The CRI Engine
Climate science, translated into financial exposure.
A technical overview of the pipeline underneath every report we deliver — from asset coordinates to Capital-at-Risk.
Live geospatial hazard render — one node per resolved asset coordinate.
The Geospatial Pipeline
Every run starts from asset coordinates, not sector averages. At site level the engine reads flood depths from the JRC river flood maps and Deltares coastal flood maps, cyclone wind from the NOAA IBTrACS track record, local flood defences from FLOPROS, water risk from the WRI Aqueduct 4.0 dataset (68,506 basin × admin-1 units), and Sentinel-1/2 satellite imagery. A further 25 parametric hazard functions (heat, drought, wildfire, sea-level rise and others) score regional exposure. The result is a facility-level hazard profile, not a country- or sector-level proxy.
Asset-Level Resolution
Flood depths, cyclone wind, defences and water risk read at each facility's coordinates.
NGFS Phase 5 Scenarios
Seven NGFS Phase 5 pathways (three models) for the Risk Analyst; the legacy CRI score runs Net Zero 2050, Delayed Transition and Current Policies.
Production Loss Modelling
Hazard exposure is converted into a production loss percentage before it ever touches a financial statement.
Financial Translation
Physical loss and transition cost are translated into the language a CFO's office already speaks. A full discounted cash flow model runs the 2026–2050 horizon with a Gordon Growth terminal value, applying a WACC uplift built from a base rate, a scenario premium, and an asset-specific exposure premium. The output is an EV haircut against baseline, not an abstract risk score.
Carbon Cost Trajectory
Scope 1 and 2 carbon costs, net of EU ETS free allocation, run against each scenario's carbon price path.
Abatement Capex (MACC)
A marginal abatement cost curve prices the capex required to hit stated decarbonization targets — modelled as capex, not double-charged against the carbon cost.
EV Haircut & WACC Uplift
Full DCF output: enterprise value under stress versus baseline, and the WACC uplift driving that gap.
Unified Data Architecture & Delivery
Eighteen commodity classes — from iron ore and thermal coal to cement, agriculture, and financial services — run through the same underlying architecture, accessed through a single API surface. Custom scenarios can be defined inline for a single run or persisted and reused across an engagement.
Modular Runs
Physical, transition, and financial modules can be run independently or as a full pipeline.
Custom Scenario Support
Carbon price paths, risk premiums, and abatement targets can be defined per engagement and saved for reuse.
Disclosure-Ready Output
TCFD, IFRS S2, and CSRD reports are generated directly from engine outputs — not reconstructed after the fact.
API Surface
Every capability above is reachable through a REST API, so the engine can sit inside your own deal workflow or reporting pipeline instead of living as a standalone tool you have to open separately.
Regulatory Endpoints
POST /regulatory/sfdr-pai, POST /regulatory/eu-taxonomy, and POST /regulatory/pcaf return the SFDR PAI pack, EU Taxonomy alignment, and PCAF financed emissions for a given portfolio.
Stress & Physical Endpoints
POST /stress/event replays a named historical event against a position, GET /stress/event/catalogue lists what's available, and POST /physical/slr and POST /physical/biodiversity return sea level rise exposure and TNFD nature risk scores.
Portfolio & Carbon Endpoints
POST /portfolio/benchmark, /portfolio/counterparty and /portfolio/export.csv for portfolios; POST /carbon/inventory (with an assurance-pack workbook), /lca/{product}, /transition-plan, /real-estate/assess, /materiality/assess and /carbon-markets/* for carbon accounting and disclosure.
ML & Agentic Intelligence
Every score also carries a forward view. A gradient-boosted surrogate model, trained on 50,000 simulated company-years whose parameters are calibrated to NGFS pathways and EM-DAT disaster losses, forecasts a company's CRI score out to 2050 under three scenarios, with confidence bands rather than a single static number. Where emissions reporting is missing, an estimator trained on observed facility emissions from the US EPA Greenhouse Gas Reporting Program, matched to parent-company revenue and evaluated on held-out companies, fills the gap with a confidence-scored estimate instead of a flat sector average. A ClimateBERT-based scanner then reads the company's own disclosures for commitment specificity, flagging the gap between what's claimed and what's backed by capex. And a tool-calling AI agent, wired directly to the engine itself, can research a company from public financials and news, run the full assessment, or work through an entire watchlist unattended and deliver the results by email.
Predictive CRI Trajectories
A gradient-boosted surrogate forecasts the 0-100 CRI score for every year to 2050 across NZE, Delayed Transition and Current Policies, with confidence bands rather than a point estimate.
AI-Filled Emissions & Disclosure Scoring
An estimator trained on EPA GHGRP observed emissions imputes missing Scope 1/2, while a ClimateBERT scanner flags the gap between a company's stated commitments and its actual capex.
Autonomous Portfolio Agents
Research, full assessment, watchlist monitoring, and batch analysis across an entire portfolio, run unattended and delivered by email or through a conversational interface.
Methodological Transparency
Every number the engine produces is traceable to a named hazard layer, a named scenario, and a named financial assumption. Nothing is a black box: we defend the CRI score and every input beneath it in front of a client's own quantitative or risk team, on request.
No Black-Box Scoring
The 0–100 CRI score decomposes into its physical, transition, and financial components on request.
Auditable Assumptions
WACC premiums, abatement targets, and carbon price paths are documented, not embedded silently in the model.
Portfolio-Level Aggregation
Value-at-Risk is computed at 95% and 99% confidence across a full portfolio, weighted by exposure.