Data health
Whether the dashboards can be trusted right now. Data counts as current for 2 days, late up to 7, and stale after that — marketplaces publish a day behind, so a one-day gap is normal.
Every tile opens what is behind it — the matching section below, or the mapping inbox.
Freshness by marketplace
Measured against the newest business day we hold, not the newest import: a connector that succeeds hourly while the marketplace serves old figures is still stale.
| Marketplace | Latest sales day | Freshness | Last successful import | Live fact rows |
|---|---|---|---|---|
| blinkit | 28 Jul 2026 | stale10 days behind | 7 Aug 2026, 7:49 pm1 run | 646 |
| zepto | 27 Jul 2026 | stale11 days behind | 7 Aug 2026, 7:49 pm1 run | 867 |
Unattributed revenue
Sales we recorded but could not tie to a SKU. The money is in the totals and missing from every per-SKU view, which is why this queue is worked by value.
Failed and stuck runs
0 failed, 0 stuck. A run is stuck when it has not changed state for 30 minutes — the tick advances one transition a minute, so anything older has lost its worker rather than being busy.
Rejected rows, by cause
Grouped by code because the shape is the diagnosis: one code dominating means a parser needs fixing, a spread across codes usually means one bad file.
Coverage and the rollup queue
Facts and their dirty markers are written in one transaction, so a backlog that grows means the tick has stopped draining — the facts are right and every dashboard is quietly showing pre-import numbers. Warn at 15 minutes old, bad at 120.