The Gold Standard of Universal Design | Checking Whole-Journey Access When Step-Level Success Rates Hide Different People

A journey is a connected task with a beginning and an outcome. At fictional Cedar Learning House, a visitor must find the workshop information, make a booking, enter the designated room and understand the starting instruction. The review defines those four steps before observing any results.

Independent completion means completing all four under the stated session conditions without the facilitator performing or explaining a required action. This is a chosen measure for the exercise. It does not imply that needing support diminishes a person or that supported participation is unimportant.

Define supported completion separately. A visitor may reach the workshop through a well-designed assistance route, which can be a valuable service outcome. The record should show that route rather than quietly treating it as independent completion.

The boundary also needs an end. Arriving at the room is not enough if the intended task includes understanding what to do there. Conversely, a review limited to booking should not claim to have evaluated the physical visit.

Write a sentence before collecting data: this session examines these people attempting these steps under these conditions, and this is the outcome we will count. That sentence prevents the definition from expanding when results look favourable or shrinking when an awkward barrier appears.

Contents · Next chapter

A component total groups results by step. It can tell you that nine of ten people found the information. A person-level record groups results by participant. It can tell you whether the person who found the information also completed the other steps.

Both views are useful. The component view helps identify where difficulties occur. The person view preserves the connections needed to describe a continuous journey. Neither should be discarded merely because the other produces a simpler report.

Imagine a table whose columns are the four steps and whose rows are anonymous participant identifiers. A successful step receives Yes. A step needing the specified assistance receives Help. A step not observed receives Unknown, with a reason where available.

Those labels are deliberately different. Help is evidence of a supported attempt under this definition. Unknown is an absence of the required observation. Treating both as ordinary Yes would overstate independent evidence; treating every Unknown as a demonstrated failure would invent a result.

Use stable identifiers within an authorised evaluation so that the rows remain connected. Do not publish personal details merely to make the arithmetic easier. The teaching tables use invented identifiers, and a real review needs appropriate consent, access controls and proportionate reporting.

Contents · Previous chapter · Next chapter

Suppose ten fictional participants attempt the Cedar journey. Each step has nine independent completions and one instance of help. A summary reports 90 per cent for finding information, booking, entering and understanding the start.

The arithmetic for each step is 9 ÷ 10 = 90 per cent. It is correct. What remains unresolved in that summary is whether the same person needed help at each step or four different people did.

The table does not identify a 90 per cent whole-journey rate. It also does not prove that the whole-journey result is poor. Several arrangements of the same totals are possible.

Ask the learner to resist adding interpretation too early. State exactly what the table establishes, then identify the missing connection. A useful answer is: nine people completed each named step independently, but the summary does not show how their results overlap.

This is a practical reason to keep a structured session record. Once the individual connections have been removed, another calculation cannot necessarily reconstruct them from the column totals.

Contents · Previous chapter · Next chapter

In Cedar’s first fictional arrangement, P01 needs help finding the information, P02 with booking, P03 with entry and P04 with the starting instruction. P05 through P10 complete all four independently.

After a difficulty, a facilitator provides a predefined bypass so the participant can attempt later steps. The record marks that help, rather than describing the resulting journey as wholly independent. Later step results describe what was observed after that support; they are not a prediction of an unsupported counterfactual journey.

Each column still has nine Yes results. Only six rows contain Yes at every step. The observed independent whole-journey count is therefore six out of ten, or 60 per cent.

An independent check counts the affected people: four distinct participants encountered at least one recorded barrier. Ten minus four leaves six. This second route reaches the same result without multiplying any percentages.

The finding applies to the stated sessions. It does not establish a general 60 per cent rate for every visitor, every disability or every later version of the service. It identifies which complete-task claim the supplied observations support.

Contents · Previous chapter · Next chapter

Now construct a second possible record. P01 needs help at all four steps, while P02 through P10 complete every step independently. Every component still shows nine independent completions out of ten.

This time nine people complete the whole journey independently, giving 90 per cent. The column totals match the first arrangement exactly, while the person-level result differs by three participants.

Neither arrangement should be chosen because it produces the preferred headline. If the row connections are available, use them. If only the totals remain, report what is unresolved.

The second arrangement also raises an important design question. A high whole-journey percentage does not make the repeated barriers facing P01 unimportant. The counts describe distribution; they do not decide whose access deserves attention.

This is where universal design remains a human subject. Knowing how results overlap helps locate both widely distributed difficulties and concentrated exclusion. A report should make those patterns useful for repair without turning one person’s experience into an acceptable loss hidden inside a strong average.

Contents · Previous chapter · Next chapter

For ten participants and four steps, each with one unsuccessful independent result, there are four recorded step failures in total. At most four different people can be affected if every failure belongs to a different person. At least six therefore have no recorded failure.

At the other extreme, all four failures can belong to the same person, leaving nine with all steps successful. The compatible whole-journey range from these complete column totals is six to nine people.

More generally, with N participants and k fully observed steps, let Fj count the independent failures at step j. The number with all steps successful is at least the larger of zero and N minus the sum of the Fj counts. It is at most the smallest step-success count.

These are finite-record bounds. They assume the same N participants, complete classification of every required step and consistent outcome definitions. They are not confidence intervals, population estimates or evidence that the midpoint is most likely.

For Cedar, the lower count is 10 − (1 + 1 + 1 + 1) = 6. The upper is 9. The two previous arrangements show that both endpoints can actually occur.

The formula is useful when only totals are available, but it is not a reason to discard richer records. Directly counting the complete rows gives a more precise description whenever the necessary observations exist.

Contents · Previous chapter · Next chapter

Someone might multiply 0.9 by itself four times and report 0.6561, or 65.61 per cent. That product would be appropriate for a particular probability model with independent step-success events and the specified probabilities.

The Cedar totals do not establish that model. Access conditions can connect across steps: the same interface style, device, language presentation or physical arrangement may affect several actions for the same person. The two constructed records already demonstrate different overlap patterns with identical marginals.

The product also predicts a probability under a model, while our table describes ten realised sessions. A result of 6.561 people is not an observed count. Rounding it to seven would not turn the independence assumption into evidence.

This does not make multiplication of probabilities invalid. It makes the conditions important. If an analyst uses a model, the report should identify it, justify its relevance and distinguish the modelled quantity from the actual session record.

For a learner, ask which piece of information multiplication supplied that was absent from the table. The answer is a relationship between events, not a new observation.

When actual row connections are available, count them. When they are unavailable, bounds may be more honest than a precise-looking product. The aim is to preserve the difference between what happened and what a chosen model expects.

Contents · Previous chapter · Next chapter

Not every table uses the same participants at every step. A natural journey can lose participants after a barrier, and a report may calculate later percentages only among those who reached that point.

Suppose ten people start, eight complete booking and six of those eight enter the room. Booking completion is 8/10 = 80 per cent. Entry among those reaching that stage is 6/8 = 75 per cent. Entry relative to the original starting group is 6/10 = 60 per cent.

These are different denominators with different jobs. The conditional 75 per cent can be useful for analysing the entry stage. It should not be described as the proportion of all starters who independently reached the room.

If a strict funnel is observed, with every later success belonging to an earlier success, multiplying the matching conditional proportions can reconstruct the end-to-end proportion: (8/10) × (6/8) = 6/10. Here the shared factor cancels because the set relationship is known.

That calculation differs from multiplying four marginal 90 per cent figures over the same starting population. A conditional stage rate is not interchangeable with a marginal rate.

Before calculating, write beside each numerator: successful among whom? Then check whether participants could re-enter, receive help or use another channel. A simple funnel model should not be assumed when the actual service permits more complicated routes.

Contents · Previous chapter · Next chapter

A participant may leave before a later step for reasons that the record does not establish. Another may encounter a technical outage, choose not to continue or have a session stopped by the facilitator. Later cells should retain those distinctions where they matter.

If P07 has three Yes results and one Unknown, the record does not establish an independently complete journey. It also does not establish that P07 could not perform the missing step. The appropriate statement is incomplete evidence for this outcome.

For a conservative demonstrated count, include only rows with every required Yes. For a compatible upper count, examine which incomplete rows could still satisfy the definition if their unknown results were successful. A row with a known Help cannot become an independent whole journey merely by resolving another cell.

Suppose five rows are complete Yes, two have one Unknown and no known Help, and three have at least one Help. Demonstrated independent whole journeys number five. At most seven could be compatible with the unresolved cells. State the missingness reasons and the limits of this calculation.

Do not fill Unknown with zero automatically when producing a learning report, and do not delete those rows silently. Both choices can change the apparent rate.

The useful next action may be to improve recording or conduct an appropriate follow-up. An invented answer would make the report tidier while making the design problem harder to understand.

Contents · Previous chapter · Next chapter

Assistance can be a planned access route, an emergency workaround or coaching introduced only during evaluation. These are different situations. A report should say which occurred and which outcome it measures.

If a visitor independently uses an advertised support channel, that may be a successful service journey under an appropriately defined supported-participation outcome. It is different from a facilitator silently taking over a task that was intended to be independently operable.

Record what help changed. Did it explain an unclear label, bypass an inaccessible control, provide physical assistance within the service arrangement, or supply information missing from the page? The repair depends on the actual barrier.

In Cedar’s matrix, Help means the specified independent step was not completed under the test definition. That label does not classify the participant’s capability in every setting. A different arrangement could remove the barrier or provide a dependable alternative.

Avoid withholding needed support merely to obtain a clean failure count. Evaluation should be respectful and appropriate to the participants and setting. The task is to learn how the design and service work, not to turn difficulty into a performance spectacle.

The broader Universal Design guide connects ordinary access and individual support. Our narrower ledger makes the assistance visible so that both the independent route and the support route can be evaluated honestly.

Contents · Previous chapter · Next chapter

Step tests conducted on different service versions may not describe one usable route. A booking screen can improve while its confirmation link changes, and a new room arrangement can alter the physical entry task.

Imagine Cedar tests finding information on Monday, booking on a revised website on Wednesday and room entry after furniture is moved on Friday. Combining those results into one complete-journey count requires evidence that the tested components join in the actual released arrangement.

Version matters even when the same people participate. They may also learn the task across repeated sessions. Their later success can be useful evidence, but it describes a different experience from an unfamiliar visitor’s first attempt.

Keep the service version, date, instructions, permitted assistance and relevant environment with the record. A change does not invalidate every earlier observation. It identifies which relationship needs another check.

For a learner exercise, give three individually successful component records with incompatible dates or destinations. Ask whether they establish one complete journey. A correct answer should identify the missing connection rather than declare all the observations useless.

The final whole-task check should follow the actual joined route. Component repairs are useful steps toward that result, but a set of successful parts from different arrangements cannot substitute for verifying the arrangement people will encounter.

Contents · Previous chapter · Next chapter

A whole-journey count can reveal an important barrier pattern, but it does not turn a small group into a complete model of all users. People with the same broad disability label can use different strategies, devices and support arrangements.

W3C WAI’s guidance on involving users cautions against generalising from limited evaluations and recommends reporting the scope and conditions. User involvement and standards-based evaluation have complementary roles.

In Cedar’s second arrangement, the same participant encounters four barriers. The useful next step is to understand those barriers and their context. It is not to infer that everyone sharing one characteristic will have exactly the same experience.

Subgroup descriptions can be helpful when relevant and appropriately collected, but small counts need care. A table of one or two people should not be presented as a stable estimate for a broad population. Privacy can also require aggregation or withholding details.

Ask what observation supports the proposed repair. A specific control may fail with a particular interaction method. That relationship is more actionable than describing an entire group as unsuccessful.

Universal design asks how the environment can better accommodate variation. The ledger supports that inquiry by preserving what happened to individuals within the task. It should not become a ranking of people or an excuse to ignore someone whose experience differs from the majority.

Contents · Previous chapter · Next chapter

Step totals alone can make two repairs look equivalent. In the first Cedar arrangement, fixing any one of the four distinct barriers could allow one additional participant to complete independently, assuming no other changes or new difficulties.

In the second arrangement, fixing only one of P01’s four barriers still leaves three others in that person’s path. The step result improves, but the independent whole-journey count may remain nine. That does not make the repair worthless; it shows why complete access may require coordinated changes.

These are conditional predictions for the fictional record, not promises about real behaviour. A revised design needs another appropriate test. Removing one barrier may also change how a person approaches later steps, so the old row cannot automatically be treated as a permanent personal profile.

Keep material severity and consequences alongside counts. A barrier affecting fewer observed people can still deserve urgent attention. Responsibilities, dignity and the actual task cannot be reduced to a simple largest-number-first rule.

For a practical planning table, name the barrier, the proposed change, related dependencies, owner and next verification. Use the overlap record to identify which repairs need to work together.

The question becomes more precise: which change helps this complete route become usable, and what evidence will show that the intended improvement survived the transition between steps?

Contents · Previous chapter · Next chapter

Suppose a later Cedar evaluation finds eight independent whole journeys out of ten, compared with six in the first record. The observed proportions are 80 and 60 per cent, a difference of 20 percentage points.

That descriptive improvement does not, by itself, establish that a particular repair caused the change. Participants, familiarity, conditions or other service changes may differ. Preserve the comparison’s limits instead of treating the new number as a complete causal explanation.

If the same people are retested, their matched records can show who changed outcome in those sessions. That is useful, but practice effects and changed conditions still need consideration. If different people are tested, avoid speaking as though two specific individuals were necessarily helped.

Also keep the outcome definition stable. A later report that counts supported completion cannot be compared directly with an earlier independent-completion measure without explaining the change.

For a learner, ask for three sentences: what changed numerically, which conditions were comparable, and what remains unresolved about why it changed. This produces a more useful account than either an unqualified success claim or a declaration that no learning is possible.

A repair programme can make progress through careful iteration. The standard is honest evidence about each step and the connected result, not a perfect-looking chart that exceeds what the evaluation actually examined.

Contents · Previous chapter · Next chapter

A useful record connects steps without collecting more personal information than the task needs. In a teaching exercise, invented identifiers are enough. In an authorised evaluation, participants should understand the purpose and how the information will be handled.

Keep the original observations distinct from later interpretations. The observer might record that a person could not locate the submit control. A later analysis may connect that difficulty with its label or focus behaviour. The interpretation should remain traceable to the observation.

Use explicit status definitions. Yes, Help, Unknown and Not Applicable should not depend on which observer completed the sheet. If a step is genuinely not required for one route, explain how the whole-task definition handles that route rather than silently converting the cell to success.

A public report can summarise patterns without exposing identifiable session details. The internal connection between rows and evidence can be protected through appropriate access arrangements. Privacy and reproducibility require proportionate design, not the assumption that every raw record belongs on a public webpage.

For a learner check, remove the participant names and retain anonymous row links. Ask which calculations still work. Then remove the row links entirely. The loss of information should now be visible: step totals remain, while exact overlap may disappear.

This demonstrates why structure matters. A smaller, well-defined record can answer more than a larger collection of disconnected percentages.

Contents · Previous chapter · Next chapter

Eighteen fictional participants attempt three required steps in the same evaluation arrangement. At each step, fifteen complete independently and three receive the defined help. Every step is observed for every participant, with supported continuation recorded where needed.

Find the largest and smallest possible number of independent whole journeys compatible with these column totals. Construct an arrangement attaining each endpoint rather than relying only on a formula.

Then consider a supplied row-level result: ten people complete all three independently. Is that compatible with the step totals? Explain why knowing this exact count gives more information than the column totals alone.

Next, a separate record contains twelve participants. Seven have Yes at every required step. Two have no known Help but one Unknown each. Three have at least one Help. Find the demonstrated independent count and the largest compatible count if the two unknown results resolve favourably.

Finally, compare the following statements: fifteen of eighteen completed each step independently; fifteen of eighteen completed the whole journey independently; ten of eighteen completed the whole journey independently. Identify which follows from the totals alone and which requires a particular row arrangement.

Write your answers with denominators and definitions. The goal is to preserve the people across the task, not merely to produce a percentage that looks plausible.

Contents · Previous chapter · Next chapter

For eighteen participants, three steps each have three independent failures. There are nine step failures. If these belong to nine different people, nine people have no failure, so the lowest possible independent whole-journey count is nine.

If the same three people receive Help at every step, fifteen complete all three independently. That is the largest possible count because no complete journey can contain more people than any one successful step.

To construct the exact ten case, distribute nine failures across eight people: one person has two failures and seven people have one each, arranging three failures in each step. The remaining ten have none. Such a table is possible, though the exact arrangement should be written and checked rather than guessed.

Here is one explicit arrangement using the invented identifiers P01 to P18. Step one has Help for P01, P02 and P03. Step two has Help for P01, P04 and P05. Step three has Help for P06, P07 and P08. All other entries are Yes. Each step has exactly three Help entries, while the affected-person set contains eight distinct identifiers. P09 through P18 therefore give the ten complete Yes rows.

Notice what the repeated P01 contributes. It adds a second step difficulty without adding another distinct affected person. Counting every Help cell as another excluded person would incorrectly subtract nine from eighteen. Counting only the largest column’s three Help entries would incorrectly assume that all difficulties belonged to the same people. The linked row record resolves both mistakes directly.

The totals support the first statement about fifteen successful people at each step. They permit, but do not establish, fifteen complete journeys. The supplied row-level count establishes ten complete journeys in that specific record.

For the separate twelve-person record, do not call the two Unknown cells observed failures or successes. Report the demonstrated count and the unresolved possibilities together.

Contents · Previous chapter · Next chapter

A learner may calculate every percentage correctly and still believe that strong step results guarantee the same strong whole-task result. Return to the rows and ask them to mark everyone with at least one Help.

If they multiply the marginal rates automatically, ask what assumption connects the steps. Have them compare the six-person and nine-person Cedar arrangements. Both satisfy the same step totals, so the totals alone cannot select one product-based answer.

If they drop missing rows, ask what question the reduced denominator now answers. A proportion among fully observed participants may be reportable, but it should not be disguised as a result for everyone who started.

If they describe assistance as personal failure, return to the task definition and the environment. The label records what happened under specified conditions. A useful next question is which design or service arrangement could make participation more dependable.

For transfer, replace the learning centre with a library visit, an online application or a museum activity. Keep the same reasoning while changing the required steps. The learner should identify the whole outcome, preserve the row connections and state a limitation independently.

This is stronger evidence of understanding than memorising a bound formula. The principle is practical: a service needs the connected route to work for a person, and its evaluation should keep that person visible across the route.

Contents · Previous chapter · Next chapter

Are step percentages useless?

No. They help identify where barriers occur. They need row connections or explicitly bounded interpretation before describing the whole journey.

Does a 90 per cent result mean the remaining barrier is acceptable?

No. The percentage describes an observed outcome. It does not settle responsibilities, severity or the need for a repair.

Is every supported journey an unsuccessful service outcome?

No. A dependable support route may be part of successful participation. State whether the measure concerns independent steps, supported participation or another defined outcome.

Can I multiply conditional stage rates?

When the observed funnel and denominators genuinely match, their product can reconstruct its end result. That differs from assuming independence between marginal rates over the same population.

Do finite-record bounds predict future visitors?

No. They describe arrangements compatible with the supplied record. Generalising to a population needs an appropriate design and further evidence.

Can this method establish WCAG conformance?

No. It is an educational analysis of task results. Standards-based evaluation and other appropriate methods remain necessary within their actual scope.

Contents · Previous chapter · Next chapter

Follow the complete service

The Universal Design root connects physical access, information, support and learning. Resolving Conflicting Access Needs in Shared Learning Spaces addresses a different question: how to respond when useful arrangements conflict. The present guide examines what linked task outcomes can establish.

Plan and evaluate respectfully

Accessibility User Research Services develops participation and task evidence. Digital Accessibility Testing Services develops reproducible checks and remediation. These specialist routes complement the finite-record reasoning here.

Check primary guidance and transfer the learning

W3C WAI’s Involving Users in Evaluating Web Accessibility and Evaluation Overview were checked on 22 October 2026 for their treatment of user involvement, scope and complementary evaluation. WCAG 2.2's complete-process conformance requirement concerns all pages in a defined web process; our mixed physical-and-service example is not a substitute for that requirement. The matrices, bounds and learner exercises are original illustrations.

Continue through Education: learning that transfers into life and How X Works. Before accepting the next collection of component percentages, ask one concrete question: do we know whether the same people can complete the connected task?

Contents · Previous chapter · Return to Universal Design