Michael Weissman

I had fun. "For a general power-law prior r_0(x) proportional to x^-a with a in the range a<3 for which divergences are not problemtic P(D=2)=(4-a)(3-a)/2^(6-a). The maximum is obtained for a = (7*ln(2)-2-(4+ln(2)2)0.5/(2*ln(2)) = 0.53 which gives P(D=2) = 0.193."The other big correction is x0.182.

Pekar 2022 was 1 of 2 key papers supporting a market origin of covid, via multiple spillovers. v3 of my arXiv fixing its math errors in a transparent way is now up, with a new histogram. tl;dr The answer from their model applied to their data is the opposite of their claim. arxiv.org/abs/2510.01484

Wait- that's all stuff from Pekar 2022, the paper with enormous math errors that you just said was irrelevant. When those errors are fixed, its 2-spill probability becomes low, which undercuts its whole story. Tho P(other zoo stories) is unaffected by market-story errors.

But I see that my "before" was an overstatement. I will change it to "around the time". Thanks, these tweaks are improvements.

The claim some (A+another mut) variants are ruled out as possible MRCAs.
Nobody seems to think that any version of B is a candidate for MRCA, tho that doesn't rule out the possibility of a separate spill.

The Data are Insufficient to Confidently Root the SARS-CoV-2 Phylogenetic Tree Abstract. Several years ago, I published a paper that described the discrepancy between outgroup and date-based methods for rooting the SARS-CoV-2 phylogen

Of course I've read that, and Bloom's response:
academic.oup.com/mbe/article/...

An Inconvenient Probability v5.11 Bayesian analysis of the probable origins of Covid. Quantifying "friggin' likely"

Here's my summary of the specific market-related evidence. No "A" cases were found in the market, giving a minor factor weighing against the specific market story, though not against general zoonosis.
michaelweissman.substack.com/i/142625697/...

That A looks more ancestral is universally agreed upon. That it was spreading widely early on is based on the Bloom paper. academic.oup.com/mbe/article/...

I have made it clear from the start that I don't know how many spillovers happened and don't think it's very relevant. A reason to think that 1 is more probable than 2 is that it is implied by the Pekar 2022 model, once its indisputable math errors are fixed.

"early" means without further downstream mutations.

Why does NE Biolabs say "The advantage of using Type IIS enzymes for assembly is that the recognition sequence can be placed in the primer on either side of the cleavage site. If placed inside, 3′ to the cleaved end, it will be retained in the construct and can subsequently be reused. " ?

The hypothesized reason for the pre-existing one would have been for convenience in repeatedly swapping in some new segments, 3-5 clones/year. Wouldn't that create different cost/benefit for RE patterns than would other uses?

Lots of things are "empirically zero", e.g. P(FCS at S1/S2| natural sarbecovirus). Or any P(highly detailed properties| some standard hypothesis). For new situations one needs a reasonable subjective estimate of P.

But that wasn't the recCA you claimed, which already treated its most likely version as 100% probable.

Not a guess. Gadboit did the sims and the current P(pattern) fits them almost perfectly.

The theory was that the number was planned. The number was 6 and the plan found later was for 6. The prior wanderings of others are not relevant to my analysis, but the subjective P(leave in|DEFUSE) is. In effect I use ~0.4 for it, but that could get revised downward after more checking.

Assuming that hypothetical ancestor also differed by ~400 nt, I'd say P(pattern|ancestor)=~0.43. That would beat or about tie engineering estimates.
If...

No. Even your recCA is off at one site, and the ~400 syn muts needed to get from it to SC2 have ~2.8% chance of getting it to the (6, <8k) criterion. Not negligible but still kind of low, depending on the competing hypothesis.

Prophetic cherry picking: called the multiple clones plan, 6 segments, BsmBI. Not ex post facto.

probability for actual existing "dogs":
RpYN06: 3/32000 0.009375%
BtSY2: 58/32000 0.1812%
BANAL-20-236: 69/32000 0.2156%
BANAL-20-52: 52/32000 0.1625%
BANAL-20-103: 386/32000 1.206%
RaTG13: 94/32000 0.2938%

and ChimericAncestor: 210/32000 0.6562%
your possible ancestor 888/32000 2.775%

This trait is transparent to selection, unlike e.g. FCS.

They corrected 3 coding errors under pressure. They have not corrected the main two remaining errors, which would reverse their conclusion. These are not subjective errors but straight math.

Unlike the more subjective P(pattern|DEFUSE) it's easier to use stats for P(pattern|Z). Overall for coronaviruses it's 0.15%. With your generous assumptions re recCA, it could go up to 2.8%.

If I understand, those sites were added but then removed by the cuts, different from the leave-sites-in procedure. That's relevant but doesn't tell us whether for many serial partial replacements leaving sites in would be convenient. Hard to get an unbiased answer ex post facto.

By that token anything by anyone who cites anything by any of the 29 authors of Pekar et al. 2022 would have to be presumed false. Let's stick to the specific points. Here the only relevant one is whether leaving RE sites in eases making a series of partial new clones.

I may misunderstand but presumably he was referring to "BsaI sites (5’-GGTCTCN|NNNN-3’) were introduced into the 3’ terminal of the Es fragment and the 5’ terminal of the Fs fragment" from Hu 2017?

I'm not using any of those side points. I am using that there would be some convenience boost for leaving the RE sites in if one planned repeated partial swaps. I'd love to see specific arguments on both sides confined to that specific technical issue.

Why use language like that either direction? It's a technical question for which a dispassionate analysis would be good.

It would be great to hear on that purely technical issue from some genomic engineer who was not emotionally engaged in either side of the covid origins issue. Unfortunately that's a rare breed.

This is the most relevant scientific issue. If there's really zero convenience boost to having the sites in ready for the repeated partial swaps, then that Bayes factor goes away. If there's some convenience boost, then the factor I give is reasonable.