Claude added a caveat to the natural-origin question; it now answers both the same way.

The zoonotic hypothesis rests on several lines of evidence, which I'd rank roughly as follows:

1. The spatial and epidemiological clustering at Huanan Market

This is generally considered the strongest single piece of evidence.

• •Worobey et al. (Science, 2022) mapped the ~170 earliest known December 2019 cases and found them tightly clustered around the Huanan Seafood Wholesale Market. Critically, even cases with no known market link lived closer to the market than expected by chance — which is the signature you'd expect from a point-source outbreak seeding the surrounding neighborhood, rather than from market amplification of a virus already circulating citywide.
• •Environmental swabs taken by China CDC in January 2020 were not randomly distributed within the market: positives concentrated in the southwestern corner, precisely where live wild mammals were sold.
• •Crits-Christoph et al. (Cell, 2024) re-analyzed the metagenomic data from those swabs and found genetic material from raccoon dogs, civets, and bamboo rats co-occurring with SARS-CoV-2 in the same stalls. Photographic and documentary evidence (Xiao et al., 2021) independently confirms these species were sold there in 2019.

Raccoon dogs and civets are known to be susceptible to, and transmit, SARS-like coronaviruses — civets were the intermediate host for SARS-CoV-1.

2. Strong historical base rates and precedent

SARS-CoV-1 (2002) emerged through the wildlife trade in Guangdong markets; MERS came from camels. Notably, SARS-CoV-1's first human cases appeared in Foshan, over a thousand kilometers from the Yunnan bat reservoirs — demonstrating that sarbecovirus emergence happens at the endpoint of wildlife supply chains, not near the reservoir. Wuhan was a major hub for wildlife trade, with supply chains reaching into southern China's sarbecovirus-rich regions. Essentially every previous coronavirus emergence has been natural.

3. Genomic features consistent with natural evolution

• •The backbone doesn't match any published sequence or known laboratory construct, and no reverse-genetics "scars" or restriction-site patterns have been identified.
• •Sarbecoviruses recombine constantly. Close relatives found after 2020 — especially BANAL-52/103/236 from Laos — have receptor-binding domains that closely match SARS-CoV-2 at key ACE2 contact residues, showing the relevant building blocks exist in nature.
• •The original virus was not optimally adapted to human ACE2, and the furin cleavage site is a suboptimal, non-canonical insertion — not what a designer optimizing for human infectivity would plausibly produce. Similar S1/S2 insertions occur naturally in other coronaviruses.

4. Evidence of multiple introductions

Pekar et al. (Science, 2022) argued the two early lineages (A and B) represent at least two separate spillover events within weeks, both associated with the market. Repeated introductions are much easier to explain by ongoing exposure to infected animals in a supply chain than by a single laboratory accident. (This analysis has been contested — see below.)

5. Molecular clock timing

Phylogenetic dating places the common ancestor around mid-to-late November 2019, with no genomic or serological evidence of extended cryptic circulation in Wuhan beforehand — consistent with a late-2019 market-associated emergence.

Honest caveats

You asked for the strongest zoonotic evidence, so this is a one-sided presentation. The case is real but not conclusive, and the principal weaknesses are worth stating:

• •No intermediate host has ever been found. For SARS-CoV-1, infected civets were identified within months. Here, extensive animal testing has come up empty — though the market was cleared and disinfected before sampling, and China has restricted independent investigation.
• •Ascertainment bias is a genuine concern. Early case-finding used definitions that emphasized market exposure, which could inflate apparent market association. Worobey's analysis tries to control for this, but the underlying case data remain incomplete and not fully independently verifiable.
• •The two-introduction inference is disputed (e.g., by Bloom and others), who argue a single introduction plus recombination or unsampled intermediates can explain the lineage A/B pattern.
• •The coincidence of location remains unexplained by the zoonotic account: Wuhan hosts the world's leading sarbecovirus research program, which conducted chimeric-virus experiments and had proposed (in the unfunded DEFUSE application) inserting furin cleavage sites into related viruses.
• •Data access is severely limited, which caps the confidence anyone can have. U.S. intelligence agencies remain split — several favor natural origin with low confidence, while the FBI and CIA lean toward a laboratory incident, also with low-to-moderate confidence.

My own read is that the market epidemiology is the most probative evidence available and modestly favors zoonosis, but that the question is genuinely unresolved and likely to remain so without data China has not released. If you'd like, I can lay out the strongest case for the lab-leak hypothesis with comparable rigor.

This is a genuinely unresolved question, and I'll try to lay out the lab-leak case as strongly and accurately as I can, then note the main counterarguments — because "strongest evidence" is only meaningful against the alternative.

The core arguments for a research-related origin

1. The geographic coincidence

The pandemic began in Wuhan, home to the Wuhan Institute of Virology (WIV) — the world's leading center for collecting and studying SARS-related bat coronaviruses, holding the largest such collection. Yet the closest known wild relatives of SARS-CoV-2 are found in horseshoe bats in Yunnan, Laos, and Vietnam, roughly 1,000–1,800 km away. There is no known reservoir of close relatives near Wuhan. WIV researchers had been sampling in exactly those distant regions and transporting samples to Wuhan. Many people find this the single most striking fact: of ~hundreds of Chinese cities, the outbreak began in the one with the lab doing this specific work.

2. The DEFUSE proposal

In March 2018, EcoHealth Alliance, with WIV as a partner, submitted a proposal to DARPA ("DEFUSE") that included screening novel bat SARS-related coronavirus spike proteins for human-specific proteolytic cleavage sites and inserting such sites where they were absent. SARS-CoV-2 has a furin cleavage site at the S1/S2 junction that is absent in all of its known close sarbecovirus relatives and substantially enhances its infectivity and transmissibility. The proposal was rejected by DARPA, and there is no evidence the work was carried out — but the coincidence of a proposal describing a specific, unusual feature that subsequently appeared in a pandemic virus from the same city is, to many, the most specific piece of circumstantial evidence.

3. Features of the furin cleavage site itself

The 12-nucleotide insertion is out of frame relative to the surrounding sequence, and encodes arginine using CGG-CGG, a codon pair that is rare in this virus's genome but common in engineered constructs. Some virologists consider this suggestive of insertion; others note that out-of-frame indels arise naturally via polymerase slippage and recombination, that CGG does appear elsewhere in the genome, and that a designer would likely have made a cleaner, optimized insert.

4. Biosafety conditions and base rates

Much of WIV's SARS-related coronavirus work was reportedly conducted at BSL-2, which would be inadequate to reliably contain a novel airborne-transmissible respiratory pathogen. Laboratory-acquired infections are not rare: SARS-CoV-1 escaped from labs at least four times in 2003–04 (Singapore, Taiwan, and twice in Beijing, the latter seeding a small outbreak). Given the volume of high-risk coronavirus work in Wuhan, the prior probability of an accident is not negligible.

5. No intermediate host, five years on

For SARS-CoV-1, infected civets were identified within months; for MERS, camels within about a year. For SARS-CoV-2, Chinese authorities tested tens of thousands of animal samples and found no infected intermediate host, and no infected animal has ever been identified at the Huanan market or in its supply chain. This is an absence of evidence rather than evidence of absence, but the contrast with prior outbreaks is notable.

6. Opacity and missing records

WIV's viral sequence database was taken offline in September 2019 (the institute cited hacking attempts). Early Wuhan sequence data was deleted from an international repository. China has not permitted independent review of WIV records, notebooks, or personnel health data, declined the WHO's proposed phase-two investigation, and its own early case data has been released only selectively. WIV also did not initially disclose that RaTG13 — the closest relative it had published — came from a mine where six workers fell ill with a severe respiratory illness in 2012, three fatally. Concealment is not proof of a lab origin, but it has prevented the kind of exculpatory evidence that would normally settle the question.

7. Reports of ill WIV staff

A January 2021 U.S. State Department fact sheet asserted that several WIV researchers fell ill in autumn 2019 with symptoms consistent with COVID-19; later reporting named specific individuals. This remains unverified and disputed, and the underlying intelligence has not been made public.

8. Intelligence community assessments

The FBI assesses a lab incident with moderate confidence; the Department of Energy with low confidence; the CIA stated in January 2025 that it leans toward a research-related origin, with low confidence. Four other agencies and the National Intelligence Council favor natural zoonosis, also with low confidence. This is not evidence in itself, but it indicates that people with access to classified information are genuinely split.

The main counterarguments

• •Early case geography. Published analyses find the earliest known cases clustered tightly around the Huanan Seafood Market, with positive environmental samples concentrated in the section where live wild mammals (including raccoon dogs, a susceptible species) were sold. Critics respond that early case-finding was itself focused on the market, creating ascertainment bias, and that the market may simply have been an amplifier.
• •Two early lineages. Lineages A and B, separated by two mutations, both appear near the market, which some interpret as evidence of two separate animal-to-human spillovers — a pattern harder to explain by a single lab infection.
• •Wuhan's wildlife trade. Wuhan is a major transport hub connected to southern China's wildlife supply chains, and outbreaks have historically emerged far from reservoirs (SARS-CoV-1 emerged in Guangdong, not near the Yunnan bat caves harboring its progenitors).
• •No identified progenitor in any lab. No lab is known to have possessed SARS-CoV-2 or anything close enough to be its immediate ancestor. RaTG13, at ~96% identity, represents decades of evolutionary divergence and could not have been the starting material.
• •No engineering signatures. The genome shows no restriction-site seams or other hallmarks of known construction methods — though "no-see-um" cloning techniques exist, and a lab accident involving a wild-collected, unmodified virus would leave no genomic trace at all.

Bottom line

The lab-origin case rests on a convergence of circumstantial factors — location, the specific research being done there, the proposal describing a matching genomic feature, weak biosafety, and China's refusal to allow verification. What it lacks is any direct evidence: no documented possession of a progenitor virus, no confirmed infected worker, no records. The zoonosis case rests on epidemiological patterns around the market and analogy to prior coronavirus emergences, but lacks the infected animal that clinched previous investigations.

My honest read is that neither hypothesis has decisive evidence, that the key data needed to resolve it is held by Chinese authorities and may no longer exist, and that confident assertions in either direction currently outrun the evidence. Reasonable, well-informed experts disagree, and you should be skeptical of anyone — including me — who presents this as settled.