An informal Bayesian write-up produced immediately after watching the debate, by an AI-alignment researcher (not a virologist/epidemiologist). Starts from a prior of 1:5 (lab-leak:zoonosis) built from base rates for zoonotic pandemics vs. WIV gain-of-function/leak/pandemic-start probabilities, then updates on the Huanan-market-centered case cluster (the biggest single factor, 3:1000 against lab-leak), the Connor-Reed/Chen early-case claims (1:1, judged unreliable), and market environmental samples (reduced to ~1/30 after adjusting for testing-selection effects). Final posterior ~9,216:225,000 ≈ 1:24, i.e. roughly 96% zoonosis — matching the “I get 96% zoonosis” summary given in his own tl;dr. Explicitly one of the case’s “six independent analyses” (per Scott Alexander’s ACX table and Rootclaim’s response to it).
relevance_note: the sixth of the six point-estimates behind the case’s “23 orders of magnitude” framing, and the most explicit about its own low epistemic confidence.