What 50 Simulations Taught Us About the Iran Peace Deal

We ran the full Iran–US peace negotiation scenario fifty times on Aurelon. Here’s what the statistical distribution revealed about the conditions for success, the tipping points that consistently broke talks, and why the median outcome is neither optimistic nor catastrophic — it’s something more interesting.

The brief was simple and the scenario was not. A delegation led by the United States sits down with counterparts from the Islamic Republic to discuss a phased return to compliance: sanctions relief in exchange for inspection regimes, ballistic missile constraints, and a public-facing de-escalation framework. European observers are present. Regional stakeholders — Israel, Saudi Arabia, the Gulf states — are not at the table but loom over every clause. We gave the agents full personas, injected live research on the current state of play, and let them run.

Across fifty independent runs, the headline number settled where few of us expected it to. The deal rate was 34%. Not the collapse you might read in a newspaper column, and not the breakthrough that optimistic analysts keep promising. A third of the time, the parties got to a framework. The other two-thirds split between structured walkaways (41%) and formal deadlocks that ran out the clock (25%).

“The most common failure was not ideological. It was procedural — which side agreed to move first.”

What was more interesting than the rate was the shape. The successful negotiations clustered around a specific structural pattern. In 28 of 50 runs, there was a moment — almost always between rounds three and five — where one side made a small, visible, costly concession unrelated to the main dispute. When that happened, the deal rate conditional on the moment occurring jumped to 71%. When it didn’t, it fell to 9%. The “four rounds to breakthrough” pattern was so consistent we started looking for it as a leading indicator.

The variable that broke more talks than any other was not nuclear at all. It was the sequencing of sanctions relief. In 18 of the 33 failed runs, the conversation stalled on whether relief would be tied to verification milestones or front-loaded as a confidence-building measure. The Iranian side, in persona, consistently framed front-loading as a dignity issue; the American side, in persona, consistently framed verification-first as a political necessity at home. Neither position was wrong. Both were structurally incompatible without a third framing.

The third framing, when it emerged, almost always came from the European observers. In runs where the EU voice was given meaningful airtime, it offered a phased calendar that decoupled specific sanctions from specific inspections — essentially a scaffolding that let both sides claim movement without conceding the principle. This is the kind of insight that only falls out of many runs. A single negotiation would have looked like luck. Fifty runs showed it was a mechanism.

“The European voice was a procedural unlock more often than a substantive one. When it was silent, the room got smaller.”

The median outcome, and this is where the analysis gets uncomfortable, is neither success nor collapse. It is a kind of managed ambiguity: a joint statement, a working group, a commitment to reconvene. Not nothing, not a deal. In 12 of the 50 runs the outcome was explicitly this shape — and in the simulations where it happened, the follow-on question (“does the second round produce a deal?”) looked very different than the first round, because the parties had already paid the cost of meeting and had something they could point to.

If we zoom back out, three things seem robust across runs. First, the framing of sequencing matters more than the substance of sanctions relief; the second-order problem keeps eating the first-order one. Second, the presence of a credible third party who can propose face-saving scaffolding raises the deal rate dramatically — the EU is structurally useful in a way that is not about their leverage but about their role. Third, the median is ambiguous rather than binary, and treating a real negotiation as having a pass/fail outcome systematically misses the most likely result.

“A third of the time, they got there. Another quarter, they ran out the clock talking. The interesting runs were the ones in between.”

None of this is a prediction. It is a map of the scenario’s topology — where the saddles are, where the cliffs are, and which ridges keep appearing no matter which trajectory you start from. We don’t know which run the real world is on. We know the terrain a little better than we did before we started.