Product & Project Delivery
Prioritize Accessibility Findings by User Impact and Reproduction Evidence
A fictional product team prioritizes keyboard and error-message barriers, investigates an unresolved chart flag, and records what a narrow repair retest does and does not establish.
Prioritize accessibility findings by the task a person cannot complete, the conditions that reproduce the barrier, and the evidence needed to verify a repair. Record the relevant WCAG criterion alongside that explanation. Use the conformance level, scanner output, and repeated instances as evidence to interpret, not as a substitute for the impact assessment. A reproducible blocker in a critical journey deserves attention even when an automated check reports no matching issue.
This LandOffer.ai guide includes a completed finding register and a manual-versus-automated evidence review. The product, people, findings, test records, and decisions are fictional teaching examples. No application, browser, assistive technology, or user study was tested for this article. Primary documentation was checked October 27, 2026. The examples show how to reason about evidence; they do not certify a product, establish legal compliance, or represent the experience of every disabled user.

Start with a blocked task, not the scanner total
An accessibility report can contain many instances of one underlying component defect, a small number of severe task barriers, and findings that need more investigation. Treating those findings as interchangeable tickets hides which tasks people cannot complete. Describe the task, the barrier, and the consequence before estimating a fix.
The UK Department for Education's guidance, reviewed September 14, 2026, recommends considering user impact when prioritizing accessibility issues. It includes journey location, whether someone can continue, and how widely an issue occurs. Its guidance also cautions that WCAG A, AA, and AAA levels do not invariably rank the size of the user barrier. This is a named organization's guidance, not a universal severity formula or a statement about your legal obligations.
Keep your chosen conformance target visible. Prioritization sequences work; it does not erase applicable requirements or make a deferred failure acceptable. If your organization has a legal deadline, contractual obligation, or release gate, bring the responsible specialist into that decision. Do not infer those obligations from this fictional product's backlog.
A rare but complete block can matter more than a frequent inconvenience. Low traffic also does not prove low impact: people may abandon the journey before completing it, and analytics generally will not tell you everyone's access needs. Record what you know about occurrence and reach without inventing a percentage of affected users.
Build a reproduction packet that another person can use
Record the product build, route, account role, prerequisite state, environment, and input method. Include exact browser and operating-system versions in real reports, plus assistive-technology name, version, and relevant settings when used. Separate the steps from the expected result and the actual result. Pair visual evidence with the interaction record: a screenshot alone cannot demonstrate keyboard operation or an announcement sequence.
Be specific about available alternatives. A mouse click is not a keyboard workaround. Asking a colleague to submit on someone's behalf may introduce dependence or sensitive-data exposure. A separate keyboard route can be relevant, but it needs to be discoverable and capable of completing the task. Do not label a control a failure merely because one expected keystroke differs from your preference.
W3C's explanation of WCAG 2.2 criterion 2.1.1 describes keyboard-operable functionality, with a specific exception for movement-path-dependent input. Its guidance also distinguishes functional keyboard access from requiring every visible control to behave identically. For an ordinary submission task, investigate whether the function is actually available through a keyboard interface before assigning that criterion.
Keep evidence references understandable. In a real tracker, link the approved recording, test script, relevant markup or computed property, and environment details. Redact personal information without removing the state needed to reproduce the issue. The invented references below describe fictional records; they are not links to recordings or tests performed by this publisher.
Completed asset: a task-impact finding register
Product owner Farah and accessibility evaluator Jules review fictional Quarry, a maintenance-request portal. Their sample covers creating a request, correcting an invalid date, and reviewing submitted requests in build 7.4. It does not cover every role, mobile device, integration, or journey. All observations in this section are invented, including the environment and repeated attempts; they are not measured product results.
Their fictional keyboard record uses Windows 11, Chrome 141, a standard keyboard, and no screen reader. These version labels define the teaching scenario, not recommended current software. Jules starts with a requester account, a valid saved draft, and no attachment. The date and description are valid before she tries to submit.
| Finding | Task and recorded barrier | Reproduction evidence in the fictional packet | Initial triage |
|---|---|---|---|
| Q-17: request submission unavailable by keyboard | A requester cannot finish a valid request without a pointer | In build 7.4, Tab moves from the description to the footer; the visible submit control receives no focus. Enter in the fields does not submit. The team identifies no equivalent keyboard route. A pointer submits the same valid draft. Record K-17 includes sequence and starting state. | First repair candidate: complete task block; criterion 2.1.1, Level A, is the evaluator's identified mapping |
| Q-18: faint invalid-date explanation | A requester has difficulty reading the normal-size error text needed to correct a rejected date | Fictional style record C-18 identifies the text and solid background, with no large-text or inactive-control exception. Evaluator records insufficient contrast under criterion 1.4.3, Level AA; rendered-state verification belongs in the repair check. | High-impact repair alongside Q-17; do not dismiss it because it is AA |
| Q-19: unreviewed chart flag | An automated result flags a low-contrast mark in an optional history chart | The initial result does not establish whether the mark communicates information, which state was scanned, or whether equivalent values are available. No evaluator has yet resolved those questions. | Investigation required; no confirmed-failure or harmless-decoration conclusion yet |
Q-17's description includes both the failed keyboard routes and the pointer comparison so a developer can distinguish a submission-validation error from an input-access barrier. It does not claim that every keyboard user or every browser has been tested. Q-18 stays important because reading the correction message is part of the same essential journey. The two findings can affect different people and need different verification.
Farah records that developer Dev will inspect Q-17's shared submission component first, while designer Min and Dev examine Q-18's error style. These are accepted investigation assignments within the fiction, not completed fixes. Q-19 goes to Jules for a bounded evidence check. The team does not assign a low priority simply because the evidence is incomplete.
The shared component may create related occurrences elsewhere. Farah links those occurrences to the same suspected cause but preserves each affected journey and state. One component repair could address several instances; that is still a hypothesis until those instances are retested. Merging duplicates should reduce repeated investigation without deleting the record of where people encounter barriers.
Resolve a conflict by comparing test scope
The automated report for Quarry says no keyboard-related issue was detected on the saved-draft route. A developer initially proposes closing Q-17 because that report is clear. Jules asks what the check actually exercised. In the fictional packet, it inspected a loaded page; it did not navigate the complete draft-submission task with a keyboard.
W3C's guidance on selecting evaluation tools explains that automated tools cannot check all accessibility aspects and can produce inaccurate or misleading results. Human judgment remains necessary. That does not make every manual claim correct or every tool result useless. It means a result needs its method and scope before it can support a conclusion.
Farah's completed evidence-review note reads:
Claim under review: The automated report contains no keyboard-related detection for this route. This is a statement about that check, not evidence that a keyboard user can submit a request.
Conflicting evidence: K-17 describes the valid starting draft, focus sequence, attempted field submission, absence of an identified equivalent route, and pointer comparison. Within the fictional case, Dev follows K-17 in the same build and environment, reproduces the barrier, and accepts the submission-function defect for repair.
Decision: Keep Q-17 open and prioritize its repair. Retain the automated report with its scope rather than changing or discarding it. Do not turn “not detected” into “passed all keyboard requirements.”
Q-19 disposition: Jules establishes that the flagged chart mark represents a series, not decoration. The corresponding data table is not available in the recorded state. Keep the item open for specialist evaluation of the relevant non-text contrast requirement and any other barriers. No final criterion mapping or severity has been approved yet.
This resolves the practical disagreement without inventing a scanner bug. The automated result and the task observation can coexist because they make different claims. If a developer cannot reproduce a manual report, record both results and compare the build, account state, extensions, settings, and exact input sequence before changing its status. Do not close an issue solely because it failed to appear in a different environment.

Use effort to plan work without hiding severe barriers
Estimates help decide whether to pair a small component repair with a larger redesign, but they should follow an explanation of impact. A quick cosmetic change should not indefinitely displace a difficult critical-task blocker. If a repair needs architecture work, record an investigation owner, a decision date, and any proposed interim access arrangement that affected people can actually use.
In Quarry's fictional triage, Dev accepts the next engineering slot for Q-17 investigation and repair. Min accepts an error-style proposal for Q-18 in parallel. Neither acceptance proves that the work fits a particular number of hours. Farah leaves delivery estimates pending until the component inspection and design review are complete, and takes the unresolved release question to the authorized release owner.
Farah also records the remaining coverage questions: other submission variants, invalid inputs, mobile layouts, and assistive-technology combinations have not all been assessed. List those coverage gaps beside the result; one successful path does not establish accessibility across the portal. Involving disabled users can add task evidence, but a participant's successful attempt is not universal proof of accessibility either.
Close the specific finding after the relevant retest
A patch, merged change, or cleaner scan is not the same as a verified repair. Name the person who checks the affected task, the build and environment, the original steps, the result, and the coverage still missing. DfE's guidance calls for recording fixes and confirmation before closing an issue; adapt that accountability to your actual review process rather than silently treating implementation as verification.
The fictional follow-up has a mixed result. Dev implements the submission change in candidate build 7.4.1. Jules repeats K-17's valid-draft task in the same recorded environment, reaches the submission control, submits with the keyboard, and reaches the request confirmation. She records that specific retest as passed and closes Q-17 for that evaluated scope. No actual code or browser execution occurred for this article.
Q-18 remains open: Min's proposed style is ready, but the rendered invalid-date state has not been independently checked. Q-19 remains under specialist evaluation. The team has not established whole-product conformance, verified every related component occurrence, or authorized a release. Farah's summary therefore says “Q-17 repaired and retested in the recorded scope; two findings remain open,” not “accessibility passed.”
Bring your real register to the people responsible for prioritization, implementation, evaluation, and release. Ask them to accept specific next actions and preserve unknowns in the record. LandOffer.ai publishes this delivery guide; it does not perform accessibility audits or certify software. Leave the review with a prioritized repair queue, accepted next actions, and a verification status that matches each finding's actual evidence.


