The Scorecard: Three Months of the US-Iran War Game, Graded

In April I ran the US-Iran confrontation as a probabilistic war game. Seventeen iterations. Fourteen actors. A three-month window. I locked the reference file on 24 April and never touched it again. That was the point. A forecast you can quietly revise is not a forecast. It is marketing.

Before the grade, the plain language. The terms carry the method, and if they stay jargon you cannot hold me to anything.

An actor is not a country. It is a decision-making bloc with its own interests, its own constraints and its own options. States contain several of them and they do not agree. This run modelled fourteen. The Revolutionary Guard command and the civilian negotiating delegation were modelled separately, because they want different outcomes and can veto each other. The White House pragmatists and the security hawks, separately, for the same reason. Then the Gulf coalition, Pakistan, the nuclear inspectorate, the European three, China, Russia, the Israeli security establishment, the Iranian bazaar and labour bloc. Treating a state as one voice is the fastest way to be wrong about it.

An iteration is one complete run of the confrontation, played forward three rounds, every actor choosing under its own incentives, from the same starting position. Run it once and you have a story. Anyone can tell a story. Run it seventeen times and you have something you can count.

That is what probabilistic means here. The output is not a prediction. It is a frequency. Seventeen of seventeen and four of seventeen are different claims, and the difference between them is the entire product. The outcome the runs converge on is the dominant pathway. Something that shows up in a handful of runs is a live risk, not a forecast, and I mark it as one. Reporting the rate is the discipline. Reporting the impression is what a pitch does.

Locking is what makes the rest of it mean anything. The file was fixed on 24 April, before any of this happened, and the numbers below were written down in advance. I did not go back and improve them once reality arrived.

That window closes on 24 July. Here is the grade. Both columns.

What it called

The instrument. The dominant pathway across the seventeen iterations ran a US naval blockade priced at roughly 435 million dollars a day in Iranian cost. On 14 July, US Central Command resumed exactly that instrument. Full coastline. Every flag. Strait transits collapsed to roughly ten ships a day against a normal of about eighty-eight, per Bloomberg, 11 July.

The actor. The model’s most dangerous pathway was an Israeli strike on Qom or the Supreme Leader’s office. It prepared that targeting package in eleven of the seventeen runs, considered and withheld authorisation in nine of those eleven, and executed it in zero. Inside the window, no such strike has fired in reality either. I should be precise about what that does and does not claim, because the file is open to anyone who wants to grade it: the model did not predict Israeli passivity. Israeli spoiler operations appear in all seventeen runs, and strikes on Iranian nuclear scientists were executed in seven of them. Israel’s last strike on Iran inside the window came on 8 June. What the model called was the escalation running through the hawk and CENTCOM track rather than through a decapitation strike, and since the war resumed in July the strikes on Iran have been American.

The shape. No full deal inside the window, and no collapse. A structural limbo in which no single actor can break the structure alone. Seventeen of seventeen iterations ended there, and so did the quarter: no signed deal, no regime collapse, no declared total war, no permanent resolution of the Strait.

The reflex. The model logged a dealmaker reflex, deals running underneath escalation, in eleven of seventeen iterations. Mid-escalation, Trump scrapped his own 20 per cent Hormuz toll under Gulf and shipping pressure and swapped it for investment deals. Strikes upstairs. Deals downstairs.

The third party. This is the call I want on the record while it is still only a report, and I am going to grade it strictly. The model flagged Saudi nuclear symmetry as a watch indicator: observable Saudi progress with China National Nuclear Corporation, logged as a threat in four of the seventeen iterations. It named the branch too. If the threat stayed verbal, the file said, Riyadh would have calculated that the American 123 Agreement path was worth more than the Chinese alternative. On 21 July the Wall Street Journal reported, and the Associated Press confirmed through two sources familiar with the matter, that Trump has approved a thirty-year civilian nuclear agreement with Saudi Arabia that may permit a uranium enrichment facility on Saudi soil, with a public announcement expected imminently. So the model called the demand, and it called the American branch. It did not call the timing: it expected this alongside an Iran deal, not in the middle of a war fought over Iranian enrichment. And four of seventeen is a modest rate, not a headline. I am logging it before it is confirmed so it can be graded either way, and I am reporting the rate rather than the impression.

What it missed

The temperature. Across all seventeen iterations the model put quiet diplomatic limbo as the dominant outcome. Nothing about this limbo is quiet. The record of the closing weeks: a re-armed blockade. By 21 July, an eleventh consecutive night of US strikes, per US Central Command. The executed strikes on Iranian bridges and on power and desalination infrastructure, 17 and 18 July, the exact escalation that had until then only been threatened. And on 17 July, an Iranian ballistic-missile and drone strike on Muwaffaq Salti Air Base in Jordan killed US service members, the first US combat deaths from direct Iranian fire since March. On 18 July Iran formally suspended its commitments under the Islamabad MoU. The diplomatic frame is not dissolved. It is suspended by one of its two sides while the shooting continues.

The model captured the structure: nobody escapes. It mischaracterised the character. Not quiet limbo. A loud, lethal stalemate, with diplomacy stalled but never formally dead.

What the market says

Hull war-risk premiums for Hormuz transits now run 3 to 10 per cent of vessel value, against roughly 0.25 per cent before the war. On a 100-million-dollar tanker that is 3 to 10 million dollars per voyage. The market is not pricing a resolution. It is pricing the structure the model found: a confrontation nobody can end and nobody can win.

The honest ledger

Grade the model the way you would grade an analyst. On the instrument, the shape and the reflex: right, and specifically right, months in advance. On the actor: right about the pathway that did not fire, and precise about what that leaves unclaimed. On the Saudi track: right on the demand and the branch, wrong on when, at a modest four of seventeen. On the temperature: wrong, and publicly wrong.

An assessment that cannot show you its misses is not an assessment. It is a pitch.

That is the discipline this publication exists for. The reference file stays locked. If the next quarter runs, it runs as a second, versioned file, graded against its own window in its turn. The track record stays public either way.

The method behind this file is the same one I run for clients against their own confrontations.

→ Request the war-game simulation