Sukat - Data Quality Monitoring with Incidents and Circuit Breakers for a Hospital Group

Dashboard, monitors, monitor detail, incident queue and incident detail. Every one of the 26 demo monitors and 14 incidents opens its own detail page.

Acknowledge and resolve targets per severity: SEV1 15 min and 4 h, SEV2 60 min and 12 h, SEV3 4 h and 48 h. Countdowns turn amber in the last 25% and red when breached.

Monitors compute row counts, null rates, distinct counts and histograms inside the warehouse. The demo data is metadata and aggregates only, and the design shows sample rows only on a DPO-approved, masked and logged request.

26 monitors across 5 types on 17 warehouse tables and models, 14 incidents, 8 data products, 8 dashboards, 3 circuit breakers, 5 hospitals and 11 staff users.

Silent Failures Look Like Results

A feed that stops does not raise an error on a dashboard: lab turnaround simply looks faster and readmissions look lower. The team needed freshness, volume, schema change, null rate and drift checks per hospital, because one Penang feed can fail while the Klang Valley passes.

No View of What a Broken Table Feeds

Nobody could say which models, data products and Power BI or Tableau dashboards a failing table reached, or whether a bad load had already been published to hospital directors, the group board or a Ministry extract.

Health Data in the Debugging Loop

Health information is sensitive personal data under the PDPA 2010, and the 2024 amendments added a data protection officer and breach notification duties. Copying patient rows to debug a load was exactly what the DPO objected to, so the tool itself had to work without them.

Clinical Code Sets That Change

Feeds carry WHO ICD-10, ICD-11, SNOMED CT and LOINC codes, and the Ministry of Health moved its own reporting to ICD-11 from 2024. Checks have to test format and membership in a loaded code set version per feed, without ever judging whether a code is clinically right.