How we determine risk classes for compulsory auctions

The risk class is a compact orientation signal: it shows how many and how severe the object-specific risk findings are that we detect in the official documents. It is not a purchase recommendation, not a separate valuation and not a legal or financing review. The original appraisal, the official notice and the court's auction conditions remain authoritative.

The value of ZVG Melder is not merely collecting auctions. The core of the platform is automated pre-analysis: we read appraisals, detect review-worthy findings, structure them by severity and show how strong the underlying data basis is. This helps you prioritise faster which properties deserve a closer review and which cases should be set aside early.

What the class expresses

We use six classes:

• 1: no material risks detected
• 2: manageable findings
• 3: medium risk
• 4: elevated risk
• 5: high risk
• 6: very high or existential risk

The word detected matters. A low class does not mean that no risks exist. It only means: based on the documents available to us, we found no or only minor evidence-backed risk findings.

What the platform detects automatically

The basis is mainly the official appraisal and other published documents. We read structured fields and text passages, including:

• market value, property type, address and valuation date
• condition, modernisation, remaining useful life and renovation needs
• damp, mould, pollutant, contamination or substance findings
• rights and encumbrances such as residence rights, usufruct, leasehold, real burdens or building charges
• tenancy, lease and eviction-related findings
• missing interior inspection or other gaps in the valuation basis
• costs, deductions or special factors not fully reflected in the market value
• partition auctions, multiple valuation units or unusual ownership structures

The class is not raised merely because a word appears somewhere. What matters is whether the finding is plausible, object-specific and supported by the source. That is the difference between the analysis and a simple keyword search: a registered right can be value-neutral, a missing interior visit can be normal for a forest parcel, and a negated sentence such as “no mould detected” is not a mould risk.

Rules and AI work together

The analysis combines two layers:

1. Deterministic rules over structured fields.
2. AI-assisted text analysis over unstructured appraisal passages.

The rule set detects clear recurring patterns. Examples include:

• a surviving residence right
• a leasehold with a short remaining term
• documented damp, mould or substance findings
• no interior inspection or only an exterior inspection
• costs explicitly excluded from the market value
• partition auctions or ownership structures with higher complexity

We also use AI models to read and structure unstructured appraisal text. This matters because appraisals vary widely. Some are concise and tabular, others contain long prose, annexes or scanned passages.

The AI score, however, is not simply the visible risk class. ZVG Melder checks whether the findings are supported by reasons and sources. A high model score without supporting evidence should not by itself raise the visible class. Conversely, hard rules or several medium findings can raise the visible class above the raw AI score.

In simplified form:

risk class = maximum of rule result and evidence-supported AI contribution

This keeps the classification more explainable than a pure model score.

Why several medium findings can become class 4

The risk class is not only the highest individual finding. Several medium risks can together create an elevated overall picture.

For example:

• needs renovation -> moderate capital expenditure
• partition auction -> legal and organisational complexity
• damp finding -> building-condition review point

Each of these points can individually sit in class 3. But if at least three distinct class-3 findings span several risk dimensions, the rules raise the overall class to 4. The statement is not “one single risk is class 4”; it is “the combination of several medium findings deserves elevated review”.

This distinction explains why the overall class can sometimes be higher than the individual dimension values.

Why information grade is separate

Alongside the risk class, we show an information grade from A to D. It describes how robust the data basis is.

• A: very good information basis
• B: good information basis
• C: relevant gaps
• D: very limited information basis

This separation is deliberate. A low risk class on a thin data basis is not an all-clear. If the data basis is too weak, we suppress low classes and show that no reliable assessment is possible instead.

What you gain from it

For investors, bidders and professional observers, the benefit appears before the detailed review:

• You detect faster which properties deserve increased attention.
• You see risk findings as structured classes rather than only as PDF prose.
• You can filter and prioritise properties by risk class.
• You see whether the data basis is strong or thin.
• You get a better review list for an appraiser, lawyer, bank or your own due diligence.

The platform does not replace the final review. It reduces manual search effort and surfaces noteworthy passages earlier.

Why the analysis has limits

Compulsory-auction data is dynamic. Documents may be missing, added later or worded inconsistently. Appraisals are unstructured text, and AI models continue to evolve. They are also not deterministic: in borderline cases, a model may weigh the same fact differently.

For that reason, every risk class is an automated orientation based on the documents available. It says which findings we detected and classified. It does not say that no further risks can exist.

We review conspicuous cases, verify findings against the original source and continuously improve rules, prompts and tests. That is how the system gets better. The final decision, however, must never rest on the risk class alone.

In brief

1. The risk class prioritises; it does not decide.
2. ZVG Melder detects risk findings automatically from official documents.
3. Evidence-backed appraisal findings matter more than isolated keywords.
4. Information grade and risk are separate signals.
5. A low class plus a thin data basis is not an all-clear.
6. Review the official original documents yourself before every bid.

Note: General orientation only; not legal, construction, financing or investment advice.