Skip to content
SENSE&VALUE
Seven-layer architecture

One data spine. Two products: operations and valuation.

Every layer below is a real engineering surface with its own protocols, contracts, and failure modes. The stack is designed so operations pays for the sensors, and valuation emerges as a signed by-product of data we were already paid to collect.

End-to-end flow

Cumulative flow across all seven layers

L-0Field sensingInputs
Physical phenomena: vibration, tilt, voltage, current, water, CO₂, temperature, motion.
Process
On-node ADC sampling → thresholded encoding → LoRaWAN uplink; gateway buffers on WAN loss.
Outputs
Compact binary frames tagged with devEUI, timestamp, and RSSI/SNR — ready for decoding.
L-1Ingestion & market da…Inputs
LoRaWAN frames from gateways + nightly market-data files and APIs.
Process
Decode → validate → dedupe → insert into hypertable; ETL normalises leases and comps onto property_id.
Outputs
Unified time-series rows: readings, leases, transactions, comparables.
L-2Asset & twin modelInputs
Component catalogue, documents, maintenance history, incoming readings.
Process
Map readings to components → apply subsystem weights → compute per-subsystem and composite BHI.
Outputs
Versioned BHI score with subsystem breakdown and confidence flags.
L-3Valuation engineInputs
BHI · NOI (measured) · comparables · cost baseline.
Process
Run three estimators independently → weight by data quality → blend → emit CI.
Outputs
Valuation + confidence interval + explainable contribution breakdown.
L-4Trust & complianceInputs
Readings, computed scores, valuation snapshots.
Process
Hash → batch → Merkle root → anchor → valuer signs.
Outputs
Signed, anchored report with verifiable hash path.
L-5ApplicationsInputs
User role · organisation · query parameters.
Process
RLS view materialises role-appropriate slice → UI composes KPIs.
Outputs
Role-scoped dashboard, alerts, exports, and signed reports.
L-6Roll-outInputs
Design-partner buildings and pilot KPIs.
Process
Instrument → measure → publish · widen coverage each phase.
Outputs
Case evidence, portfolio benchmarks, open signed city index.
InputsProcessOutputsTap a row to open its layer
100%
L-3

Valuation engine

Triangulated: ML comparables + measured NOI + cost/BHI. Ships a value with a confidence interval, updated on events.

ML comparablesα · price/m²Measured NOIβ · yield capCost × BHIγ · depreciationV$284.1MCI ±3.4%±bandV = α·ML + β·NOI + γ·(Cost × BHI) · event-driven recompute
100%

Valuation is a triangulation, not a single model. Three independent estimators run in parallel: ML comparables (learned from local transactions and hedonic features), measured NOI capitalised at a market yield derived from real leases, and cost-approach depreciated by the current BHI. The engine blends them with data-quality-weighted coefficients and emits a value with an explicit confidence interval. Recomputes are event-driven — a new lease, a new comp, a BHI drop past a threshold — so the number on the page always reflects the latest signal, not last quarter's snapshot.

Specifications
  • Formula: V = α·ML + β·NOI + γ·(Cost × BHI)
  • Confidence: interval derived from input dispersion
  • Triggers: new reading · new lease · new comp · BHI Δ
  • Cadence: near-real-time on event; batch nightly
  • Explainability: per-input contribution logged
  • Guardrails: outlier clamp · human-review flag
Inputs

BHI · NOI (measured) · comparables · cost baseline.

Process

Run three estimators independently → weight by data quality → blend → emit CI.

Outputs

Valuation + confidence interval + explainable contribution breakdown.

4 / 7