StreetMirror Banks & Insurance POC — Theoretical Rationale

Document: Theoretical and methodological defence of the Banks & Insurance capture schema Version: 2026 Companion file: StreetMirror Banks & Insurance - POC Database - 2026.xlsx

1. Purpose of this document

This document explains why the Banks & Insurance capture schema is structured the way it is. It is the methodological defence of the model — the answer to "why these variables, and not others?" — for use in sales conversations, regulatory questions, partnership discussions, and ongoing internal calibration.

Every variable in the schema earns its place by reference to one of two anchors:

1. A record category that banks and insurers already maintain for a financed or insured property, or
2. A validated academic framework with peer-reviewed empirical support.

Most variables anchor to both.

2. The unifying thesis

StreetMirror produces independent, ground-truth intelligence on physical assets at city scale. For Out-of-Home advertising, the lens is visibility. For banks and insurers, the lens is value.

The shared engine: structured visual census → atomic indicators → repeat over time. The shared gap: organisations with financial interest in physical assets have no scalable, lawful, external way to verify that the assets on their books still match reality.

Banks rely on a bond file recorded at origination. Insurers rely on a policy snapshot recorded at inception. Neither maintains a routine re-verification process. StreetMirror is that process.

3. The 3C framework

Three pillars. Each one mirrors a record category that every bank and insurer already maintains:

The 3C frame is commercially aligned, not academically invented. It is the minimum structure that allows position-by-position comparability between our row and the buyer's record.

4. Condition — theoretical grounding

The Condition pillar captures the physical state of the asset itself. Its nine variables derive from three convergent bodies of work.

4.1 The Fannie Mae UAD C1–C6 condition scale

Fannie Mae's six-point condition rating (C1 = newly constructed, C6 = uninhabitable) is the international standard for residential property condition reporting, used by every UAD-compliant appraiser and mirrored by appraisal practice in most lending markets including South Africa.

Critically, the C-rating is determined on an absolute basis — the property's own merits, not relative to the neighbourhood. This matches the external-observer position exactly. We capture absolute, visible condition signals from public space and roll them up to a UAD-compatible C-rating.

The nine Condition variables are the field-observable inputs to that rating:

• Roof state (variables 3, 4, 5) — missing elements, rust, sag
• Dwelling wall state (variables 6, 7, 8) — cracks, peeling paint, plaster damage
• Window integrity (variable 9) — broken or boarded panes
• Perimeter envelope (variables 1, 2) — boundary wall and fence integrity

A property accumulating multiple Condition flags is, by UAD definition, drifting from C3 toward C5 or C6. Our model exposes that drift quarterly.

4.2 The "broken windows" tradition (Wilson & Kelling, 1982; Skogan, 1990)

Wilson and Kelling's thesis is that visible physical disorder is a leading indicator of further decline — for crime, social cohesion, and property value. Skogan's 1990 operationalisation produced a set of binary indicators that have become standard in urban observation research:

• Broken windows
• Boarded-up or abandoned buildings
• Damaged or unmaintained façades
• Overgrown / unmaintained property
• Litter / debris

Our Condition pillar adopts the broken-windows operationalisation directly. The variable "Broken window" is not a metaphor — it is the literal indicator from this tradition, validated across forty years of empirical research as a predictor of property-value decline.

4.3 The NIfETy 14-item physical disorder scale (Furr-Holden et al., 2008)

NIfETy is a validated, peer-reviewed instrument for systematic observation of neighbourhood physical disorder. It is composed entirely of binary observable indicators and demonstrates Cronbach's α = 0.825 across published studies — high inter-rater reliability for field data.

NIfETy validates that the binary-atom approach to physical disorder produces statistically reliable, inter-rater consistent measurements at scale.

4.4 The external-to-internal capture flow

The ordering of the nine Condition variables — perimeter → roof → dwelling walls → windows — follows the CPTED (Crime Prevention Through Environmental Design) observation sequence used by police-trained surveyors and adapted in modern systematic social observation protocols (Sampson & Raudenbush, 1999). The capturer's eye and the supervisor's photo review traverse the property in a single repeatable arc, which:

• Minimises missed observations
• Standardises capture across surveyors
• Mirrors how potential buyers, loss adjusters and credit officers actually inspect properties

5. Construction — theoretical grounding

The Construction pillar captures the physical composition of the asset. Three frameworks justify its inclusion.

5.1 The Fannie Mae UAD Q1–Q6 quality rating

Parallel to the C-rating, Fannie Mae's Q-scale rates construction quality on a six-point scale driven by material composition. Our wall and roof material picklists (v10, v11) provide the field-observable inputs to a Q-scale rollup.

5.2 Standard insurance construction-class tables

Every short-term insurer in South Africa maintains a construction-class lookup that translates wall material × roof material into a fire-risk band and a per-square-metre replacement cost. The same lookup is used at policy inception and is rarely re-verified thereafter.

Capturing the current roof and wall materials lets us flag any property where the inception assumption has been superseded by an alteration or extension.

5.3 Informal subdivision (v12) and SA-specific moral hazard

The "multiple dwellings on one stand" binary is grounded in two sources:

• South African municipal zoning and land-use research documenting the high prevalence of informal subdivision in residentially-zoned suburbs since the early 2000s.
• Insurance moral-hazard literature establishing that policy class and sum-insured assume a single primary occupant household; multi-tenant informal subdivision materially changes claims behaviour and is almost universally unreported.

A residentially-zoned, residentially-bonded property that has been informally subdivided is one of the largest invisible risks on a South African bank's residential book. Variable 12 makes it visible.

6. Continuity — theoretical grounding

The Continuity pillar captures whether the asset is still being used as recorded. Four sources ground its seven variables.

6.1 Occupancy underwriting practice

Vacancy is a primary moral-hazard and physical-hazard signal in insurance. A vacant property has materially higher fire, theft and water-damage exposure than an occupied one, and most policies contain explicit vacancy clauses that void cover after a specified unoccupied period. Yet insurers almost never independently verify occupancy after policy inception.

Variables v13 (curtains), v14 (driveway overgrown), and v15 (window boarded) are the three most reliable street-observable occupancy proxies in the published literature on vacancy detection. They are validated in the NIfETy instrument and used in academic vacancy studies across the US, UK and EU.

6.2 Use-class drift and zoning compliance

A property bonded or underwritten as residential but operating as commercial constitutes a material misrepresentation under most standard South African bond and insurance contracts. Detection of use-class drift is a recognised gap in lender and insurer risk management — flagged regularly by financial regulators but operationally unaddressed at portfolio scale.

Variables v16 (commercial signage) and v17 (institutional signage) provide the externally-observable, lawfully-captured signal that closes this gap.

6.3 Construction-activity signalling

The presence of visible construction work — whether active or stalled — indicates that the asset's footprint, sum-insured exposure and structural risk profile have moved away from the inception file. Variable v18 captures this state without requiring the capturer to distinguish active from stalled (which would invite subjectivity). The temporal trajectory (progressing vs frozen) is derived from quarterly re-walk comparisons.

6.4 Market-distress signalling

A visible "For Sale" board is a leading indicator of upcoming property turnover and, in aggregate, of neighbourhood-level market pressure. Variable v19 captures this signal from public space; aggregated to street level, it produces a near-real-time market indicator that lags the deeds registry by months rather than weeks. This is commercially valuable to banks tracking concentration risk and to insurers managing book-level renewal pricing.

7. Methodological grounding

7.1 Why atomic binaries

The schema deliberately avoids ordinal scales. The trade-off is well-documented in field-instrument design:

• Ordinal scales (e.g. 0/1/2 severity) offer more granularity per variable but introduce capturer subjectivity, lower inter-rater reliability, and require longer training.
• Atomic binaries offer marginally less granularity per variable but maximise inter-rater reliability, minimise supervisor disputes, and allow defensible per-variable statistical modelling.

Furr-Holden et al. (2008, 2011) demonstrated that binary-only field instruments achieved inter-rater reliability of α > 0.80 across 350+ block faces. Ordinal-rich instruments in the same domain struggled to achieve α > 0.65. We adopt the same constraint.

7.2 Why 19 fields produces sufficient signal

Each captured variable is statistically independent and individually observable. The total field count (19) is comparable to other validated instruments in this domain:

Our model is more granular than the validated academic instruments (because we need three commercial pillars rather than one) and substantially more disciplined than typical commercial property condition assessments (which capture 80+ items but rely heavily on subjective ordinal ratings).

7.3 Why supervisor verification works

Every binary has a single-clause decision rule referencing a visible feature. The supervisor's QA workflow is mechanical: open the photo set, re-tick the boxes independently, compare to the capturer's row. Disagreements are factual (the feature is in the photo or not), not interpretive (the feature is "moderate" or "severe"). This is the operational definition of objectivity in field-data capture.

8. The C1–C6 translation layer

The final design choice — deriving an overall C1–C6 rating from the nine Condition variables — solves the buyer-comparability problem.

Banks and insurers do not buy spreadsheets of binary indicators. They buy ratings they can drop into their existing models. By publishing both layers:

• The atomic 19-field row (for analysts, statisticians, modellers, regulators)
• The derived C1–C6 rating per property (for underwriters, loss adjusters, credit officers, executive summaries)

…every stakeholder receives the abstraction layer they need.

The Fannie Mae C-rating is the most widely-understood condition framework in lending and insurance globally. Aligning to it makes our output interpretable to any qualified appraiser without explaining a proprietary scale.

9. Summary citation table

10. Closing position

Every variable in the StreetMirror Banks & Insurance schema is defended by reference to:

1. A bank or insurer record category it maps to, and / or
2. A peer-reviewed academic framework with empirical support.

No variable is invented for novelty. No variable is captured because it is interesting. Every variable is in the schema because it either fills a gap in the buyer's existing record or because it has been validated as a reliable predictor in the published condition / disorder / occupancy literature.

This is what makes the model defensible against scrutiny — from underwriters, from credit risk teams, from regulators, and from academic reviewers if and when peer-review becomes commercially useful.