Synthetic CRM case study

From an AI-proposed MERGE to a reviewable engineering plan.

This example shows the nine canonical planning deliverables generated for a synthetic CRM scenario. It demonstrates the structure, specificity and transparency of the handoff — not a production implementation.

← Return to the product overview

The scenario

Two CRM records described as one continuing legal person.

An AI agent proposes merging CRM-ACCT-2097 into canonical CRM-ACCT-1042. The supplied descriptions identify the same official registry ID (FIC-REG-004821) and a legal-name change effective 2024-04-01. On those descriptions, identity continuity is supported conditionally on authenticity and customer-policy eligibility. The Planner did not verify the original registry extract, gateway verification record or 2025 agreement, so the evidence is not yet authorization-eligible. A negative control also rejects similarity-based MERGE for related companies with different registry IDs.

DeliverableWhat it containsSource
00 — Plan overviewScenario, recommendation status, provenance, validation record and handoff summary.Open Markdown ↗
01 — Mutation scopeApplication and operation scope, affected state classes and operation boundaries.Open Markdown ↗
02 — Guardrail specificationTwelve named controls, their triggers, required actions, enforcement points, failure behaviour and validation links.Open Markdown ↗
03 — Evidence requirementsClaims, evidence roles, provenance, admission, promotion and conflict requirements.Open Markdown ↗
04 — Commit-boundary designPre-persistence controls, transaction and write-set questions, version requirements and failure behaviour.Open Markdown ↗
05 — Implementation blueprintProposed components, responsibilities, logical interface, implementation sequence and operational concerns.Open Markdown ↗
06 — Validation scenariosProposed tests linked to operations and controls, including failure and authorization paths.Open Markdown ↗
07 — Assumptions and gapsUnresolved assumptions, blocking customer decisions and dependencies that must be resolved.Open Markdown ↗
Machine-readable planJSON representation of the guardrail specification.Open JSON ↗
Important interpretation notes

The example plan is explicitly marked not_implementation_ready. It contains 12 proposed controls, 31 customer-operation test scenarios and three proposed Planner-level checks; none of the tests or Planner checks was executed. Eight customer decisions (GAP-01 to GAP-08) remain blocking. Draft 2020-12 JSON Schema and format validation were not run in the original generation environment. On the supplied descriptions, identity continuity is supported, but the Planner did not independently verify the source documents and the evidence is not yet authorization-eligible. Commit status is HOLD; the plan does not execute or authorize a live operation. All names and account identifiers are synthetic.

What this example demonstrates

The connection between an investigation, an evidence-based recommendation, explicit customer-side controls, implementation responsibilities, validation scenarios and declared unknowns. The output is a starting point for customer engineering and policy review, not a turnkey production recipe.