Summary view
This market asks whether any nuclear weapon is detonated in a given calendar
year, counting declared tests, wartime use, or accidental yield-producing
detonations. The yearly buckets are independent: probabilities
reset each January rather than accumulating across years.
This version is base-rates only. It does not use an inside
view about current crises, named countries, or live geopolitical catalysts.
The probabilities come from historical reference classes for years in which at
least one qualifying nuclear detonation occurred.
Reference classes
1. Broad modern reference class: post-CTBT-era detonation years
The broad relevant reference class is the modern test-ban era rather than the
Cold War. CTBTO notes that fewer than a dozen nuclear weapon test explosions
have been declared or detected since the Comprehensive Nuclear-Test-Ban Treaty
opened for signature in 1996. Counting calendar years with at least one
qualifying detonation gives six detonation years between 1997 and 2025:
1998, 2006, 2009, 2013, 2016, and 2017.
That raw frequency is materially above zero, but it likely overstates the
current annual risk because it includes the short cluster of detonation years
before the present long no-test stretch had fully emerged.
2. Narrower conditioning reference class: the current long quiet spell
CTBTO also reports that, as of January 14, 2026, the world has gone more than
eight years without a declared or detected nuclear weapon test explosion, the
longest such period since 1945. A naive recent-frequency estimate from that
quiet spell alone would be zero, but zero is not a defensible annual forecast.
So we smooth it rather than taking the raw recent frequency literally.
3. Pooled base-rate choice
We therefore use a pooled base-rate approach:
- The broad post-1996 detonation-year base rate is the upper anchor.
- The eight-year no-test streak is the lower anchor.
- A smoothed midpoint of those base rates gives a 10.0% full-year annual probability.
This keeps the estimate historical and mechanical. It uses only observed
detonation-year frequencies plus smoothing for a run of zero recent events.
Probability framework
We model each calendar year separately. Within a year, monthly submarkets are
the cumulative probability that at least one qualifying detonation has occurred
by the end of that month. Across years, the buckets are independent and reset
to January.
The annual probabilities used for the staged update are:
| Year bucket |
Annual probability |
Interpretation |
| 2026 (Apr-Dec remaining) |
7.6% |
Derived mechanically from the same 10.0% full-year base rate after conditioning on no detonation in January-March 2026. |
| 2027 |
10.0% |
Full calendar-year base rate. |
| 2028 |
10.0% |
Same full calendar-year base rate; no inside-view adjustment between years. |
Within each year we use a constant monthly hazard implied by the 10.0%
full-year base rate, then convert that into cumulative month-end
probabilities. For 2026, the same monthly hazard is applied only to the
remaining April-December months because March 11, 2026 has already passed
without the event.
Interpretation notes
- The probabilities are intentionally simple and stable across adjacent years.
- The 2026 bucket is lower only because less of the year remains.
- No country-specific or crisis-specific adjustment is applied.
Sources