A
-
ABAC (Attribute-Based Access Control)
An access-control model that grants or denies access based on attributes of the user, resource, action, and environment - rather than only on assigned roles. The reference framework is NIST Special Publication 800-162.
-
Audit Trail
A record produced by Colrows for every query that captures the graph version, identity context, resolved entities, proven join paths, and compiled SQL. Enables point-in-time reproducibility - the ability to re-run a historical query with the exact definitions in force at that moment.
-
Autonomous Maintenance
The continuous process by which agents detect schema or content drift, propose updates to the semantic graph, and route them for human approval - keeping the graph current without a manual catalogue-update process.
See: Agents that maintain your data systems, The rise of autonomous semantic systems.
C
-
Compile-Then-Execute Pipeline
The Colrows runtime architecture: intent is first compiled - resolved against the semantic graph, planned with proven joins, governance-injected - and only then executed against the warehouse. The opposite of retrieval-and-hope-it-works text-to-SQL.
-
Compile-Time Governance
Enforcement of RBAC, ABAC, and row/column-level predicates inside the compiler, before any SQL leaves the planner. Unauthorised queries fail compilation; the data is never read. Governance is structural, not advisory.
-
Consensus Layer
The Colrows mechanism that resolves conflicts between sources of meaning - documentation, dbt definitions, observed usage, human input - to produce a single agreed-upon value for each concept in the semantic graph.
-
Constrained Planning
Query planning that searches only the typed semantic graph for valid join paths. Cannot fabricate joins or invent entities; failed planning produces a structured error rather than a guessed answer. The structural antidote to LLM hallucination on enterprise data.
-
Context Resolution
The step in the compile pipeline where every term in the request is resolved against the semantic graph using the requesting identity, role, persona, and scope. The same string ("revenue") resolves to different concepts depending on who is asking and from which scope.
D
-
Dialect-Perfect SQL
SQL emitted by the Colrows compiler that is specialised for the target engine's dialect (Snowflake, Databricks, BigQuery, Postgres, and so on) - syntactically correct, semantically faithful, and optimised for that engine. The same intent against different engines produces different SQL.
See: SQL as a compiler target.
-
Drift Detection
The autonomous process of identifying when the underlying data estate (schemas, columns, foreign keys, business definitions) has drifted from the current semantic graph. Triggers proposed updates for human approval rather than silent breakage.
I
-
Intent
The semantic content of a request - what a human or AI agent wants to know - independent of the SQL or API call used to express it. The Colrows compiler accepts intent and emits SQL: the input is meaning, the output is execution.
J
-
Join Path Proof
A formal proof that a join path between two entities exists and is valid under the typed semantic graph's relationships. Required before SQL emission; failed proofs abort compilation. Eliminates the fabricated joins that ungoverned text-to-SQL systems produce.
K
-
Knowledge Drift
The gradual divergence between the documented meaning of business concepts and how they are actually used in queries, dashboards, and agents. The dual problem to schema drift; addressed by drift detection and autonomous maintenance.
M
-
Multi-Scope Semantics
The property by which the same concept resolves to different definitions at different scopes -
global,datastore,persona,user. Lets a single semantic graph encode legitimate per-tenant or per-team variation without duplication. -
Multi-Tenant Semantic Isolation
The combination of scoped resolution, partitioned embeddings, and policy isolation that prevents one tenant's meaning from leaking into another's queries in a multi-tenant deployment - even when the underlying agents and embeddings are shared.
-
Multi-Vector Embeddings
Three embedding vectors maintained per concept in the Colrows graph: definition (the formal description), usage (how the concept actually appears in queries and dashboards), and combined. Powers context-aware retrieval and disambiguation between similarly-named concepts.
P
-
Point-in-Time Reproducible
A property of Colrows audit trails: a historical query can be re-run with the exact graph version, definitions, and policies that were in force at the time, producing the same result. The basis for safe operation in regulated environments where audits arrive months after the query.
R
-
RBAC (Role-Based Access Control)
An access-control model that grants permissions to roles and assigns roles to users. Simple and well-supported, but suffers from role explosion at enterprise scale - typically combined with ABAC in modern deployments.
-
Row/Column-Level Predicates
Governance predicates that restrict access to specific rows or columns inside a table based on user identity or attributes. In Colrows, these predicates are injected at compile time as additional
WHEREclauses or column projections in the generated SQL - never as a post-query filter.
S
-
Semantic Decay
The slow loss of fidelity between business concepts and their representations as the data estate evolves - renamed columns, drifting metric formulas, contradictory definitions. The technical-debt cousin of knowledge drift.
-
Semantic Execution Layer
An infrastructure layer that compiles enterprise intent into governed, deterministic, dialect-perfect SQL. The category Colrows defines: a runtime - not a metric API or BI feature - that sits above the warehouse and below the prompt.
-
Semantic Graph
The typed, versioned, multi-scope representation of business meaning that backs every Colrows query. Encodes entities, metrics, relationships, constraints, and governance in one place. The system of record for meaning, parallel to the warehouse as the system of record for data.