[eli-r-ph]

💡 Motivation and Context

A review of the 8.0 migration guide against the v1 code found upgrade risks it did not mention. The biggest one: person processing turned off in a context or in super_options now reaches alias and group_identify, so PostHog drops their $create_alias and $groupidentify events with no on_error report.

Changes

• Every capture timestamp (event timestamp, created_at, PostHog-Request-Timestamp) now always has six fractional digits. isoformat() dropped the fraction on whole seconds, which breaks the at-least-milliseconds rule every v1 SDK follows.
• Warn about the personless context or super_options setting on $create_alias and $groupidentify.
• sync_mode now retries up to 4 attempts in the calling thread (7.x made one).
• capture() does not move $ai_ events to the AI endpoint, and capture_ai with a non-$ai_ name is dropped as misrouted_event.
• An unreadable option value is dropped as invalid_options.
• Smaller fixes: CaptureMode in the removed imports, gzip= as a keyword, reason tags in the loss line, the removed Error in capture: sync line, the 7.x first retry wait, NaN in $set, and when trace ids are added.

💚 How did you test it?

New test_wire_timestamps_keep_fractional_digits_on_whole_seconds covers sync, background and async sends with whole-second clock and caller times; it fails without the fix. Guide lines were checked against the code on v1 and 7.x on main, and the sync_mode attempts were checked with a local stub server.

📝 Checklist

• All commits are signed (see Commit signing)
• I reviewed the submitted code.
• I added tests to verify the changes.
• I updated the docs if needed.
• No breaking change or entry added to the changelog.

If releasing new changes

• Ran sampo add to generate a changeset file

🤖 Agent context

Autonomy: Human-driven (agent-assisted)

• Written with Claude Code (claude-opus-5-5) from a review of the guide against v1 and the capture service. Merges cleanly with fix(ai): treat an empty or whitespace distinct id as missing in AI integrations #1158 and docs: ask users to review before_send hooks for the v1 event shape #1159. CI needs ci: pin Python 3.11 by minor version for Ubuntu 26.04 runners (v1) #1160 first, like the other open PRs.

🤖 Generated with Claude Code

[turnipdabeets]

[suggestion] Targeted mutation tests fails on _rfc3339_timestamp: astimezone(None) survives, likely because the runner is on UTC. Could we add a case with a non-UTC TZ to test_rfc3339_timestamp_always_writes_the_fraction_in_utc? The rest looks good, and we can approve once that's green.

[eli-r-ph]

updated, thank you!