Municipal holdings management
Portfolio overview with a source for every figure.
Municipalities and groups steer dozens of holdings across scattered reports. Anyone who needs a portfolio overview today consolidates by hand — and loses exactly what a resolution requires: proof of where each figure comes from. cNode consolidates the holdings into one audit-proof, evidence-backed view in which every figure stays traceable to its source.
Context & background
A municipal holdings portfolio routinely spans utilities, housing companies, transit operators, hospitals and special-purpose associations — each with its own accounting, reporting cycle and system landscape. Holdings management must forge from this a single steering view that also satisfies the municipal code, the statutory holdings report and the council committees.
This task is not primarily a computation problem but an evidence problem. The council, the audit office and the public may expect every reported figure to be backed and the path to it to be traceable. As the EU AI Act phases in, this expectation shifts from good practice to a regulatory requirement: systems that prepare decisions must provide logging (Art. 12) and transparency (Art. 13).
Mid-market firms and public bodies face the same dilemma: they need the speed of software but must not give up the provability of a manual paper trail — and certainly not trade it for a black box.
The problem
The usual route is manual consolidation in spreadsheets. Figures are retyped or copy-pasted out of PDF reports, financial statements and ERP exports. Every one of these steps severs the link between figure and source: in the end a number sits in a cell, but no one can prove within seconds from which document, period and calculation it originates.
The reflex to solve this with generative AI makes it worse. A language model that 'summarizes' reports produces plausible-sounding numbers with no verifiable compute path and can invent values. In a holdings context that is untenable: a figure that cannot be backed is not a figure but a liability.
Concrete failure points
- Holding reports arrive in dozens of formats and cycles.
- Figures are retyped by hand — error-prone and quickly stale.
- Decisions lack a traceable compute path down to the source.
- Generative summaries invent numbers and cannot be audited.
The causal chain
Why the end product is a non-audit-proof resolution is not a matter of individual diligence but a chain of cause and effect. Each link produces the next:
- Scattered, inconsistent sourcesforce manual consolidation
- Manual consolidationsevers figure from evidence and introduces transfer errors
- Missing evidence + errorsmake every figure contestable
- Contestable figureslead to non-audit-proof resolutions
Where cNode breaks the chain
cNode acts not at the last link but at the root. Because the sources are ingested deterministically and anchored in a knowledge graph with a prescribed ontology, manual consolidation never arises in the first place. The transfer error disappears, the evidence chain stays attached to every figure — and the decision paper is audit-proof because it already is so at the data root.
How cNode solves it
In the first step cNode ingests the relevant sources deterministically — financial statements, ERP exports, annual accounts, commercial-registry extracts. 'Deterministic' means: the same input always yields the same structured representation, without a model interpreting freely.
The ingested data is transferred into a knowledge graph whose ontology is prescribed. The system does not 'guess' what a figure means — its meaning is fixed by the ontology. Every figure is thereby unambiguously linked to a unit, a period and a document.
On this graph, deterministic computations run: figures, aggregations and portfolio views are derived by rules and reproducibly. For every derived statement cNode retains the source and the full compute path.
The language model enters only at the very end — and only to cast the already-computed, evidenced results into readable language. It does not compute, it invents nothing, it phrases. The figure stays provable and the explanation stays legible.
The deterministic process
- Ingest data deterministically (financial statements, ERP, annual accounts, registry).
- Knowledge graph with a prescribed ontology — every figure unambiguously anchored.
- Deterministic computation — reproducible via a fixed seed.
- Source + compute path stored per statement.
- Audit trail per data point; the LLM only verbalizes.
The outcome
The outcome is a current portfolio overview instead of a patchwork of spreadsheets — and, more importantly, an overview that withstands any query. For every figure, source and compute path can be called up in one click; the decision paper carries its evidence chain within it. What today costs weeks of manual reconciliation becomes a repeatable, documented derivation.
Methodology & verifiability
Provability here is not an after-the-fact feature but a construction principle. Every reported statement references the document it comes from and the compute path it arose through. Because the computation is deterministic and reproducible via a fixed seed, the same data basis provably yields the same result — the prerequisite of any audit.
The audit trail per data point maps to EU AI Act Art. 12 (record-keeping) and Art. 13 (transparency); data processing follows the data-processing agreement under GDPR Art. 28, either in Frankfurt or fully on-prem up to air-gapped. Customer data is not used for training.
Sources & further reading
- EU AI Act, Art. 12 — record-keeping (logging).
- EU AI Act, Art. 13 — transparency and provision of information.
- GDPR, Art. 28 — processor obligations.
- Related: Compliance & audit, Public administration.