Grid Intelligence.
KRITIS-classified metropolitan grid. ~6,700 transformer stations. PSIcontrol SCADA, GE Smallworld GIS, and SAP PM unified into a single asset condition layer. No system replaced.
A large Southern German municipal utility — structured as a multi-utility Konzern serving approximately 1.6 million people across electricity, gas, and district heating — maintained world-class grid reliability while operating with no unified view of asset condition. Transformer load data resided in SCADA. Cable routing lived in GIS. Maintenance history was tracked in SAP. A unified asset condition layer was deployed connecting all three systems without modification. The MV transformer fleet now carries a single health index per asset, and grid anomaly detection surfaces systemic patterns invisible to any individual system.
The grid.
One of Germany's largest metropolitan distribution grids: 8,500 km of low-voltage cable, 3,700 km of medium-voltage cable, 580 km of high-voltage lines, and 69 km of extra-high-voltage overhead, feeding through approximately 6,700 local transformer stations from 136 HV/MV connection points. Gas network: 6,400 km. District heating: 800 km. Peak electrical load exceeding 1,000 MW. Annual throughput roughly 6 TWh.
German unbundling law requires distribution operators with more than 100,000 connected customers to legally and operationally separate grid operations from retail. The grid subsidiary operates under BNetzA incentive regulation, where revenue caps decline annually and every cost decision is measured against the next efficiency comparison. The IT-Sicherheitskatalog classifies the grid as KRITIS critical infrastructure, requiring ISO 27001-certified security with energy-sector controls under ISO 27019.
The operational IT landscape was assembled over two decades from three primary vendors: PSIcontrol for SCADA, GE Smallworld for GIS, and SAP for maintenance and billing. Each represented a different data model, a different update cycle, and a different definition of what constituted an asset, an event, or a fault.
Multi-utility Konzern. KRITIS critical infrastructure classification. ~1.6M customers served.
Three systems. No shared view.
Maintenance scheduling was time-based — transformers received inspection cycles calibrated to manufacturer recommendations and regulatory minimums, not to actual operating conditions. Anomaly detection was alarm-based — individual threshold breaches triggered individual responses, with no mechanism to correlate multiple sub-threshold signals across systems into an emerging pattern. Fewer than 5% of German distribution operators use true predictive maintenance. This utility was not among them.
The §14a EnWG framework added urgency. Every new heat pump above 4.2 kW, every EV wallbox, every battery storage system now connects with the operator's obligation to accept it. The transformer fleet was designed for unidirectional load patterns. It now faces bidirectional flows from rooftop PV, simultaneous charging peaks, and heat pump surges — without visibility into which assets approach operational limits and which have hidden capacity.
The absence was not technical — it was definitional. Three systems held three partial views of the same physical infrastructure, but no shared commitment existed to what a transformer's condition meant when load history, maintenance state, and network position were considered together.
Stathon engagement assessment
PSI offered operational analytics within PSIcontrol. SAP proposed an S/4HANA migration path. GE Vernova's GridOS Data Fabric promised a federated data layer. A major system integrator concluded that a unified condition model would require either multi-year platform consolidation or acceptance of permanently siloed analytics — and recommended the latter.
GIS used geospatial feature IDs, SAP used equipment numbers, SCADA used point addresses. Approximately 12% of transformer station records showed naming divergences between GIS and SCADA — stations renamed during network restructuring in one system but not updated in the other.
No structural definition existed for what constituted an “asset condition event” evaluable across all three systems. A SCADA alarm, a SAP work order closure, and a GIS topology update each described a fragment of reality. No schema linked them into a unified condition assessment.
SCADA operated in near-real-time. GIS topology updates lagged by weeks depending on surveyor backlog. SAP work orders closed within days. During fault investigation, the dispatcher saw outdated topology in at least one system. The answer was stale by the time it was assembled.
Each vendor offered analytics within its own silo. PSI offered analytics within PSIcontrol. SAP proposed S/4HANA migration. GE Vernova’s GridOS promised a federated layer. A system integrator concluded that a unified model would require multi-year consolidation — or permanent silos.
Two phases.
Vault-tier deployment. Non-negotiable — KRITIS classification requires that every data access event, every inference operation, every model query is logged, attributable, and retrievable under regulatory audit. Aegis was active from Week 1.
Definition & Integration Spine
Weeks 1–8First Live Capability
Months 3–5Single health index per asset derived from four input streams: SCADA load history (15-min resolution, 12-month window), thermal trends (where available — ~15% of fleet), SAP PM maintenance history, and GIS topology context. Rule-based escalation layered beneath probabilistic scoring. Netzführung adopted within three weeks. Instandhaltung required six weeks of parallel operation — three cases where condition score advanced an inspection were confirmed on physical inspection.
MV network anomaly detection entered 90-day parallel-run validation. Advisory mode — flags correlated patterns for human review. Initial false positive rate ~40% above target, primarily from planned switching operations classified as faults. Two tuning cycles reduced FP rate by roughly one-third. Entering final validation phase.
Three adjacent MV/LV transformer stations in a high-heat-pump-density district individually showed normal parameters. Athena identified all three simultaneously trending toward upper quartile of historical load distribution during evening peak. GIS revealed a shared 1987-vintage MV feeder cable. The pattern represented a single systemic overloading condition invisible to any individual system. Feeder reclassified for accelerated inspection.
A 110/20 kV power transformer with online dissolved-gas analysis showed gradual hydrogen concentration trend correlated with atypical load cycling. Flagged approximately 4 hours before emergency threshold. Operations scheduled controlled load transfer during overnight window, avoiding high-urgency switching during next day’s peak.
What changed.
Seven-month deployment window. Two measurable structural shifts, one capacity reallocation, and one CAPEX prioritization capability in production.
The utility did not receive a predictive maintenance platform. It received a structural definition of what its grid assets are, how their condition is constituted, and how events across three previously disconnected systems relate to a single operational reality. The scoring engine is not the infrastructure. The decision to define what condition means is.
Every transformer in the fleet now carries a condition score informing the Netzausbauplan investment process. Data-driven deferral where assets retain useful life, accelerated intervention where condition trends indicate earlier-than-expected degradation. Financial quantification planned for Q3 2026 review.
What this infrastructure is.
Beneath the existing landscape.
PSIcontrol continues to manage SCADA operations. Smallworld GIS continues to hold the network topology. SAP PM continues to process work orders. None were modified, replaced, or re-architected. What changed is that a definitional and continuity layer now exists between them — an entity graph resolving what a transformer is across all three, an event schema classifying condition-relevant changes regardless of source.
Sovereignty as architectural foundation.
The Vault deployment reflects the operational reality of German critical infrastructure. This is not a commercial tier distinction. In a KRITIS-classified environment, sovereignty over what is known, by whom, and when, is not a feature to be enabled later. It is the architectural foundation upon which every integration, every model, and every operational decision rests.
Structural intelligence, not performance.
The grid was already among the most reliable in the world. What it lacked was not performance but structural intelligence — the capacity to know, in continuous time, what the condition of its assets is across the full depth of its operational data, without requiring a human to serve as the integration layer between vendor-locked systems.
The integration is not visible from the surface. Its absence would be.
Stathon deployment conclusion
Forward roadmap.
Cable segment risk scoring
Extending the entity graph and condition model to the MV cable network. Two pilot feeder districts selected for their mix of aging PILC cable segments from the 1960s–1970s and newer XLPE installations. Integrates partial discharge test results with GIS cable vintage, SAP failure history, and SCADA load profiles.
Low-voltage grid observability
Dependent on smart meter (iMSys) rollout achieving sufficient density. With approximately 23% of mandatory installations completed nationally and Bavaria tracking below federal average, meaningful LV visibility requires ~50% penetration.
§14a load balancing intelligence
Architecturally designed for controllable devices — heat pumps, EV chargers, battery storage. Deployment contingent on low-voltage observability providing sufficient data density for informed curtailment decisions.
Deployment record.
Stathon · Definitional Infrastructure Company. Client identity withheld by agreement. Deployment metrics reflect production conditions as of March 2026.