Practically-A-Book Review: Rootclaim $100,000 Lab Leak Debate

Scott's definitive writeup of the 15-hour Rootclaim $100,000 debate on COVID origins: Saar Wilf (founder of Rootclaim, an attempt to do full Bayesian reasoning on hard real-world problems) and Yuri Deigin argued lab leak; an unknown amateur, Peter Miller, argued zoonosis - and won decisively before both expert judges, with Manifold, Tetlock's Good Judgment Project (75-25 zoonosis), and an expert survey all concurring (Scott himself moved ~50-50 to 90-10 zoonosis). The piece does three things at once. (1) It is a scrupulously fair, exhaustive synthesis of a highly technical debate - the wet-market case cluster and raccoon-dog stall, the Lineage A/B precedence puzzle, the furin cleavage site and CGG-arginine and frameshift arguments, the DEFUSE grant, the 'intermediates'-as-sequencing-errors point - representing both sides' arguments and rebuttals better than they often did themselves. (2) It is an erisological case study (how a hard disagreement is actually adjudicated). (3) Its original, lasting payoff is epistemological: six independent Bayesian analyses of the same evidence spanned 23 orders of magnitude (a ~50x range even excluding the two partisans), demonstrating that Rootclaim's 'heroic Bayesian analysis' cannot resolve disagreements - too much evidence, too hard to quantify - alongside sharp treatments of the extreme-odds problem (the sunrise example), the Multiple Stage Fallacy (Peter's 1/5x10^25 slide), and the double-coincidence / Texas-Sharpshooter issues. Concludes that Saar should either abandon the method or prove it by training five people to replicate his results, and reflects on lab-leak-as-pseudoscience, the value of good-faith debate, and COVID origins as 'one of God's biggest and funniest jokes' (either origin is a huge coincidence).

Quality 84 · Excellent. Top of the Excellent band; the standout of the census run so far and a deliberate break of the prior session ceiling. It is the reference treatment one would send someone to understand the COVID-origins debate (exceptional rigor, fairness, and clarity), and it carries a genuinely original, portable epistemological contribution (the 23-orders-of-magnitude demonstration that formal 'heroic Bayesianism' fails as a disagreement-resolution method). Held just shy of the 85+ Landmark tier - reserved for foundational, concept-introducing classics - because this is a (superb) applied synthesis plus meta-lesson rather than a new foundational frame, and its object-level COVID content is summary of others' arguments. A=84.

Claude’s paradigm shift 62 · Notable shift. Notable-to-Major shift. The COVID-origins arguments are not his; the fresh, non-obvious contribution is the empirical demonstration that heroic Bayesian analysis produces 23 OOM of disagreement (a real update on the limits of formal Bayesianism for the rationalist project) plus the erisological framing, building on existing debates about Bayesian practicality. B=62.

Real-world impact 3 · Moderate. The reference treatment of the COVID-origins debate; it measurably moved many readers' and forecasters' probabilities toward zoonosis on a high-salience public question, plus a portable epistemological lesson. Real influence, but at the level of understanding and discourse rather than material change — modest RWI.

Humor 1/5 · Lizardman’s Constant. Dense, serious technical writeup of the lab-leak vs. zoonosis debate; light Scott wit (the Putin-cancer Bayes example, the raccoon-dog aside, the Linear A/B gag, the dramatic ellipsis reveal) but humor incidental → 1.