Posts

Posts

Short pieces, posted elsewhere, collected here.

• Mind the gap · August 2026
  The conformal information gap named: a hard deck that split conformal can only approach, never fly below. The Holloway Gambit, its interpretations, and the log-optimal adversary betting against you at your own quoted odds. Read on LinkedIn.

• Negative association is not enough · July 2026
  A correction: negative association in the sampling distribution does not guarantee per-level tightness of conformal guarantees. Counterexamples found, paper updated. The summed-over-levels proof stands, and for most smoothly parameterized copulas it is morally true. Read on LinkedIn.

• A century of balanced walks · July 2026
  Any zero-sum population of bounded vectors can be walked without the running sum drifting. Steinitz's ordering, Riemann and Lévy behind it, Tijdeman's chairman, Welling's herding, and conformal coverage residuals as the same ledger game — balancing, not estimation. Read on LinkedIn.

• Small-sample comfort · July 2026
  The beta-law coverage result, well known under independence, conjectured to hold in far greater generality — with an average-over-levels version proved, and a demo to poke the conjecture with. Read on LinkedIn.