This slot holds only what the person said or did — their words, their emphasis, what they had already built or bought. Nothing that begins "which suggests".
Superhuman's method is instructive here because the score is the least useful thing it produces. A single percentage is an average, and averaging is exactly the operation that destroys the information you needed: it tells you that something is wrong and nothing about what. Their actual work was segmenting the responses and reading the open text underneath — the score was a trigger for looking, not the finding.
The reason for the strictness is that interpretation is lossy in one direction only. From the quote you can always re-derive the conclusion, including a different conclusion six weeks later when you know more. From the conclusion you can never re-derive the quote. A discovery programme storing only conclusions has thrown away the thing that would have let it change its mind.
Two kinds of evidence outrank everything else people tell you, and both belong here verbatim:
- What they already built or bought. A spreadsheet, a script, a contractor, a competing subscription. Spend beats stated preference, and a hack beats a complaint.
- The last time it happened. "Last Thursday, and here is what I did" is an event. "It happens all the time" is a self-assessment, and people are poor at those.
A good note here is one that would interest someone who does not share your hypothesis.