Accuracy
The agent measured: 7 replayed days, Sat to Fri. Measured against 249 injected incidents (incidents_truth) between the agent's first and last tick, Sat 19:00 to Fri 23:01 IST, 7 replayed days. Incidents from before it was switched on are not counted. A detection matches a planted incident of a kind it reports, on the same store, merchant or integration, within 20 minutes.Precision0.972680 detections
Recall1.000of 249 injected
Resolved automatically94.2%auto rate, 7-day mean
Reverted by people0actions a person undid
Cost per resolved incidentINR 0.00395 resolved · INR 0.00 model spend
Time to resolve4 min 10 smean, all resolved · stuck orders to the re-send 56 s · 46 min 17 s when a person decides
Precision per detector, per day
the 6 busiest detectors; the rest are in the table belowReverted actions
who undid the agent, and why| When (IST) | Store | Kind and call | By | Why | |
|---|---|---|---|---|---|
| No reverts in this window. | |||||
Drift alarms
precision down 15 pts week on week| Fri | green | all detectors | Week-on-week alarms need 14 replayed days; the ledger has 7. waiting for data |
| Fri | green | model | No drift: answers usable, confidence steady, few reverts. recal.drift |
Per detector, 7 days
precision: of what it flagged, how much was a planted incident| Detector | Detections | Precision | Reports |
|---|---|---|---|
| acceptance_risk | 19 | 0.632 | acceptance_risk |
| aggregator_outage | 3 | 1.000 | aggregator_outage |
| churn_risk | 36 | 1.000 | churn_signal |
| courier_no_show | 55 | 0.982 | courier_no_show |
| duplicate_order | 11 | 1.000 | duplicate_orders |
| failing_integration | 34 | 0.941 | integration_version_regression, pos_auth_expired |
| menu_drift | 18 | 1.000 | menu_sync_drift |
| offline_printer | 31 | 1.000 | printer_offline |
| overdue_invoice | 37 | 1.000 | invoice_overdue_45 |
| overloaded_branch | 80 | 0.900 | branch_overload |
| pos_link_down | 325 | 1.000 | pos_auth_expired |
| pos_token_expiring | 14 | 1.000 | pos_token_expiring |
| release_regression | 2 | 0.500 | integration_version_regression |
| stuck_order | 15 | 1.000 | pos_auth_expired |
Planted incidents, by kind
recall: of what was planted, how much it found; the counts add up to 249| Planted kind | Planted | Found | Recall | Reported by |
|---|---|---|---|---|
| courier_no_show | 54 | 54 | 1.000 | courier_no_show |
| branch_overload | 40 | 40 | 1.000 | overloaded_branch |
| invoice_overdue_45 | 37 | 37 | 1.000 | overdue_invoice |
| churn_signal | 35 | 35 | 1.000 | churn_risk |
| pos_auth_expired | 21 | 21 | 1.000 | failing_integration, pos_link_down, stuck_order |
| menu_sync_drift | 18 | 18 | 1.000 | menu_drift |
| pos_token_expiring | 14 | 14 | 1.000 | pos_token_expiring |
| printer_offline | 14 | 14 | 1.000 | offline_printer |
| acceptance_risk | 12 | 12 | 1.000 | acceptance_risk |
| aggregator_outage | 2 | 2 | 1.000 | aggregator_outage |
| duplicate_orders | 1 | 1 | 1.000 | duplicate_order |
| integration_version_regression | 1 | 1 | 1.000 | failing_integration, release_regression |
| total | 249 | 249 | 1.000 |