Method / Overview
Methods and evidence How Greenophy analyses company disclosures and physical climate hazards, with examples, data sources and limitations.
01 / Business Model Engine
The engine identifies the activities a firm describes, who it serves and the quantities it reports. Findings retain the legal entity, reporting period and supporting evidence.
Data sources. Official identity records; annual reports, Pillar 3 and relevant sustainability disclosures; and the firm’s official business and product pages. Availability varies by firm. Website evidence may have no reporting year.
Find and scope Match the entity and collect sources. Keep the bank, subsidiaries and wider group distinct. Read and extract Capture passages and tables: activities, customers, locations and reported amounts. Classify and check Map supported descriptions to business lines; check evidence, scope, units and period. Retain review items. Show the evidence Return a profile with source links and gaps. A missing disclosure is not a zero or an absent business.
A country total and an industry total do not disclose their intersection. Without a joint disclosure, the country–industry amount is not disclosed .
From public disclosures to a business model A 36-second authored, fictional Northbank example, not a live engine run. Establish the reporting entity, read passages in context, identify the provider, activity and customer, and map disclosed activities to standard business lines. Borrower industries retain company-wide scope. Fictional operating locations connect London to the head office, Manchester to corporate lending and Leeds to asset finance. These are separate from the UK-wide mortgage book and do not establish property locations. A 2025 table reports loans and advances in GBP billions: home loans 8.2 and business finance 1.4. The business finance balance is not split between corporate lending and asset finance. Select a completed finding to inspect its illustrative evidence. UK outline: Natural Earth, public domain.
FICTIONAL EXAMPLE
Northbank
Profile 07 / 07
REGISTER Northbank Building Society Legal entity United Kingdom Northbank OUR BUSINESSES ANNUAL REPORT Northbank 2025 Entity accounts BUSINESS REVIEW 04
Northbank provides home loans to households secured on homes
LENDER ACTIVITY CUSTOMER COLLATERAL
ACTIVITY Home loans Mortgage lending
REPORTED SEGMENT Business finance Corporate lending Asset finance and leasing Business loans Equipment leasing
BORROWER INDUSTRIES · COMPANY-WIDE Housebuilders Onshore wind farms
SECTORS Building construction Onshore wind power Business-line allocation not disclosed OPERATING LOCATIONS London Head office Manchester Corporate lending Leeds Asset finance ILLUSTRATIVE OFFICES Loans and advances £ billion · year-end 2025 2024 Home loans 7.8 Business finance 1.4 1.2 Combined loans & leasing Split not disclosed MORTGAGE LENDING Loans · UK · 2025 BUSINESS ACTIVITIES Mortgage lending Home loans · UK · 2025 Business finance £1.4bn Corporate lending · Asset finance and leasing Shared balance · split not disclosed BORROWER INDUSTRIES · COMPANY-WIDE Building construction · Onshore wind No business-line allocation £8.2bn Offices · London, Manchester, Leeds
Select a result for evidence.
Close Illustrative disclosure · 2025
Play 36 / 36s Replay
Business model / Examples
Follow a finding back to its source The Northbank animation is fictional and illustrates the workflow. The full guide uses real HSBC UK report passages and a saved review item to show what the evidence can establish.
Explore the HSBC UK worked example → Source passage · Interpretation · Checks · Result
02 / Climate Hazard Engine
The engine samples observed records and modelled data for a resolved location. Each hazard has its own calculation and units; these readings are not combined into a single risk score.
Data sources. Published flood maps; NASA NEX-GDDP-CMIP6 climate projections; NOAA cyclone tracks; NASA fire detections; Hydro-JULES drought data; IPCC sea-level projections; Copernicus coastal water levels and soil moisture; and ISRIC soil properties.
Resolve the place Use a point or stated boundary. Record the source resolution and any nearby-point fallback. Select the input Keep the observation period or projection scenario, horizon and baseline with the data. Calculate the measure For example, count wet cells, average threshold days, or summarise depths. The reduction depends on the hazard. Report the limits Show the value, units, coverage and source. No data is distinct from a measured zero.
Hazard describes a physical condition. Exposure locates people, assets or activities in that condition. Risk also depends on vulnerability and consequences; these hazard readings do not estimate financial loss.
From a place to a traceable hazard reading A 73-second explanation of the Climate Hazard Engine. Type London, resolve its real GHS urban boundary and select appropriate sources. Real examples from the methodology page: JRC 100-year undefended river flood, 44,140 wet cells out of 347,789, 12.69%, median wet-cell depth 1.758 metres; England FIRMS 2025, 2,017 vegetation-fire detections out of 3,979, density 0.0423 per 1,000 square kilometres per day; London SoilGrids 60–100 cm, 418 predicted cells out of 1,864, 22.4% coverage and 35.3% median clay; the nearest central London Hydro-JULES cell, first-decade mean SPEI -0.0124 and last-decade mean 0.3354, change +0.3478. Point and boundary readings remain distinct. Cyclone uses the existing real IBTrACS Vamco example near Manila in November 2020: two recorded positions within 100 km count as one storm, retaining the closest recorded pass at 64.9 km. The coastline, search-radius geometry and track positions are copied from the full method page. Connecting segments add no observations. Heat days and soil-water shortfall use explicitly hypothetical method illustrations, not local hazard results. The coastal section is schematic, with actual Hull example values: AR6 SSP2-4.5 in 2100 has medium-confidence median rise 0.586 m and low-confidence median 0.573 m, each with its own 17th–83rd range. The GTSM 100-year height is 5.735 m; adding the medium-confidence rise gives a conditional 16.2-year recurrence under fixed storm climate. GTSM and AR6 baseline periods differ and the offset is unadjusted. Neither illustration maps inundation or flood defences. Flood defences are not represented in the JRC example. The final three graphics retain each example’s spatial scope: boundary flood cells, sorted clay coverage cells and nearest-point SPEI change. Source references remain on the full methodology page. No composite score is calculated. Place Urban boundary London London · urban footprint Hazards Nine hazard categories Flood Extent & depth Heat Hot days Drought Water-balance anomaly Wildfire Thermal detections Cyclone Nearby storm tracks Sea level Mean sea-level rise Extreme water Height & recurrence Subsoil clay Clay at depth Soil deficit Below a local threshold Flood London · 100-year model 44,140 wet cells 347,789 boundary cells 12.69% wet cells 1.76 m median wet-cell depth Undefended river flood Heat Illustrative week Dry bulb > 35 °C 32 35 36 36 38 38 34 37 37 35 3 / 7 days Wet bulb > 31 °C 28 30 31 32 32 29 33 33 31 2 / 7 days Strictly above the threshold Wildfire England · 2025 3,979 detections 2,017 vegetation fire 1,962 other · excluded 130,541.3 km² × 365 days 0.0423 detections / 1,000 km² / day Detections ≠ distinct fires Subsoil clay London · 60–100 cm 418 predicted 1,446 unpredicted 35.3% median clay 22.4% coverage Sorted cells · not a geographic map Map unshaded Drought London · nearest point cell Mean SPEI 1981–1990 2013–2022 −0.012 +0.335 0 = local normal +0.348 SPEI change · wetter Point reading · not a boundary average Soil deficit Method illustration Soil-water indicator Local 5th percentile Relative to local history Total shortfall Tropical cyclone Vamco · Manila · November 2020 Manila 100 km 2 nearby records 1 storm Closest recorded pass · 64.9 km Nearby observations · not a local wind or damage map Sea level Hull · 2100 · SSP2-4.5
Reference mean level
Mean rise +0.586 m Hull coast Medium- and low-confidence projections Medium confidence 0.586 m Median · 17th–83rd range Low confidence 0.573 m Median · 17th–83rd range Schematic section · mean change, not flood depth Extreme water level Hull · 2100 · SSP2-4.5 · medium confidence
Reference mean level
Same extreme height · 5.735 m Mean rise +0.586 m Return period at the same height Present day 100 years With mean sea-level rise 16.2 years Schematic section · conditional recurrence Fixed storm climate · baseline offset unadjusted London Selected results Flood 12.69% Wet cells within the boundary 100-year · undefended river flood Predicted / unpredicted cells Subsoil clay 35.3% Clay · predicted cells only 60–100 cm · 22.4% coverage First Last First vs last decade · SPEI Drought +0.348 SPEI Water-balance change Nearest point cell · 1981–2022 Play 63 / 63s Replay
Climate hazard / Examples
See how the result is calculated In the London flood example, 44,140 wet cells divided by 347,789 sampled boundary cells gives 12.69%. This is a cell-count share of an undefended flood model, not a share of a bank’s assets.
Inspect the London inputs and calculation → JRC · Present-day river flooding · 100-year return period The animation also contains labelled method illustrations. The guide distinguishes them from measured source outputs.
Seven places a model reads, each checked and labelled An 86-second authored, fictional Northbank example, not a live engine run. Seven places where a language model reads, each with its source, the model’s reading, one code check and the label it earns: finding a missing Pillar 3 report and refusing a candidate whose cover names another company; identifying the reporting entity of a sustainability report by exact name; withdrawing an asset management activity that rests only on a template row; attributing life cover to Harbour Life Assurance Ltd after a word-for-word quotation check and a second reading; counting Leeds as an office but not Spain, mentioned in passing; drafting a climate target whose values are checked against the quotation, with an unsupported “on track” claim dropped; and a note on a London flood reading whose figures must all appear in the reading. The film ends on a workbench list of the seven labelled results; select one to read its evidence.
FICTIONAL EXAMPLE
Northbank
Finding reports 01 / 08
COVERAGE CHECK · 2025 EXPECTED · NOT FOUND Pillar 3 disclosures MODEL · SEARCH Northbank Building Society · Pillar 3 Northgate Bank plc · Pillar 3 Kept only if its cover names Northbank Found by a model · cover checked by code SUSTAINABILITY REPORT 2025 · FIRST PAGE “This report covers Northbank Building Society and its subsidiaries.” MODEL · REPORTING ENTITY Northbank Building Society Same name as the register, letter for letter Northbank’s own report · may be used PILLAR 3 TEMPLATE · £BN · 2025 Residential mortgages 8.2 Asset management — MODEL · OBJECTION A template row, not a business All its evidence shown; the objection quoted Activity withdrawn · labelled inferred ANNUAL REPORT 2025 · PASSAGE 2 “Mortgage customers can add life cover, provided by our partner Harbour Life Assurance Ltd.” MODEL · WHO CARRIES IT ON Harbour Life Assurance Ltd Found word for word; a second reading agrees Two-pass AI review · not human-verified ANNUAL REPORT 2025 · TWO PASSAGES “Our Leeds office supports asset finance.” “Cardholders pay no fee in Spain.” MODEL · WHAT EACH PLACE IS Leeds — an office Spain — a passing mention Each role quotes its passage Leeds counted · Spain not counted CLIMATE REPORT 2025 “We aim to cut Scope 1 and 2 emissions by 50% by 2030, from a 2019 baseline.” MODEL · DRAFT RECORD 50% · 2030 · baseline 2019 Progress: on track Every value is in the quotation “On track” is not in the source, so it is dropped AI-assisted summary · source checked LONDON · RIVER FLOOD · 100-YEAR, UNDEFENDED Wet cells within the boundary 12.69% Median wet-cell depth 1.76 m MODEL · NOTE “12.69% of the boundary is wet; where wet, the median depth is 1.76 m.” Every figure is in the reading; no rating words Model-written note · figures stay with code WORKBENCH · NORTHBANK · AI CONTRIBUTIONS Pillar 3 disclosures 2025 Found by a model; cover checked Search record kept Sustainability report 2025 Northbank’s own report Reporting entity · quoted Asset management Template row only Withdrawn · inferred Life cover Harbour Life Assurance Ltd Two-pass AI review · not human-verified Leeds An office supporting asset finance Inferred Emissions target 50% by 2030 · Scope 1 and 2 AI-assisted summary · source checked London flood reading Note on 12.69% wet, 1.76 m Model-written note Select a result to read its evidence. Source Model reads Code checks Label
Close Illustrative disclosure · 2025
Play 86 / 86s Replay
03 / Use of AI
Language models read company disclosures where fixed rules cannot. Every answer must point to its source, is checked by code and is labelled as AI in the workbench. Hazard figures, amounts and identifiers stay with code.
Example · Where AI is used · Checks · Labels
