Data quality

Catch bad data before anyone uses it

Attach checks to assets. A blocking check stops bad data from reaching the published layer. A warning check tells you and lets the data through. Records that fail are quarantined, not lost.

Request access Assets in the docs

Checks on assets

Declared next to the data they protect

Checks are part of the asset definition, so they are versioned, reviewed and deployed together with the code. Built-in checks cover the common cases, and custom checks are plain Python functions.

Schema: required fields, types and allowed values.
Freshness: the asset must be updated within a time window.
Uniqueness: keys must not repeat.
Custom: any rule you can express in code, such as totals matching.

pipelines/checks.py

@sy.asset(
    layer="staged",
    checks=[
        sy.checks.schema(required=["org_id", "name"]),
        sy.checks.unique("org_id", severity="error"),
        sy.checks.fresh(within="26h", severity="warn"),
    ],
)
def company_profiles(ctx, raw): # ...

@sy.check(asset=company_profiles, severity="warn")
def address_not_empty(ctx, rows):
    bad = [r for r in rows if not r["address"]]
    return sy.CheckResult(passed=not bad, detail=f"{len(bad)} rows without address")

Blocking or warning

You choose how strict to be

	Blocking check	Warning check
On failure	Run stops before the published layer	Run continues
Failed records	Quarantined in staged	Flagged and kept
Alert	Sent to the on-call channel	Sent to the team channel
Best for	Keys, required fields, types	Freshness, soft ranges, completeness

Quarantine

Bad records wait for a decision

When a check fails, only the failing records are moved to a quarantine area in the staged layer. Good records continue. You can inspect the quarantined rows, see which check rejected them and the original raw payload, then fix the rule or the source and release them.

Quarantined records keep their provenance, so you know where they came from.
Releasing records is a replay-safe operation.
Counts of quarantined records are shown on the asset page.

app.synchronoid.com/acme/assets/company_profiles

Overview

Runs

Assets

Sources

Lineage

Alerts

Settings

Checks for company_profiles

schemaBlockingPassed

unique(org_id)Blocking14 quarantined

fresh(26h)WarningLast 27h

address_not_emptyWarningPassed

Tip. Start with warnings on a new asset. After a week you know which rules are realistic, and you can promote the stable ones to blocking.

See where bad values came from Storage layers Alerts

Ready to stop babysitting pipelines?

Synchronoid is in private beta. Tell us about your data and we will get you a workspace.

Request access Book a demo