The inference is Bayesian, not deductive. Under the hypothesis that an infected live animal shed SARS-CoV-2 at the market, one predicts viral RNA to be enriched precisely where that animal (and hence its shed mtDNA) was, i.e. at the wildlife stall — which is observed (three independent spatial signals converge on stall A; all five positive stall-A samples carry susceptible-mammal mtDNA; nutria/foxes appear only in negatives). Under the competing hypothesis that only humans shed the virus, viral RNA should track human presence/handling and be spread across the market like human mtDNA, not localise to one wildlife stall while humans are not even the most abundant mammal there. So the spatial pattern favours a wildlife source. The argument is explicitly bounded: because the environmental samples were taken about a month after human transmission was already widespread, virus could also have been deposited by infected humans onto animal-DNA-rich surfaces, so co-location raises but does not settle the probability that the animals themselves were infected — a limitation the authors concede.
Step 6 verdict — approved (checked)
Reconstructed step: a likelihood-ratio comparison. P(observed spatial pattern | infected animal shed at stall A) is high, because that hypothesis predicts viral RNA enriched exactly where the animal’s shed mtDNA is; P(same pattern | only humans shed) is lower, because human shedding predicts RNA tracking human presence and spreading like human mtDNA rather than localizing to one wildlife stall where humans are not the dominant mammal. Given the premises (three converging spatial signals on stall A; all five positive stall-A samples carry susceptible-mammal mtDNA), LR > 1, so the pattern favours a wildlife source. Probed the strongest undercutting defeater: post-outbreak human deposition of virus onto animal-DNA-rich surfaces (samples taken ~1 month after human transmission was widespread) — this is exactly what would break an inference to “the animals were infected,” and the statement is already hedged to “raises the probability… does not prove infection.” The hedged conclusion is immune to that defeater. Traced from the premises, so checked.