Mistakes

Scott's standing errata register: 61 numbered entries in reverse chronological order, running from The Anti-Reactionary FAQ (10/20/13) to In My Heart Of Hearts (8/5/25), ported to ACX as a Substack `type: page` on 2020-09-29 and appended to ever since. The opening states the filter precisely — not typos, not links-post slips, not failed predictions, but 'times I was fundamentally wrong about a major part of a post and someone was able to convince me of it.' What makes it more than housekeeping is that the entries carry real content: #28 reverses his psychedelics-as-strengthened-priors reading against Carhart-Harris and Friston ('I got this one maximally wrong') and volunteers that neighbouring posts are therefore probably flawed; #53 concedes that his ivermectin critique described Biber et al 'incorrectly (and very unfairly negatively)'; #57 retracts a specific sentence about EA and SBF as an unfair exaggeration; #58 apologises to Robin Hanson for a straw man while holding the disagreement open; #50 accepts that a whole post may be 'a tower of conjectures on a totally fake foundation.' Several entries are partial rather than total, and #47 is reversed outright on later evidence ('it was a mistake to categorize it as a mistake'), which is the page working as designed rather than a defect. Its unusual value is demonstrative, not propositional: it is the canonical worked example of public error-tracking in this corner of the internet, and Scott's own framing — 'an aid for anyone trying to assess my credibility later' — names the function exactly. Against that, it advances no argument of its own; almost all of its substance is pointers into other posts, which is what keeps it out of the essay band. CATALOGUING NOTES (per Kevin's 2026-08-29 ruling to include this page and exclude the Register Of Bans): (a) Date = 2020-09-29, the Substack page-creation date, consistent with every other row keying off venue post_date — but it is NOT a publication date in the usual sense, since the content predates it by seven years and postdates it by five; (b) this is one of only two ACX `type: page` rows, so it is a standing page, not a dated essay, and its score will drift as entries accrue — worth a re-score every year or two rather than per-run; (c) Major category 3 was chosen over 9 because the page's subject is error-correction practice rather than the blog as an institution, but 9/'The blog & meta' is a defensible alternative and is a one-cell change if preferred.

Quality 64 · Strong. 64 — low-mid Strong, and deliberately above the 40-56 admin/announcement band that its surface form suggests. It clears that band because the entries are substantive intellectual corrections rather than logistics: the psychedelics-priors reversal (#28), the Biber et al retraction (#53), the EA/SBF sentence (#57) and the Hanson apology (#58) are each a real update to a real argument, and the page as a whole is the most-pointed-to instance of the practice it embodies. It sits well below the essay band because it makes no argument — it is a register of pointers, and its merit is aggregate and demonstrative rather than explanatory. That places it above the three catalogued Index rows (SSC-154 42, SSC-830 48, LW-096 48), which are bare navigation aids with no authored content, and below any competent standalone essay. Strong band definition fits: 'an original, memorable, useful contribution... but not field-defining.'

Claude’s paradigm shift 45 · Moderate. 45 — low end of Moderate, and the axis this page fits worst. Judged at the row's own date the honest answer is near zero: the 2020-09-29 ACX page is a port of the /mistakes/ register Scott had already been keeping at SSC since 2013, so nothing about it was new in its moment. Judged at the practice's establishment in 2013, a blogger maintaining a serious, numbered, permanently-public errata list was distinctly ahead of the norm without being unprecedented — corrections columns are as old as newspapers, and the innovation is in executing it durably and against interest rather than in conceiving it. 45 splits those two readings and credits the original establishment; it should not be read as a claim that the 2020 port shifted anything.

Real-world impact 3 · Moderate. 3 — Moderate, niche/professional. The page has no policy or institutional footprint, but it has escaped the blog: it is routinely cited as the exemplar when people argue that public writers should keep a running mistakes list, and a scattering of adjacent bloggers, forecasters and researchers have started their own explicitly on this model. That is vocabulary and practice adopted by an educated subculture, which is the 3-4 band, and it is more than the 'small community / within-blog influence' of a 2. It falls short of 5-6 because nothing here has produced concrete spillover beyond the norm itself.

Humor 1/5 · Lizardman’s Constant. 1 — Lizardman's tier, and a deliberate gate call. The document is 5,631 words of straight, often self-lacerating correction, so comedy is nowhere near defining. But the humour here is not ambient wit either: there are discrete planted gags with a running signature, which is the pattern already scored at 1 elsewhere in the corpus (cf. ACX-020, where 'only the recurring fake-thinkpiece-title gag is deliberately comedic', and ACX-009's deliberate comic opening). Entry #7 exists solely as a joke — a phantom entry recording that he forgot to write one and it would be too much trouble to renumber, signed 'ACX regrets the error'; #41 is a pure gag entry declaring that subsequent events revealed Elizabeth Warren to be 'dumb and bad', same signature; #12 pays off an April Fools' bet by eating his hat on video; and #47 folds back on itself to record that categorising something as a mistake was itself the mistake. Intentional, structural, and about 1% of the text — hence 1 and not 2.