected upstream and downstream nodes.
All four fields must refer to the same version. Updating a mockup without the route record makes it impossible to tell whether an old defect is fixed. Updating code without the experience promise turns “the control responds” into a false claim that the viewer understood it. Editorial, interaction, engineering, and QA are accepting a repeatable journey, not a screenshot.
Put the requirement into the production workflow
First, inventory the main route, short route, backtrack route, and no-input route. Mark every choice, playback control, and essential piece of information. Second, give each task a default and an alternative method that reach the same narrative result. Third, bind the rule to stable scene IDs, asset IDs, and state fields so that copy edits do not destroy traceability.
Fourth, test cheaply before polishing: complete a route by keyboard, at enlarged text size, with motion reduced, and after pausing and returning. Fifth, run cross-device and assistive-technology checks while recording steps, expected results, actual results, and evidence. Generate the matrix from a rule inventory, then let editorial, accessibility, and engineering remove true duplicates together. Retain screen evidence, state logs, and exact steps. Automation handles stable repetition; humans judge understanding, sound, order, and narrative leakage.
A useful record does not say “looks fine.” It says: “Enter at H02, enable the alternative setting, choose the left action, verify clue_letter=true, and confirm that H03 acknowledges it.” That record covers entry, action, feedback, and exit. When it fails, the team can locate the fault in input, presentation, or narrative state.
Most failures occur at system boundaries
Sampling by ending count misses unique mid-route controls. Pass/fail without steps cannot be reproduced. Retesting only the repaired node misses downstream state damage. Coverage should name the rules covered instead of hiding gaps behind one percentage.
Accessibility controls also fail when they are buried in a settings page that an affected viewer cannot reach. Provide a discoverable entry at launch, at the first choice, and in the pause menu. A changed preference needs immediate perceivable feedback and must remain active at the next node. If it disappears after reload, fix it or disclose the limitation before release.
Automation can flag missing names, some contrast problems, and some focus defects. It cannot decide whether choice copy is understandable, whether audio description spoils another branch, or whether pausing breaks narrative comprehension. Route testing and participation by disabled users remain essential. Return content defects to editorial, media defects to sound or post-production, state defects to engineering, and scope conflicts to the accountable owner.
Release gates must cover the complete process
Every unique interaction rule, media alternative, locale boundary, and state migration has an end-to-end case; no release-blocking defect remains; every known limitation has an owner and disclosure; and failures reproduce from entry to incorrect exit.
The minimum release packet contains the feature scope, route inventory, preference defaults, interface copy in all five locales, devices and input methods tested, known limitations, defect owners, and one end-to-end record from opening to ending. W3C sources provide standards and explanatory guidance. The tables, node labels, and fictional case here are an editorial method, not a mandated W3C template.
A matrix reduces repetition but does not replace exploratory testing or research with disabled users. When a component, player, locale, or route changes, recalculate risk and representatives instead of inheriting an old green result.
For the final check, ask someone outside the production team to reproduce the experience from the record. If they still need to ask what to press, how much time remains, what a cue means, or where a resumed session begins, the handoff is incomplete. Write the answer back into the rule and its evidence before publishing.
Sources: W3C WCAG 2.2; W3C WAI: Making Audio and Video Media Accessible.
