Validation

Grounded in your actual data, not documentation.

ClariLayer probes your live warehouse to show whether metric logic compiles, reaches the expected sources, and survives bounded data checks before it hits a dashboard or AI agent.

Start Free
ClariLayerValidation EngineIntegrity probes · 30-day window · Tier gatesWAREHOUSESnowflake · Databricks · BigQueryVALIDATION RESULTSCompilationpassNull checkpassUniquenesspassRow countwarn

Documentation drifts. Your warehouse doesn't.

The wiki says

Catalogs and wikis document what should be true. That documentation is useful, but it can drift from the actual SQL, schemas, and data behavior running in the warehouse.

The warehouse shows

Upstream schema changes, column renames, and data type shifts break metric logic without warning. No one catches it until the board meeting.

ClariLayer proves

We execute metric logic against your actual data within a bounded test window. When checks run, validation evidence is attached to the metric version. No guessing.

How validation works

01

Integrity probes

Automated checks for source availability, SQL compilation, null values, uniqueness assumptions, and data type mismatches. Catches the obvious failures before anyone sees them.

02

Bounded testing

Validation is restricted to a 30-day data window to keep costs low and execution fast. Enough to prove the logic, not enough to break your budget.

03

Hard gates

Financial-tier metrics cannot be promoted without the required validation evidence. Experimental metrics get lighter checks. The rigor matches the stakes.

04

Validation evidence

Each validated metric version carries its validation report. Auditors, AI agents, and team members can see when it was last checked against real data.

Evidence Surface

Validation is the bridge between stated intent and warehouse execution.

A metric can sound right in a meeting and still fail when it meets current schemas, null behavior, or data type changes. ClariLayer records validation evidence alongside the definition so reviewers, auditors, and AI consumers can inspect the proof instead of trusting a stale description.

The evidence is intentionally attached to the metric version, not left in a one-off test log. When a definition is released, later consumers can see which checks ran, what warehouse surface was used, and whether the metric still needs additional review before it can support higher-stakes decisions.

Shared warehouse path

Catalog browse, validation probes, direct deploy, and rollback are shared across Databricks, Snowflake, and BigQuery. Observe/query-history ingestion remains Databricks-only today.

Bounded proof, not open-ended spend

Validation focuses on targeted checks: SQL compilation, source availability, basic integrity assumptions, and bounded data windows. The goal is enough evidence to govern the metric without turning every edit into a warehouse bill surprise.

Databricks-only observe caveat

Validation is broader than Observe. Snowflake and BigQuery share the catalog, validation, deploy, and rollback paths, but they do not yet share the live query-history loop.

Trust signal for consumers

Validation status gives BI users and AI agents a concrete reason to treat one metric differently from another: proven, unverified, stale, or blocked from promotion until the evidence is refreshed.

Validation becomes most powerful when it is paired with approval workflows that match metric risk and the Metric Registry that exposes current trust signals.

Key capabilities

Live warehouse probing

Validates metric logic against your actual Databricks, Snowflake, or BigQuery data. Not assumptions, not documentation — bounded execution against real tables.

Automated integrity checks

Probes for source availability, SQL compilation, null behavior, uniqueness assumptions, and data type mismatches. Catches problems before they reach production.

Tier-gated validation requirements

Experimental metrics can ship with lighter checks. Financial-tier metrics cannot be promoted without a passing validation report. Different rigor for different stakes.

Platform-neutral

Catalog browse, validation probes, direct deploy, and rollback are shared across Databricks, Snowflake, and BigQuery. Observe/query-history ingestion remains Databricks-only today.

Ready to close the context gap?

Join the companies building a trusted context layer for their AI agents and business teams.

Start Free