A participant asks for a comparison dashboard. Another cannot find a specification needed to continue.
Both notes reach the backlog. The dashboard gets a detailed ticket because it sounds like a feature. The missing specification gets a vague request to improve content.
Start with the attempted task. Preserve what the person did, what they said, and what the observer inferred. Then decide what needs to change or be examined next.
GOV.UK's research analysis guidance separates observations, findings, and actions. The worksheet below applies that distinction to fictional B2B website sessions. Its triage categories are a proposed editorial method, not a validated scoring model.
Keep a Traceable Record of the Attempt
Use a fictional supplier of workshop storage cabinets. The research asks whether buyers can find requirements, compare shortlisted products, and prepare an inquiry. Every participant identifier, action, and result below is invented.
The prototype is version R1. Inquiry transmission is disabled. Participants receive synthetic measurements and product details. The note-taker records which tasks each person attempts and which help the moderator provides.
Here is the initial observation set. The statements column contains synthetic paraphrases, not real quotations.
| Reference | Task and observed sequence | Statement or context |
|---|---|---|
| P01-R1-T1 | Opens dimensions sheet, returns to product page twice, finds clearance information after a moderator hint | Says they expected all placement requirements in one document |
| P02-R1-T1 | Opens installation sheet directly and identifies clearance without help | Has prior experience reviewing installation documents |
| P03-R1-T1 | Inspects dimensions and stops without identifying clearance | Says the displayed measurements seemed sufficient |
| P04-R1-T2 | Opens both product pages and makes a shortlist using the supplied constraints | Requests a dashboard after completing the comparison independently |
| P05-R1-T3 | Reaches the inquiry mock, encounters a disabled attachment control, and cannot demonstrate adding the fictional plan | Attachment behavior is absent from this prototype |
The five rows describe different evidence conditions. Keep them visible. The disabled attachment control does not establish a defect in a working form. The dashboard request does not establish an unsuccessful comparison.
Separate Four Layers Before Writing Tickets
Give each note four fields: observation, participant statement, researcher interpretation, and proposed change. Store the evidence reference beside all four.
For P01, the observation is a sequence of document visits followed by assisted completion. The statement concerns the expected location of requirements. The interpretation is that requirements may be hard to locate from the product page.
The proposed change might be a short requirements summary linking to the authoritative installation sheet. That proposal needs review. It is not something P01 demonstrated was sufficient.
For P04, the observation is an independently completed comparison. The statement is a request for a dashboard. The interpretation might concern convenience, confidence, or an unobserved later task. None is established yet.
Ask what information a dashboard would preserve or expose. Keep the answer separate from the original task outcome. Avoid turning a named feature into the definition of the problem.
GOV.UK's observation guidance recommends recording what was seen or heard before interpreting it. Apply that discipline even when the proposed solution seems obvious. If a note combines action and explanation, split it before grouping.
Use customer interviews to investigate earlier work when that context is missing. A follow-up account can inform interpretation. It should remain identifiable as reported experience, separate from behavior observed on R1.
Group by Task and Obstacle
Create a requirements group containing P01, P02, and P03. Give it a provisional title: finding and checking installation clearance.
That title identifies the work. “Bad product page” names a verdict before the evidence review. “Content problems” is too broad to guide a decision.
Within the group, preserve the distinction between finding the document and interpreting it. P01 eventually finds the right information with assistance. P03 stops earlier with an incomplete answer. Those paths may need different remedies.
Keep P02 as a counterexample. A participant reaches the answer directly under the same prototype version. Examine the entry path and relevant experience before deciding what the contrast means.
If the team proposes a revised document label, ask whether it addresses both difficult attempts. A clearer label could help P01 locate the sheet. It may not help P03 recognize that the sheet is needed.
Keep the dashboard request in a comparison group. Keep the disabled control in an environment-limits group linked to inquiry preparation. Do not merge everything into one feature initiative because the same designer will work on it.
Use a Rubric That Exposes Uncertainty
Review four dimensions independently. The proposed rubric uses descriptive categories so a reviewer can inspect the reasoning. Do not add the categories into an automatic priority score.
| Dimension | Proposed categories | Question to answer |
|---|---|---|
| Task impact | Blocked, incomplete or incorrect, recovered difficulty, no observed obstacle | What prevented or changed the intended outcome? |
| Recurrence within this sample | Relevant attempts and affected attempts, with session references | Which comparable attempts show the obstacle? |
| Evidence strength | Recorded sequence, contemporaneous note, participant report, inference only | What can another reviewer verify? |
| Unresolved context | Known test limitation, missing context, conflicting explanation, none identified | What could change this interpretation? |
The categories describe the current record. They do not measure the commercial value of a revision. A blocked task may deserve investigation even when it appears once. A frequent complaint may remain poorly explained.
For the requirements group, report that P01 needs a hint, P03 stops without the specified clearance, and P02 completes independently. Those are three relevant attempts with different outcomes. Do not combine assisted and incorrect outcomes into a claim that all participants fail.
If another participant never attempts requirements, exclude that session from this task's comparison. Keep the omission visible. The number of people invited is not automatically the number exposed to an obstacle.
Treat observed frequency as a description of this round. It does not estimate the share of all buyers affected. Participant selection, task wording, starting pages, and synthetic inputs define the limits of the observation.
Distinguish an Obstacle From a Feature Request
Write two separate questions for the fictional dashboard request. What problem does P04 believe the dashboard would address? What evidence would show that the current comparison path obstructs that problem?
The observed task supports a narrow statement: P04 completes this shortlist comparison without help. It leaves open whether comparing more products, sharing the shortlist, or returning later creates difficulty.
A discovery action fits that gap. Ask the participant to describe the later work, then design a relevant task if the question remains important. Do not claim a dashboard is necessary from the request alone.
The clearance obstacle has a different record. Two task attempts lack independent completion of the specified requirement check. The next decision concerns a bounded explanation and revision, with P02's successful path retained for comparison.
This distinction does not dismiss suggestions. It gives each suggestion an evidence requirement. A feature request can identify valuable work that the current session never asked someone to perform.
Keep conversion optimization claims outside this register unless separate measurement supports them. Task difficulty establishes a research finding under stated conditions. It does not establish lost inquiries, missed revenue, or the outcome of a redesign.
Create a Findings Register With Owners
Use one row per bounded finding. Keep the detailed observations in the controlled research record. The register needs references that an authorized reviewer can resolve.
| Finding and evidence | Affected task | Bounded recommendation | Owner | Follow-up study question |
|---|---|---|---|---|
| F01: clearance check lacks independent completion in P01-R1-T1 and P03-R1-T1; P02-R1-T1 succeeds | Requirements | Prototype a requirements summary linked to the installation sheet | Content lead, with product specialist review | Can relevant participants find and apply clearance requirements without a hint? |
| F02: P04-R1-T2 completes comparison and requests a dashboard | Comparison | Investigate which later comparison activity the request concerns | Research lead | Does sharing or revisiting a shortlist introduce a distinct obstacle? |
| F03: P05-R1-T3 cannot use the unimplemented attachment control | Inquiry preparation | Provide a safe attachment simulation before evaluating that step | Prototype owner | Can participants identify and attach the supplied fictional plan in the revised test? |
Give each row a status, decision date, and next review point in the working document. Useful statuses include investigate, prototype, ready for review, held, and retest. Mark an unassigned owner explicitly.
For F01, the product specialist checks that the proposed summary preserves the fictional requirements accurately. The content lead prepares the revision. The research lead checks whether the next task still tests finding and applying information.
Keep delivery work bounded through consulting project management. A recommendation needs an acceptance condition and an owner before it becomes a delivery commitment. The register records the decision. It does not approve a wider website rebuild.
Record the Decision, Including a Decision to Hold
Review evidence before discussing implementation preference. Read the task, examine the session references, and inspect the counterexample. Ask which interpretation the available record supports.
For F01, the fictional team chooses a prototype revision and another observed attempt. It records that the summary is a hypothesis. Completion of the design ticket will establish that the revision exists, not that it resolves the obstacle.
For F02, the team holds dashboard implementation pending context about later comparison work. State what evidence would reopen it. “Held until we understand shortlist sharing” is reviewable. “Low priority” hides the missing question.
For F03, the team repairs the test condition before making a usability judgment about attachment behavior. Preserve the first session as limited evidence. Do not rewrite its outcome after the prototype changes.
Disagreements belong in the register too. If reviewers favor different explanations, name both and identify the observation that would distinguish them. A meeting should not erase an unresolved question merely because someone needs a ticket.
Close a Finding With New Evidence
Use a new version identifier for the revised prototype. Retain the original task goal. Record changes to task wording, starting pages, or recruitment so the next round can be interpreted on its own terms.
GOV.UK's moderated testing guidance includes neutral task instructions and watching attempts. Keep that observation boundary when evaluating the proposed change. A moderator explaining the new summary would contaminate an attempt to assess its discoverability.
For F01, record whether each participant finds the requirement, applies it correctly, and needs assistance. Keep any new obstacle alongside the original question. If the summary creates conflicting answers, reopen the content decision.
Close the row with a bounded conclusion about the new sessions. Leave wider market and business outcomes unresolved unless other evidence addresses them. Keep the original notes available under the agreed access conditions.
An observed obstacle earns a precise next decision. Preserve its evidence.
Sources
- GOV.UK Service Manual: Analyse a research session. Distinguishing observations, findings, and actions during research analysis. Checked 2026.10.28.
- GOV.UK Service Manual: Using moderated usability testing. Task observation and neutral instructions. Checked 2026.10.28.