The Only Clause Available

by

Here are two statements about the same agreement, issued within twenty-four hours of each other, by officials who were present at its creation.

Donald Trump, April 8, 2026: “There will be no enrichment of Uranium, and the United States will, working with Iran, dig up and remove all of the deeply buried Nuclear ‘Dust.'”

Iran’s published 10-point proposal, same week: “acceptance of enrichment.”

The ceasefire that both sides announced as a success contains — must contain — a clause that simultaneously means two incompatible things. This is not a translation error. It is not diplomatic sloppiness. It is, as I will argue, the only clause that could have been agreed to. Understanding why requires a brief detour through a paper published in 2001 that most people outside formal political theory have never heard of, and that those inside it probably underuse.

The paper is Tim Groseclose and Nolan McCarty’s “The Politics of Blame: Bargaining Before an Audience,” published in the American Journal of Political Science in 2001.1 The formal setup involves Congress making a take-it-or-leave-it offer to a president, both observed by an electorate that is imperfectly informed about the president’s preferences. The institutional context is specific — a unilateral proposal game with a single uninformed audience observing one side — but the theory it produces is considerably more general than the setup suggests.

What the theory says is this: when two negotiators bargain before an audience, the audience changes what is rational to propose and rational to accept — in ways that produce systematic, predictable distortions. Congress has an incentive to write bills designed to be vetoed, because a veto signals that the president’s preferences are more extreme than the audience believed. The president, wanting to appear moderate, may sign bills that are worse than the status quo from his own perspective. Bargaining fails not because the parties can’t find an agreement, but because the signals that failure sends are sometimes more valuable than the deal itself.

The central result: gridlock can be an equilibrium outcome even when both parties have perfect information about each other’s preferences and even when mutually beneficial agreements exist. The audience’s imperfect information is sufficient to make apparent failure the rational strategy. The breakdown is not a malfunction. It is the game working as designed.

Now consider what the Iran ceasefire actually required. The Groseclose-McCarty model has one audience, watching one side. The Iran situation has two audiences, each watching their own side — and neither audience is watching the other’s. Trump was negotiating before a domestic audience that needed to hear “total victory on enrichment.” Iranian officials were negotiating before a domestic audience that needed to hear “our right to enrich is preserved.” These audiences are separated not just by geography but by information: neither reads the other’s domestic press closely enough to arbitrate between incompatible readings of the same clause.

As a two-sided game of incomplete information, this is a much harder problem than the one Groseclose and McCarty solved (which Article I, Section VII of the US Constitution partially justifies in terms of sequencing at least), and I think the extension is worth sketching. Think of the United States and Iran as two parties negotiating a contract, with Trump and the Iranian negotiator as their respective lawyers. The Groseclose-McCarty model describes one lawyer making a proposal to the second, observed by the second lawyer’s client. The first lawyer is, in effect, trying to get the second lawyer fired: rejection signals to the client that the second lawyer’s principals have extreme preferences, damaging their standing. The second lawyer, to avoid being fired, sometimes rejects proposals that would be good for both clients — taking a worse policy outcome in exchange for the reputational benefit of not appearing to capitulate.

Now give the first lawyer a client, too. In the original model, voters are assumed to know Congress’s ideal point — a tractable but asymmetric assumption. Once both lawyers have clients who observe proposals and responses, you have two simultaneous problems of the same kind running in parallel. My conjecture is that in this extension, a policy outcome is achievable in equilibrium only if both clients trust their lawyer enough not to fire them after observing the relevant proposal-and-acceptance sequence — and the order of proposal and acceptance is not symmetric. Policy X might be achievable when proposed by Iran and accepted by the United States, but not achievable when proposed by the United States and accepted by Iran, because the signaling value of “who proposed it” differs across the two client relationships. The same deal, navigated differently, produces different reputational outcomes for each lawyer with their respective client.

For negotiations to always reach Pareto-efficient outcomes, an alignment condition must hold: for every Pareto-efficient policy, there must exist at least one proposal-and-acceptance sequence that is itself Pareto-efficient for both lawyers relative to their respective firing constraints. This condition is not automatically satisfied. And when it fails — when no sequence exists that keeps both lawyers employed while delivering a mutually beneficial outcome — the result is not merely gridlock. It is a clause whose meaning must remain unsettled, because settling it in either direction destroys one lawyer’s relationship with their client.

The enrichment clause did not mean two things because the negotiators were careless. It meant two things because that was the only clause available. A clause that unambiguously said “no enrichment” was a clause the Iranian negotiator could not propose, and could not accept without being fired. A clause that unambiguously said “acceptance of enrichment” was a clause Trump could not accept without the same outcome. The clause that both sides signed is the clause that survives both press conferences — settled enough to produce a signature, unsettled enough that neither lawyer has conceded anything their client can see.

The ceasefire expires Monday. I am writing this on Friday, which means I am making a prediction in public, with a short enough horizon that it can be checked. Here it is: the deadline will not produce clean resolution. (Ed.: Oh, that’s just perfect…all this for “more of the same”?) It will not produce a genuine deal — a text whose meaning on enrichment is settled in the same way by both audiences simultaneously. And it will also, I believe and hope, not produce a “clean” resumption of hostilities, which would force both lawyers to account for why the ceasefire failed on their watch.

What it will produce, if my own read of what the Iranian and American “clients” want of their “lawyers” in this scenraio, is a renegotiated form of ambiguity — an extension, a “framework for continued talks,” a statement that both sides can describe as progress while the underlying question remains precisely as unsettled as it was before the clock ran out. Deadlines in two-audience bargaining games do not force resolution. They force signals about what each side is willing to accept. Both sides have already sent those signals. The Islamabad talks broke down not because the parties couldn’t find a clause, but because the clause they needed doesn’t exist in unambiguous form.

If I am wrong — if Monday produces either a genuine deal with settled meaning or an unambiguous resumption of hostilities — then the alignment condition must hold in some way I do not expect that it does, and that would itself be worth understanding.2

With that, I leave you with this.

1 Groseclose, T. & McCarty, N. (2001). “The politics of blame: Bargaining before an audience.” American Journal of Political Science, 45(1), 100–119. https://doi.org/10.2307/2669362. A freely accessible working paper version is available at the Stanford GSB.

2 To be clear, the Iranian leadership’s “client” is very unclear. I think most people actually agree that the current conflict is at the very best inefficient (and I think it is far worse than that, personally). So, given the “randomized control trial being conducted by the US and Israel,” these negotiations are not only incredibly important, they arguably offer an “identification opportunity” with respect to the questions of “who are the new Iranian leadership’s clients?” and/or “what do the members of the leadership believe that their clients want them to sign onto?” That’s a pretty macabre take on causal inference, but it is what it is…