Elicitation with examples that distinguish interpretations

1. Make the distinction needing clarification explicit

While preparing a fictional migration, APS says some files need manual review. Two characteristics may explain selection: a new source, represented by N, and corrected content, represented by C. The analyst records four hypotheses describing current processing: review if N; review if C; review if N or C; review only if N and C.

These are hypotheses, not approved requirements. Both variables are Boolean and the exercise admits only these four explanations without observation errors. Actual work may involve other conditions, unknown states and exceptions.

Writing hypotheses down supports choosing a question that separates them. A vague question about “the right procedure” can produce four verbally similar answers without exposing the difference. Also record the decision needing this information: preserving behavior, correcting a defect and proposing a change are distinct purposes.

2. Recognize an example that does not distinguish alternatives

A file from a known source without correction has N=false and C=false. All four hypotheses predict processing without review. A corrected file from a new source has N=true and C=true; all four predict review.

Showing only these two cases can create apparent consensus but does not select a hypothesis. Repeating the same combination many times does not resolve this logical distinction either. Those examples may still help with other questions, such as confirming vocabulary or observing a complete path.

An activity’s value depends on the uncertainty it should reduce. For this exercise, tabulate each hypothesis’s predicted result before the session. Select at least one combination with differing predictions instead of choosing only the easiest example to demonstrate.

3. Adapt the second example to the first result

The first question presents a new source without correction: N=true and C=false. “Review if N” and “N or C” predict review; the other two predict processing without review. If the confirmed result is no review, “review if C” and “N and C” remain.

The second question can present a known source with correction: N=false and C=true. If review occurs, only hypothesis C remains compatible with both observations. If it does not, N and C remains.

The conclusion is conditional on the hypothesis set and observation reliability. It does not establish which rule the organization should adopt. If an outcome contradicts every hypothesis, do not choose the “least wrong” one: revisit conditions, evidence or the explanation set. Allow room for that discovery.

4. Do not turn observed behavior into authorization

An authorized replay in a demonstration environment shows a corrected file proceeding without review. The business owner says review should occur. Possible explanations include a defect, particular configuration, valid exception or misinterpreted example.

Replay answers what that environment did with those inputs and configuration. The statement addresses intent or the rule the participant believes applies. Record both without rewriting either to make the conflict disappear.

Ask for the basis of the expectation and confirm scope, version and possible exceptions with relevant participants. Information can be accurately recorded while describing undesirable behavior. Confirming elicitation supports further analysis; the change decision and requirement approval have their own scope.

5. Distinguish multiple copies from multiple observations

A report, a presentation and three messages repeat that corrections need manual intervention. Following their references shows that all derive from one interview note. There are five documents but one origin for the claim.

They are not five independent process observations. Retain the origin, date and relevant transformations, especially when a summary drops a condition. The original note said “corrections to already published files”; the slide says only “corrections”.

That shortening expanded scope without new confirmation. Return to the example and the person able to clarify the condition. This purpose does not require retaining complete personal or production files: an authorized, minimized example can preserve the relevant relationship.

Traceability should help explain content, not merely collect hyperlinks.

6. Retain conditions making the comparison useful

To clarify whether a correction triggers review, the team compares two examples. The first uses a known source and initial file in environment A with configuration X. The second uses a new source and corrected file in environment B with configuration Y.

A different outcome cannot isolate correction’s role: source, environment and configuration changed. Where feasible and authorized, construct examples changing only the relevant characteristic and record remaining conditions. This improves interpretation of this comparison but does not turn a limited trial into a universal guarantee.

If context cannot be controlled, state the limitation and combine observation with additional information. In L3 support, a record without configuration version may be insufficient to reconstruct an exception described by a colleague.

7. Confirm meaning through a concrete decision

In an international meeting, two teams accept the phrase “the correction replaces the file”. One understands that the original remains queryable as history; the other understands that it disappears from every query. Instead of asking again whether everybody agrees, present a case: a correction arrived today, and tomorrow APS needs to explain content used for yesterday’s close.

Ask each team what should remain visible and why. The example distinguishes interpretations without making someone guess a term’s meaning. Record the conclusion in the working language and confirm translation of critical terms.

“Replaced for current processing” may need a separate sentence about historical access. Confirming understanding does not turn the analyst’s suggested wording into an approved policy. Also distinguish enabling publication on request from performing it automatically: a file without a request can remain pending, while rejecting a valid request may contradict the action that should be available.

8. Exercise: prepare two useful questions

Before reading the solution, choose two N and C combinations distinguishing the four hypotheses from section 1 and write what each result permits you to conclude. Solution: use (N=true, C=false) and (N=false, C=true). In that order, outcomes are review/no review for N, no review/review for C, review/review for N or C, and no review/no review for N and C.

Each pair identifies one hypothesis within the model. All four cases are unnecessary for this distinction; this also does not establish coverage of every actual service situation. Add who confirms examples, which context must be retained and what to do if an outcome falls outside the model.

For APS handover, prepare a summary separating the selected hypothesis, evidence limits and the outstanding business decision.

Common pitfalls

Counting documents as independent sources; choosing only consensual examples; changing several conditions without recording them; confusing observed implementation with an approved rule; treating a finite model as coverage of reality.