Phase 1 — Confirm what actually happened
Pull the stored SEM-O raw payload for meter 8131 around 2026-02-11 (same window as the trace in the ticket).
In that raw data, check whether two rows share the same StartTime (duplicate interval in one response vs a single row).
Optionally verify meter metadata in DB (granularity, expected interval type, resource name) so odd config is ruled out, even though identical starts usually point at duplicate rows, not granularity.