A claim without its source is a guess

2.3aImagination you can auditsrc 01

Design always runs partly on imagination. Bulk demographics are too broad to design from, long immersion is too slow, and some users — a CEO, an astronaut — you will never reach. The discipline is to feed imagination with what you saw and were told, and to keep each claim's provenance tag · nhãn nguồn gốc beside it:

• Typed claims look alike. Once written, "she pauses at the fee" and "she shops weekly" read with the same confidence. !Nothing in the sentence shows which one was seen.Surprising: I expected a guess to give itself away in the wording. In a finished persona it reads as flatly as a fact, so only the tag tells them apart.
• Interpretation needs a check. Contextual inquiry, developed by Hugh Beyer and Karen Holtzblatt, asks the researcher to share an interpretation of the work with users so they can confirm or correct it. I read that check as needing the interpretation marked as one.

2.3bSix tags, strongest firstsrc 03

I rank a claim by how far it sits from something a user did:

1. [OBSERVED] — seen in a session, a recording or the product.
2. [STATED] — the user's own words. What people say and do are "very often" different (Rohrer).
3. [SECONDARY] — logs, tickets, published studies: records made for another question.
4. [PROXY] — support agents, sales: a proxy · nguồn đại diện speaking for users, filtered by their job.
5. [INFERRED] — reasoned from the tags above it.
6. [ASSUMED] — ?guessed, with nothing beneath it.Confused: If every assumed claim must become a question, how many can a persona hold before it is only a proto-persona with a photo?

The line falls between proxy and inferred. Above it, someone saw or said something; below it, the claim is my reading.

2.3cFour moves that keep the source attachedsrc 01

1. Tag every line as you write it. Bracket it at the end of each statement in a persona, scenario or requirement. A source costs one word to record now and an interview to recover later.
2. Split fact from reading. I keep four columns: fact, hypothesis, implication, design idea. In the fact column, an interpretation borrows authority it never earned.
3. Turn every [ASSUMED] into a question. ★Copy it into the research planActionable: My next research plan: one row per assumed line, with who could answer it., because a written question gets scheduled and a quiet guess gets built.
4. Book the proxy loan. Treat proxy insight as a loan against real-user research: note who said it and when you will check it with users.

2.3dAt work: four situations

• "We can't talk to users before launch." Imagination is now the method. Write the persona anyway, tag it, and take the [ASSUMED] list to the first real contact.
• "Sales says every client wants dashboards." That is a proxy claim. Tag it, ask which client said it and in what words, and book the check.
• An AI drafts a persona from a one-line prompt. →Every line starts as inferred or assumed.Connection: Same test as the AI quality checks: output I cannot trace to an input is a claim to verify. Promote a line only when I can point to its source.
• A requirement reads "Users need offline mode." Ask for its tag first. ★No tag means [ASSUMED].Actionable: Requirement reviews: a tag column in the spec template, filled before sign-off.

2.3eGOV.UK Service Manual, 2016src 04

The manual's guide to analyzing a session asks for a single note per observation, written exactly as seen or heard. What the team concludes goes later, as a finding on a sticky note of a different color.

2.3fWikipedia's citation needed tagsrc 05

An editor who doubts a statement tags it [citation needed]: a request for another editor to supply a source. The page adds that the tag alone never improves an article.