Benefits

Simple by design. Stays in your Fabric.

You do not learn a new observability product UI. You grant workspace access — results appear in your Lakehouse and the deployed Power BI report.

Radically simple to operate Two setup steps, then scheduled scans. No YAML rule catalogs, no PK/FK definitions, no Spark notebooks for analysis.
Nothing leaves customer Azure / Fabric Results write back to the Lakehouse you choose. The report runs Direct Lake in your tenant — no external data plane.
No new tool to learn Analysts stay in Fabric + Power BI. You only add service principal access to monitored workspaces and the output Lakehouse.
Marketplace-priced $49 / container / month for the software offer — without an enterprise observability contract. Billed through your existing Azure subscription, so you do not need to onboard a new vendor. Get it on Marketplace.

OneLake Guard Analysis chart of anomalies over time — Same Fabric + Power BI UX — Analysis charts included

Detectors

What we detect — and why it matters

Zero-config detectors over OneLake Delta tables. Built for trust signals and Direct Lake health.

Detector	What it catches	Why detect it
`NEW_TABLE`	Table appeared that was not in the previous completed scan	Spot unexpected creates or new landing zones early
`DELETED_TABLE`	Previously monitored table is no longer discoverable	Catch drops, renames, moved paths, or lost Viewer access
`SCHEMA_DRIFT`	Column add/remove/type changes between Delta versions	Stops silent breaks in reports, Direct Lake models, and jobs
`VOLUME_SPIKE`	Daily row volume unusually high vs baseline (Z-score)	Flags backfills, duplicate loads, or upstream surges
`VOLUME_DROP`	Daily row volume unusually low vs baseline (Z-score)	Often failed/partial ETL or a silent pipeline stop
`NULL_RATE_SPIKE`	Share of nulls in a column jumped vs recent history	Missing joins, broken mappings, source quality regression
`VALUE_RANGE_SHIFT`	Column min/max moved outside the seen envelope	Unit/currency changes, bad casts, new outlier populations
`MEAN_SHIFT`	Column mean shifted vs baseline (light-scan)	Business metric drift or ETL scaling/unit issues
`DISTINCT_SHIFT`	Approximate distinct count shifted vs baseline	Cardinality explosion or collapse (broken keys / duplicates)
`STALE_TABLE`	No commits longer than the stale threshold	Abandoned table, stopped pipeline, or delayed landing
`COMMIT_STORM`	Unusually many Delta commits in one day	Micro-batching / row-by-row writes → latency & small files
`LARGE_REWRITE`	Single commit rewrote/deleted a large number of rows	Expensive overwrite/MERGE patterns; storage & compute cost
`SMALL_FILE_BLOAT`	Too many tiny Parquet files relative to total files	Protects Direct Lake latency and OneLake list/read cost
`STORAGE_BLOAT`	Large tombstone / unreferenced data past VACUUM retention	You pay for GB queries no longer use until VACUUM
`VACUUM_OVERDUE`	Removed files older than retention still present	Maintenance debt — schedule VACUUM to reclaim space

Anomaly list in OneLake Guard Overview report — Overview — schema drift, volume drop, small-file and storage bloat

OneLake Guard Help page listing anomaly detector types — Help — every detector explained inside the report

Two steps to turn it on

Service principal + workspace access — then automatic scans in Fabric.

Open Marketplace How to