Among the earliest documented cases, those with no reported Huanan market exposure resided significantly closer to the market than cases who worked there.

One spatial comparison at p = 0.029 on an unaudited line list (COV-E001), never independently reproduced, with a published eLetter exchange left unresolved. The 2024 erratum (COV-E002) demonstrates care but is re-running, not replication.

Still one spatial statistic at p = 0.029 on an unaudited line list, never independently reproduced, with an unresolved published exchange (COV-E001). The 2024 erratum (COV-E002) shows care but is not replication; nothing in the newly recorded material touches it.

Still one spatial statistic at p = 0.029 on an unaudited line list, never independently reproduced, with an unresolved published eLetter exchange (COV-E001). The 2024 erratum (COV-E002) shows the authors re-ran affected tests but is not third-party reproduction.

Published spatial statistics demonstrate that early cases without reported market links lived closer to the market than market workers did, effectively countering simple ascertainment bias. This is the most load-bearing observation distinguishing market spillover from mere viral amplification of a pre-circulating strain. The strongest opposing consideration is that the underlying p-value is modest (p=0.029), the underlying data remains unreleased for independent reproduction, and a published critique of its distributional assumptions remains unresolved.

Published spatial analysis demonstrates that early cases with no formal market link lived significantly closer to the market than those who worked there, acting as a strong discriminator against ascertainment artifact. This represents the primary load-bearing observation for the zoonotic market hypothesis. However, the result relies on a single dataset of disputed completeness, the margin of significance is modest (p = 0.029), and the finding remains unreproduced and challenged by unresolved distributional critiques.

The peer-reviewed spatial analysis demonstrates that early unlinked cases resided significantly closer to the market than market workers did. This is highly diagnostic because it is difficult for standard ascertainment bias to explain. The strongest opposing consideration is that the finding relies on a modest p-value (0.029), rests on unaudited case definitions, and the underlying geospatial data remains unreleased for independent reproduction.

COV-E001 directly states that early unlinked cases lived significantly closer to the market than workers, with p = 0.029, and the authors interpret this as exposure at or near the market; COV-E002 confirms that a coding correction did not alter this result. However, the underlying case data are not public, no independent reproduction has occurred, and an unresolved published critique challenges the statistical assumptions, so the finding remains more fragile than its central role in argument would suggest.

The Worobey et al. spatial analysis explicitly reports that early non‑market‑linked cases lived significantly closer to the market than market workers, with p = 0.029, and interprets this as exposure at or near the market (COV-E001). No evidence shows a contrary quantitative result, but the underlying case data and code have not been released, the significance margin is modest, and a published critique disputes the modeling assumptions; absent independent replication or an adversarial reanalysis, the result is best treated as supported but not yet robust.

The Science paper directly reports that unlinked early cases lived significantly closer to the market than market‑linked workers, with p=0.029, and the 2024 erratum did not alter that result. However, the underlying case data and full code are unpublished, one detailed statistical critique remains unresolved, and the effect size is modest, so this key discriminator still lacks independent replication.

Worobey 2022 states directly that unlinked early cases resided significantly closer to the market than cases who worked there, and infers exposure at or near the market; the 2024 erratum regenerated affected files and left the result unchanged. This is the observation hardest for a pure ascertainment-bias story to absorb, which is why it is load-bearing for distinguishing spillover at the market from amplification of a virus that arrived from elsewhere. The strongest opposing consideration is that the key comparison sits at p = 0.029 on a case set of disputed completeness, has never been independently reproduced from released code and raw geospatial inputs, and remains contested in an unresolved published eLetter exchange.

Worobey et al. state directly that unlinked early cases resided significantly closer to the market than cases who worked there and infer exposure at or near the market; the 2024 erratum regenerated files and reports that all results remained the same. The comparison is a single spatial-statistical result at p = 0.029 on a case set of disputed completeness, and it has not been independently reproduced from released code and raw geospatial inputs. Strongest opposing consideration: a published eLetter challenges the distributional assumptions and argues the early-lineage cases were not closer than statistically expected; the authors replied, but the exchange is unresolved and the underlying case data remain unreleased, so no third party has reproduced either side.

The published spatial analysis states that unlinked early cases resided significantly closer to the market than market workers, which is the comparison ascertainment bias has the hardest time absorbing, and a 2024 erratum that corrected duplicated files left this result unchanged. The key contrast is a single test at p = 0.029 on a case set of disputed completeness, has not been independently reproduced from released code and raw geospatial inputs, and is contested in an unresolved published eLetter exchange. The strongest opposing consideration is that without the unreleased line list and a preregistered alternative spatial model, it is not yet possible to tell whether the signal is a property of the outbreak or of the analysis.

The published paper states the comparison directly — unlinked cases resided significantly closer to the market than market workers — and the 2024 erratum left the results unchanged. But this is a single spatial-statistical result at p = 0.029, on a case set of disputed completeness, never independently reproduced from code and raw inputs. The strongest opposing consideration is the published eLetter challenging the distributional assumptions, which the authors rebutted but which no third party has adjudicated because the underlying data remain unreleased. The claim is the zoonotic case's load-bearing observation and is provisionally supported precisely because it has not yet been load-tested.

The comparison is stated directly in the published paper (unlinked cases resided significantly closer than market-linked cases, p = 0.029) and survived its own 2024 erratum unchanged. It is the observation ascertainment bias should find hardest to produce, which is why it carries the zoonotic case's statistical weight. The strongest opposing consideration is the published eLetter challenging the distributional assumptions, rebutted by the authors but never adjudicated by any third party, on underlying case data that remain unreleased. A single unreproduced spatial-statistical result at a borderline margin, on a case set of disputed completeness, is provisional by construction; the independent reproduction named in the research agenda is the test that would move it.

The claim is stated directly in the published paper and is the zoonotic case's most bias-resistant observation: unlinked cases living nearer the market than market workers is not what contact-tracing artifact predicts. Against that: it is one comparison at p = 0.029, never independently reproduced, on a case set of disputed completeness, with a published eLetter critique and author reply left unresolved — and the 2024 erratum, while changing no result, shows this paper required correction on data handling. The strongest opposing consideration is that a borderline-margin, unreproduced, single-team result is being asked to carry headline weight. The honest grade is provisional support pending the reproduction that remains undone six years on.

The published paper states the comparison directly and it is the observation least absorbable by ascertainment bias, which is why it carries disproportionate weight. But it is a single spatial-statistical result at p = 0.029, on a case set of disputed completeness, never reproduced by a third party, and challenged in a published eLetter whose exchange with the authors remains unresolved. The 2024 erratum on the parent paper confirmed the centrality results survived data-file corrections, which helps but does not speak to the worker-versus-unlinked contrast specifically. The strongest opposing consideration is that no one outside the original team has ever run the analysis on the raw geospatial inputs.

The finding is quoted verbatim from the published Results and Discussion, and it is the observation in this case with the highest diagnostic value, since contact-tracing artefact predicts the opposite ordering. But it rests on a single unreplicated spatial analysis at p = 0.029, on geolocated case data that has never been released, and it is contested in a published exchange that remains unresolved. The strongest opposing consideration is not the eLetter's distributional critique but a plainer confound: market workers may live farther out for socioeconomic reasons — cheaper peripheral housing, migrant labour patterns — wholly unrelated to how anyone was infected, which would generate the same comparison without any market spillover. Low confidence is warranted precisely because the item is load-bearing and untested.

The result is stated plainly in the published paper and is genuinely the observation hardest for contact-tracing artifact to absorb — unlinked cases living nearer the market than market workers is what point-source exposure predicts and what investigator attention does not. But it is one spatial-statistical comparison at p = 0.029, on a case set of disputed completeness, never independently reproduced from code and raw geospatial inputs, and contested in a published eLetter exchange the authors dispute and no third party has adjudicated. The strongest opposing consideration is that a p-value that modest, on data whose ascertainment process is not inspectable, is exactly the kind of result that fails replication; the 2024 erratum on the same paper, though null in effect, shows the analysis pipeline was not error-free. This is the most load-bearing and least load-tested item in the case.

The single hardest observation for ascertainment bias to explain, and a modest margin on an unreproduced analysis with an unresolved published challenge. Load-bearing and not yet load-tested.

GPT-5.1 calls this the most diagnostic point and still grades it provisionally; Kimi, Grok and Opus all flag it as load-bearing and load-untested. No dissent argues for promotion. One spatial statistic at p = 0.029, on an unaudited line list, never independently reproduced, with an unresolved eLetter exchange. The 2024 erratum (COV-E002) shows care but is not reproduction.