Ask your sessions the questions your product depends on.

Generic questions get generic answers

A generic summary can tell you that someone "browsed and left". It cannot tell you if the admin sent an invite. That is the type of question your product depends on. In Sequential, the questions are yours.

How it answers: when a session has had no new event for 30 minutes, the judge reads a short text list of its events. It does not watch the video. It gives each of your questions a probability. It writes no text. Sequential makes each chip and sentence from these numbers, your events and your words.

What you get

• An answer on each judged session. In the replay player, the Insights tab shows your verdicts, a chip for each signal that fired, and moments that you can go to. For example: "They hit noticeable struggle at 0:35 on «Apply code» — rage clicks, form left unfinished, went in circles."
• A ranked list to watch. Give a signal or a scale a Watch score %, and its answer adds to the session's Watch score. Watch uses that score to order the sessions in each of its groups.
• An agent that works in your words. With MCP, your agent can read your questions in full, and how often the judge was right when it gave an answer. Sequential finds and answers. Your agent acts in your tools.

How it works

1. 01

   Start from twelve.

   The default questions ask what people came to do, how it ended and how much friction they hit. They also ask six yes or no signals, such as Rage clicks, and three moments, such as Where they gave up.

2. 02

   Make them yours.

   Owners and admins can reword, switch off or add questions, up to 24. Each question is a choice (Intent), a level (Friction), a yes or no (Hit an error) or a moment (Where they gave up). The editor is at Settings, Projects, your project, Insights.

3. 03

   Try it, then save.

   Run your draft on a real session from the last 7 days and read the answers before you save. Each run is a real judgment and uses the project’s daily allowance for Try it.

Example: Northlight Outfitters, our fictional demo shop

Findings finds problems from exact evidence. In our example, it shows that 38 people get stuck at the shipping address, and that the shop rejected a postcode with a space. The team's agent files ticket WEB-231.

An admin adds a signal:

• Question: Did the shop reject a shipping address the person typed?
• Yes when: an address_error event follows the postcode field.
• Watch score %: 20

Try it on a session from this morning gives "Address rejected 92%". After the save, each new judged session gets an answer to this question. When the answer is likely yes, the session shows an "Address rejected" chip, and the answer adds to its Watch score.

Change the question. Keep the past.

Each answer stays tied to the questions that made it. When you edit, the editor says "Unsaved changes · 1 added — applies to sessions judged from now on".

• New wording applies forward. Earlier sessions keep their answers. Watch marks them "judged with an earlier rubric".
• Labels and weights apply at once. This includes sessions that are already judged, because the judge never reads labels and weights.
• Today’s questions on an old session? Open its recording in the replay player. The Insights tab judges it again with your current questions. Its Watch row keeps the first answer.

The docs tell you what each change does and what it resets.

Good to know

• It knows what your events show, and nothing more. The judge always answers. When your events cannot settle a question, a guess can look like a measurement. Signals from 40% to 70% show as "unclear". Below 40%, there is no chip.
• You check how often it is right, on your own sessions. Calibration compares the judge's answers with your team's thumbs up or thumbs down on Worth watching. Today, Worth watching is the only question that collects thumbs. Thumbs never re-judge past sessions.
• What the judge gets. A text list of at most 255 events. It does not get the video or the user id. A third-party service runs the judge.
• Keep emails and ids out of event properties. Background judging removes property names that look personal. Judging from the replay player does not.
• Which sessions are judged. Sessions with 3 or more events that ended in the past 7 days and had no new event for 30 minutes. The project’s daily allowance must have room. The allowance starts again at 00:00 UTC.
• Changes apply to new judgments only. You cannot re-judge earlier sessions in bulk today.
• New wording restarts Calibration for every question. Make related rewordings in one save.
• Try it is a real judgment. Each run uses the daily allowance for Try it. Each person can run it up to 10 times a minute.

Questions