Methodology

Data types

Every value carries a label saying where it comes from. We never mix types in a single figure without saying so.

Data freshness

Each source has an expected publication frequency. We compare the date of the last successful update with that window and show one of these statuses. Dates include the time zone.

We don't use the “real time” label unless the provider guarantees it and we've verified the latency. More detail inhow we measure freshness.

When a source fails

• We keep the last valid value and mark it as stale if it exceeds its window.
• We never replace a value with zero or with made-up data.
• We log the error, retry within limits and alert the team if it repeats.

Calculations

The simulators use deterministic formulas with automated tests based on reference cases calculated by hand. Current version of the financial library: finance@1.0.0. Every report will store the version it was calculated with so it can be reproduced.

• We keep the original unit and document every conversion.
• We don't convert currencies without stating the date and source of the exchange rate.
• When a metric doesn't exist for some data (for example, an IRR with no sign change), we say so instead of showing a number.

Market metrics (MarketPulse)

Version market@1.0.0. They're calculated on the monthly averages published by the World Bank, in nominal dollars.

• Monthly and year-on-year change: percentage change against the previous month and against the same month a year earlier. If either month is missing, it isn't calculated.
• Historical volatility: sample standard deviation of monthly log returns over the last 36 months, annualised (× √12). Requires at least 12 returns.
• Maximum drawdown: largest fall from a previous peak within the window (60 months).
• Correlation: Pearson coefficient of log returns over the months in common (at least 12). It does not imply causation.
• Base 100: each series is divided by its value in the first common month of the period.
• Euros: the dollar price divided by the ECB's monthly average EUR/USD rate (only months with at least 10 published days). Unit conversions use fixed factors (e.g. 31.1035 g per troy ounce).

Property investment (Property Intelligence)

Version property@1.0.0 for the calculator and housing@1.0.0 for the housing index.

• Gross yield = annual rent / purchase price. Net yield = first-year operating income / total investment (price, taxes, costs and refurbishment).
• Loan payment with the French amortisation system; cash flow = operating income − loan payments. All before income tax.
• IRR and NPV on the equity cash flows, with the sale at the end of the period (price × (1 + appreciation)^years − selling costs − outstanding debt).
• House prices: INE Housing Price Index (registered sales, constant quality, 2025 average = 100). We always distinguish a transaction index, asking price, valuation and rent; today we only use the first.
• Estimated value of a property = price paid × current index / index of the purchase quarter. It's the average trend in the area, not a valuation.

Logistics (SupplyChain Atlas)

Versions supply@1.0.0 y searoute-marnet50@1.0.0.

• Sea routes: shortest path on Eurostat's public SeaRoute network (50 km resolution), with strategic passages you can close at will. Arctic routes are closed by default.
• Air, road and rail: straight-line distance multiplied by a visible, editable route or detour factor.
• Lead times: distance divided by an assumed speed plus handling days; by road, 9-hour driving days (Regulation (EC) 561/2006). They're estimates, never quotes or real times.
• Landed cost: customs value under the Union Customs Code (arts. 70 and 71) and import VAT base under Directive 2006/112/EC (arts. 85 and 86). Each item is marked as estimated or confirmed.
• Supplier concentration: Herfindahl-Hirschman index on spend shares. Stock to cover a delay = annual spend / 365 × days of delay.

Energy (EnergySim)

Versions energy-finance@1.0.0 y wind@1.0.0.

• Solar PV: monthly output from PVGIS 5.3 (European Commission, JRC) for 1 kWp with the chosen losses, technology and angles, multiplied by the capacity. It's queried from the server and reused for up to 30 days per location and configuration.
• Wind: monthly mean wind at 50 m from NASA POWER, extrapolated to hub height with the power law, a Weibull distribution and a generic power curve with density correction for altitude. It's a first-order calculation for screening sites.
• Pre-tax financial model with annual cash flows: NPV, project and equity IRR, simple payback, debt service coverage and LCOE (it matches the public calculator when there's no escalation or variable OPEX).
• One-year P90: mean − 1.2816 × the year-to-year deviation from PVGIS, assuming a normal distribution. It doesn't include model uncertainty.
• The initial economic values are examples and are labelled as such; grid connection isn't assessed because we have no capacity data.

Competition (CompetitorIQ)

Versions competitor-matrix@1.0.0 y review-themes@1.0.0.

• Company profile: data from GLEIF's LEI register (CC0), with the LEI validated by its check digits (ISO 7064). The group structure is what each entity declares; if it doesn't report it, the exception is shown.
• Matrix: the latest observed value for each criterion, with its date, marked as verified (with a link to the source) or estimated. Changes are consecutive pairs of different observations in the same cell.
• Score: each numeric criterion is normalised between the group's minimum and maximum depending on whether higher or lower is better, and weighted; a missing value is excluded (it doesn't count as zero) and coverage is shown. It shows which criterion would change first place.
• SWOT: each item is either a cited fact (with source and date) or an inference; the report keeps them apart.
• Review themes: document frequency of words and word pairs, without stop words, with the average rating of the reviews that mention them. No AI, in the browser.
• Monitoring: every day the name, status, legal form, addresses, parents and successor are compared with the last check; the update date on its own doesn't count as a change.

Industrial costs (IndustrialCost)

Versions industrial-cost@1.0.0, industrial-budget@1.0.0 y industrial-lifecycle@1.0.0.

• Capacity: planned hours × OEE (availability × performance × quality) × rate × machines per stage; the line is limited by the slowest stage.
• Unit cost: materials with scrap, labour, energy (variable and base load), water, straight-line depreciation, maintenance, logistics and fixed costs, with overheads and contingency. Every value carries its origin: quote, published price, historical estimate, index or assumption.
• Make or buy: relevant cost (variable plus avoidable fixed costs) versus the purchase cost, and the indifference volume.
• Budget: PERT per line item, sum of independent items and P10/P90 by normal approximation. Expired quotes don't count as quoted.
• Life cycle: present value of costs in constant euros and equivalent annual cost with the capital recovery factor.
• References: Eurostat, only for the EU, EFTA and candidate countries; gas published per GJ is converted to kWh (1 GJ = 277.78 kWh).

Geographic risks (GeoRisk)

Versions georisk-seismic@1.0.0, georisk-climate@1.0.0 y georisk-exposure@1.0.0. Hazard, exposure, vulnerability and risk are distinguished: only exposure is measured.

• Earthquakes: USGS ComCat catalogue, magnitude 4.5 or above since 1976, within the chosen straight-line radius (haversine). The frequency per year is observed, not a probability; if 5,000 results are reached, the period starts at the oldest earthquake received and this is stated.
• Climate: NASA POWER daily data (MERRA-2 reanalysis, cells of about 50 km) for the last 20 complete calendar years; years with fewer than 90% of days with data are discarded. Extreme heat: maximum ≥ 35 °C; tropical night: minimum ≥ 20 °C; heavy rain: ≥ 20 mm (R20mm); dry spell: consecutive days with less than 1 mm (CDD). The comparison between the first and second half of the period is descriptive.
• Portfolio: for each asset, number of earthquakes, frequency, largest magnitude, distance to the nearest and date of the latest; assets without data are counted separately, never as “no earthquakes”.
• Alerts: once a day the earthquakes of the last 3 days are stored and at most one alert per rule and day is created, with the strongest earthquake above the magnitude within the radius, only if it occurred after the rule was created.
• Floods, wildfires, landslides and climate projections aren't included: we link to the official flood zone viewer (SNCZI) instead of showing data without a clear licence.

Agriculture (AgriTech Intelligence)

Versions agri-margin@1.0.0, agri-inputs@1.0.0 y agri-water@1.0.0. Yields, prices, costs and subsidies are user assumptions: there's no calibrated yield model.

• Margin: revenue (yield × price + by-products) minus costs (quantity per ha × price), with and without subsidies. Break-even yield and price; ±20% sensitivity of one variable at a time; yield-by-price table from −30% to +30%.
• Comparator: the same adverse scenario for every crop, and margin per m³ of water and per hour.
• Sowing: seeds per m² = target plants / (germination × emergence); kg/ha using the thousand-seed weight. Fertiliser: units = rate × nutrient content.
• Climate: NASA POWER daily data for 20 complete years; Hargreaves ET₀ (FAO-56, equation 52) with FAO-56 extraterrestrial radiation; FAO monthly effective rainfall; net irrigation = max(0, Kc × ET₀ − effective rainfall) and gross = net / efficiency. Frost: minimum at 2 m of 0 °C or below; degree days with a 10 °C base.

Reports

Every report will include: objective, scope and period; methodology and calculation version; sources with observation date and licence; facts, calculations, estimates and assumptions kept apart; scenarios and sensitivity; uncertainties and missing data; identifier and version.

Scores and comparisons

• Weights are editable and visible, and you can see which change alters the ranking.
• Missing data is flagged; it isn't turned into zero.
• A score is not a probability of success.
• We don't call anything the “best option” without a defined objective.

Use of AI

When we add AI assistants, they'll be used to interpret questions, set up analyses and summarise results by calling the deterministic calculations. They'll show their sources and won't make up data. See AI and limitations.