By Vellora Labs · Updated October 11, 2026
Use one row per observation to connect a finding to its source, context, interpretation and possible counterexample. The download starts blank. It is a working record, not an automated analysis or a completed study.
Download blank CSVDownload field guide (.txt)
Keep four layers separate.
| Layer | Record | Why keep it separate |
|---|---|---|
| Source | Participant code, source type, timestamp, question and stimulus version | Lets an authorized reviewer find the original context. |
| Observation | Exact words or action; label quotation, observation or paraphrase | Prevents an interpretation from posing as a direct quote. |
| Interpretation | Your explanation and an alternative | Makes the reasoning open to review. |
| Claim and action | Proposed claim, limits, review state and next step | Keeps an untested hypothesis from becoming a product promise. |
A clearly fictional example
Invented for teaching: in a prototype session, P03 pauses before sharing a draft and asks who can see it. The source reference is session-demo, 04:12; the task is “prepare a draft for a colleague.” No actual participant or recording underlies this example.
Observation: a pause and a question about audience. Interpretation: visibility may be unclear. Alternative: the person may understand the control but want reassurance for this particular document. Next action: test the audience explanation with a fresh task. This does not support a broad claim that all users find sharing confusing.
Use a review state with an explicit scope.
Start a proposed claim as unreviewed. A reviewer should inspect the original material, check the context and permission, and record whether it is supported within a stated scope, needs more evidence, or is contradicted. Retain opposing observations and people absent from the sample.
Protect the record without losing traceability.
Use participant codes and restrict source access. A source link should not automatically grant public access to a recording. Keep identity keys and payment details elsewhere. Record permission for the intended reuse; a research interview is not automatically a public testimonial.
When pasting outside text into spreadsheet software, set the destination to plain text. Some applications interpret values beginning with formula characters as executable formulas. The downloadable CSV is blank and contains no participant text.
AI summaries and simulated answers
Keep the original evidence reference alongside an AI summary and inspect both before approval. Label synthetic output as synthetic, even when based on real customer material. It is not a new interview. If you cannot locate the source behind a claim, leave it unreviewed.
Choose a workspace using the analysis tools collection. Then use the research-to-story guide to decide what can responsibly appear in a product explanation. TapVid is our team’s product video tool; it does not validate the evidence in this log.