When someone finishes a usability task after the facilitator reveals a useful clue, the completion is real. It also answers a different question from finishing before that clue. A results table needs to preserve both the outcome and the assistance history, rather than flatten them into one success percentage.
The short answer is to define the task, record when task-revealing assistance occurs, and report completion before and after that point separately. This lets a team learn from recovery without pretending that the original barrier disappeared. It also avoids treating normal access arrangements as evidence that the participant could not use the service independently.
This article uses an original fictional community-workshop website and invented session records. The examples are teaching accounts, not observations of a real service or estimates of a population success rate. They show how a researcher can turn a session timeline into an honest result and a useful design question.
eduKateSG · User Experience
Find your next learning step
Preserve the assistance timeline before interpreting task-completion percentages.
ROUTE 1 · CHAPTERS 1–4
Define the research outcome
Distinguish the task, access arrangements and assistance rule.
ROUTE 2 · CHAPTERS 5–8
Read timelines and proportions
Reconstruct observed outcomes with explicit denominators.
ROUTE 3 · CHAPTERS 9–12
Preserve the evidence limits
Track interventions, missing records and nonstarts.
ROUTE 5 · CHAPTERS 17–21
Practise and continue
Solve changed records, check ambiguity and choose the next route.
Chapter contents
Define the research outcome · Chapters 1–4
1. Name the question before naming success
2. Distinguish access from task-revealing assistance
Read timelines and proportions · Chapters 5–8
5. Read one session as an ordered sequence
6. Build mutually exclusive outcome categories
Preserve the evidence limits · Chapters 9–12
9. Do not infer the causal effect of the hint
10. Preserve the reason and content of each intervention
Compare and improve · Chapters 13–16
13. Compare rounds with the assistance conditions visible
14. Report time with its assistance boundary
Practise and continue · Chapters 17–21
17. Solve a changed-context record independently
18. Choose a practical learner route
19. Audit ambiguous prompts without hiding the uncertainty
CHAPTER 1 OF 21 · Define the research outcome
1. Name the question before naming success
Imagine a fictional team testing a workshop website. The participant’s task is to find the cancellation deadline for a selected workshop and identify the correct date. The interface contains the necessary information, but the team does not yet know whether its organisation and wording make that information findable.
One research question is whether likely users can discover the answer without task-specific clues from the facilitator. Another concerns what support helps someone recover after becoming stuck. A third asks whether the complete service, including a planned support channel, can deliver an understandable answer. These questions overlap, but their outcomes are not interchangeable.
Start by stating the first question narrowly. The participant receives a neutral scenario and the same information that the intended user would normally have. The facilitator observes the attempt. Completion means finding the relevant information and identifying the right deadline, rather than merely opening a page with a promising title.
This boundary makes a later intervention interpretable. If the facilitator names the exact page containing the deadline, the discovery conditions have changed. The participant may still demonstrate useful understanding after reaching it. The result now needs two statements: what happened before the clue, and what happened afterwards. Neither statement should erase the other.
CHAPTER 2 OF 21 · Define the research outcome
2. Distinguish access from task-revealing assistance
A participant’s usual screen reader, keyboard, magnifier or other access arrangement is not automatically a task hint. It may be the ordinary way that person uses the service. Calling every such arrangement assistance would confuse access to participation with information about how to solve the task.
For this article, unaided completion means completion before the facilitator supplies task-revealing information. It does not mean working without assistive technology, agreed communication support or the access conditions needed for a fair session. Record those conditions because they matter to interpretation, while preserving the separate question of whether the answer or route was revealed.
Consider two fictional events. In one, a facilitator repeats the scenario in an agreed accessible format without naming a control. In another, the facilitator says that the answer is under the workshop terms link. The second provides information about the solution path. The first may simply make the original task available to the participant.
Do not decide from a label alone. An interpreter’s clarification could preserve the original task or accidentally supply new clues; a supposedly neutral prompt could draw attention to a particular area. Preserve what actually happened and examine its relevance to the research question. A useful distinction should make the evidence clearer without turning participants’ ordinary ways of working into deficits.
CHAPTER 3 OF 21 · Define the research outcome
3. Define completion through the intended outcome
The workshop task has several observable stages. A participant can locate a terms page, find a cancellation section, interpret which deadline applies, and state the correct date. Opening the page is an intermediate event. If the participant selects the wrong workshop’s deadline, the defined information task is not complete.
Choose this outcome before reviewing the results. Otherwise the team may quietly change the rule after seeing where participants stopped. A difficult round can appear more successful simply because the definition moves from correct answer to page opened. A later round using the stricter rule would then be difficult to compare.
The definition also needs a scope. We are not testing an actual cancellation or payment transaction in this fictional exercise. We are testing whether the participant can find and interpret a date in supplied material. Do not expand a successful reading task into proof that a complete real service works.
Record partial achievements because they can explain the barrier. Someone who finds the correct section but misreads a date has a different difficulty from someone who never locates the terms. Both can be incomplete under the final outcome rule while offering different evidence for improvement. A binary summary is useful only when the richer record remains available to explain what the binary result leaves out.
CHAPTER 4 OF 21 · Define the research outcome
4. Prepare the intervention rule before the session
A moderated session may need to move on after someone becomes stuck. Preserving an unaided outcome does not require leaving a participant distressed or spending the whole session on one barrier. It requires a plan for how intervention changes the interpretation of what follows.
In our fictional study, the team defines a task-specific clue as any facilitator statement that identifies a relevant control, destination, search term or answer not already supplied in the scenario. The record marks the first such event. Later completion is categorised as completion after a task hint, even if the participant performs the remaining steps independently.
The team also records why it intervened and the wording used. A predetermined intervention condition can improve consistency, but it does not make every session identical. Participants may request clarification, choose to stop or encounter a prototype limitation. Those events need their own descriptions rather than being forced into a convenient failure category.
The GOV.UK moderated-usability-testing guide supports neutral tasks and observation while recognising that helping someone recover can allow further learning. Our completion categories and numerical accounts are original examples. The important consequence is to keep the change in conditions visible, so a practical decision to continue the session does not silently rewrite the earlier evidence.
CHAPTER 5 OF 21 · Read timelines and proportions
5. Read one session as an ordered sequence
Consider a supplied session record. At time zero, the task begins. At forty seconds, the participant opens a workshop overview. At seventy seconds, they say that they cannot find the cancellation information. At ninety seconds, the facilitator names the terms link. At one hundred and twenty seconds, the participant identifies the correct deadline.
The final outcome is correct completion. The assistance history shows that it occurred after a task hint. The record does not demonstrate unaided completion within this observed attempt. Nor does it demonstrate that the participant would never have finished without the hint; the unaided observation ended when the facilitator intervened.
Two durations can be reported with clear labels. Total observed time from task start to correct answer is one hundred and twenty seconds. Time from the recorded hint to correct answer is thirty seconds. The second duration is not an independent task-completion time because it begins after earlier exploration and after new information was supplied.
This chronology is the foundation for the whole article. If a summary preserves only completed at one hundred and twenty seconds, it loses the findability barrier and the intervention. If it preserves only failed, it loses what the participant could do afterwards. The ordered record supports both learning questions while keeping their boundaries intact.
CHAPTER 6 OF 21 · Read timelines and proportions
6. Build mutually exclusive outcome categories
Suppose twenty-four fictional participants attempt the same defined task. Fifteen reach the correct answer before any task hint. Six reach it after a recorded hint. Three do not reach the answer within their observed attempts. For this complete illustrative dataset, each participant belongs to exactly one of those categories.
The categories must neither overlap nor leave a hidden remainder. A participant who finishes after a hint should not also be counted among unaided completions. A participant who never attempted the task should not be inserted into these twenty-four attempts merely because they were invited to the study.
| Observed category | Participants | Share of 24 attempts |
|---|---|---|
| Completed before a task hint | 15 | 62.5% |
| Completed after a task hint | 6 | 25% |
| Did not complete in the observed attempt | 3 | 12.5% |
| Total | 24 | 100% |
This table is a partition of the supplied records. It does not assign a cause to any outcome. In particular, the six later completions are not six proven rescues that would have been impossible without help. They are six completions observed under changed conditions.
Before calculating a headline percentage, check that the counts add to the denominator. This simple reconciliation catches duplicate classification and omitted attempts. It cannot establish whether the category definitions were sensible, but it makes their application inspectable.
CHAPTER 7 OF 21 · Read timelines and proportions
7. Calculate the two completion proportions
Unaided completion in the supplied dataset is 15/24 = 62.5 per cent. Completion with the recorded assistance history is (15 + 6)/24 = 21/24 = 87.5 per cent. Both describe these sessions, with different outcome definitions.
The difference is twenty-five percentage points. That is the six post-hint completions expressed as a share of all twenty-four attempts. Relative to 62.5 per cent, the numerical increase is forty per cent. The distinction between percentage points and relative percentage change remains important even when neither comparison supports a causal interpretation.
A defensible summary gives the counts alongside the percentages and names the conditions. For example: fifteen of twenty-four participants completed before a task hint; another six completed after one; three did not complete during the observed attempt. A reader can reconstruct either proportion and see where assistance enters.
Avoid a heading saying the website achieved an 87.5 per cent unaided success rate. That would use the broader numerator with the narrower label. Equally, calling every post-hint completion meaningless would discard evidence about interpretation and recovery. The point of separate proportions is to retain more useful information, not to award a single score to the design or participant.
CHAPTER 8 OF 21 · Read timelines and proportions
8. Keep the recovery denominator distinct
Nine participants did not complete before a hint in this fictional account. Six subsequently completed. The conditional observed recovery proportion is therefore 6/9, approximately 66.67 per cent. This answers a different question from the six post-hint completions divided by all twenty-four attempts, which is twenty-five per cent.
The denominator should match the wording. Among the nine participants without pre-hint completion, six later completed. Across all twenty-four attempts, six completed after a hint. Those sentences describe the same records from different perspectives without contradiction.
The first sentence also requires care about eligibility and observation. In a real dataset, not everyone who lacks an unaided completion necessarily receives a comparable hint or continues afterwards. If some stop voluntarily, encounter a technical interruption or receive different assistance, they cannot silently be treated as nine identical recovery opportunities.
For our example, the supplied complete categories make the arithmetic simple. For an actual report, preserve who received what assistance and whether the subsequent task was observable. A conditional proportion is useful only when the condition is clear. Writing recovery rate without describing the relevant group can make an apparently precise number impossible to interpret or reproduce.
CHAPTER 9 OF 21 · Preserve the evidence limits
9. Do not infer the causal effect of the hint
The six post-hint completions show what happened after intervention. They do not reveal what those same participants would have done without intervention over the same subsequent period. That unobserved alternative is necessary for a causal claim that the hint produced six additional completions.
One participant might have found the terms link moments later without help. Another might have remained stuck. The observed post-hint sequence alone cannot distinguish those histories. Our classification does not need to decide between them because its question is descriptive: when did completion occur relative to the assistance event?
A team may still learn something concrete. Several participants searched the overview and missed the terms link; after the destination was named, they could interpret the deadline. That pattern suggests investigating findability separately from date comprehension. It provides a design hypothesis and a route for another test, without proving a particular causal effect size.
If a study is designed to compare support conditions causally, its design, assignment, outcomes and analysis require their own justification. Do not retrospectively turn an ordinary moderated session into such a study by subtracting two descriptive proportions. The arithmetic is correct; the causal interpretation would be an additional step whose evidence has not been supplied.
CHAPTER 10 OF 21 · Preserve the evidence limits
10. Preserve the reason and content of each intervention
Two hints can differ substantially. One names a page. Another points to an exact line containing the answer. A third explains the date rule itself. Combining all three under assisted completion can be adequate for one broad summary while concealing important differences for the design decision.
Keep a short event record with the trigger, wording, timing and subsequent action. Was the participant confused by a label, unable to operate a control, unsure which workshop the terms applied to, or interrupted by a prototype fault? The assistance event should connect to the observed barrier rather than float as an unexplained tick box.
In the workshop case, a destination hint followed by correct interpretation suggests a different next question from an answer-revealing hint. After the answer is explicitly supplied, repeating the date demonstrates little about independent comprehension of the original text. The team should not give those events the same explanatory role simply because both end with the right spoken date.
A useful report can retain an overall post-hint category while adding a small qualitative breakdown. It should avoid rankings that imply one participant needed more help because of personal ability. The session investigates the fit between person, task, interface and conditions. The content of the intervention helps identify which part of that relationship deserves attention.
CHAPTER 11 OF 21 · Preserve the evidence limits
11. Treat missing outcomes as missing information
Now consider a separate fictional record with twenty-four starts, fifteen unaided completions, six known post-hint completions, one known noncompletion and two sessions whose endings were not recorded. The missing endings cannot automatically be assigned to failure or success.
If the task and category rules are otherwise compatible, observed overall completion is at least twenty-one and at most twenty-three out of twenty-four. Those logical limits are 87.5 per cent and approximately 95.83 per cent. They do not tell us which value is most likely.
The unaided-completion limit needs more chronology. If both missing sessions had already received task hints before recording stopped, neither can later become a pre-hint completion under our definition. The unaided count remains fifteen. If their hint histories are also missing, as many as two additional unaided completions may be possible, giving an upper limit of seventeen out of twenty-four.
This distinction shows why a missing endpoint and a missing intervention history are different information gaps. Reporting only the number of usable recordings can hide their consequences. A transparent report states which facts remain known and gives an appropriately limited summary. Do not manufacture completed records to make a table add neatly, and do not describe a logical range as a confidence interval with an unstated probability guarantee.
CHAPTER 12 OF 21 · Preserve the evidence limits
12. Keep invited people separate from attempted tasks
Suppose thirty people were scheduled and twenty-four attempted the workshop task. Six did not start it. The attempt-based unaided proportion remains fifteen out of twenty-four under the earlier complete dataset. Dividing by thirty would answer a broader question that includes nonstarts, with a different definition.
Nonstarts can matter greatly. A participant might be unable to access the session, the prototype might fail before the task, or a person might choose not to continue. These events may reveal barriers outside the defined task attempt. Excluding them from an attempt-based calculation should not make them disappear from the study account.
Write a simple flow: scheduled, attended, eligible for this task, started, observed to a defined endpoint. Different research questions can use different stages, but the stages need compatible counts and explanations. The final denominator should not be selected merely because it produces a more attractive rate.
Task selection also matters when participants receive different scenarios. If a person never receives the cancellation task because another task is more relevant, their absence from its denominator is not a cancellation-task failure. Record the assignment rule. A useful metric has a population boundary that another reader can reconstruct, rather than a denominator assembled from everyone whose name appeared somewhere in the research calendar.
CHAPTER 13 OF 21 · Compare and improve
13. Compare rounds with the assistance conditions visible
Imagine two new fictional rounds of twenty-four attempts each. Round A contains fifteen pre-hint completions and six post-hint completions. Round B contains eighteen pre-hint completions and three post-hint completions. Both have twenty-one overall completions and three observed noncompletions.
The overall completion proportion is 87.5 per cent in both rounds. Unaided completion changes from 62.5 to seventy-five per cent, a difference of 12.5 percentage points. A table with only overall completion would conceal that descriptive change in assistance history.
The pattern does not prove that a design revision caused the change. Participant familiarity, task wording, facilitator behaviour, prototype differences or other conditions may differ. It also matters whether the same people repeat the task and remember the route. The counterbalancing article develops that separate learning-history problem.
For a useful comparison, show the category counts, intervention rules and relevant differences between rounds. Then connect the observed behaviour to the proposed repair. If the revised label appears easier to find before a hint, investigate that mechanism directly. Stable overall completion can coexist with a meaningful change in how completion occurs, but the evidence still needs to support the explanation assigned to that change.
CHAPTER 14 OF 21 · Compare and improve
14. Report time with its assistance boundary
A mean completion time can combine very different histories. Suppose three fictional completions occur at sixty seconds without a hint, ninety seconds without a hint, and one hundred and twenty seconds after a hint at ninety seconds. Their combined mean is ninety seconds.
That mean describes time to completion under the recorded session conditions. It does not describe mean unaided completion time. Among the two unaided completers, the mean is seventy-five seconds. For the assisted completer, total time is one hundred and twenty seconds, with thirty seconds between the hint and the correct answer.
Do not compare the thirty-second interval directly with the two full task times as though all three began with the same knowledge and exploration history. The assisted participant had already spent ninety seconds on the task and then received new information. Resetting the clock does not reset that history.
Time among completers also excludes those who did not finish. If a revision enables slower participants to finish, the mean among completers can rise even while access to the outcome improves. The User Experience root works through that broader success-and-speed distinction. Here the additional task is to preserve the intervention boundary so a duration’s label remains faithful to the sequence it summarises.
CHAPTER 15 OF 21 · Compare and improve
15. Recognise when support belongs to the intended service
A service may be designed to include human support. In that case, evaluating the complete supported journey can be a legitimate research question. The issue is whether the support observed in the session resembles the support the intended user would actually have, and whether the report names that boundary.
For the fictional workshop website, a planned help channel might explain cancellation terms after a user requests assistance. A researcher supplying a clue instantly in a test room is not automatically evidence that this real support journey will work. Availability, discovery, waiting, continuity and the quality of the eventual answer are separate parts of that service.
Likewise, a question about self-service findability can remain useful even when human support exists. The two findings can inform different decisions: improve the information route and maintain a dependable supported alternative. There is no need to force every outcome into one universally preferred path.
State which support is part of the proposed design and which support was supplied only to continue the research session. This prevents a prototype from borrowing the facilitator’s knowledge without recording the dependency. It also lets a team recognise the value of assistance honestly, as a service component to investigate rather than an invisible contribution to a self-service score.
CHAPTER 16 OF 21 · Compare and improve
16. Turn the observed barrier into a testable repair
The workshop records show several participants looking on the overview page for a cancellation deadline while the relevant terms link is hard to recognise. A plausible repair is to make the relevant information route clearer at that decision point. This is a hypothesis grounded in observed behaviour, not proof that one particular label is the final answer.
Define what the next study should reveal. Can likely users locate and interpret the deadline before a task-specific hint under the revised conditions? Preserve the same intended outcome while checking that the task wording does not reveal the new label. Otherwise the revised test can accidentally remove the very findability question it was meant to investigate.
Also inspect the quality of the information reached. A prominent link to ambiguous terms can improve page visits without improving correct answers. The task outcome should retain both discovery and interpretation where both matter. Record newly introduced difficulties rather than counting only evidence favourable to the proposed repair.
The next study can be modest and still useful when its scope is clear. It may reveal a persisting barrier or support further refinement without estimating a population effect. The goal is a defensible next decision: what changed, what was observed, what remains uncertain and which part of the experience should be examined next.
CHAPTER 17 OF 21 · Practise and continue
17. Solve a changed-context record independently
A fictional library catalogue study contains thirty task attempts. Eighteen people find the correct resource before a task hint, eight complete afterwards, and four do not complete during the observed attempt. Calculate unaided completion, overall completion and the conditional post-hint completion proportion among those without unaided completion.
Unaided completion is 18/30 = sixty per cent. Overall completion is 26/30, approximately 86.67 per cent. Twelve participants did not complete before a hint; eight later completed, giving 8/12, approximately 66.67 per cent in this complete illustrative account. The post-hint contribution to all attempts is 8/30, approximately 26.67 percentage points.
Now suppose the eight hints identified the exact resource title. What can a later correct selection establish? It can show what happened after that information was supplied. It cannot establish unaided discovery of the resource or the number of completions caused by the hint. Additional understanding may still be observable, depending on the remaining task.
For a final variation, two of the four noncompletions were actually unobserved endings. The known overall completion count remains twenty-six, with an upper limit of twenty-eight if both missing endings were completions. Explain which extra chronology would be needed before changing the unaided count. This tests the evidence structure rather than merely repeating a familiar percentage operation.
Begin with a single timeline if the distinction between pre-hint and post-hint completion is unclear. Mark the task start, first task-revealing intervention and completion. Ask the learner to write two true sentences about the record, then one stronger claim that it does not support.
Move to a category table when the timeline is understood. The learner should reconcile every attempt, choose a denominator for each question and explain why six out of nine differs from six out of twenty-four. Keep the language attached to the quantities throughout; a percentage without its group is an unfinished explanation.
For extension, add missing endings, changed intervention rules or a planned support channel. Ask which earlier conclusions survive. A strong response may retain the arithmetic while narrowing the interpretation, or revise the calculation because the denominator no longer matches. Both show understanding when the reason is explicit.
The Gold Standard of Education provides the wider learning route. The transferable skill is preserving the history behind an outcome. A student can use it when evaluating a prototype, reading an assessment report or comparing service data, without assuming that every context shares the same rules for assistance or completion.
CHAPTER 19 OF 21 · Practise and continue
19. Audit ambiguous prompts without hiding the uncertainty
Sometimes the record preserves a prompt but its classification remains debatable. A facilitator might say, what are you looking for now, without naming a destination. Another might say, have you looked in this area, while pointing to a control. The words and accompanying action matter together. A transcript that omits the pointing can make a directional intervention look neutral.
Return to the original twenty-four attempts, but suppose three of the fifteen apparently unaided completions contain unresolved prompt events. Twelve are definitely pre-hint completions, six are definitely post-hint completions, three are ambiguous between those categories, and three did not complete. Overall known completion remains twenty-one. Its count does not depend on resolving the prompts.
The unaided count can range from twelve to fifteen under this revised evidence state. The corresponding proportion lies between fifty and 62.5 per cent. This range reflects classification uncertainty in three known completions, rather than missing final outcomes. If the records later establish that two prompts revealed task information and one did not, the unaided count becomes thirteen and the post-hint count becomes eight.
Keep the original event record, the coding rule, the disputed interpretation and the final resolution distinguishable. A second reviewer can help identify inconsistent coding, but agreement alone does not turn a poorly chosen rule into a good answer to the research question. Reviewers need to consider what information became available to the participant.
This is also a useful check on a report’s robustness. If the practical decision changes depending on how three ambiguous events are classified, the uncertainty deserves attention. If a substantial observed barrier remains under either classification, the team may already have a clear reason to investigate that barrier further. State that relationship directly instead of forcing every prompt into a confident category for the sake of a tidy chart.
An ambiguity range should remain tied to the known events. It is not permission to speculate about every unrecorded possibility. The aim is to preserve a specific uncertainty that could change the stated result, and identify the evidence or definition needed to resolve it.
CHAPTER 20 OF 21 · Practise and continue
20. Questions about hints, outcomes and fair interpretation
Is helping a stuck participant a research mistake?
Not automatically. It can allow the session to continue and reveal what happens beyond the barrier. Record the help and interpret subsequent outcomes under those changed conditions. A useful recovery observation should remain distinguishable from evidence of unaided discovery.
Does a request for clarification make the task assisted?
The content matters. Clarifying the scenario without revealing a solution can differ from naming a control or answer. Preserve the actual exchange and apply the declared coding rule. If an exchange is ambiguous, report that ambiguity rather than relying on the word clarification alone.
Should assistive technology be removed to measure unaided use?
No such rule follows. The intended conditions can include the person’s ordinary assistive technology and access arrangements. This article’s unaided category concerns task-revealing information from the facilitator, not a requirement that everyone use one input method or work without accessibility support.
Can small-session percentages describe all users?
They describe the supplied sessions under their task and recruitment conditions. A broader prevalence claim needs an appropriate design and uncertainty account. Counts and percentages can organise qualitative research records without turning a small purposive round into a population survey.
Is completion after a hint evidence of learning?
It may provide evidence about what happened after new information, but it does not establish durable independent performance in a changed context. A separate transfer task can investigate that question. Repeating an answer immediately after it was revealed is a much narrower observation.
CHAPTER 21 OF 21 · Practise and continue
21. Follow the connected research and design routes
Keep the experience and task history together
Return to The Gold Standard of User Experience for needs, journeys, prototypes and meaningful outcomes. Continue to Why Counterbalancing Does Not Erase Task Learning when repeated tasks or design order change what participants already know. That question concerns learning across exposures; this article concerns assistance within an observed attempt.
Use the primary practice source within its scope
The GOV.UK moderated-usability-testing guide is the first-party practice source for neutral tasks, facilitation and suitable participation arrangements. It was last updated on 3 October 2017 and its live link was checked on 26 October 2026. Our workshop, timelines, categories, percentages and missing-information exercises are original fictional teaching constructions, not results reported by that guide.
Continue into wider subject questions
The Inclusive User Research Services guide develops participation and trustworthy findings. The Accessibility User Research Services guide connects real tasks with respectful access arrangements. Use them when the unresolved question concerns whose experience the sessions represent or how participation conditions are provided.
The Service Design guide follows outcomes across human and digital support. Human–Computer Interaction provides the broader interaction discipline. These routes help keep a facilitator’s temporary intervention distinct from a support journey that the service actually promises to deliver.
Use the verified How X Works hub to choose the related design or measurement discipline that matches the unresolved question. Return to learning that transfers into life for the broader educational purpose. The dependable habit is to preserve what the person knew, what the facilitator supplied and what outcome was actually observed.