page-05.html / Workflow Sim
Page 06 page-07.html GitHub
The Autonomous Enterprise  ·  Domain Suite  ·  06

Workflow
Simulator

Run live agent pipelines across three financial scenarios. Inspect SHAP explanations, make HITL decisions, and watch every step land in an immutable audit trail.
ScenariosIC · Treasury · AP[cite: 15]
EntitiesDE · NL · CA[cite: 15]
HITL PatternApprove / Override[cite: 15]
ComplianceEU AI Act Art. 13[cite: 15]
System Architecture[cite: 15]

Three pipelines.
One control plane.

Each financial scenario routes through a dedicated agent pipeline — IC Reconciliation, Treasury/FX, and AP Exception — converging on a shared HITL gateway, write-guard layer, and immutable audit store[cite: 15]. SHAP explanations are computed before any human checkpoint per EU AI Act Art. 13[cite: 15].

SOURCE SYSTEMS SAP GL / ERP BANK API AP INBOX IC PIPELINE TREASURY PIPELINE AP PIPELINE SAP Extract Feature Store Anomaly Score SHAP Explain Bank API Fetch FX Feed Enrich Cash Forecast Confidence Check Document AI Parse PO Match Exception Classify SHAP Explain HITL GATEWAY ⊙ HITL Approve / Override Write Guard Audit Trail Store EU AI Act Art.13 SHAP before HITL ✓
Interactive Simulation[cite: 15]

Run a live pipeline
end to end.

Select a scenario and entity. Press Run. Watch data events arrive, the pipeline light up sequentially, and make the HITL decision when the agent pauses for your review[cite: 15].

Scenario
Entity
Data Event Stream IDLE
⬡
Awaiting simulation start
Agent Pipeline IDLE
Agent Reasoning Stream IDLE
Time to Detection — awaiting run
HITL Decision Time — awaiting decision
Audit Record ID — not yet written
EU AI Act Art. 13 — SHAP before HITL
EU AI Act Art. 14 — Human oversight
EU AI Act Art. 17 — Quality log written
Scenario Anatomy

Three domains,
one decision model.

Each scenario is a distinct event class with its own ingestion format, feature set, and write-back target — but all converge on the same SHAP → HITL → Audit pattern required by regulation[cite: 15].

SAP Extract Feature Store Anomaly Score SHAP Explain HITL-IC-01 Write Guard Journal Post Audit Trail
Regulatory Compliance

EU AI Act —
Article 13 by design.

Every pipeline is architected so that SHAP feature attributions are computed and committed to the audit store before any HITL checkpoint is presented. No human decision is made without a machine-readable explanation. This is not retrofitted — it is the sequence[cite: 15].

EU AI Act · Art. 13 Transparency & Explainability SHAP feature attributions are written to the immutable audit store before the HITL checkpoint is opened. The human reviewer sees the top-3 drivers of every model decision before acting[cite: 15].
Implemented · SHAP before HITL[cite: 15]
EU AI Act · Art. 14 Human Oversight Every pipeline contains a mandatory HITL node. No write-back to SAP, the bank, or the ERP can occur without an explicit Approve signal from an authorised user. Override routes to a manual correction queue[cite: 15].
Implemented · All paths gate on HITL[cite: 15]
EU AI Act · Art. 17 Quality Management & Logging Every simulation produces an immutable Audit Record with: agent ID, scenario, entity, model score, SHAP values, HITL timestamp, decision, override reason (if any), and write-guard result[cite: 15].
Implemented · Full audit trail per run[cite: 15]