Map an enterprise data boundary before building retrieval

A retrieval diagram that begins at chunking has already skipped the hardest question: which data may cross the system boundary? Search indexes, embedding stores, caches, traces and evaluation datasets are derived copies. They inherit obligations from the source even when they live in a different platform.

Build the decision artifact

Inventory each source and derived store. Record the business owner, data class, permitted principals, purpose, region if relevant, retention, freshness and deletion route. Draw trust boundaries around ingestion, indexing, retrieval, generation and observation. For every arrow, name the identity used and the minimum fields transferred.

The enterprise delivery lab makes this reasoning executable with synthetic data:

from enterprise_cases import data_boundary_case

result = data_boundary_case()
assert result["unowned"] == []
assert result["derived_copy_count"] == 1
assert len(result["assets"]) == 3
assert result["assets"][0]["copy"] == "search index"

The lab finds one derived copy: an internal policy corpus becomes a search index. Invoices and support tickets remain outside that path. Every asset has an owner and an allowed role, making an unowned data source a blocking defect instead of an implementation detail discovered late.

Test the failure path

Common boundary leaks include exporting broad database rows for convenience, storing prompt bodies in general logs, and using an evaluation dataset copied before a user’s deletion request. A data map should be testable: sample records must be traceable from source to every derivative and back to the deletion contract.

Save evidence that another reviewer can inspect:

• source and derivative inventory
• trust-boundary and identity-flow diagram
• field-level purpose and retention table
• lineage plus deletion test for one synthetic record

This exercise aligns with the enterprise delivery, secure integration, operations and leadership work in the FDE for Professionals course. The course link describes the pathway; this local draft and its synthetic verifier do not claim that a client deployment, security approval or production result occurred.

Practice task

Map one document source, one structured source and every derived copy. Remove any arrow whose identity, purpose or deletion behavior you cannot name.

Continue learning

This article is part of the Enterprise AI delivery and architecture sequence. Use the neighbouring tasks when you need the prerequisite or the next application.

• Review the prerequisite or neighbouring task in Choose rules, classical ML or an LLM for a client workflow.
• Continue with Design tenant isolation across retrieval, tools and caches.

Reference: NIST AI RMF Playbook.