Skip to main content

Glossary

Terminology used throughout PLACEHOLDER Cloud. Definitions match what the code calls things — when the docs say "asset" we mean the DataAsset row, not "any data in your warehouse."

Datasource​

One logical place data lives — a Postgres database, a Snowflake account, an S3 prefix. Backed by the Datasource table. Holds connection parameters (encrypted) and an execution_mode (direct or agent).

Data asset​

One table, view, query, or parquet dataset inside a datasource. Backed by the DataAsset table. Rule suites and profile runs are scoped to a data asset.

Rule suite​

A named set of expectations that PLACEHOLDER Cloud runs against a data asset. Backed by the RuleSuite table; stored as a portable Great-Expectations-compatible JSON blob. The unit of authoring.

Expectation​

A declarative, single-assertion check — "column email has no nulls", "row count is between 100 and 10,000,000". Lives inside a rule suite as an entry in RuleSuite.suite.rules. PLACEHOLDER Cloud ships a curated set of expectation types and (planned) supports custom SQL + plugin classes.

Checkpoint​

The unit of execution. Bundles one or more (asset, rule-suite) pairs and runs them as a single job, either manually or on a cron schedule. Backed by the Checkpoint table.

Validation result​

What a checkpoint run produces. One row per (asset, rule-suite) inside the checkpoint. Backed by the ValidationResult table. Holds a per-expectation pass/fail outcome, failing-row samples where available, and a top-level success boolean.

Profile run​

A statistical snapshot of a data asset — row count, schema hash, per-column stats. Backed by the ProfileRun table. Feeds anomaly detection and AI rule suggestions.

Anomaly​

A statistical deviation in a tracked metric (row count, null percent, schema hash) relative to a rolling baseline of profile runs. Backed by the AnomalyEvent table. Surfaced in the Anomalies route with severity, baseline / observed values, and an AI explain panel.

Alert rule​

The configuration for "when X happens, tell Y." Backed by the AlertRule table. Has a scope (checkpoint or workspace-wide), a trigger (on_failure / on_anomaly / on_success / always), a channel (email / slack / teams / webhook / pagerduty / jira / linear), and a destination (channel-specific config, sensitive fields encrypted under the per-tenant DEK).

Alert event​

One occurrence of a fired alert rule — one row per dispatched message. Backed by the AlertEvent table. Has a status (pending / sent / failed), a sent_at, and an error field for failed delivery.

Agent token​

A dqa_… opaque bearer token used by an on-prem DQ Agent to register itself with PLACEHOLDER Cloud and execute checkpoints inside a customer network. Backed by the AgentToken row in the public schema. Scoped to one organisation and (optionally) one workspace. Revocable.

Workspace​

A logical grouping of datasources, assets, rules, and members inside an organisation. Backed by the Workspace table. Used to separate environments (production, staging) or departments (marketing, finance) inside one org.

Organisation​

The top-level tenancy boundary. Backed by the Organisation table in the public schema. Every org has its own tenant_<org_id> schema for its data.

LLM config​

Per-organisation Bring-Your-Own-LLM configuration. Backed by the LLMConfig row in public. Holds provider (openai / anthropic / ollama / azure_openai / custom), model identifier, encrypted API key, and optional base URL.

DQ Agent​

A separate, on-prem process that registers datasources with PLACEHOLDER Cloud and executes validation work inside the customer network — so the customer's data connection string never leaves their environment. Authenticated by a dqa_… agent token.

Batch definition​

A named slice of a data asset that scopes a validation run to one partition, interval, or the whole table. Backed by the BatchDefinition table (per-tenant); referenced by ValidationConfig.batch_definition_id. Three kinds: whole_asset, column_partition, interval. See Batch definitions.

Asset dependency (lineage edge)​

A directed edge in the lineage graph — either between two DataAsset rows in the same workspace, or between a DataAsset and an external system (dbt, airflow, external). Backed by the AssetDependency table. Carries a source (manual / dbt) so re-ingest can purge its own rows without touching human-typed edges. See Lineage.

Custom expectation​

A great_expectations.Expectation subclass discovered from a Python package listed in EXPECTATIONS_PLUGIN_PACKAGES. Backed by the CustomExpectation row in public. Surfaced in the rule-suite editor and GET /api/v1/organizations/{org_id}/custom-expectations (FEATURE-35).

Identity provider​

A per-organisation SAML or OIDC configuration backing single sign-on. Backed by the IdentityProvider row in public. Status is one of enabled / enforced / disabled; enforced disables password login for the org. See Configure SSO.

Invite​

A dqi_… one-time credential that grants a recipient access to an organisation (or one of its workspaces) when redeemed against POST /auth/accept-invite. Backed by the Invite row in public. Expires after INVITE_EXPIRY_DAYS (FEATURE-38).

Audit event​

A single row recording a security-relevant action — successful and failed logins, logout, tenant provisioning, agent-token mint or revoke, LLM-config / datasource / alert-rule / rule-suite / data-asset writes and deletes, AI suggestion bulk actions, catalog-sync or data-docs config changes. Backed by the AuditEvent row in public. Queryable via GET /api/v1/organizations/{org_id}/audit-events, optionally scoped to one resource via ?resource_type=...&resource_id=... (HARDEN-21, FEATURE-8). The audit tables are append-only at the database layer: a trigger blocks any UPDATE, and DELETE is permitted only for the dated retention prune — so a compromised app credential can't silently rewrite or erase the trail.