[NOTE THAT THAT GEMINI INCLUDED THE ERRATUM BASED ON WHAT YOU HAD UNCOVERED] This study investigates the early genomic diversity and ancestral rooting of SARS-CoV-2 using 787 high-coverage human viral genomes (plus 20 putative C/C and T/T intermediate genomes) collected globally through 14 February 2020. The authors (i) audit the sequencing depth and shared synapomorphies of the 20 intermediate C/C and T/T genomes to argue that early transitional haplotypes between lineages A and B are sequencing artifacts or convergent C-to-T mutations; (ii) reconstruct a 15-fragment non-recombinant common ancestor (recCA) across 31 sarbecoviruses to quantify C-to-T reversion bias across the early pandemic; (iii) implement a nonreversible random-effects substitution model (HKY+RE) in BEAST v1.10.5 under unconstrained, recCA-rooted, outgroup-rooted, and fixed-haplotype rooting schemes; and (iv) couple BEAST tMRCA posterior distributions with individual-based SAPHIRE epidemic simulations on a 5-million-node Barabasi-Albert contact network (FAVITES-COVID-Lite) via rejection sampling. Synthesizing the inferred ancestral haplotype probabilities with simulated tree topologies (as updated in the 2023 Erratum), the authors report Bayes factors of 4.2 to 4.3 favoring at least two separate zoonotic introductions (lineages B and A) in mid-to-late November 2019 over a single introduction, with minimal cryptic human circulation prior to December 2019. The manuscript makes several lasting empirical contributions: its forensic audit of read-level coverage at sites 8782 and 28144 in early C/C and T/T sequences, its demonstration that C-to-T mutational bias repeatedly generates revertant haplotypes (decisively ruling out derived haplotypes such as WA1/A.1 and 20SF012 as the pandemic MRCA), and its open deposition of BEAST XMLs and simulation code on Zenodo. However, the central quantitative claim that phylodynamic rooting and epidemic simulations favor multiple zoonotic introductions over a single introduction (BF = 4.2–4.3 after the 2023 Erratum) rests on an asymmetric hypothesis-testing framework, an invalid conditional probability normalization in Equation (1), and a 10-fold mismatch between simulated sequencing density (~5,000–7,500 tips per tree) and empirical sampling (787 tips). These methodological tensions—along with residual inconsistencies between the main text and corrected Supplementary Materials—can be resolved through targeted mathematical and simulation refinements without additional wet-lab work. ------------------------------------------------------------------------ 2. SPECIFIC CONCLUSIONS AND ASSESSMENT OF SUPPORT * Rejection of Derived Revertant Haplotypes (A.1 / WA1 and A+C29095T / 20SF012) as the Pandemic MRCA: Well Supported. By reconstructing the mosaic sarbecovirus progenitor (recCA) across 15 GARD and 22 3SEQ non-recombinant segments and modeling directional C-to-T and G-to-T mutation biases via the nonreversible HKY+RE substitution model, the authors convincingly demonstrate that substitutions toward the recCA are 2.8-fold enriched (6.04% vs. 2.17%) and heavily dominated by C-to-T transitions (82.6%). All phylodynamic rooting schemes (unconstrained, recCA, RaTG13, and BANAL-20-52) reject A.1 and A+C29095T with BF > 100. * Artifactual Nature of Early Transitional C/C and T/T Genomes: Moderately Supported. Read-level inspection andshared derived synapomorphies account for 14 of 16 C/C genomes and all 4 T/T genomes sampled before 14 February 2020 as low-coverage consensus call errors (<=10x depth) or homoplasies. However, two early C/C genomes sampled in Beijing in late January / early February 2020 lack raw read data and share no derived mutations with lineages A or B (SM p. 17). More importantly, showing that *sequenced* C/C tips in GISAID are artifactual or homoplastic does not prove that a true transitional C/C (or A/B) infection never occurred in Wuhan in late November / early December 2019, when fewer than 1–2% of infections were sequenced. * Bayes Factor Support for Two Zoonotic Introductions over a Single Introduction (BF = 4.2–4.3): Fragile / Methodologically Biased. Even after the October 2023 Erratum corrected the coding bug that had inflated the Bayes factor by an order of magnitude, the remaining BF of 4.2–4.3 in Equation (1) (SM pp. 12–14) is an artifact of three mathematical and simulation asymmetries: 1. Asymmetric topological filtering between I1 and I2: For a single introduction (I1) to match the empirical tree (tau_1C or tau_2C), the simulation must simultaneously produce two polytomies of >=100 lineages separated by 1–2 mutations with zero sampled intermediates AND a relative clade-size split between 30% and 70% of all tips (matching the observed 35.2% Lineage A / 64.8% Lineage B split). For two introductions (I2), the authors require only that two independent simulations each have a basal polytomy of >=100 lineages (tau_P, occurring in 47.5% of runs, giving P ~ 0.475^2 = 22.6%), without requiring that the combined two-introduction epidemic exhibit a 30%–70% clade-size balance or a 2-mutation separation between introduced lineages. Because two asynchronous introductions in a superspreading network usually result in the earlier lineage dominating by orders of magnitude (Fig. 4B), omitting the 30%–70% clade-balance filter from I2 severely inflates P(tau | I2) relative to P(tau | I1). 2. Invalid normalization of P(S_MRCA | tau) in Equation (1): On SM p. 13, the authors define P(S_MRCA | tau) by normalizing over the number of "compatible" haplotypes (assigning 1/2 to S_A and S_B under I1's tau_1C, 1/4 to each of the 4 haplotypes under I2's compatible root-shape pairs, and 0 to all haplotypes for incompatible root-shapes). In probability theory, every simulated tree has an ancestral haplotype with probability 1 (sum_{S_MRCA} P(S_MRCA | tau) = 1 for all tau); setting the sum to 0 for incompatible topologies conflates the prior distribution P(S_MRCA | tau, I_n) with the topology likelihood P(tau_obs | S_MRCA, I_n). Moreover, Equation (1) plugs P(S_MRCA | Y)—estimated in BEAST under a single-tree human coalescent model (I1)—directly into the numerator for I2, where the MRCA of A and B resides in an animal reservoir rather than the human Skygrid tree. 3. Over-sequencing in simulated trees (~5,000–7,500 tips vs. 787 empirical tips): In the FAVITES-COVID-Lite pipeline (SM pp. 8–11), every ascertained individual among the first 50,000 infections is sequenced (15% ascertainment -> ~7,500 sequenced tips per tree; 96.7% of trees have >5,000 taxa), whereas the empirical dataset has only 787 genomes worldwide. Sequencing ~10-fold more tips in the simulations than in the empirical alignment makes it far more likely that a transient 1-mutation intermediate lineage is sampled (artificially depressing tau_1C and tau_2C under I1) while simultaneously inflating the probability that a basal polytomy exceeds the fixed threshold of n = 100 lineages (which enters squared in the numerator for I2). * Discordance Between Unconstrained Clock Rooting (Lineage B) and Outgroup/recCA Rooting (Lineage A): Overinterpreted. Under unconstrained rooting, Lineage B is favored over Lineage A (80.85% vs. 1.68%, BF = 48.1), whereas recCA rooting favors Lineage A over Lineage B (77.28% vs. 8.18%, BF = 9.4). The authors interpret this tension as independent evidence that A and B could not have arisen from a single introduction. However, their sensitivity analysis excluding Wuhan genomes (SM p. 19 and Data S3) shows that unconstrained support for Lineage B drops substantially and Lineage A is no longer rejected (BF < 10), demonstrating that the molecular-clock preference for Lineage B is heavily driven by dense early sampling of the Huanan market Lineage B superspreading cluster under a strict clock and panmictic Skygrid coalescent prior. * Timing of the Primary Cases (Mid-to-Late November 2019) and Minimal Prior Cryptic Circulation: Well Supported. Across all rooting strategies, ancestral haplotypes (including counterfactual single-introduction models rooted on A, B, C/C, A.1, or A+C29095T; Tables S14–S15), doubling times (2.65–4.45 days), and ascertainment rates (5%–25%), the primary infection consistently dates to mid-to-late November 2019 (95% HPD spanning late October to early December), with a median of 3 cumulative infections at the tMRCA and 0 hospitalizations before 1 December 2019. This conclusion is robust to whether SARS-CoV-2 entered humans once or twice.