This working paper reports the first-stage empirical validation of the Aurevo ten-module non-invasive residential risk assessment framework, applied to a 1,000-case cross-typology sample drawn from all 22 counties of Taiwan between 2023 and 2026. The framework scores each dwelling across ten functional domains — electrical safety, long-term care and ageing-in-place, circulation, daylighting, air quality, building materials, bathroom safety, household composition, behavioural habits, and community environment — and issues both an overall risk grade (A–D) and a binary indicator of whether the case would be underestimated by standard property evaluation. Three findings are reported. First, 69.2% of the cohort met the operational criterion for underestimation by standard evaluation, rising to 100% among cases graded moderate-high or high (n = 482). Second, the proportion of such cases increased non-linearly with building age, from 17.4% at 0–10 years to 99.6% at 31–40 years and 100% beyond 50 years, consistent with a threshold-blindness hypothesis in which conventional state-based valuation fails systematically once cumulative invisible decay crosses a critical point. Third, the modular architecture of the framework captured a distribution of dominant risks — led by building materials (20.5%), electrical systems (14.5%), and circulation (13.4%) — that cuts across property value tiers and challenges the assumption that risk is co-extensive with observable condition. The results are descriptive and classificatory rather than predictive, and rest on single-assessor scoring with retrospective coding criteria; inter-rater reliability and longitudinal calibration against realised loss outcomes remain open and are the subject of ongoing work. With those limits acknowledged, the cohort is, to the author's knowledge, the largest cross-typology non-invasive residential risk sample reported to date in the publicly accessible literature, and the observed pattern is sufficient to warrant revisiting institutional due-diligence methodology for residential assets in jurisdictions with a large stock of dwellings aged thirty years or more. Cross-market replication in Japan, Singapore, and the United Kingdom is in progress. This Zenodo record contains the working paper (v1), the full de-identified analytical dataset (1,000 cases × 44 fields), a Python reproduction script that regenerates Tables 1–3 from the raw data, and a README documenting the package. The dataset and code are released under CC BY-NC 4.0; commercial licensing of the dataset or framework may be requested from the author.
Yao-Hui Huang (Fri,) studied this question.