Semantic namespace
01

Principal: organization, project, role

02

Meaning: domain, concept, version

03

Data: source, scope, watermark

04

Runtime: session, retrieval, cache

05

Evidence: resolution path and conflict

A term is unique only inside an explicit context

Retail sales may mean paid order value, while finance means recognized revenue. Another business unit may apply tax or currency differently. A globally merged synonym can produce technically valid but semantically wrong queries.

Return namespace_id, concept_id, definition_version, organization, scope, and effective period. Narrow candidates through the current project and ask for clarification when peers still conflict.

Model semantic objects, not a keyword dictionary

A metric needs names, formula, grain, dimensions, date role, currency, filters, owner, and version. A dictionary that equates GMV with sales cannot express refunds, status, tax, or day boundary.

Give concepts, entities, relationships, enumerations, and metrics stable IDs. Bind retrieved documents to namespace and access scope rather than relying on display text.

Carry the scope key through every component

A practical key combines organization, project, domain, locale, and policy version. Parsing, candidate retrieval, planning, caching, and audit must carry it without falling back to unscoped global search.

A group metric may be inherited from a parent namespace. Project overrides require lineage and priority, not copied definitions that silently diverge.

Govern inheritance, override, and conflict

Priority can move from explicit project definition to domain, group standard, and controlled default. Overrides need reason, approval, effective period, and impact analysis. Equal candidates produce an ambiguous state.

Choosing whichever table name looks closest is a counterexample. New physical tables would then change old answers without a semantic decision.

Isolate conversation context

When a user moves from project A to B, inherited region, currency, customer scope, and metric must not follow automatically. Persist the namespace and source message for every confirmed slot.

Copying a question into a new project should trigger new resolution. Cross-project comparison is an explicit operation requiring both authorizations and a comparable metric mapping.

Partition caches and retrieval

A text-only cache key can reuse one project’s monthly sales in another. Include namespace, identity scope, semantic version, and data watermark, then authorize again on every hit.

Vector partitioning, metadata filters, and post-retrieval authorization are complementary. Prompt instructions are not an access control, and deletion must reach index replicas.

Map concepts before cross-domain analysis

Combining sales, inventory, and finance requires declared grain, entity keys, time, currency, and choice of group or parallel definitions. Report non-comparable components and residuals.

Do not force two revenues into a single total. A plan should expose each sub-result namespace, conversion, and conservation check.

Test conflict, switching, and unauthorized candidates

Cover same-name different-meaning, synonym, override, effective-date change, missing project, context switch, dual-role user, revocation, and cache hit.

Insert a more similar unauthorized document: it must not be selected. Measure unscoped requests, cross-namespace retrieval, false inheritance, and clarification outcomes.

Assign operating ownership

Group data owners govern shared concepts, domain owners govern local meaning, project owners approve overrides, and security defines cross-domain access. Every change receives dependency analysis and regression.

The goal is not one vocabulary at any cost. Reuse genuine standards, preserve legitimate difference, and make conversion explicit.

AskTable.ai boundary

Pilot real terminology conflicts and verify stable IDs, scope propagation, inheritance, context switching, cache isolation, retrieval authorization, and audit evidence.

AskTable.ai may be evaluated as a project-aware natural-language analytics entry point. This article does not establish full namespace inheritance, index isolation, or every conflict policy in the current release; each runtime boundary requires verification.

Public references

Ready to help your team start?

Talk through a real scenario and see how AskTable.ai can fit your business.

Book a demo