feat(wizard): name the workspace group, and stamp events on capture

[itelo]

Two fixes found by reading a real run's events in PostHog, both in analytics.ts.

1. The workspace group had no name

Events already join a workspace group, but nothing ever called groupIdentify, so every workspace showed as a bare UUID in exactly the breakdowns the funnel dashboard is built on:

workspace_id                          runs
1579fa6a-1496-4c15-a952-b77b57633995     7

Now, when a run settles on a workspace:

client.groupIdentify({
  groupType: 'workspace',
  groupKey: workspace.workspace_id,
  distinctId: current.distinct_id,
  properties: { name: workspace.name, is_sandbox: workspace.is_sandbox },
})

Two deliberate choices:

• Only name and is_sandbox. Group properties are shared with everything else that reports to that workspace (the Console, webviews, server events), so the wizard writes only what it knows first-hand and doesn't stomp on anyone else's fields.
• distinctId is the install's own id. Without it, PostHog mints a synthetic person for the workspace as a side effect of the identify.

This also settles the "how many times has each workspace run the wizard" question at the group level rather than needing a property breakdown — wizard_connected broken down by workspace now reads as a name.

2. Events could be stamped out of order

The SDK assigns a timestamp inside an async path (prepareEventMessage), so two events captured in the same tick can be stamped out of order — and PostHog orders funnel steps by timestamp. A real run reported:

screen_index	screen	timestamp
1	init	18:27:17.181
2	method	18:27:17.045

The wizard's own ordering was right (screen_index proves it) and the stamps were inverted. That's enough to make an ordered screens funnel read a step as skipped. track now passes timestamp: new Date(), taken where the event is captured.

Testing

• $groupidentify payload asserted against what the SDK actually sends ($group_type, $group_key, $group_set, and the install's distinct id) — verified first against a local ingest server rather than guessed.
• A test pinning that each event's stamp falls inside the window of its capture call, and that capture order is timestamp order. It documents the invariant the funnels depend on; it can't reproduce the race deterministically, which is worth knowing when reading it.
• One existing test selected events positionally and now picks them by name, since $groupidentify lands between the two it was checking.

npm run typecheck, npm run lint and npm test (199 tests) pass.

Note

Nothing here changes distinct_id, which stays the anonymous install. Making it the workspace was considered and rejected: the pre-connect events — wizard_run_started, the welcome/init screens, wizard_init_finished, wizard_connect_method_selected, wizard_connect_failed — happen before any workspace is known, and those are the drop-off the funnel exists to measure. Switching mid-run would split a single run across two persons and break every person-scoped funnel; alias would fix that but merges irreversibly, so one machine connecting to a sandbox and then a production workspace would permanently merge those two workspaces.

🤖 Generated with Claude Code