Enterprise identity and access

The enterprise surface compiles an operator-authored structural import into a deterministic, safely fictional identity/access universe with physically separate public product input and evaluator truth. It also provides bounded reference benchmark packs and standards-shaped projections.

SynthWorld is not a directory service, IGA workflow system, policy decision point, or runtime enforcement component.

Author, validate, compile.

synthworld scaffold-enterprise-access --format yaml --output private-enterprise.yaml
synthworld validate-enterprise-access --input private-enterprise.yaml
synthworld compile-enterprise-access \
  --input private-enterprise.yaml \
  --seed 20260804 \
  --output compiled-enterprise

Treat the authored blueprint as operator-private. Importing enterprise structure is not anonymization, and structural keys or namespace material can remain sensitive even when no production person records are present.

Validation reports stage-local diagnostics rather than silently accepting partial structure. Compilation then writes separate public/ and evaluator/ trees. Canonical account-to-principal binding truth belongs only in the evaluator tree; public account records do not expose that linkage.

Determinism and seed semantics. Compilation is reproducible from the saved import plus the explicit seed and versioned compiler/schema inputs. Structural identifiers such as tenants, organisations, units, principals, groups, roles, targets, and permissions are derived from the authored namespace and logical structure. Account allocation is seed-driven, so account-related subjects and access atoms can move between seeds while principal-level structure remains stable.

Do not generalize one generator’s seed behavior to another SynthWorld surface. Record the complete configuration and version inputs with any result.

What the compiled universe represents. The public universe contains bounded entity inventories and access atoms. It is a test input, not a live directory topology or policy engine. The normative enterprise contract defines supported authoring formats, selector rules, cross-tenant constraints, output layout, and other limits:

Enterprise identity/access contract

Persistent access declarations and observed account bindings are intentionally tenant-local. A cross-tenant reference in either surface is malformed input, not a scoreable denied case. To test a cross-tenant authorization guard, keep the compiled access atom structurally valid and express the evaluated subject/resource tenant divergence in the cell’s public ABAC facts and policy. The bounded enterprise contract does not model a persistent federated cross-tenant grant; a consumer must not bypass validation to manufacture one.

Authorization and benchmark surfaces. The repository contains directory/RBAC, ABAC, ReBAC, identity-fabric, enterprise-agentic, contextual-access, authority-governance, and continuous-assurance surfaces with different APIs and maturity levels. Do not infer that every family has the same CLI or accepts an arbitrary compiled universe.

Some tasks expose command-line generation/evaluation; others are Python-only. Use the CLI reference, the relevant contract README, and the generated capability catalogue for the current boundary.

For the released-package Python workflow from this compiled universe through a public-only authorization adapter and independent scorer, use Build and score an enterprise authorization experiment.

Reference packs are conformance fixtures. Their evaluator truth may be published in the repository, so physical separation prevents accidental oracle use but does not turn a public reference pack into a blind test. A perfect reference score is evidence of conformance to the declared cases, not generalization.

Standards-shaped projections. SCIM, OpenFGA, AuthZEN, and Shared Signals/CAEP mappings are documented in Standards profiles. These are bounded offline mapping surfaces; a mapping declaration is not proof of protocol transport, signing, interoperability, or deployed enforcement.

Evaluation boundary. Give a system under test only the public artifact required by that task. Normalize its output into the versioned prediction or trace contract, then score against separately loaded evaluator truth.

See Evaluating a system, DATA_DICTIONARY.md, and the normative enterprise contract for exact schemas and metric semantics.

For the end-to-end topology, planning, lifecycle, and future agentic-composition journey, see Enterprise Identity Planning.