From a hunch to a number, in one loop.

Session replay and analytics

Walk into the roadmap review with a number.

Thursday: your roadmap review starts in ten minutes. Someone will ask a question. Did the checkout change from the last sprint make a difference?

Without Sequential, you have a dashboard that says traffic is up. You have forty recordings that nobody watched, and a ticket marked Done. You say "I think so".

With Sequential, the problem was counted days ago. Your agent filed it, with the evidence. After the release, Sequential counted the page again. You walk in with one line.

• Example: Northlight Outfitters, our fictional demo shop. Measured after the release: 41% → 12% of sessions on the shipping page.

That is a measurement, not a verdict on your code. And you did not spend the week to get it.

How Sequential finds problems

Findings come from fixed rules over the events that your SDK already sends. There are five kinds of problem: exceptions, failed requests, dead clicks, rage clicks and dead ends. Sequential counts them exactly, every 15 minutes.

Separately, a judge reads a short text summary of a session. It never watches the video. It answers your questions with a likelihood. Watch uses these answers to rank the sessions that are worth a look. Ask uses them to find sessions from one plain sentence.

Who does what

• Sequential finds, counts and measures. It finds problems and counts who they hit. It records the releases that you send it. It measures before and after, and it answers questions.
• Your agent acts in your tools. This is the coding assistant that you already use, such as Claude Code or Cursor. It files the ticket and opens the pull request, with only the access that you give it. Sequential connects to none of your tools.
• You decide. You review, approve and merge.

The loop, in four beats

1. 01

   Find: see where people get stuck.

   A JavaScript error or a server failure becomes a finding from the first session it happens in. Other problems become a finding when they repeat: at least 3 sessions from at least 2 people in the last 7 days. Each finding has sample sessions to replay. Northlight: one failed request on the shipping page, in 52 sessions from 38 people, is finding F-12. A replay shows why: the form rejects a postcode with a space.

2. 02

   Hand off: the evidence goes with the work.

   Every finding has a fix brief. It tells what users hit, where, the evidence, since when, and how Sequential will check. Your agent files it, or you copy it. Northlight: the agent filed WEB-231 in Jira, with three sessions attached. A coding agent opened a pull request: "Accept spaces in postcodes".

3. 03

   Ship: one call in your deploy starts the count.

   A person merges. Your deploy, or your agent, records the release and names the finding.

4. 04

   Measure: the same page, counted again.

   About every hour, Sequential compares the page of the finding before and after the release. It gives a range for each rate. It states what changed, never why. The finding moves through these stages: New, Handed off, Fix linked, Fix released, Measuring, then a result such as Improved. Northlight, a week later: "After 2026.10.09, 10 of 82 sessions had it (7–21%). Before, 123 of 300 (36–47%)."

Your part: an SDK, and one call in your deploy

Add sequential.js. It is one tag for analytics and session replay.

Or keep a supported analytics SDK that you already ship. Change a key and a host to point it at Sequential. You do not fork it, and you do not instrument again. With a third-party SDK, web replays need one more script line.

Then add the release call to your deploy, or let your agent make it. Sequential works with websites and with iOS and Android apps. iOS is tested with a real app on the simulator. Android is not tested with a real app yet.

Where people stay in control

• Tokens. Only owners and admins connect agents. A token covers one project and the permissions that you pick. Then it expires.
• Approvals. An agent can ask for a fix draft. Nothing is sent until an owner or an admin approves it. No token can dismiss a finding or remove a fix link.
• Your users. Web replays are masked on the visitor's device. The Strict preset is the default.

Good to know

• It measures. It does not explain. A result states what the rate did. It gives ranges and notes on other changes that it can see. It cannot see all of them.
• A release must name the finding. A merge, a ticket or a fix link alone does not start the count.
• Quiet pages take longer. A finding can stay at Measuring until there are enough sessions to tell.
• The hosted preview keeps 7 days of data by default. A before-and-after window can use only that much data.
• Not every session is judged. A session needs at least 3 meaningful events, within a daily allowance.
• Ask finds. It does not count. A short or empty answer does not mean that nothing happened.
• Browser and supported mobile SDKs only. Sequential refuses events from a server or an HTTP library as bot traffic.

Questions