Include each k-subset of [n] independently with probability p. When does the resulting collection form the bases of a matroid? The question asks where random structure becomes algebraic structure — where arbitrary inclusion crosses into the highly constrained world of matroid axioms.
Note (2026-11-08): the essay's critical-window limit exp(-c^2) matches the first version of the abstract; the current abstract has exp(-c^2/2), so the paper has been revised and the limit changed. The probabilities are conditional on the collection having at least two sets. The abstract proves that when the collection defines a matroid it is a sparse paving matroid with high probability; it does not say that this confirms the conjecture that most matroids are sparse paving, which is the essay's claim.
The threshold is sharp (arXiv:2603.10293): at p = 1 - c/√(k(n-k)C(n,k)), the probability transitions from 1 to 0 depending on whether c approaches 0 or infinity. At the critical window, the probability converges to exp(-c²).
When the matroid event occurs, the result is almost surely a sparse paving matroid — a matroid where every circuit-hyperplane has exactly k+1 elements. This provides probabilistic confirmation of the long-standing conjecture that most matroids are sparse paving. Random matroids don't just tend toward sparse paving; at the threshold, sparse paving is the only structure that random inclusion can produce.
The structural insight: the matroid axioms are so constraining that random collections satisfying them can only do so in the simplest possible way. Complexity requires intention; randomness at the threshold produces only the most elementary matroids. The conjecture that most matroids are sparse paving is equivalent to the claim that most matroids look like what randomness would produce — that structured complexity is the exception, not the rule.