fix(api): order session records by producer time so turns stop interleaving

[mmabrouk]

The problem

A conversation rebuilt from the record log came back with its turns interleaved. A three-turn run reconstructed as user, user, assistant, user: the reply to the first question landed after the second question.

get_records ordered by (created_at, record_index). Two facts make that wrong:

• record_index restarts at 0 on every turn, so it only orders records within a turn.
• The ingest worker batches records from adjacent turns into a single write, so those rows share a created_at.

When both happen together the tiebreak sorts the next turn's first record ahead of the previous turn's later ones. Straight from the table, with turn 2's user record jumping ahead of turn 1's agent record:

record_source | record_index | turn_id  | timestamp     | created_at
user          |            0 | turn-2   | 07.644        | 07.880223
agent         |            3 | turn-1   | 07.544        | 07.880223

Same created_at, so record_index 0 wins over 3, and turn 2's question sorts before turn 1's answer.

This ordering predates the last-message-only work. It did not matter while records fed a viewer. It matters now that the model's context is rebuilt from them.

The fix

Order by the producer-stamped timestamp first. The runner already sets it on every record, and unlike the per-turn index it is monotonic across turns. Ingest time and the per-turn index stay as tiebreaks, and rows written before timestamp existed sort last within their batch rather than jumping the queue.

Testing

Verified against the live table for a three-turn conversation: ordering by producer time reproduces the true conversation order, where the previous ordering did not. Found during the differential QA on #5486.

Known follow-up

This fixes the ordering. It also surfaced a separate issue that is not fixed here: the worker writes a turn's records to Postgres after the next turn has already read them, so the immediately preceding turn's reply can still be missing at reconstruction time. That is a read-after-write question about the ingest pipeline, and it needs its own decision.

[vercel]

The latest updates on your projects. Learn more about Vercel for GitHub.

Project	Deployment	Actions	Updated (UTC)
agenta-documentation	Ready	Preview, Comment	Jul 24, 2026 10:11pm

[coderabbitai]

Important

Review skipped

Auto reviews are disabled on base/target branches other than the default branch.

Please check the settings in the CodeRabbit UI or the .coderabbit.yaml file in this repository. To trigger a single review, invoke the @coderabbitai review command.

⚙️ Run configuration

Configuration used: Organization UI

Review profile: CHILL

Plan: Pro Plus

Run ID: c36b1032-4168-4d89-b5b5-fb8ab107fac0

You can disable this status message by setting the reviews.review_status to false in the CodeRabbit configuration file.

Use the checkbox below for a quick retry:

• 🔍 Trigger review

✨ Finishing Touches

🧪 Generate unit tests (beta)

• Create PR with unit tests
• Commit unit tests in branch fix/sessions-record-ordering

---

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share

• X
• Mastodon
• Reddit
• LinkedIn

_{Comment @coderabbitai help to get the list of available commands.}