How to read it

Kahn arranged 44 kinds of action in a nuclear confrontation from mild to catastrophic, grouped them in seven bands, and marked six thresholds (the red dashed lines) where he thought the character of a crisis changes, from "Don't Rock the Boat" at the bottom to "City Targeting" near the top. RAND's Dangerous Thresholds notes that more than half of the 44 rungs involve at least some use of nuclear weapons. The names on this page are Kahn's, quoted exactly.

New to the page? Press Learn to play for a five-minute guided first game on the Cuban Missile Crisis. Otherwise, pick a crisis in the panel and step through it with Back and Next (or the arrow keys). Each step is a dated event with its source. The blue number on a rung shows which step reached it; the highlighted rung is the current step. Choose a second crisis under Compare with to overlay its path in orange. Click or tap any rung to see which crises reached it.

At the end of a crisis you can retry it from the allied side: make the U.S. or allied decisions yourself, among options that were really on the table, and watch a notional model (placeholder numbers for thinking, not estimates) shift the odds of four endings: a negotiated settlement, a frozen standoff, limited war, or nuclear use. The colored bar is your score; the orange and red at its right end are "war or worse." Pick an option that did not happen and you leave the record: the crisis keeps going as a model-generated branch, with dice whose seed (a fixed starting number) makes a shared link replay the same rolls; "Run 1,000 futures" then estimates where such branches end. See the method notes below.

No crisis here crossed the No Nuclear Use Threshold. Most of them spent their dangerous hours on rungs 5 through 11: shows of force, confrontations, and alerts. Several also moved down, which a ladder drawn as a climb does not show well.

Method and coding

• Rung, group and threshold names are transcribed from Kahn's figure (On Escalation, 1965, p. 39) as reproduced in RAND R-3235 (1984), Fig. 2.
• Every event is dated and linked to a source that was opened and checked on 2026-09-29. Placing an event on a rung is marked coding by the author of this tool unless a source makes the placement itself.
• Kahn wrote the ladder for a U.S.-Soviet confrontation. Applying it to India and Pakistan, North Korea, or Russia in Ukraine stretches it; the coding picks the closest rung name, and the step notes say why.
• Where one side's move and the other's response happened days apart, they are separate steps. Rungs reflect the most escalatory action in each step, not the whole crisis.

Method: Retry from the allied side

At the last step of a crisis, or from the button under the crisis list, you can replay it as the U.S. or allied side. At each decision point you choose among two to four options. One is what actually happened, marked after you choose. The others were considered or publicly debated at the time, and each cites a source that was opened and checked on 2026-09-29. No option was made up to fill a slot.

• Who you play. Where the United States or an ally was a party, you play that side. Kargil and Balakot had no U.S. or allied party; you play India, the U.S. partner and the democracy defending its side, because the sources document India's options (RAND on crossing the LOC, Mukherjee on India's menu after Pulwama) and record none that Washington weighed. In the 1969 Sino-Soviet crisis you play the United States, the third party Moscow probed.
• Rungs. The historical option moves the crisis to the rung of the next historical step. A rung for an option that did not happen is coding by the author of this tool, with the reason shown after you choose.
• Leaving the record. Once you pick an option that did not happen, the later historical decisions are gone; they assumed a history that no longer holds. The crisis continues as a Model-generated branch: notional, not history, drawn with dashed purple borders so it cannot be mistaken for the sourced steps.
• Branch mechanics notional. Round 1 is your departure; its move type (de-escalate, hold, escalate one step, major escalation) is read from how far it moved the rung. In every round the opponent's response is one seeded roll on the response table below, keyed by the rung band after your move and the move you made: backs down, holds, matches or escalates, each moving the rung by a set amount. A second roll, the end roll, then checks for settlement (more likely when either side softens, and lower on the ladder) and, at rung 12 or higher, for fighting that locks in as a limited war. A rung of 21 or more, past Kahn's No Nuclear Use Threshold, ends the branch as nuclear use. If nothing ends it, the branch is scored after the set number of rounds: limited war at rung 12 or higher, otherwise frozen standoff. From round 2 you choose among four generic moves worded for the period. Where a wording repeats an alternative a source records as considered, the source is linked; the model's response is never sourced. The log shows each roll, the chance it was rolled against, and the result, as in Penghu Gambit.
• Dice. The generator is Mulberry32 with the seed shown on the page (default 163). The hash carries the historical choices, the branch moves and the seed, so a shared link replays the same rolls. "New dice" draws a new seed.
• 1,000 futures. Replays the branch 1,000 times from the moment you left the record, with your departure fixed and a new seed for each run derived from the page seed. Later rounds follow a policy: your own choices in order (repeating the last), always de-escalate, always hold, tit-for-tat (answer a back-down with de-escalation, a hold with a hold, a match or escalation with one step up), or always escalate one step. The key-drivers table changes one notional number at a time and replays the same seeds, so a difference comes from the lever, not the dice. Because a war that locks in ends the branch, the model does not follow a war upward; raising the lock-in chance can lower the nuclear share. That is a limit of this model, not a finding.
• Default branch numbers notional, all editable under "Edit the model":
• Outcomes. Four ways the crisis could end: negotiated settlement, frozen standoff (it ends or fades without agreement), limited conventional war (fighting that widens beyond the crisis, defined per crisis under "Why this side?"), and nuclear use. They follow two breaks on Kahn's ladder: the move into large conventional war (rung 12), and the No Nuclear Use Threshold.
• The model notional. It starts from a distribution over the four outcomes. Each round of the crisis, a historical step or a choice, sits on a rung. The rung's Kahn group sets a chance of sliding one outcome up (settlement → standoff → war) and one down; a war slides into nuclear use at a set share of the slide-up. Climbing several rungs in one move adds risk per rung; stepping down adds relief per rung. An option that is itself a deal pulls some standoff toward settlement; one that leaves the dispute in place pulls some settlement toward standoff. The arithmetic is exact, with no random draws, so the same choices always give the same numbers.
• Grounding. The structure, not the numbers, follows the literature this tool already cites. Hazards change at Kahn's group boundaries because Kahn put his thresholds there. The slide-up grows faster than the slide-down higher on the ladder because Dangerous Thresholds (RAND, 2008) holds that escalation is usually easier than de-escalation. Large single jumps cost extra because Hersman's "wormhole" escalation stresses sudden skips. No source gives any of the numbers. Every one is a placeholder, and you can change it under "Edit the model."
• Kennedy's estimate. On the Cuba run, a hatched band marks Kennedy's own estimate that the chance of escalation to war was "between 1 in 3 and even," as reported by Graham Allison (Foreign Affairs, 2012). The page uses Allison's wording; it did not open an earlier source for the estimate. The band is a reference point. The model was not fitted to it, and at the default numbers it runs lower.
• What the model cannot see. It knows only rungs. It cannot price Soviet troops arriving in Egypt in 1973, or the Soviet reading of Able Archer that made 1983 dangerous. The ladder has the same blind spot, which the critiques above describe.
• Left out.

Crises and highest rung coded

Sources