The number is on your ICAAP clock, and your risk team already knows it. The ECB set a ~300bp CET1 depletion floor this year under the 2026 SSM geopolitical-risk thematic, applied across 110 significant institutions SOURCE 219. Your submission defends the path.

A reverse test asks your risk team for a world your engine cannot compute.

When the 2026 SSM thematic reached your desk, it came without a scenario file. The ECB set the depletion floor at ~300bp CET1 and left everything else to your book. Your risk engine ran the same intake it always runs, inputs and weights and transmission paths, and returned nothing, because a reverse spine does not take shock parameters as inputs. It takes a failure as a fact and asks your book to reason back. Not what your model computes when you feed it a stress. What world produces a stress of this size against a balance sheet composed the way yours is, and through which channels in your own exposure that world reaches you. That is self-diagnosis, and the supervisor chose it this year because compound geopolitical exposure does not decompose into isolable shocks a forward engine can process in parallel.

So what your submission actually defends is the world, not the number. The world is compound, one entangled geopolitical event rather than five signals your risk committee reads down a list. Reading the signals separately is what a forward engine forces. Reading them as one geopolitical event is what the reverse spine asks. And the difference decides which channels in your book see the transmission first, and at what pace it arrives.

Your book does not meet five shocks. It meets one geopolitical event.

The five signals your risk committee reads down every quarter (Iran and the US, Hormuz, Russia and Ukraine, Xi and Trump, Turkey) are not five hits landing in parallel on your book. They are one loop that sustains itself, and reading them as five separate items misses the mechanism that keeps the loop closed.

Iran closed the strait on 28 February, and the US responded without forcing it back open. Hormuz has carried the physical transmission ever since: commercial shipping displaced onto longer routes, insurance markets repricing, P&I cover withdrawn. Underneath that sits an economic incentive the disruption does not dissipate. Russia's flow finds better terms in a market shaped this way than in a market not shaped this way, and as long as that holds, the closure has an economic underwriting no single-signal analysis names. Trump and Xi modulate what escalates and what does not, because sanctions posture and trade posture depend on both sides in every direction. Turkey moves along whatever routes the four leave open, on regional interests that do not need to pick a principal side. So what looks like five geopolitical events your risk committee reads down every quarter is one loop with an economic core that keeps it closed. Take the war-economy incentive out of the loop and the closure loses its underwriting inside a quarter. Leave it in place and the transmission through your book arrives on the loop's clock, not on the news cycle of any single signal.

Hormuz today: the observed state, not a spring snapshot

Hormuz is not a forecast. It sits at entrenched closure, Day ~159 from the 28 February action, after a full arc across spring and summer: Islamabad MoU on 17 June, retracement, collapse between 7 and 12 July, the strait declared closed until further notice. Effectively shut to commercial shipping, war-risk insurance at roughly eight times pre-crisis, P&I cover withdrawn SOURCE 204 · 205 · 207. That is the state your book is exposed to now.

The read hands you a map, not a number to recompute.

Your risk team already holds the book, and your risk system already runs the quantification for any exposure it has named as material. What neither holds is the reasoning that connects one entangled geopolitical event to which of your specific counterparty groups reach stress first, in what order, and at what pace. The supervisor named that gap this year: the inability to quantify group-wide exposures to specific client groups in a crisis. SOURCE 071

So the division of labour this read holds to is the one the supervisor's data-quality framing already assigns. BearingA never touches your book data. What BearingA runs against your audited posture is the reasoning: calibrated so its output arrives in the language your team already uses, against the precedents your team treats as comparable, the channels your team already reads, the thresholds your submission is already framed against. Your team then runs the quantification inside its own audited framework, because that is the framework the submission has to defend from.

Read at the generic altitude, this geopolitical event is the industry baseline. Read at the desk-analyst altitude, it is what your team already runs day-to-day. Sitting between the two, tailored to your posture but composed outside your books, is the altitude the supervisor reads the submission against.

Four channels, each read only as deep as the evidence carries.

Your exposure meets this geopolitical event through the ECB's own transmission taxonomy: financial market, real economy, safety and security, with a policy-reaction layer over all three SOURCE 073.

The reason the shape matters for your depletion is asymmetry. Across 47 years of BIS confidential banking statistics over roughly 12,000 country pairs, negative geopolitical events cut cross-border credit 10 to 20 percent more between blocs than within them, while positive events produce no matching increase, because the trust international bank credit requires does not rebuild at the pace it breaks SOURCE 068. Capital retreats hard and returns slow. That is the shape your ~300bp is sized against.

Which way this geopolitical event breaks, and what tips it.

Your team needs to know the direction the geopolitical event breaks before your book registers the move, and which channel from the four above carries the transmission.

Phase identification runs on HMM regime characterisation at p<0.001; the phase-by-cluster pattern is observable in the event's own trace. Three directions can tip the current phase, each through a specific geopolitical trigger onto a specific channel.

One path shown. One method you run yourself.

Your submission needs a number your risk committee can stand behind, and it needs the method that produced the number more than the number itself. Any specific figure a narrated path yields is illustrative; the method your analysts run against your own book is what the submission actually stands on.

The path narrated below goes from this geopolitical event to a depletion of the ~300bp order, in enough detail for your team to see the shape. What matters more than any specific figure the path lands is the method that produced it: the reusable procedure your analysts run against your own book, feeding ICAAP Principles 4, 6, and 7 (risk identification, quantification, stress-testing) SOURCE 065. Take the narrated path as a shape reference. Take the method as what the submission stands on.

Where the arithmetic quietly misleads

The 2026 results credit mitigating actions with a real offset: CET1 15.5 percent at end-2025, 12.1 before actions, 13.6 after. The report itself flags them as potentially over-optimistic under a systemic crisis where many banks reach for the same actions at once SOURCE 219 · Chart 5, the fallacy-of-composition caveat the reverse exercise names but does not compose: mitigating actions are shared. You cannot all sell the same assets into the same stressed market, raise capital from the same investors, or reprice into the same demand-compressed book at the same moment your peers do.

Where BearingA stops and your book begins.

Your supervisor cares exactly where the line falls between what a third party can read for you and what only your own team can compose, and it is drawn in the open.

BearingA reads this geopolitical event and its transmission through your book, at production cadence, with every load-bearing claim drilled to a primary source. Your team holds the book data, runs the quantification inside your audited framework, and owns the submission. The division is not BearingA's boundary; it is the supervisor's data-quality framing, and the read composes to it SOURCE 071.

A diligence CRO will test one distinction, and it lands here. The ECB's own research function has built an LLM that reads financial news and scores an aggregate indicator SOURCE 220. Same category of composition, different substrate: an aggregate news feed produces a directional index that could apply to any bank in scope. This read composes the other way: from gated primary sources through a position-specific cascade to a named CET1 line on your book. That is the primary-source, position-specific version the supervisor's firm-specific ask requires.