Methodology · TAS — Total Asset Score

Four pillars. One rating.

Arctura combines four risk pillars into a single A–F rating for physical assets. Every score is explainable and traceable. The pillars are computed deterministically; ML models enrich the forecasts; the weightings are trade-secret — an approach modeled on how established rating agencies publish their methodology while keeping the underlying weightings proprietary.

Pillars
4
battery · usage · market · compliance
Rating
A–F
deterministic scoring
Weightings
Trade-secret
methodology public, weights proprietary
Residency
EU only
Hetzner Helsinki · DORA · NIS2
Book a pilot conversationAPI documentation →
Four pillars · weighted into one composite

What goes into a rating.

Battery health

Components
SoH trajectory, degradation model, charge history, temperature anomalies
Data sources
OEM telematics, BMS data, charge-point data
ML model
Gradient-boosting predictors, time-series forecast

Usage & behaviour

Components
Km overrun, drive profile, DC fast-charge share, seasonality
Data sources
Telematics, GPS, OBD-II
ML model
Statistical anomaly detection, time-series forecast

Contractual & market

Components
RV-gap analysis, contract term, brand / model factor, market-price delta
Data sources
Contract data, market APIs, Schwacke / Autovista
ML model
ML-based residual-value forecast

Environment & compliance

Components
CSRD readiness, ESG scoring, Battery-Passport coverage, AFIR compatibility
Data sources
Regulatory registries, company data, certificate registry
ML model
Rule-based scoring + EU-deadline weighting
The A–F scale · one a CFO already knows

Five grades, one composite.

A
Best

Excellent. Minimal risk.

B
Good

Good. Normal monitoring.

C
Elevated

Elevated risk. Active monitoring.

D
High

High risk. Action required.

F
Critical

Critical. Immediate intervention.

Machine-learning models · deterministic TAS, ML enriches the forecasts

The scoring is deterministic. The forecasts are learned.

Residual-value forecast

gradient boosting

A gradient-boosting model predicts residual value from thousands of data points. It enriches the Contractual & Market pillar.

Mileage forecast

time-series

A time-series model projects contract-period mileage and flags overruns three to six months ahead. It feeds the Usage & Behaviour pillar.

Anomaly detection

Z-score

Continuous deviation detection in temperature, charge pattern and usage profile relative to the fleet mean. It feeds the Battery Health pillar.

Explainability

AI Act Annex IV

A regular model review covers validation discipline, drift monitoring and fairness checks. The exact pillar-weight blend is trade-secret, but the input pillars and validation methodology are auditably transparent.

Data sources · six universal adapters, one engine input

Where the inputs come from.

Telemetry & OEM API

  • OEM REST — a direct link to the manufacturer: SoH, charge, km. BMW · Tesla · Polestar and others.
  • Telematics poll — Wialon · Samsara · Flespi · SignalK · Tesla. One telematics_telemetry table per partner.
  • MQTT + OBD-CSV + file-upload + BESS adapter — six universal-pattern adapters in total.
  • iPaaS-compatible — enterprise iPaaS platforms (Frends, MuleSoft, Boomi) act as an integration rail, not a required path.

Market & compliance

  • Spot-hinta.fi · Fingrid · FMI — electricity spot prices, weather points, optimal charge timing.
  • Helsinki public OpenAPI — 10 curated demo buildings live (a wider dataset is not offered).
  • Digitraffic AIS — 525 road-weather stations, weather-camera imagery, 7 vessels tracked (Marine).
  • Schwacke / Autovista — residual-value APIs; C-segment uplift €450/vehicle (Schwacke/Autovista 2024 research citation).
EU regulatory readiness · all data in the EU

Sovereign by construction.

Deadlines ahead

  • Battery Passport18.2.2027EU 2023/1542 — digital product passport
  • NIS2In forceCybersecurity directive — Kyberturvallisuuslaki 124/2025
  • AFIRIn force 13.4.2024Charging infrastructure — data requirements
  • AI Act2.12.2027Annex IV — explainability requirement
  • CSRD2026Sustainability reporting — Scope 3 emissions

Infrastructure

  • ComputeHetzner Cloud — Helsinki (Hetzner hel1-dc2, EU)
  • DatabaseSupabase EU (eu-west-1, Ireland)
  • AuthSHA-256 + HMAC-SHA256 JWT
  • RLS100% of tables (80+)
  • WAF11 rules, rate limiting
  • AI sovereigntyEU-only or Global, selectable
Validation Packet · modeled on rating-agency disclosure

The methodology is public. The weights are not.

A Model-Risk-Management function audits the inputs and the validation methodology — not the proprietary weighting. The same split an established rating agency applies to its own methodology.

Public — in the DD pack

Methodology + inputs

  • +Methodology paper (pillars + score-mapping)
  • +Four input pillars, auditable record
  • +Model review (validation discipline, drift, fairness)
  • +Sub-processor list + data-flow diagram
  • +3rd-party validation roadmap (Y1 EU audit firm)
  • +Source-code escrow (Strategic tier)
Trade-secret — not disclosed

Pillar weights

  • ·Pillar weightings
  • ·Score-mapping formula constants
  • ·Internal cross-validation cohort splits
  • ·Drift-trigger threshold values

The same way an established rating agency does not publish its sovereign-rating weighting formula.

SANDBOXEngine is in sandbox today — no live customers. Grades shown across the site are illustrative.

Want to see TAS on your own data?

Book a pilot conversationAPI documentation →