Answering Questions

The questionnaire is the input source a test plan is built from. Answering it is the bulk of the work in Discovery. The config scan can settle part of it from your system's configuration, and what remains needs a person. The agent asks, records, and keeps track of what is left.

How the questionnaire is organized

The questionnaire has two phases, and they serve different purposes:

1. Orientation establishes how the system is used at all. These are broad questions about the interfaces people work with, the functions the operation performs, and how the system connects to others. Answer orientation first, because these answers determine which of the detailed sections apply to you.

2. Exploration covers the detail within each functional area that orientation opened up. These questions are specific to how your operation runs.

Sections and individual questions can depend on earlier answers. Anything whose dependencies are not met is never asked. This is why the questionnaire has no fixed length: it is scoped to the system you described, so an operation that uses a handful of functions answers far fewer questions than one that uses all of them.

Working through the questions

Before asking, the agent presents a summary of the applicable sections and how many questions remain in each. Use it to see the shape of the work ahead and to decide where to spend your attention.

The agent then asks in small batches, at most four questions at a time, and keeps every batch within a single section. It repeats until no applicable question is left unanswered. In Claude Code, each batch arrives as a set of selectable questions. See Picking from the agent's choices.

Questions the config scan already answered are not asked again. If you ran the scan first, what you work through here is what the configuration could not settle.

Each question arrives with its choices, and a choice can carry a short note explaining what picking it would mean or that it is the questionnaire's likely default for your system. Read those notes. They are where a question that looks ambiguous usually becomes answerable.

You do not have to finish in one sitting. Progress is saved as each answer is recorded. See Resuming Discovery.

Recording an answer

Each question can be left in one of four states:

State	Use it when
answered	You know the answer. Multiple-choice questions accept more than one selection.
unsure	You do not know yet and want to come back to it.
inapplicable	The question does not apply to your operation.
unanswered	You want to clear an answer you already gave.

Only answered counts toward completion

Marking a question unsure or inapplicable does not complete it. Those questions stay outstanding, so the questionnaire remains incomplete and no test plan can be generated. Resolve them before you generate.

Answering a gating question

Some questions gate others, and answering one of those opens up the questions behind it. When that happens, the config scan looks at the questions your answer just made applicable and settles the ones your configuration export can settle. The agent reports which questions that covered. Answering one question can therefore complete several, so read what the agent reports rather than assuming the outstanding count dropped by one.

This pass reaches only questions that are still unanswered, so it never replaces an answer you gave. It also requires a configuration export to read. A project without one is unaffected.

Changing an answer

Ask the agent to review a section, and it reports every applicable question there with its current value and whether you or the config scan supplied it. A question the scan answered also carries the configuration it read, so you can judge the answer rather than take it on trust. This is read-only, so you can look without disturbing anything. Change any answer you want from that review.

Changing an answer can clear others

When you change an answer that other questions depend on, those dependent answers are cleared. This is deliberate: an answer that is no longer visible must never shape a test plan. An entire section can reopen this way, and a questionnaire that was complete can become incomplete again.

Check the hub after a change of this kind so you can see what reopened.

Checking progress

Ask the agent for discovery status at any point. It reports answered and applicable counts with a percentage, broken out per section, per phase, and overall. Percentages count applicable questions only, so a question that was never asked does not count against you.

The report takes roughly this shape:

Phase        Section                              Answered / Applicable
-----------  -----------------------------------  ---------------------
orientation  Interfaces                           3 / 3
orientation  Functional Areas                     6 / 6
exploration  Inbound                              2 / 5
exploration  Outbound                             0 / 10
exploration  Inventory                            0 / 8
exploration  retired by your orientation answers  0 applicable

Overall: 11 / 32 applicable questions answered

Section names come from the questionnaire for the system under test, so the ones you see will differ. The row for retired sections is the useful one: it shows what your orientation answers took off the table, which is why the applicable total is smaller than the questionnaire's full length.

Example

You say:

I want to work on the questions.

The agent:

1. Takes you into the questionnaire and shows the applicable sections with the number of questions outstanding in each.
2. Asks the first batch, all from one orientation section, no more than four at a time.
3. Records each answer as you give it, clears any dependent answers that a change invalidated, and reports any question the config scan settled once your answer made it applicable.
4. Continues through the remaining batches, opening exploration sections as your orientation answers make them applicable.
5. Returns you to the hub, with the questionnaire status updated.

You say:

What is my progress so far?

The agent:

1. Reports completion per section, per phase, and overall, counting applicable questions only.

Finishing Discovery

• Making a Test Plan: Generate the recommended tests once the questionnaire is complete.
• Ingesting Documents: The third input source, which runs independently of your answers.