One run is a vivid story. Fifty runs are a distribution — and the distribution is where the real geometry lives.

Negotiation Arena was built because the most useful thing about a negotiation is almost never the words anyone says first. It is the path the room takes from the opening positions to the eventual deal, deadlock, or walkout.

Most consequential negotiations are not really decided by interests, constraints, or BATNAs in isolation. They are decided by sequencing. A small concession in round two unlocks an offer in round six; a clumsy opening in round one poisons the table before the actual issues are even named.

The Arena treats the unfolding as the object of study. Each party is given goals, style, walk-away thresholds, and what it does and does not know. They exchange proposals, harden, soften, posture, climb down, occasionally walk out.

The most honest thing a simulator can do is refuse to summarize the negotiation, and let the sequencing speak.

• Define the actors: each party gets goals, style, a plausible bottom line, and informational asymmetries.
• Let them interact: no shortcut to a summary.
• Repeat the run: Monte Carlo repetition converts one anecdote into a distribution.
• Read the structure: the useful output is the recurring shape.

A single run can already shift how a brief gets written. The deeper picture only shows up at fifty. That is where the deal rate stabilizes, where the consensus terms harden, where the failure modes sort themselves, and where the settings start to matter.

The easiest mistake a negotiation simulator can make is to be polite. Models trained on conversational data have a built-in bias toward agreement. The Arena is built to resist it.

• Deal rate
• Rounds to resolution
• Consensus terms
• Failure modes
• Sensitivity to settings