upsert rolls back workflow state

Symptom

"I definitely marked this as processed, and it's back in the list."

Cause

The collector's records carried 'status': 'new'. The collector cron runs every six hours with a merge-duplicates upsert, and it reset every status a human had changed back to the initial value. Already-published items were resurrected as candidates. One more day and a duplicate publish would have gone out.

Rules

• Strip workflow-state fields from collector records right before upsert (r.pop('status', None))
• The consumer owns state, not the collector. The collector writes facts only.

How to recognize it

"I marked it and it came back" → compare against the sync interval.

---

---