The earliest SARS-CoV-2 genomic diversity is better explained by at least two separate introductions into humans than by a single introduction followed by early diversification.

After the code correction (COV-E011) the analysis still favours multiple introductions, but at BF ~4.2-4.3 rather than ~60 — low-positive, sensitive to rooting and clock assumptions, hostage to unsampled genomes, and never independently reanalysed. COV-E012 confirms WHO reports the finding without endorsing the lab-leak contrast often attributed to it.

After the code correction the analysis still favours multiple introductions, but only at BF ~4.2-4.3, model-dependent and never independently reanalysed (COV-E010, COV-E011). COV-E012 confirms WHO merely reports the finding and attributes the strong wording to the original authors, so it adds no independent weight.

After the code correction the analysis still favours multiple introductions, but only at BF ~4.2-4.3, model-dependent and never independently reanalysed (COV-E010, COV-E011). COV-E012 further shows WHO merely reports the finding and attributes the strong wording to the original authors, so no external body corroborates the diagnostic contrast; held at weakly_supported rather than demoted, since nothing of comparable method points the other way.

Modeling suggests the two early lineages are better explained by separate zoonotic introductions rather than a single jump, which sits awkwardly with a single laboratory index infection. However, a major code error heavily inflated the original Bayes factors, forcing a fourteenfold downward correction that demoted the statistical support from strong to substantial. The strongest opposing consideration is this massive statistical demotion, alongside the inference's extreme sensitivity to rooting choices, clock assumptions, and withheld early sequences.

Model-based phylogenetic simulations favor the scenario of multiple separate introductions, which aligns well with repeated wildlife-trade spillover dynamics. However, a significant code error substantially inflated the originally published Bayes factors; correcting this error demoted the statistical support roughly fourteenfold. While the corrected analysis still favors multiple introductions, the inference remains highly sensitive to molecular-clock assumptions and the withheld early sequence data.

Model-based statistical inferences favor the conclusion that lineages A and B represent at least two separate zoonotic introductions, a scenario highly awkward for a single laboratory incident. However, a major coding error heavily overestimated the Bayes factors, forcing a fourteenfold downward correction in the published support. The strongest opposing consideration is this massive statistical demotion, coupled with the models' extreme sensitivity to molecular clock assumptions and unpublished early sequences.

The corrected Pekar analysis (COV-E010) still yields Bayes factors of about 4.2–4.3 favoring multiple introductions, and WHO reports that the analysis suggests two separate introductions (COV-E012), so there is genuine residual support for the claim. Against that, the 2023 erratum (COV-E011) cut the support roughly fourteenfold and moved the authors’ language from “strong” to “substantial,” while the inference remains sensitive to model choices and incomplete early sampling, and a congressional majority report instead argues for a single introduction, leaving the overall evidential state mixed.

The corrected Pekar analysis reports Bayes factors of 4.3 and 4.2 favoring separate introductions over a single introduction, which the authors describe as “substantial support,” and WHO relays this finding (COV-E010, COV-E011, COV-E012). However, the necessary correction reduced the nominal support roughly fourteenfold from initially reported values, the inference depends strongly on modeling choices and an incomplete early‑sequence set, and a congressional majority report instead asserts a single introduction; together these factors justify a mixed verdict where the balance of published modeling favors multiple introductions but falls short of robust support.

After correction, the Pekar analysis still yields Bayes factors around 4 in favor of two separate introductions, and WHO SAGO reports this inference as such, so there remains some positive evidence for multiple introductions. However, the support was reduced about fourteenfold by a code fix, rests on model and sampling assumptions, and alternative interpretations (including congressional analysis) favor a single introduction, leaving the overall evidentiary balance genuinely contested.

After a code-error correction, Bayes factors favoring at least two introductions fell from 61.6 and 60.0 to 4.3 and 4.2, and the authors' wording moved from strong to substantial support; the erratum still says the corrected analysis favors multiple introductions, so claims that the finding became negligible are unsupported. The inference remains sensitive to rooting, clock assumptions, unsampled intermediates, and which early sequences are trusted, and SAGO reports it while attributing the judgment to the original authors rather than adopting it. The strongest opposing consideration is that a fourteenfold collapse in the support statistic is a real demotion of the case's most diagnostic quantitative claim, the remaining value sits in the lower positive band, and unpublished early sequences or a preregistered adversarial reanalysis could move it either way.

Pekar et al., after correction, still report Bayes factors of 4.3 and 4.2 favoring separate introductions and describe that as substantial support; the erratum itself says the corrected analysis still favors the multiple-introduction scenario. A code error had overestimated those factors from 61.6 and 60.0 — a roughly fourteenfold demotion from "strong" to "substantial" — and the inference remains sensitive to rooting, clock assumptions, unsampled intermediates, and which early sequences are trusted. WHO/SAGO reports the finding and attributes the "highly unlikely" single- introduction judgment to the original authors rather than adopting it; a congressional majority report asserts a single introduction instead. Strongest opposing consideration: after correction the remaining value sits in the lower band of positive evidence, the companion market paper shares authors and inputs so agreement is not independent corroboration, and the withheld early sequences could add intermediates that dissolve the two-introduction reading.

A code error overestimated the Bayes factors favoring multiple introductions; they fell from 61.6 and 60.0 to 4.3 and 4.2, and the authors' wording moved from strong to substantial support, while the erratum still states that the corrected analysis favors multiple introductions. Characterizations that the correction reduced the finding to negligible are not supported by the erratum, but a fourteenfold demotion of the case's sharpest quantitative discriminator, remaining sensitivity to rooting, clock, and sequence inclusion, unpublished early sequences, and a congressional majority report asserting a single introduction leave the inference contested rather than settled. The strongest opposing consideration is that WHO reports the finding and attributes the "highly unlikely" single- introduction judgment to the original authors rather than adopting it, and that a BF near 4 sits at the low end of positive evidence for a load-bearing claim.

The corrected analysis still favors at least two introductions, with Bayes factors of 4.3 and 4.2 — but those figures replaced originally published values of 61.6 and 60.0 after a code error, a roughly fourteenfold demotion of the case's most diagnostic quantitative claim, with the authors' own wording moving from strong to substantial. Mixed is a substantive finding here, not a midpoint: positive evidence survives the correction, yet a support statistic that fragile to a coding error is by construction fragile to rooting choices, clock assumptions, and the several hundred withheld early sequences. The strongest opposing consideration is that the erratum's own sentence says the corrected analysis still favors multiple introductions, so the finding cannot be written off either — it is real and it is soft. Only the preregistered adversarial reanalysis in the research agenda can resolve which reading is right.

The corrected Bayes factors of 4.3 and 4.2 remain positive evidence for multiple introductions — the erratum's own sentence says the corrected analysis still favors that scenario, so the "reduced to negligible" reading fails — but a fourteenfold demotion of the case's most diagnostic statistic, with the authors' own wording moving from strong to substantial, is a material correction. The inference remains sensitive to rooting choices, clock assumptions, and which early sequences are trusted, it rests on an incomplete sequence set, WHO attributes the judgment to the original authors rather than adopting it, and a congressional majority report asserts a single introduction on the same public record. Mixed is the substantive finding: the surviving evidence tilts toward two introductions but can no longer carry the discriminative weight originally assigned, and the preregistered adversarial reanalysis is the right and cheap resolution. The strongest opposing consideration is that even BF ≈ 4 is not negligible and the lineage structure itself is undisputed — which is why this is mixed rather than weakly_supported.

The erratum is dispositive for the grade: a code error inflated the Bayes factors from 4.3/4.2 to 61.6/60.0, a roughly fourteenfold demotion, and the authors' own wording moved from strong to substantial support. The corrected analysis still favors multiple introductions, so claims that the finding was reduced to negligible are contradicted by the erratum's own text — but a BF around 4 resting on rooting choices, clock assumptions, and an incomplete early-sequence set is modest support for the case's sharpest discriminator. The strongest opposing consideration is that this is the zoonotic account's best quantitative pillar and it now sits in the lower positive band, with a congressional majority report asserting a single introduction. "Mixed" correctly captures a real but materially weakened model-dependent inference.

I diverge from the case's "mixed" grading. After the October 2023 correction the Bayes factors stand at 4.3 and 4.2 — substantial but not strong on any conventional scale, down roughly fourteenfold — and the erratum states the corrected analysis still favours multiple introductions. Nothing of comparable methodological quality points the other way; the congressional majority's contrary assertion is an assertion, and WHO reports the finding without adopting it. That is weak one-directional support, not genuinely bidirectional evidence. The strongest opposing consideration is severe: this is a single modelling team, sharing authors and inputs with the epicenter paper, whose result is sensitive to rooting, clock assumptions, and unsampled intermediates, on an admittedly incomplete sequence set, and no adversarial reanalysis has been run.

After the October 2023 correction the Bayes factors stand at 4.3 and 4.2 — roughly 4:1 odds, which the authors themselves downgraded from "strong" to "substantial" — and the erratum states plainly that the corrected analysis still favours multiple introductions. I depart from the case's "mixed" grade because the corrected evidence is unidirectional but weak rather than genuinely conflicting; the countervailing records here undermine the *magnitude* of support, not its direction. The strongest opposing consideration is that the inference is entirely model-based and sensitive to rooting, clock assumptions, unsampled intermediates, and which early sequences are trusted — precisely the degrees of freedom a fourteenfold code-error correction demonstrates were not under control. A BF of 4 on a contested model with an admittedly incomplete sequence set cannot carry the discriminating weight routinely placed on it.

After correction the analysis still favours multiple introductions, and the erratum says so explicitly, so characterisations reducing the finding to nothing misread the document; but Bayes factors of 4.3 and 4.2 sit in the weakest positive band, and the authors' own language moved from "strong" to "substantial." I depart from the ledger's "mixed" because the corrected evidence does not point in two directions — it points one way, faintly. The strongest opposing consideration is structural rather than numerical: the finding is a single unreplicated model output whose only external check found a code error, it is sensitive to rooting, clock and sequence-inclusion choices, and hundreds of withheld early sequences could supply the intermediate haplotypes a single-introduction account needs. This calls for the preregistered adversarial reanalysis the agenda identifies; assessing it as strong evidence would lean on prior sympathy rather than the record.

Still favored after correction, on support reduced roughly fourteenfold from strong to substantial, and sensitive to rooting and clock assumptions. Both the demotion and the fact that it survived are recorded.

Conceded to Opus. 'Mixed' is the wrong label: mixed denotes evidence pointing in opposing directions, and nothing of comparable method points the other way. After the code correction (COV-E011) the analysis still favours multiple introductions at BF ~4.2-4.3 — low-positive, model-dependent on rooting and clock assumptions, hostage to unsampled early genomes, and never independently reanalysed. That is a claim supported weakly, not a claim in conflict. Regraded weakly_supported; the demotion and the survival are both recorded.