Skip to main content

​
A

Access Token A credential used to authenticate programmatic access to Prizm. Prizm supports two token types: API tokens (for REST API calls) and MCP tokens (for AI tool access via the Model Context Protocol). Tokens are scoped to specific tool sets and expire based on the configured TTL.
Agent A lightweight process deployed inside customer infrastructure that connects outbound to Prizm SaaS over WSS/443. Each data source type has its own agent (Snowflake, Databricks, dbt, Tableau, Spark). Agents never receive inbound connections — they subscribe to NATS JetStream and execute tool calls on behalf of the Prizm platform.
Alert An automated notification generated when a monitored metric value breaches its expected threshold range. Alerts are produced by the Observability job and classified by priority: LOW, MEDIUM, HIGH, or CRITICAL, based on how many standard deviations the value falls outside the normal band. Alerts are the raw signals; one or more related alerts may be grouped into an Issue.
Anomaly Detection The engine that establishes an adaptive, statistical baseline for each observability signal (volume, freshness, schema) and flags values that fall outside the expected range. Anomaly Detection converts raw Observability Job signals into Alerts, using standard-deviation bands rather than fixed thresholds.
Application A semantic grouping used to tag an asset with the business application or system it supports — for example, Order Management or Billing. Applications are assigned in the Semantic Context section alongside Domain and Product, and can be used to scope search, dashboards, and alert routing.
Asset Any data object that Prizm has discovered and cataloged from a connected source. Assets include tables, views, columns (attributes), dashboards, pipelines, and more, depending on the connector. Each asset gets its own detail page showing quality score, metrics, lineage, usage, and context.
Attribute An individual column within a table or view. Prizm profiles each attribute to compute null rate, distinct count, distribution, and pattern statistics. Attributes have their own quality scores and can have metrics defined against them independently.

​
B

Bastion VM A jump host VM deployed inside the same VNet or VPC as a private Kubernetes cluster. All kubectl, helm, and cloud CLI commands are executed from the Bastion VM — it is the only machine with network access to the private cluster API server.
Behavioral Metric A business metric type that tracks how values change over time — for example, week-over-week growth in order count. Behavioral metrics alert when trends deviate from expected patterns rather than when a single value breaches a threshold.

​
C

Catalog Job A scheduled job (default: daily) that discovers structural metadata from a connected source — databases, schemas, tables, views, columns, tags, functions, procedures, streams, and pipes. The Catalog job populates the asset inventory in Prizm.
Category A classification label used to group similar assets or glossary terms for easier navigation and reporting — for example, grouping tables under “Transactional Data” or terms under “Revenue Metrics.” Categories are managed independently of Tags and Domains and are primarily used for organizing the catalog.
Connector The integration that lets Prizm connect to and run jobs against a specific data platform (e.g., Snowflake, Databricks, Postgres). Each connector defines which job types (Catalog, Profile, Observability, Performance) are supported and what credentials or roles are required.
Context Completeness A score (0–100%) shown in the Context panel of an asset detail page. It measures how completely the asset’s context has been filled in — covering description, owner, glossary terms, domain, and audience. 100% means all context fields are verified.
Context Panel The right-hand panel on an asset detail page that shows an AI-generated narrative summary of the asset, its context completeness score, and freshness timestamps for when the description, terms, and ownership were last verified.
Conditional Metric A business metric type that evaluates a rule only when a filter condition is true — for example, checking that all orders with status = 'shipped' have a non-null tracking number.
Criticality A classification assigned to an asset (e.g., CRITICAL, HIGH, MEDIUM, LOW) that controls profiling depth, alerting sensitivity, and priority in issue triaging. Critical assets receive more thorough monitoring and alert more aggressively on deviations.

​
D

Dashboard A saved, shareable view in Prizm that surfaces metrics, scores, and trends across a chosen set of assets, domains, or products. Dashboards can be built from built-in widgets or custom queries and are commonly used for executive or team-level reporting.
Data Product A curated bundle of assets, metrics, and documentation packaged around a specific business use case or consumer — for example, a “Customer 360” data product spanning CRM, billing, and support data. Data Products give consumers a single trustworthy entry point instead of requiring them to discover and assemble individual assets themselves.
Domain A business grouping used to classify assets by ownership or subject area (e.g., Customer Insights, Sales, Finance). Domains are assigned in the Semantic Context section of an asset and are used to filter, scope, and route alerts and issues.
Distribution Metric A structural metric that tracks the statistical distribution of values in a column over time — detecting shifts in mean, percentiles, or value frequency that may indicate data quality degradation.
Docker Swarm The container orchestration mode used by Prizm’s self-hosted deployment. install.sh initializes Swarm automatically and deploys all Prizm services as a single stack (docker stack deploy). Not the same as Kubernetes.
Drift A status indicating how far an asset’s current data has deviated from its historical baseline, shown on alerts as Low, Medium, or High. Drift is calculated by comparing current values against established patterns for a column or measure and complements the point-in-time Quality Score by showing directional change over time.

​
E

.env File The primary configuration file for Prizm’s self-hosted Docker deployment. It controls registry credentials, domain names, secrets, image tags, and runtime parameters for the stack. Populated from .env.example before running install.sh.

​
F

Fingerprint Column The column Prizm uses to detect when an asset’s data was last changed — typically a timestamp field such as updated_at or transaction_date. The fingerprint column feeds the Freshness signal and is configured per asset during onboarding.
Freshness An observability signal measuring the time elapsed since a table or view was last modified (based on LAST_ALTERED in Snowflake). Prizm tracks freshness over time and alerts when a table goes stale beyond a configured SLA window.
Frequency Metric A structural metric that tracks how often specific values appear in a column. Alerts when the proportion of a value changes beyond the expected range — for example, if a status column that was 90% ‘active’ drops to 60%.

​
G

Glossary (Business Glossary) The centralized library of business terms and their approved definitions — for example, Customer or Net Revenue. Glossary terms can be linked directly to assets and attributes so the same business meaning is visible everywhere the data appears.
Governance The set of Prizm capabilities — glossary, terms, domains, tags, ownership, and policies — that keep data assets accurately defined, classified, and accountable over time. Governance entities can be extracted or recommended automatically from source metadata and documentation, then reviewed and approved by a steward.

​
H

Helm A Kubernetes package manager used for the AKS (Azure) agent deployment. Prizm provides a Helm chart (customer-agents) that packages all agent deployments with a shared values.yaml configuration. Helm manages release versioning, upgrades, and rollbacks.
Hybrid Agent Outbound Prizm’s standard deployment architecture. The Prizm SaaS platform is hosted on Azure. Customer-side agents are deployed inside customer infrastructure (cloud or on-premises) and connect outbound to Prizm SaaS over WSS/443. No inbound connections to customer infrastructure are required.

​
I

Impact Analysis The forward-looking use of Lineage to identify every downstream asset, report, or pipeline affected by a change or failure at a given asset. Impact Analysis is the counterpart to root-cause investigation, which looks upstream instead.
Issue A validated, actionable problem created from one or more related alerts. Unlike an alert (which is an automated signal), an issue is a tracked work item that can be assigned to an owner, commented on, and resolved. Issues are created when alerts meet priority or correlation criteria — for example, when multiple signals fire simultaneously on a critical asset.

​
J

Job A unit of work that Prizm runs against a connector on a schedule or on demand. The four main job types for Snowflake are: Catalog, Profile (Quality), Observability, and Performance. Each job type collects a different category of metadata.

​
L

Lineage A directed graph showing the upstream sources and downstream consumers of a data asset. For Snowflake, Prizm derives lineage from the OBJECT_DEPENDENCIES view and the GET_LINEAGE API, both of which require Snowflake Enterprise edition or above — on Standard edition, no lineage is produced. Lineage is visualized in the Lineage tab of each asset detail page.
Lookup Metric A reconciliation metric type that verifies referential integrity by checking whether values in one column exist in a reference dataset or lookup table.

​
M

MCP (Model Context Protocol) An open protocol that allows AI assistants (such as Claude) to call tools exposed by Prizm. Prizm implements an MCP server that exposes data quality, lineage, metric, and governance tools. Access is authenticated via MCP access tokens.
MCP Access Token A specialized JWT token with token_type: mcp_authorization used to authenticate AI clients connecting to Prizm via the Model Context Protocol. Configured as a mcpServers JSON block in the AI client — not as a bare bearer token.
Measure A specific, configured check applied to an asset or attribute — for example, “null rate on email must stay below 5%.” Measures are what generate Alerts when their threshold is breached, and one or more measures roll up into an asset’s overall Quality Score.
Metric A defined quality or observability check applied to an asset or attribute. Metrics can be built-in (distribution, freshness, volume, null rate) or custom (user-defined SQL rules). Each metric produces a score and can fire an alert when its value breaches the configured threshold.

​
N

NATS / NATS JetStream The messaging backbone used for communication between Prizm SaaS and deployed agents. Agents subscribe to NATS JetStream subjects and receive tool call messages. NATS provides at-least-once delivery, message retention, and redelivery on failure. All NATS communication uses WSS (WebSocket Secure) over port 443.
NATS Environment The tenant identifier (NATS_ENVIRONMENT) that scopes all NATS message routing to a specific Prizm tenant. Agents and the Prizm platform must share the same environment value for tool calls to route correctly.

​
O

Observability Job A scheduled job (default: hourly) that collects time-series signals from connected sources — volume (row count), freshness (last altered time), and schema change (DDL diff). The Observability job feeds the anomaly detection engine, which computes adaptive thresholds and fires alerts on deviations.

​
P

Performance Job A scheduled job (default: daily) that queries ACCOUNT_USAGE in Snowflake to collect query execution times, warehouse compute credits, storage costs, and access history. Requires IMPORTED PRIVILEGES on the SNOWFLAKE database.
Pipeline An asset type representing a data pipeline job that moves or transforms data between sources, tables, or systems. Prizm monitors pipeline run status and can surface it as Active, Running, or Paused, and links pipelines into Lineage graphs.
PRIZM_ROLE The recommended Snowflake role to create for the Prizm service account. It is granted the minimum permissions required to run all four job types. The name is a convention — any role with equivalent grants will work.
Pattern Metric A structural metric that validates whether values in a column conform to an expected format — such as email addresses, phone numbers, date strings, or custom regex patterns.
Profile Job A scheduled or manually triggered job that runs statistical analysis on in-scope tables and columns — computing null rates, cardinality, min/max, mean, distribution, and pattern conformance. Profile results feed the quality score for each asset.

​
Q

Quality Score An overall percentage (0–100%) representing the health of a data asset, computed across all active metrics. The score is shown prominently on the asset Overview tab and tracked over time so degradation trends are visible.
Query Metric A business metric type where a custom SQL query defines the check logic. The query result (a number or boolean) is evaluated against a configured threshold to determine pass or fail.

​
R

Ray The distributed computing framework used by Prizm’s self-hosted deployment to run AI workloads and large-scale data processing jobs. Ray runs as a service in the Docker Swarm stack and has its own dashboard URL (https://<ray-domain>/).
Reconciliation Metric A metric category that compares data across systems to verify consistency — covering Comparison (cross-system value checks) and Lookup (referential integrity) metric types.
Report An asset type representing a reporting object or materialized output, such as a BI report or dashboard export, that Prizm catalogs and includes in lineage and context alongside its source tables.

​
S

Schema Change An observability signal that fires when Prizm detects a DDL-level change on a table — such as a column being added, removed, renamed, or having its data type changed. Schema changes are detected by comparing LAST_DDL timestamps between Observability job runs.
Scope The defined set of databases, schemas, or tables a connector is configured to catalog and monitor. Assets outside a source’s configured scope are not discovered or profiled by Prizm.
Semantic Context Metadata that describes the business meaning of an asset, including its Domain, Application, Product, Tags, Glossary Terms, Owners, and Audience. Semantic context is managed in Prizm and, for Snowflake connectors with bi-directional tag sync enabled, can be written back to Snowflake.
Semantic Model An asset type representing a logical data model or business-facing definition layer — such as a dbt model or BI semantic layer — that sits between raw tables and the metrics or reports built on top of them.
Source A connected data platform or system (e.g., a Snowflake account, Databricks workspace, dbt project). Each source has its own connector configuration, credentials, asset scope, and job schedules in Prizm.
Self-Hosted A Prizm deployment mode where the entire Prizm platform (UI, API, database, Ray) runs on customer-managed infrastructure using Docker Swarm. Contrast with Hybrid Agent Outbound, where only agents run in customer infrastructure and the Prizm platform is SaaS.
SSO (Single Sign-On) Prizm supports Single Sign-On with major identity providers, enabling secure, seamless authentication and centralized identity management. When SSO is configured, users authenticate through the identity provider rather than the standard Prizm login form.
Statistics Metric A structural metric that computes and tracks statistical properties of a column (mean, standard deviation, min, max, percentiles) and alerts when values deviate significantly from historical norms.
Steward A person responsible for data quality and governance. Prizm additionally supports two other ownership roles: Business Owner (accountable for the data’s business meaning) and Technical Owner (responsible for the pipeline or engineering). All three Owner fields are user pickers that resolve only against people holding a licensed Prizm account — not free-text roles or teams. Owners are shown on the asset Overview tab.

​
T

Tag A lightweight classification label applied to an asset for filtering and grouping. Prizm supports six tag types: Status, Domain, System, Data, Priority, and Security. Tags differ from Glossary Terms, which define business meaning rather than classification.
Tenant A Prizm customer environment. In the Hybrid Agent Outbound model, each tenant has a unique NATS_ENVIRONMENT value that scopes all agent communication. In self-hosted deployments, the tenant is the organization configured during the initial seed step.
Term An entry from the Business Glossary that defines what a piece of data means in business language — for example, linking a column to Customer Identifier. Terms can be extracted automatically from source metadata or documentation and are then reviewed and linked to assets by a steward.

​
V

Vault A secrets manager integration (e.g., AWS Secrets Manager, Azure Key Vault, HashiCorp Vault) used to store connector credentials outside of Prizm. When a connection is created with Vault, credential rotation happens in the secrets manager and Prizm picks up the new credentials automatically — without any connector reconfiguration.
Volume An observability signal tracking the row count of a table over time. Prizm alerts on unexpected volume drops (e.g., a pipeline failure that stops loading data) or spikes (e.g., duplicate ingestion). Volume is one of the four key metrics shown on every asset’s Overview tab.

​
W

WSS (WebSocket Secure) The transport protocol used by Prizm agents to connect to the NATS JetStream broker. WSS runs over port 443, which means no special firewall rules are needed beyond standard HTTPS egress — agents connect outbound only, no inbound ports required.