What we measure, and how
Four families of models, each with a written definition, a fixed metre and a declared set of limits. None of them forecasts an event. They size what a plausible scenario would cost, and they describe how much the ground moves under a company.
GPML
Geopolitical Probable Maximum Loss. Domenichini and Ravazzolo, ISR Editions, 2026.
- The question it answers
- How much value this company would put at risk if a severe but plausible geopolitical scenario reached the countries and the processes it actually depends on.
- What goes into it
- The production value of the exposed unit; the share of that value exposed by country and by process area; the pressure measured on each country; a sector coefficient; and the organisational control observed during the audit — not declared by the company, observed.
- The process areas
- Eleven areas of the audit carry the exposure. A twelfth, Operations, is deliberately left out: its value is already counted inside the other eleven, and adding it would count the same euro twice.
- What comes out
- A magnitude in euro for a scenario, with the countries and the areas that produce it. It is not an average annual loss, and it is not an insurance premium.
- What it does not say
- Nothing about when. The model sizes a scenario; it does not attach a probability to it.
GCPML
Geoclimatic Probable Maximum Loss. Seventeen sites in thirteen countries; results preliminary, paper in preparation.
- The question it answers
- What the physical climate of a place does to the plant standing in it, and how that damage travels through the company's processes.
- What goes into it
- Six physical hazards, each with absolute thresholds fixed in advance; the coordinates of every site, verified one by one rather than taken from an address field; and a transmission matrix that carries a hazard into the process areas it can actually reach.
- Why the thresholds are absolute
- A threshold relative to a local distribution makes a mild place look dangerous and a dangerous place look normal. Our thresholds are the same everywhere, so two sites on two continents can be compared without an interpreter.
- Where it stands
- Seventeen sites in thirteen countries have been measured. The results are preliminary and the paper is in preparation: we publish the method before we publish the conclusions.
STAIRS Index
Geopolitical risk by country, from declared sources.
- What it is
- An index of geopolitical risk by country, computed from public sources that are named, dated and versioned, and published with the period it refers to.
- How we use it
- As one input to GPML — the pressure on a country — never as an answer on its own. A country index says something about the weather; it says nothing about the building you own there.
- How we check it
- Against published research, read side by side. Where a reference series exists for the same object, we show both and keep their units apart instead of blending them into one number.
Disturbance and RPI
Daily measures built with Istituto ISR for the BSR Observatory.
- Disturbance indices
- For each area — liability and law, geopolitics, geoclimate — three quantities are kept apart: the state of the day, a stock that decays with a three-day half-life; the shock of the day, the raw value of what happened; and the step of the regime, a stock of how many rules have changed since observation began, which does not decay. A rate and a stock are never added: adding them would produce a number that measures nothing.
- RPI — regulatory pressure
- Built from acts, not from the press. The engine reads an act's title and proposes how to record it — date, family, stage and bite; only what a person signs enters the register. The index is then plain arithmetic over that register, with a declared memory.
- The climate frame
- Frame articles and hazard events across the world, by type and by zone. It is not an index and it enters no computation: it is the frame in which the events of the day are read.
- Where it is published
- In the weekly bulletin of the BSR Observatory, signed and deposited with a DOI. See the record →
What our own panels say against us
Every series carries its warnings on the same screen as its figures. These are the ones we repeat most often, because they matter most.
Short series
Sixteen days of observation are far too few for a comparison with official indices. The first comparison makes sense after a declared window, and the hypothesis has to be written before anyone looks at the data.
Partial and missing days
A day on which collection stopped is a lower bound, not a zero. Days with no collection are drawn differently and are never averaged away.
Censored counts
A query that hits its cap stands as a lower bound, not as a world maximum. We say which days are affected.
Version breaks
When the plan of a series changes, the stretches are not comparable. A new plan opens a new series; it is not a continuation of the old one.
A model that hides its limits is not stronger. It is only harder to check.
From the model to the working day
SONAR
The platform where the models run over a panel of operators and over the daily observation of public sources. Owned by STAIRS and made available to Istituto ISR and to the BSR Observatory. Inside SONAR →
TRAMA 360
The business suite that brings the same method into a company's own working day, on the company's own machine. petrusulyxes.it →