1. LandOffer
  2. Blog
  3. All articles
  4. Accessibility Portfolio for Tech Jobs: Show an Audit, a Fix, and a Verified Retest

Job search

Accessibility Portfolio for Tech Jobs: Show an Audit, a Fix, and a Verified Retest

An owned local HTML disclosure and input provide actual browser keyboard, DOM, label and retest evidence with explicit accessibility scope limits.

10 min readLandOffer team

An accessibility portfolio for tech jobs should show a traceable chain: an authorized sample, a reproducible problem, a specific fix and a recorded retest. Explain what you checked and what remains untested. A screenshot of a better-looking component or an automated score alone cannot establish that the original interaction now works.

LandOffer publishes this guide and offers job-search tools. As you browse recent tech roles on LandOffer.ai, choose evidence that matches the role’s actual design, engineering or audit responsibilities. We created and tested the owned local HTML sample below on October 21, 2026. No private website, user account, assistive-technology session or participant research was accessed; the sample is teaching code, not a client project.

A keyboard connects an issue report to a corrected disclosure model awaiting a recorded retest.

Define the tested sample, states and methods

Begin the audit case with permission and a concrete scope. Use your own page or obtain the authorization needed for someone else’s site. Record the version tested, selected pages or components, relevant states, browser and methods. Explain why the sample is useful without implying it represents every part of a larger product.

Our sample contains one “Delivery notes” disclosure and one room-code input. Those two interaction types let us examine keyboard activation, exposed state and a persistent input label. The page has no submission, storage or external resources. Before and after files are preserved separately so a reviewer can reproduce the change without depending on a live client site.

Testing used a Chrome browser tab at a recorded content viewport of 1728 × 835 pixels. We recorded the active DOM element, hidden-panel state, ARIA attribute values, native accessibility snapshots and screenshots. The evidence identifies Chrome and the 1728 × 835 content viewport; its exact browser build and operating-system version were not collected, so exact-version reproduction remains a documentation gap. We did not resize the browser for this sample or run a screen reader, user study or accessibility scoring tool.

This is a deliberately small, purpose-built sample. It does not cover a full website’s navigation, media, error recovery, zoom, contrast, mobile interaction or complete processes. Keep those untested conditions visible beside the corrected findings.

Connect a finding to the relevant requirement

W3C’s Understanding Keyboard guidance explains SC 2.1.1’s keyboard-operability requirement. It also distinguishes the requirement from common button conventions: supporting both Enter and Space is a useful native-button behavior, but the criterion is not a universal rule that every custom control must use both keys.

Understanding Name, Role, Value discusses programmatically determinable component names, roles and user-changeable states. A control that looks like a button needs its behavior and semantics considered together. Adding an ARIA role does not, by itself, implement keyboard interaction.

These Understanding pages are informative explanations of WCAG criteria. We use them to frame the issue, not as a certification checklist for the whole sample. Do not label a local retest “WCAG compliant” when you have examined only a few conditions.

A useful issue report names the affected task and exact reproduction steps. Separate the observed failure from an inferred user impact. “Enter left the notes hidden” is an observation. “Every keyboard user would abandon the page” would be an unsupported claim.

Worked asset 1: a complete keyboard disclosure issue and fix

Issue LOCAL-A01 concerns the owned before sample. The disclosure was a focusable div with a click listener, rather than a native button. Its text read “Delivery notes”; the associated paragraph started hidden. There was no keyboard handler or alternate keyboard action for that disclosure in this sample.

The complete reproduction record is:

Report field Observed local evidence
Task Reveal the sample delivery notes without a pointer.
Starting state Before file loaded; notes paragraph hidden.
Steps Focus the Start sample link, press Tab to the notes control, then press Enter. Repeat with Space while the control remains focused.
Actual result Active element was DIV with id notes-toggle. Enter and Space both left the paragraph hidden.
Pointer comparison Clicking the same visible control opened the paragraph.
Expected task result A keyboard path should make the same notes available.
Proposed priority Fix this blocked task before polishing the component; a local triage choice, not a universal severity rating.

The before markup contained:

<div class="toggle" id="notes-toggle" tabindex="0">Delivery notes</div>
<p id="delivery-notes" hidden>Demo room notes: use the side entrance.</p>

The actual fix changed the interactive element to:

<button id="notes-toggle" type="button"
  aria-expanded="false" aria-controls="delivery-notes">Delivery notes</button>

The click listener still toggles the paragraph’s hidden property and now updates aria-expanded to match. Using a native button supplies its normal keyboard activation behavior rather than adding a second, improvised key-handling path. The panel id and the button’s controlled-element reference remain aligned.

We then retested in the same browser. Tab from the starting link reached BUTTON with id notes-toggle. Enter opened the paragraph and changed aria-expanded to true. Space closed it and set false; a second Space reopened it, and Enter closed it again. The four observed transitions matched the paragraph’s visible or hidden state.

A further keyboard activation was preserved in a full native snapshot: “Delivery notes” appeared as an expanded button and the notes text was present. The corresponding screenshot showed the focused control and revealed content. These records establish the tested browser interaction and exposed snapshot, not what a particular screen reader would announce.

The completed issue decision is original keyboard failure no longer reproduced in the tested after sample. It is narrower than “disclosure accessibility finished everywhere.” Other browsers, assistive technologies, complex panel content and integration with a real application remain outside the check.

Retest the task, not only the changed line

Keep the original reproduction steps in the report and run them against the corrected artifact. If the fix changes the control type, inspect what receives focus and whether the state follows the actual content. Pair the code diff with the retest so a reviewer can connect the implementation change to the task that originally failed.

Check adjacent behavior that the change could affect. In our after sample, forward Tab moved from the control to the input and then the End sample link. Shift+Tab returned from that link to the input. We recorded those focus targets rather than inferring a successful navigation path from source order alone.

A passing local path is still a bounded result. This sample has no modal, validation errors or dynamic routing, so the observations do not cover keyboard traps elsewhere. Keep the browser context, artifact version and exact steps attached to the retest so another person can challenge or extend it.

A permanent label stays connected to an input after its example placeholder is replaced by a value.

Worked asset 2: input-label repair and a completed retest matrix

Issue LOCAL-A02 concerns the separate room-code input. Before the fix, it had only the placeholder RM-DEMO, with no associated label or persistent instruction. We focused the field and typed the fictional value RM-TEST. The sample then displayed the value without a visible field-purpose label above it; DOM inspection returned an empty input-label collection.

We did not claim that the before field had no name in every accessibility API. The native snapshot listed a text field using the room-code identifier. That observation is not a screen-reader announcement or a full accessible-name computation. The specific reported problem was the absent persistent visible label and absent label association in this markup.

W3C’s Labels or Instructions explanation distinguishes visible input guidance from programmatic association and accessible naming, which relate to other criteria. An invisible name alone does not provide a visible label for all users. Those distinctions help keep the report focused instead of treating every labeling concern as the same test.

The correction added a visible label whose for value matches the input id. It also added a persistent hint, associated through aria-describedby:

<label for="room-code">Room code (demo)</label>
<p id="room-hint">Enter a fictional code such as RM-DEMO. Nothing is submitted.</p>
<input id="room-code" aria-describedby="room-hint"
  placeholder="RM-DEMO" autocomplete="off">

The filled retest matrix records both findings and their limits:

Check Before After and completed decision
Notes activation from focused control Enter and Space left notes hidden; pointer opened them. Both keys toggled the after button; close LOCAL-A01 for this tested path.
Disclosure state No aria-expanded attribute on the div. True/false matched the panel across four transitions; native expanded-button snapshot saved.
Input after entering RM-TEST No associated labels; no persistent purpose text above the input. Room code (demo) remained visible; labels collection contained that text. Close LOCAL-A02’s tested label condition.
Input hint association No persistent hint. aria-describedby pointed to room-hint; DOM contained the expected fictional-code instruction.
Neighboring keyboard route Not assessed as a complete before navigation audit. Forward Tab reached End sample; reverse Tab returned to the input. Record this limited after check.
Screen reader and user comprehension Not tested. Still not tested; retain as open evidence, not an automatic pass.

The after native snapshot named the text field “Room code (demo)” with the entered value. The saved after-field screenshot preserved the visible label, hint and RM-TEST value together. That is visual evidence of persistence in this sample, not a comprehension result from a user. We checked the DOM association separately from appearance; neither record is presented as speech output from assistive technology.

The completed decision is to retain the button and label repairs in the owned sample, preserve the original files and close only the reproduced conditions. A local consistency script checks that our saved observation records agree with those conclusions. It validates the record’s assertions, not a new browser test or an automated WCAG score.

Keep the reproduction environment reviewable

Record how the interaction was driven. Here browser automation sent the key events and captured native browser snapshots and DOM observations. No human participant evaluated the task. The pointer action was a separate comparison on the before sample; it was not substituted for the failed keyboard steps. The after result came from the button’s browser behavior, not a script setting its expanded attribute to a desired answer during the test.

Preserve the starting state with each sequence. A notes panel that is already open cannot establish that Enter opened it. Likewise, a screenshot taken before typing cannot establish that a field’s label remained visible afterward. Our records distinguish the initial hidden state, the key sequence, the typed field and the later expanded-button snapshot. The files describe the state they capture instead of being presented as interchangeable proof.

When a later change touches the listener, control type, panel identifier or label association, rerun the affected paths. A formatting-only article edit does not change the sample, but changing its behavior or markup can invalidate the earlier retest. Keep the earlier result as historical evidence and attach the new artifact version to the new result rather than overwriting the failure record.

Package evidence without overstating the outcome

Give the portfolio reviewer an index to the scope note, before and after HTML, issue report, patch, reproduction steps and retest records. Keep source hashes or another stable artifact identifier so a later edit does not silently detach the screenshot from the code it supposedly proves.

If you worked on a permitted real site, describe your role and the people responsible for implementation and acceptance. A recommendation is different from a merged change, and a local fix is different from a deployed release. Do not invent a client endorsement or a disabled participant’s experience to make an independent project sound like employment.

Avoid a percentage improvement in accessibility based on two selected issues. The sample was constructed to demonstrate those problems; its fixed count is not a website-wide defect rate or a user-success measure. Additional checks may uncover new issues without invalidating the narrow results already recorded.

Before applying, check that the original steps, corrected source, retest record and remaining checks are all available and clearly tied to the same sample version. Explore LandOffer.ai to organize applications while keeping the audit and verification claims tied to their evidence.

Sources and Further Reading