By Vellora Labs · Reviewed October 12, 2026
Record what happens during a task in order: an action, a question, a system response, and any help. This sheet is for the event timeline within a session. Use the separate evidence log when you later connect observations across sessions to a claim.
Download blank CSVDownload printable text version
Prepare the header before the attempt.
Write a participant code, task ID, product or stimulus version, starting state and observable success condition. Agree on the start of the timer and how the observer will record interruptions. Use safe test data and confirm the actual participation and recording arrangements.
One row for each meaningful event.
| Field | What belongs here |
|---|---|
| Elapsed time | Time since the agreed start; note restarts rather than silently resetting. |
| Event type | Action, words, system response, assistance or interruption. |
| Observation | What you saw or the exact words; label a paraphrase explicitly. |
| System state | Visible errors, loading, missing data or other relevant conditions. |
| Assistance | What the moderator or another person explained and when. |
| Outcome | Evidence for the success condition, with assistance and setup failures separated. |
A fictional sequence
| Elapsed | Event | Record |
|---|---|---|
| 00:18 | Action | Participant opens a sharing panel. |
| 00:31 | Words | Invented quote: “Who will be able to see this?” |
| 00:42 | Assistance | Moderator explains the audience label. |
| 01:03 | Action / outcome | Participant saves the intended audience after assistance. |
This sequence is invented for teaching. It supports recording completion with assistance. It does not show independent discoverability, nor establish that the participant felt anxious. If the panel failed to load, record that system problem separately.
Interpret after the attempt.
While observing, avoid replacing a concrete action with a diagnosis such as “confused user.” After the task, ask what the person expected, write your interpretation in a separate note, and consider another explanation. Keep the source location for any statement you may use in a later finding.
Use outcomes consistently.
- Completed unaided: evidence meets the predefined condition without route-revealing help.
- Completed with assistance: record the help and subsequent result.
- Not completed: the attempt did not meet the condition; record what happened rather than assigning blame.
- Not assessable: setup or system failure prevented a meaningful attempt.
Define these labels for your actual study before comparison. Small qualitative sessions do not establish a population success rate. Elapsed time that includes thinking aloud or interruptions should not be presented as an ordinary production-use benchmark.
Keep the working files controlled.
Use codes, restrict source links and remove unnecessary personal details. When importing outside text into spreadsheet software, set the destination to plain text to avoid formula interpretation. These downloads contain no collected responses.
Prepare a task with the task builder, study AI product onboarding, then move reviewed observations into the cross-session evidence log.