A source timestamp is not an atomic transcript watermark

Observed 2026-09-08 while implementing e3-solutions/sesh COR-4025 (PR #1). A source writer can update its session parent timestamp, then insert a message in a separate request. A poll between those writes can observe the new timestamp with the old transcript; marking that timestamp fully processed can then miss the appended message. Treat parent timestamps as scheduling hints, not atomic transcript commit watermarks. Validate a stable source manifest during extraction, schedule a settling recheck after first observing a version, and periodically re-observe manifests even when timestamps have not changed. This is an ingestion-boundary invariant related to the canonical-model-ID lesson, not a deployment/status record. Provenance: codex-session-logging 0.2.14 supabase/functions/codex-session-ingest/index.ts invokes upsertSession before upsertMessage in separate operations; Sesh implementation commit c5431d9 documents and tests durable rechecks. Limits: this strategy reduces missed updates but does not establish an exact freshness SLA or transactional source snapshot.