Nothing in the semantic view enforces its metric on a bypassing consumer: base-table grants are the only fence, and Snowflake's own agent path requires those grants
The question
"What actually enforces a Snowflake semantic view's metric definition for a third-party BI or agent consumer that does not query through the semantic layer?"
Context: open follow-up from [[2026-10-02-snowflake-semantic-views-cortex-gateway-agent-identity-audit]], and the load-bearing question under the founder's draft "Agents don't need your data warehouse. They need your definitions." ([[2026-10-02-agents-need-your-definitions-v1]]). A definition is a governance control only if something stops a consumer from recomputing the metric off base tables.
Headline: For a consumer that does not query through the semantic view, nothing enforces the definition. The only fence is ordinary role-based access control (RBAC) on the base tables. Semantic views do run with owner's rights, so you can build a role that sees only the semantic view (the "secure view" pattern). But Snowflake's own docs say Cortex Agents need SELECT on both the semantic view and its underlying tables, and the Cortex Analyst access table lists SELECT on "the tables mentioned in the semantic model." So the agent path Snowflake sells cannot be sealed this way. Even a sealed role can build its own ad-hoc aggregates from public facts and dimensions, and dimensions cannot be made private. Verdict: by default a semantic view is a convenience. It is a partial contract only for SQL and BI roles that are denied the base tables, and even then it controls vocabulary, joins and row policy, not the arithmetic.
What we already know (from the vault)
- Status is settled. The core semantic view object is GA (three release-note dates, mid-2025 to 2026-03-02). The export surfaces, Open Semantic Interchange / Apache Ossie YAML and Tableau
.tds, are open preview. [[2026-10-02-snowflake-semantic-views-cortex-gateway-agent-identity-audit]] - The parent brief offered this as a "cleaner supporting fact than the slogan": querying a semantic view needs
SELECTon the view only, not on the underlying tables. True for direct SQL. But the parent brief missed the Cortex Agents exception on the same docs page (see Contradictions). [[2026-10-02-snowflake-semantic-views-cortex-gateway-agent-identity-audit]] ACCESS_HISTORYrecords base columns read through a view, withagents_infonaming the agent chain (Enterprise Edition). That is a detective control, and it turns out to be the strongest enforcement-adjacent fact available. [[2026-10-02-snowflake-semantic-views-cortex-gateway-agent-identity-audit]]- Authoring already needs base-table
SELECT, which the vault flagged as a structural limit on business-user self-serve. [[2026-09-23-semantic-studio-self-serve-authoring-erosion]] - Misteli's framing names the exact gap: "Semantics without execution is documentation." He splits a semantic contract (what the metric means) from an execution contract (what must not vary at runtime). Snowflake ships the first. Enforcement of the second exists only inside the semantic view's own query path. [[2026-08-10-misteli-semantic-layer-control-plane]], [[semantic-layer]]
What the web says
- Owner's rights is documented, so a "semantic-view-only" role is possible. "Semantic views use owner's rights, meaning that a user with access to a semantic view does not require separate access to its underlying tables; the view's owner (role) controls access. ... This behavior is consistent with the privileges required to query standard views." And: "If you want to restrict users strictly to semantic views, you can omit the grants on tables and views and only grant privileges on the semantic view objects." (best practices; querying: "You only need the SELECT privilege on the semantic view itself.")
- The very next sentence breaks this for agents. "However, note that Cortex Agents that use semantic views require the role that is executing queries to have SELECT privilege on both the semantic view and its underlying tables." (best practices). The Cortex Analyst access-control table agrees:
SELECTon "The tables mentioned in the semantic model" (Cortex Analyst). The same page's own example grants future tables and views "to support scenarios where users might need direct access to the underlying data objects ... outside of the semantic layer." Any role that can run the agent can also query the base tables directly and compute the metric its own way. - Even through the semantic view, consumers can write their own metrics. Querying docs: you can "define and select ad-hoc metrics by passing a dimension or fact to any aggregate function"; in standard SQL, "you can specify an expression that uses a metric" (example
AGG(order_average_value) * 10) (querying). Joins, window functions,QUALIFY,PIVOTand correlated subqueries are blocked, so the view does enforce join paths. It does not enforce the arithmetic. PRIVATEis the only in-object lever, and it is partial. Private facts and metrics "cannot be queried or used in a query condition," but "You cannot mark a dimension as private. Dimensions are always public" (CREATE SEMANTIC VIEW; SQL guide). Private items still appear inDESCRIBE SEMANTIC VIEWandGET_DDL. TheINFORMATION_SCHEMAviews list them only for roles withREFERENCESorOWNERSHIP.- Masking and row access policies are the one control that binds on both paths. They "can't be set directly on semantic view attributes, but if they are set on underlying tables and columns, they propagate to semantic views and are enforced" (sample values stored as metadata are not masked) (best practices). Because they sit on the base table, they also bind a consumer who bypasses the view. But they govern which rows and values, not what the metric means.
- The OSI / Ossie export carries the definition, not the enforcement.
SYSTEM$READ_OSSIE_YAML_FROM_SEMANTIC_VIEWis "Preview Feature — Open" and needs only "SELECT or USAGE" on the semantic view. It exports names, expressions, equi-join relationships andai_context. It silently drops non-equi relationships and omits data types. Private status travels only as anaccess_modifierinside the Snowflake vendor extension, and "External tools can ignore these extensions or pass them through unchanged." Grants, owner, and masking or row policies are not in the format (function page). The import procedure was renamed from..._FROM_OSI_YAMLto..._FROM_OSSIE_YAML. - Tableau keeps the definition only when the query runs back through Snowflake.
SYSTEM$EXPORT_TDS_FROM_SEMANTIC_VIEW(open preview) maps each numeric metric to "a calculated field [that] passes the value of the metric to the Snowflake AGG function." So a Live connection does evaluate the governed definition. But numeric facts becomeSUMmeasures, which exposes ad-hoc aggregation by default. "You cannot create an extract from a semantic view" (SQL guide, TDS export). The listed BI integrations are Sigma, Omni, Honeydew, Hex and ThoughtSpot (best practices). Each is a vendor integration that queries the view; none is an enforcement mechanism.
Convergences and contradictions
- Correction to the parent brief (log in Errors & Corrections). [[2026-10-02-snowflake-semantic-views-cortex-gateway-agent-identity-audit]] cites "
SELECTon the view only" as the clean governance fact. It holds for SQL and BI roles but is wrong for the agent consumer the draft is about. The same docs paragraph says Cortex Agents need base-tableSELECTtoo. Cite the two sentences together or not at all. - The vault and the docs agree on the shape: definition and enforcement are separate layers. Misteli's two-contract split and the concept note's "governed translation layer" both describe this. Snowflake enforces the semantic contract only inside its own query path. Outside that path, enforcement is ordinary grants (preventive), base-table policies (row and value level only) and
ACCESS_HISTORY(detective). - The draft's wording happens to be safe, but only barely. Draft v1 says agents and humans "start from one written definition" and puts access policy in a separate bullet. That is accurate. Upgrading it to "enforces one definition" would be false on the current docs.
Synthesis for RDCO
The honest verdict is the negative one: a Snowflake semantic view is a convenience that can be partly hardened into a contract, and the hardening is unavailable on the agent path. The single fact the backlog expected to decide this, whether a role can get the view without the base tables, resolves "yes" for SQL and BI roles. Owner's rights work like a standard view. But Snowflake's own docs then exempt the case the thesis is about. Cortex Agents and Cortex Analyst need the executing role to hold SELECT on the base tables. Any principal who can ask the agent can also run SELECT ... FROM base_table and compute readmission rate their own way. Even a sealed BI role can build ad-hoc aggregates from public facts and dimensions inside the view. PRIVATE cannot hide dimensions. The exports, Ossie YAML and Tableau TDS, are previews that carry text, not policy. The access modifier rides in an extension that consuming tools are told they may ignore.
This changes the draft's argument rather than killing it, and the change makes it sharper. "They need your definitions" survives as a claim about consistency of intent. One written definition that agents and dashboards default to beats one definition per prompt, and the readmission example still proves that. It does not survive as a claim about control. The accurate three-layer version, each layer a GA primitive: (1) the semantic view publishes the definition and constrains join paths. (2) Base-table grants plus row and masking policies fence who can bypass it and what they see. Today that fence cannot exclude agent roles from base tables. (3) ACCESS_HISTORY with agents_info detects bypass after the fact, by showing which base columns a principal read directly versus through the view. So for the agent consumer, governance of meaning on Snowflake today is audit, not prevention. A post that says so plainly is more credible to a platform team than one that implies the semantic view is a fence. It also answers the bear case the draft already concedes.
For phData and CAF delivery, this converts into a concrete design pattern and an honest caveat. Pattern: issue BI and service consumers semantic-view-only roles, with no base-table grants. Mark component facts PRIVATE where a raw numerator or denominator should not be re-aggregated. Run a recurring ACCESS_HISTORY query that flags any principal reading the base columns of a governed metric outside the view. Caveat to state up front: agent users keep base-table access by design, so the governed-metric guarantee for agents is "default plus audit," and the client should know that before treating a semantic view as a compliance control for a regulated metric. That caveat is a differentiator: competitor content that sells semantic views as "the single source of truth" does not mention it.
One inference flagged as unverified. Wrapping the agent's data access in an owner's-rights stored procedure or custom tool, rather than the built-in Cortex Analyst tool, might restore a sealed agent path. No doc reviewed here states this. It is a follow-up, not a claim.
Why this is in the vault
This decides how [[2026-10-02-agents-need-your-definitions-v1]] (a founder-pinned piece under [[2026-09-29-nov30-credibility-targets]]) may describe semantic views. It can say "publish," not "enforce," and the agent-path caveat must sit beside the owner's-rights fact. It also gives phData semantic-layer engagements a specific seal-plus-audit design pattern, with a documented limit to disclose to regulated (healthcare) clients.
Open follow-ups
- Does a Cortex Agent custom tool backed by an owner's-rights stored procedure let an agent answer from a semantic view while its role holds no base-table
SELECT? Do Snowflake docs or Snowflake-Labs samples document this pattern? - Is the Cortex Agents base-table requirement tracked anywhere for removal (release notes, Summit 2026 sessions, roadmap statements), or is caller's-rights execution a deliberate design?
- Does the Apache Ossie specification draft (ossie.apache.org / GitHub) define any normative policy, access-modifier or conformance field that consuming tools MUST honor, or is the spec purely descriptive by design?
Related
- [[2026-10-02-snowflake-semantic-views-cortex-gateway-agent-identity-audit]] - parent brief; GA status map and the
ACCESS_HISTORYaudit fact; corrected here on the agent privilege point - [[2026-10-02-agents-need-your-definitions-v1]] - the draft this verdict constrains
- [[2026-09-29-agents-need-your-definitions]] - original draft outline and bear case
- [[2026-09-29-nov30-credibility-targets]] - publish target the draft serves
- [[2026-09-23-semantic-studio-self-serve-authoring-erosion]] - base-table
SELECTas an authoring gate - [[2026-08-10-misteli-semantic-layer-control-plane]] - semantic contract vs execution contract; Apache Ossie origin
- [[semantic-layer]] - concept article; should gain a "publish vs fence vs detect" line
- [[2026-07-08-cortex-sense-semantic-layer-wedge-caf]] - the "governed correctness" wedge this qualifies
Sources
Vault:
- ~/rdco-vault/06-reference/research/2026-10-02-snowflake-semantic-views-cortex-gateway-agent-identity-audit.md
- ~/rdco-vault/01-projects/personal-brand/drafts/2026-10-02-agents-need-your-definitions-v1.md
- ~/rdco-vault/01-projects/personal-brand/drafts/2026-09-29-agents-need-your-definitions.md
- ~/rdco-vault/06-reference/research/2026-09-23-semantic-studio-self-serve-authoring-erosion.md
- ~/rdco-vault/06-reference/2026-08-10-misteli-semantic-layer-control-plane.md
- ~/rdco-vault/06-reference/concepts/semantic-layer.md
- ~/rdco-vault/06-reference/research/2026-07-08-cortex-sense-semantic-layer-wedge-caf.md
Web (all fetched 2026-11-14):
- https://docs.snowflake.com/en/user-guide/views-semantic/best-practices-dev - owner's rights; "restrict users strictly to semantic views"; Cortex Agents base-table requirement; policy propagation; BI integrations (verbatim checked against raw page)
- https://docs.snowflake.com/en/user-guide/views-semantic/querying - SELECT-on-view-only; ad-hoc metrics; standard-SQL restrictions
- https://docs.snowflake.com/en/sql-reference/sql/create-semantic-view - PRIVATE/PUBLIC; create privileges incl. base-table SELECT
- https://docs.snowflake.com/en/user-guide/views-semantic/sql - private fact/metric visibility rules; Tableau TDS export mapping and limitations (open preview)
- https://docs.snowflake.com/en/sql-reference/functions/system_read_ossie_yaml_from_semantic_view - Ossie YAML export contents, vendor-extension "can ignore" (open preview)
- https://docs.snowflake.com/en/user-guide/snowflake-cortex/cortex-analyst - access-control table: SELECT on tables in the semantic model
- https://docs.snowflake.com/en/user-guide/snowflake-cortex/cortex-agents - caller's-role tool access
- https://docs.snowflake.com/en/release-notes/preview-features - preview status cross-check