Monitors + SLOs
Declare watched workloads with cost, freshness, and quality floors
Mission control for running data/ML workloads
Watch everything in the air. Intervene with governance.
Fabric Radar is a governed mission control for running data and ML workloads on Databricks. It watches live pipelines, experiment runs, and model deployments, detects when reality breaches declared SLOs, and makes every intervention — pause a pipeline, cancel a run, quarantine a model — a governed, auditable action.
SLO monitors
Cost, freshness, quality ceilings
Evaluator evidence
Anomalies with proof attached
Governed writes
Every intervention gated + audited
Verified recovery
The loop closes itself
Fabric ecosystem
Shared foundations across the Fabric family
Databricks-native detection
Built around the workloads data teams actually run
Radar extends native Databricks — it does not duplicate it. System tables, MLflow, and Unity Catalog stay the source of telemetry and truth. Radar adds what a workspace does not give you on its own: declared SLOs, an anomaly lifecycle, and a governed intervention path with an audit trail.
Anomaly inbox
Threshold crossings become triageable anomalies with evaluator evidence
Interventions
Pause pipelines, cancel runs, quarantine models — all governed
Runs ledger
A platform event log projection links every run to what shipped it
The loop
Declare, observe, detect, triage, intervene, verify
Every state transition is a Fabric Platform action invocation. Console clicks, agent tools, schedules, and webhooks all enter through the same path: actor, action, policy, state machine, handler, adapter, event, projection. There is no side door for writes.
Read the architectureDeclare and observe
A monitor declares a workload and its SLOs as a governed action. Telemetry flows append-only into Lakebase metrics tables — no actions involved.
Detect and triage
Fabric Experiments evaluators over system tables and MLflow cross a threshold and raise an anomaly with evidence. Humans or a finite triage agent work the inbox.
Intervene and verify
An intervention is a governed Platform action with blast-radius policy gates. An evaluator confirms recovery and the anomaly resolves.
The moat
Governance under every mutation, evidence under every signal
Indie mission-control dashboards can show you a red pipeline. Radar can act on it — because the Fabric governance pipeline sits under every write, and Databricks-native detection sits under every signal.
See the anomaly lifecycle01
Every intervention governed
Pause, cancel, and quarantine are Platform actions with policy and a state machine.
02
Evidence on every anomaly
Threshold crossings carry the evaluator verdict that raised them.
03
Blast-radius gates
Intervention policy scopes how much a single action can touch.
04
Agents are first-class
A Harness triage agent gets read-only tools plus one governed mutation tool.
The Fabric family
Radar watches everything in the air.
Harness builds the agents. Experiments judges them. Runway ships them. Tower flies the missions. Radar watches everything in the air. Platform keeps everyone honest. Radar owns the monitoring domain and delegates the rest to the products that own them.
Fabric Platform
Governs every intervention: durable invocations, policy evaluation, the state machine, and the immutable event log.
Fabric Harness
Owns Databricks transport and durable workflows: quarantine via Unity Catalog, intervention workflows via Temporal.
Fabric Experiments
Owns detection: evaluators over system tables and MLflow whose threshold crossings raise anomalies with evidence.
Start watching
Declare a monitor over a live workload today
Point Radar at a running pipeline or model deployment, declare its SLOs as a governed action, and let the detection loop start watching.
Fabric Radar is built and supported by TechFabric.
Contact TechFabric