A manual check answers one question: can a person complete the full route in a specific environment, from the initial state to a confirmed result, including an error and its recovery. It complements automated diagnostics and is carried out on a real interface with real input devices and assistive technologies.
The six basic objects of checking are described in the component cards. The normative basis and the rules for recording defects are set out in the evaluation methodology and in the WCAG 2.2 level A and AA protocols.
What counts as manual evidence
Tabtransitions test keyboard reachability and focus order. They do not test the screen reader virtual cursor, reading of the structure, or announcement of changes.- Synthesised speech does not confirm output to a braille display. The braille route is tested separately on a real display, recording its model, braille table, abbreviations, and connection method.
- The accessibility tree, responsive design mode, the CSS high-contrast class, and emulation of browser settings help to find the cause. They do not replace a run with a real assistive technology and a real system setting.
- An expert run with a screen reader is a manual technical check. It does not become user evidence unless a user takes part in it under a protocol defined in advance.
- The verbatim synthesiser phrase is not set as a universal reference. What is checked is the availability of the name, role, state, relationships, result, and next action.
Run passport
Before the check, the passport is filled in. If any of the significant versions or settings has changed, that is a new run.
| Field | What to record |
|---|---|
| Object | Address, screen, component, product version, or build |
| Initial state | User role, signed in or signed out, data, open panels, error state |
| OS | Name, version, significant updates |
| Browser | Name and full version |
| Assistive technology | Name, version, speech synthesiser or braille display |
| Input | Keyboard, touch screen, mouse, voice, switch, or a combination of them |
| Settings | Language, speech verbosity, navigation mode, zoom, text size, colours, reduce motion, orientation |
| Scope | Task, starting point, test data, expected end result |
| Evidence | Step record, screenshot, video, log, or link to a defect, without personal data |
Results from different combinations of OS, browser, and assistive technology are not merged into a single mark. An untested combination receives the status "not tested" rather than an assumed result.
Recording one step
Every state change is recorded separately.
| Field | Entry |
|---|---|
| Action | One command, gesture, or user input |
| Expected | Observable result and expected focus position |
| Observed | Actual visual, programmatic, and operational result |
| Focus and return | Where the focus moved, how the user continued or returned |
| Notification | What changed without a focus move, and how it was detected |
| Hint or assistance | None; neutral repetition; guiding hint; familiar support; technical assistance |
| Source | Direct link to the criterion, template, or requirement |
| Date | Date of observation, not the date the report is written up |
Step status: "passed", "failed", "not applicable", or "not tested". For a negative result, the point at which the route breaks, the consequences, and the re-check after the fix are stated.
General order
- Record the passport and safe test data.
- Return the interface to the defined initial state.
- Complete the task with one selected interaction method, without substituting the mouse or another method.
- Check the loading, empty result, error, correction, removal of the outdated error, pending, and success states, where these are possible.
- For every action, record the state before, the action, the state after, the focus, the notification, and the assistance.
- Close the temporary layer or complete the operation and check the return to a logical place.
- After a fix, repeat the whole affected route within the same environment passport.
The focus order must preserve meaning and operability according to WCAG 2.2, criterion 2.4.3, and the current focus must be visible according to criterion 2.4.7.
Routes by interaction method
Screen reader
At least one desktop or mobile run is carried out in a combination that falls within the declared support scope.
Desktop route:
- Start a real screen reader and record the version, browser, synthesiser, language, and speech verbosity.
- Open the page from the beginning. Check the heading, the landmarks, the section headings, the links, and the form elements with the structural navigation tools.
- Find the component with the virtual cursor or the browse mode, not with the
Tabkey alone. - Switch to the interaction mode, perform the action, and check the name, role, state, and relationships before and after it.
- Raise an error, go to the message and the field, correct the value, make sure the outdated error is no longer announced, and then confirm success.
- Check the return of focus after closing the layer or completing the step, and the ability to continue the route.
Mobile route:
- Record the device, the OS, the browser or the built-in view, the screen reader, the language, and the gesture settings.
- Repeat the route with successive swipes, and separately explore it by touch.
- Check that visually available items are reachable by both means, that the order is understandable, and that the activation gesture performs the stated action.
- Go through the error, the correction, the disappearance of the error, the notification, and the return, without turning the screen reader off.
The name, role, state, and value are assessed against WCAG 2.2, criterion 4.1.2. Messages that appear without a focus move are checked against criterion 4.1.3.
Braille route: if braille support is part of the scope, the key scenario is repeated with a real display. Lines and abbreviations, cursor position, state change, error message, and confirmation of the result are recorded separately. A speech check does not close this route.
Keyboard without a mouse
- Remove the mouse and touch input from the route.
- Start before the first component and complete the task with the
Tab,Shift+Tab,Enter,Space,Escapekeys and the arrow keys, where these are provided for by the component model. - For every step, record visibility, order, and focus position. A step cannot be marked as passed if the action is available only through a bypass element.
- Open and close all temporary layers, raise an error, correct it, and complete the task.
- Check the reverse route with
Shift+Tab, the exit from a composite component, and the return of focus after closing or removing an item.
Basis of the check: WCAG 2.2, criterion 2.1.1 "Keyboard". A keyboard check does not confirm that a screen reader works.
Enlargement and screen magnification
Two different mechanisms are checked.
Browser zoom and reflow:
- Record the window size and the zoom level. Check text enlargement to 200 percent against criterion 1.4.4.
- Check the presentation at a width of 320 CSS pixels or an equivalent width, without loss of content or actions, against criterion 1.4.10.
- Go through the full route with the keyboard: open the component, raise an error, correct it, wait for the notification, and close the layer.
- Check that the focus, the error, and the new result do not end up outside the visible area without a clear way to find them.
Real screen magnification:
- Switch on the system or a separate magnifier and record its version, magnification factor, window or full screen mode, focus following, and pointer behaviour.
- Go through the same route with the usual input method.
- Check the reading of long labels, the movement of the viewport, the stability of the focus position, and the detection of messages outside the current screen fragment.
Browser zoom and screen magnification give different evidence. One run does not replace the other.
Voice control
- Switch on the real system voice control and record the OS, version, language, microphone, and the enabled overlays of names, numbers, or the grid.
- Find the action by its visible label and pronounce it. The accessible name must contain the visible text according to WCAG 2.2, criterion 2.5.3.
- If labels repeat, check the selection by the overlay or another accessible refinement.
- Go through the input, the error, the correction, the removal of the error, the confirmation, and the return, without a mouse.
- Separately record the commands that required a number, the grid, or a manual workaround.
For real commands, Voice Control in macOS or Voice Access in Android is used in the declared environment. Dictation enters text, but by itself it does not test interface control. Screen reader speech is not voice control either.
Forced colours and high contrast
- Switch on forced colours or high contrast in the OS settings and record the theme, the palette, the OS, and the browser.
- Reload the page and make sure the browser has actually applied the system mode.
- Go through the full route and check text, links, pictograms, field borders, visible focus, and the selected, expanded, disabled, and error states.
- Make sure the meaning is not conveyed by background, shadow, or colour alone, and that background images do not hide the needed content.
- Correct the error and check that both the visual and the programmatic marking of the error has disappeared.
The system palette may replace author colours, borders, shadows, and part of the images. This behaviour is described by forced-colors. CSS emulation without the system mode remains a diagnostic and is not the result of this route.
Reduced motion
- Record the OS, the browser, and the initial motion setting.
- Go through the scenario in the normal mode and note automatic movement, blinking, parallax, transitions, and focus movement.
- Switch on the system reduce motion setting, restart the browser or reload the page if necessary, and then repeat the same scenario.
- Check that non-essential motion is removed or reduced, and that the content, the state, the focus, and the confirmation of the result are preserved.
- Separately check the possibility to stop moving or automatically updating content according to criterion 2.2.2.
The site receives the system setting through prefers-reduced-motion. Media query emulation helps the developer, but does not replace a run with the real OS setting. Reduced motion also does not replace a separate check of flashes and of media control.
Six components
Each route below is repeated with the selected interaction methods. The keyboard and state models are checked against the WAI-ARIA Authoring Practices Guide, but the result is determined by the actual behaviour of the product.
C-01. Button
- Record the enabled, disabled, pressed, or pending state before the action.
- Find the button with the selected method. Check the visible label, the accessible name, the role, and the state.
- Activate it with the standard action. Check exactly one expected result, the new state, and the focus position.
- For a long-running action, check the pending state, protection against an accidental repeat, the error, the retry, and success.
- For a toggle, repeat the action and confirm the reverse change of the visual and the programmatic state.
C-02. Form with validation
- Start with an empty form. Find the labels, the required state, the format, and the instructions before entering anything.
- Fill in several fields and submit a knowingly wrong value.
- Check the summary or the message, the link between the error and the field, the focus position, and the preservation of the remaining data.
- Return to the field, correct the value, and make sure the outdated visual and programmatic error states are removed.
- Submit the form again, check the pending state, the service error or success, and the ability to continue. Form notifications are checked against the W3C Forms Tutorial.
C-03. Modal dialog
- Record the closed state and the element that opens the dialog.
- Open the dialog. Check the role, the name, the initial focus, and the inaccessibility of the background layer.
- Go through the content forwards and backwards, without letting the focus leave the boundaries of the dialog.
- Raise an error inside the dialog, correct it, and check the notification without a loss of context.
- Close the dialog with the button,
Escape, and the other supported methods. The focus must return to the opening element or to the next logical place if that element has disappeared. The expected model is set out in the Modal Dialog Pattern.
C-04. Tabs
- Record the selected tab, its panel, and the position of the focus.
- Enter the tab list with a single
Tabtransition, then use the provided arrow keys,Home, andEnd. - Distinguish focus movement from tab selection. Check the consistency of the visual and the programmatic selected state.
- Move to the panel, raise a loading, empty result, or error state, if they are provided for, and then return to the list.
- Check the exit from the composite component and the recovery after a tab is removed, if removal is supported. The model is described in the Tabs Pattern.
C-05. Disclosure
- Record the collapsed state and the inaccessibility of the hidden content for navigation.
- Activate the control and check the change of the visual and the programmatic state.
- Move to the expanded content, perform a nested action, or read it with the selected method.
- Collapse the block and make sure the hidden content no longer enters the route, and that the focus has not remained in the hidden area.
- Repeat the opening after an error or a page reload. The expected states are given in the Disclosure Pattern.
C-06. File upload and notifications
- Before the file is selected, find the types, the size, the restrictions, the security requirements, and the way to cancel.
- Check a correct file, a wrong type, an exceeded size, and a network error, using safe test data.
- For each variant, record the state before, the selection, the validation, the progress of the upload, the outcome, and the focus position.
- Check determinate or indeterminate progress, cancellation, retry, and protection against a repeated submission.
- After a fix, make sure the previous error is removed, and that the success and the next step are available without a forced focus move. A notification must not take the focus only in order to be announced.
How to write up the result
The result for each component contains:
- the exact environment passport and the date;
- the full route and the initial state;
- the expected and the observed for every step;
- the focus position, the return, and the way notifications are detected;
- the hint, the assistance, and the workaround separately from independent completion;
- the direct source of the expectation;
- evidence without personal data;
- the status of the defect and its consequences;
- the result of the re-check after the fix.
One successful run confirms only the stated combination of versions, settings, task, and state. It does not confirm the whole site, all assistive technologies, or conformance with all criteria.
User pilot of the Dostupno site
Object of the pilot: the public Dostupno site, its methodological materials, the training report, and the safe training form. The manual technical check prepares the routes and the test stand, but does not replace sessions with participants.
The pilot covers five tasks:
- Find the material for checking the accessibility of a site before release.
- Save a document and find in it the check of work with the keyboard.
- Find the explanation of a selected requirement and determine the check that is needed.
- Choose a problem in the training report and show the next step after the fix.
- Correct an error in the training form and determine whether the action has completed.
The first iteration provides for 6-8 individual sessions of 30-45 minutes. The participants include people who use a screen reader, enlargement, or colour settings, who work without a mouse, or who need especially clear instructions and a predictable sequence. Voice control is checked separately where possible. The working roles of the visitors are also taken into account: the owner of the resource, the author, the developer, and the checker. One person may combine several ways of working; a small sample is not treated as representative of all users.
For each task, the technical expectation, the actual action of the participant, the hint or assistance, the observed outcome, the interpretation, and the decision on the fix are recorded separately. An expert forecast is not recorded as user evidence. The actual sessions have not been held yet, so there are no user results so far.
