A Form Is a Cheat Sheet, a Blank Page Is an Exercise

Thirty minutes in, the screen looked right.

Precondition filled. Three steps, numbered. One Expected Result, phrased carefully. Every field had something sensible in it.

And I had no way of telling whether any thinking had happened.

The Exercise I Thought I Was Running

The target was the step before a test case exists. Take a requirement, decide what is worth checking at all, get that list out of your head. Conditions, not cases. No steps yet, no expected results yet, no numbering.

To run it I opened our test management tool and created a new test case. That single click was the entire design error, and it took me two sessions to see it.

An empty form is not empty. Precondition, Step, Expected Result — those labels are an outline somebody else already wrote. They announce that there will be a setup, that there will be ordered actions, that there will be exactly one outcome worth naming. Every structural decision I wanted made in the room had been made in advance by whoever designed that screen.

Watch what the form does to the order of work. The first thing typed into it is a step, because the first field that invites a sentence is a step field. Scope has already been settled by then, silently, by whatever came to mind first. There is no moment where the question "how many separate things is this?" gets asked, because there is no box for the answer.

So the intermediate step did not get skipped. It got deleted. There was nowhere on the screen for it to live.

Take the Fields Away

Second run, same requirement, different tool. A plain text editor. Nothing but a cursor.

The instruction was small. Write down everything this requirement makes you want to check. Any shape. Fragments are fine, questions are fine, order does not matter yet.

That version was harder in the useful direction. A blank page starts asking all the things the form had been answering on our behalf. What comes first? Is that one condition or two? Is that a precondition or a case in its own right? Does anything here depend on anything else here?

Every one of those is the work. The form had not made them easier — it had made them invisible, which is not the same thing and is considerably worse.

So the rule I took out of that session: when you want to force an intermediate step, do not ask for it. Change the tool to one that has no place to put anything else. A form suggests a structure. A blank page requires one.

There was a second benefit I had not planned for. A condition written in a text editor has to carry its own context, because no field label is standing next to it explaining what kind of sentence it is. That is the same discipline as writing every test case for a stranger — nothing around the line is doing work on its behalf, so the line has to do it.

The Same Error, Twice More

Once the shape of the mistake was clear, I found it in two other exercises I had built.

One was classification. I had a checklist of categories in a document and the artifact to be classified in another window. The exercise I believed I was running was about recognising types. The exercise I actually built measured how well someone holds six category names in memory across a tab switch. The repair is physical: the checklist goes in the same frame as the thing being classified, and the decomposition rule gets said out loud before the first question. Otherwise the result is a reading on access to the list, not on understanding.

Then there was "find the bug". I opened a screen and asked what looked wrong on it. No candidate list anywhere — not on paper, not on screen, not stated. When the room went quiet, the only instrument I had left was hints, and every hint narrows the space a bit more. At that point the exercise has stopped measuring anything about testing. It measures how long a person will sit still while I feed clues. So the candidate failure types now get written out, in view, before the screen is opened.

Three exercises, one error underneath all of them. In each case I decided what would be on the screen before I decided what the exercise had to measure. Once the screen was chosen, it quietly made the rest of the decisions for me, and it made them well enough that nothing looked broken.

That is the part worth keeping. None of these exercises failed loudly. They produced filled fields, confident answers and a finished artifact every time. The tool was generous enough to cover for a design I had not finished thinking through, and a covered failure is the expensive kind.

Which of your exercises would survive being run on a blank page?