Your last incident could train your next expert.

A postmortem is a document, not a rehearsal

Who actually practises the investigation before the next incident?

A static postmortem is written for a reader who already has the timeline. It records the cause and the fix, but not the moment where the evidence was ambiguous. A new engineer cannot test whether they would have noticed the missing successful-payment filter, the wrong join grain or the timestamp tie — they can only read that someone else did.

Reading is not practice, and nothing is graded, so the same misunderstanding can survive into the next incident.

What a guided replay would add

1. Start from a reviewed incident

   A published replay would be derived from an incident postmortem that already passed review — not a story invented for marketing.

2. Follow a labelled transcript

   The story would present a clearly fictional transcript with the timeline, the claims and the evidence, so a learner can tell narration from production data.

3. Answer checkpoints with real queries

   At a checkpoint the learner would run SQL against pinned synthetic fixtures and be graded on the actual result, including the cases that look right but are wrong.

4. Separate watching from playing

   Watch and Play would be different modes. Watching, replaying or seeking would never create completion credit; only a real checkpoint result would.

None of this is playable today. It is the intended design, and the status above says so.

Released today: one standalone graded exercise

Delete duplicates, keep the lowest IDAvailable

A graded challenge over nine pinned fixtures: remove duplicate customer rows, keep the lowest ID, and handle NULL keys safely. The DataTug app grades the SQL you actually run.

Difficulty: Beginner to intermediateEstimated 10 minutesPinned to demo-project commit 2367ff8, challenge revision 1, nine fixtures, SQLite 3.54.1

Open the graded exercise →

This is a graded challenge in the released app, not an Incident Replay story, and it is not Phantom Orders. The lesson Run links elsewhere on the site remain ordinary SQL handoffs.

Private team learning is later

Private team challenges — invitations, rosters, shared progress and comparison — are proposed and unreleased. They need authenticated membership and access rules that do not exist here yet.

There is deliberately no team signup, invitation, roster, share or "add to workspace" control on this page. None is offered as a placeholder, and no anonymous Work item is created or written to.

Synthetic training data, not production evidence

The one released exercise uses a pinned public demo project and nine fixtures at SQLite 3.54.1. The data is synthetic and versioned so a result can be reproduced; the challenge revision and the demo-project commit are fixed in the link, not "latest".

Incident Replay would keep the same boundary: training happens on synthetic or explicitly published sample data, never on a visitor's production records. This page never asks for credentials, a connection string or a private database, and the graded exercise is executed by the DataTug app, not by this static page.

Real incident evidence stays in the normal reviewed private workflow. A training replay would not publish that evidence, and nothing on this page implies that it does.

Interested in piloting Incident Replay?

If a guided replay of your own reviewed incidents would help your team, write to support@sneat.co with the subject “Incident Replay interest”. Tell us what you would want to practise and which data boundary you need.

That is an ordinary email to the existing support address. It starts a conversation; it does not add you to a mailing list, and there is no subscription form on this page.