How Japan's outbreak looked from the inside

By late March, the public narrative held that Japan had weathered the coronavirus crisis. Official statistics showed fewer than 1,000 infections on March 19th, and the national consensus — echoed by domestic experts, international researchers, and the media — was that prompt action and cluster-based management had contained the virus. The discussion centered on when life would return to normal, and whether a major sporting event would proceed.

That consensus was wrong. The outbreak was geographically distributed and worsening through March, and the evidence was discoverable to anyone willing to look carefully. This is the account of an independent research project that reached that conclusion before official acknowledgment.

Timeline of a rapidly evolving crisis

The path from "Japan is fine" to "Japan declares a nationwide state of emergency" took about a month. Key milestones, including the March 22nd and 25th entries below, were not widely known at the time.

  • March 19th: An NHK panel of experts outlines the national consensus on the coronavirus situation, the response plan, and its risk factors.
  • March 21st: Osaka Governor Yoshimura distributes an official assessment of the situation in Osaka and Hyogo on live television. Few grasp its significance until days later.
  • March 22nd: I post a hash on Twitter at 11:24 AM, establishing a pre-registered record of a prediction.
  • March 23rd: Tokyo Governor Koike warns that a lockdown may be necessary if infections rise rapidly, but insists it must be avoided at all costs.
  • March 24th: A sporting event is postponed.
  • March 25th: I post a second hash at 1:31 PM. Later that day, Governor Koike urges voluntary social distancing beginning that Saturday.
  • March 26th: The New York Times publishes "Japan's virus success has puzzled the world. Has its luck run out?" Economist Tyler Cowen publishes "The coronavirus situation in Japan is probably much worse than you think," covering claims from an anonymous "working group." The government's expert panel formally assesses that the virus is likely "rampant" nationwide.
  • March 27th: NHK covers why "a powerful American newspaper" questions Japan's containment story.
  • March 28th: The Prime Minister addresses the nation, declaring there is no need for a state of emergency. [Correction: my notes indicate the "state of national hardship" remark was actually made on April 1st.]
  • March 29th–April 5th: Prefectural governments, the Japanese Medical Association, and researchers escalate calls for a state of emergency.
  • April 6th: The Prime Minister announces a state of emergency for seven prefectures, including Osaka and Tokyo.
  • April 9th: Aichi requests inclusion in the emergency declaration.
  • April 16th: The declaration widens to the entire nation, with 13 prefectures flagged as particular concerns. Tokyo hospitals begin turning away suspected coronavirus patients.
  • April 17th: The Prime Minister asks citizens to cut human-to-human contact by 70–80%.
  • April 20th: Confirmed cases exceed 11,000; some measure of improvement appears, but the situations vary across regions. The National Police Agency reclassifies suspicious deaths, including a man who died on a Tokyo street, as coronavirus-related.
  • April 21st: Experts warn that given the virus's spread, elimination within a year is implausible, and far stricter measures may be needed to avoid overwhelming the medical system.

The white paper and its path to publication

On March 26th, Tyler Cowen published an essay based on his correspondence with an anonymous "working group" in Tokyo. That group concluded Japan had a geographically distributed epidemic and predicted a public health crisis in April. The group had been privately circulating evidence to policy actors; once officials publicly acknowledged the epidemic, the group published its white paper anonymously to help the public prepare.

I initiated the working group and wrote the white paper's primary draft. Cowen's essay linked to the copy we published anonymously via a third-party website moments earlier. The reception was mixed: some found the claims concerning given the official consensus, some organizations acted on them, and some dismissed the work as uncredentialed amateurs posting anonymously.

The circulated version — sent privately to several parties for review and feedback, including Cowen — is archived and matches the March 25th 1:31 PM hash. Its conclusions are substantially identical to the published version, with only minor wording differences.

The published version is reproduced below, unedited. My own commentary continues after it.

Japan’s Reported Numbers Are Missing Asymptomatic Carriers

National statistics released by the Ministry of Health, Labor, and Welfare for PCR-test-diagnosed patients on six dates (March 10–14 and March 23) consistently show asymptomatic patients at roughly 10% of all diagnoses, ranging from a low of 10.24% to a high of 11.29%. That is far below what should be expected for covid-19.

The Diamond Princess experience offers a reference point for what extensive testing reveals. The National Institute of Infectious Diseases reports that 48% of infected passengers were asymptomatic at the time of sample collection. Iceland’s chief epidemiologist, whose testing program can cover almost the entire population, reports a similar figure: 50% of infections there are asymptomatic.

A pre-print in the Journal of Infectious Diseases, from a Japan-based team, used Japanese citizens evacuated from Wuhan who were exhaustively tested upon return and calculated an asymptomatic ratio of 30.8%. A Eurosurveillance pre-print, using more sophisticated modeling of the Diamond Princess data, estimates the true asymptomatic rate at 17.9%. Even the most conservative of these estimates implies Japan is undercounting cases given its policy of largely refusing to test asymptomatic individuals except for those with deep, direct contact with an infected person (濃厚接触者).

If virtually all symptomatic patients are successfully identified, a generous assumption, then the true infection count can be estimated by dividing the number of diagnosed symptomatic infections by the true rate of symptomatic infections. Using the Ministry of Health data from March 22nd:

907 / (1 - 0.48) =~ 1,750 true infections

That is roughly 70% higher than the number of detected infections on that date, under the most optimistic set of assumptions. Every missed symptomatic infection is also likely accompanied by a missed asymptomatic one.

Deaths Imply a Much Higher Case Count

A pre-print from the Center for Mathematical Modeling of Infectious Diseases, “Using a delay-adjusted case fatality ratio to estimate under-reporting,” estimates Japan’s detection rate at between 15% and 35% by reasoning backward from observed deaths. That suggests Japan is undercounting cases by a factor of approximately 3X to 6X.

Testing the Traveler Pipeline

Singapore has traced four coronavirus cases to individuals who had recently been in Japan; assuming three were contracted there to be conservative, a rate calculation emerges. Singapore’s airport statistics show approximately 64,000 passengers arrive from Japan over a two-week period. Three infections in that population is about 47 basis points, 5X the 9 basis point infection rate Japan reports domestically.

The implication is either that tourists are disproportionately stumbling into surveilled clusters or that uns surveilled community transmission is widespread. If tourists encounter clusters at the same rate as residents, the official count is understated by roughly 5X. If they contract the virus through community spread only, the undercount could be 25X or more.

Containment Strategy Has Failed

Japan’s official strategy, reiterated at the March 19th Panel of Experts session on NHK, is cluster-based containment: rapidly identify clusters, surge medical attention to diagnosed patients, and ask for voluntary behavioral changes. Yet prefectural authorities are beginning to acknowledge that containment is not holding.

In a document dated March 16th and shown on TV by Gov. Yoshimura on March 21st, experts in Osaka Prefecture and Hyogo Prefecture — currently epicenters of the outbreak — acknowledged that new infections without detected links to existing clusters indicate community transmission is underway. The document states: “It is believed that infections without surveilled chain to a cluster continue to increase and that therefore a rapid increase in infections has already begun.” (見えないクラスター連鎖が増加しつつあり、感染の急激な増加がすでに始まっていると考えられる。)

Aichi Prefecture (Nagoya) reported on the 22nd that 103 patients were hospitalized, with 27 of them untied to either of the prefecture’s known clusters. Asahi Shinbun reported on March 11th that over half of the prefecture’s 161 high-grade infectious disease beds were already in use.

What the Numbers Project for April

Osaka’s expert document forecasts 3,374 infections (including 227 severe cases) before April 3rd, compounding at more than 6X per week. New York, with aggressive measures in place, is compounding at approximately 2X per week.

If official counts were accurate but containment has failed, applying the optimistic doubling rate of peer nations yields roughly 3,000 cases, including more than 200 severe cases, in each of Nagoya, Osaka, and Tokyo by the end of April. Using the Osaka working group’s own growth estimates without aggressive mitigation, those figures could be ten to one hundred times higher.

These scenarios both point to an “overshoot” — a breakdown in the provision of care that Japan’s own experts have warned about — leading to sharply increased mortality among covid-19 patients and among patients with other conditions as medical resources are exhausted.

Tokyo announced on March 23rd that it had 118 beds suited for high-level infectious disease care, planning to add 700 beds for severe patients and 3,300 for moderate ones. But the bottleneck is not likely to be beds — it is skilled medical personnel, of which Japan has an ongoing shortage. Reliable surges from unaffected areas are unlikely, as those personnel will be needed for outbreaks at home.

Given that official counts are likely a gross undercount, the true picture is materially worse. A projection of over 500,000 infections, including more than 5,000 severe cases, with breakdowns in care across Nagoya, Osaka, and Tokyo before the end of April, is consistent with the available evidence.

The Government Has Not Acted With Urgency

As of March 19th, government policy was to wait and see before asking the public to voluntarily restrict social activity (自粛) in outbreak regions. Asia Times reported one unidentified official describing the approach as a “don’t ask, don’t tell” strategy based on minimal testing. Governor Koike of Tokyo announced on March 25th a voluntary stay-at-home order starting March 28th, after floating the possibility of a lockdown without indicating one would be pursued.

Schooling was suspended nationwide on February 27th, and large events have been discouraged. But aside from those measures and individual citizens adopting modest precautions, life continues largely as usual. Mask wearing in central Tokyo is below 30%, reporters have not observed social distancing at press conferences about the epidemic itself, and domestic travel remains essentially unimpeded — the Tokkaido Shinkansen saw only an 8% drop in usage in February despite the outbreak.

Japan is not weathering this crisis well. The statistical evidence indicates community transmission is already established, and the official response has not risen to meet it. April will be a month of national public health emergency unless immediate and aggressive policy interventions are implemented.

Judging the white paper against its claims

The white paper advances many falsifiable claims about a deeply complex situation. Some, particularly its forward-looking statements, necessarily depend on how the response effort played out. Its central assertion — that Japan was experiencing a geographically widespread coronavirus epidemic that containment could no longer stop — stood or fell on observable events.

It fell the right way. Japan's April case data made the core result unambiguous. The additional prediction that the country would face its most serious public health crisis since the war also proved accurate. Not every claim in the document will survive close scrutiny; the authors were non-experts working quickly under pressure. The paper itself acknowledges the limits of its sourcing and reasoning, and accepts responsibility for errors. But on the question that mattered most, the result held.

Novelty, utility, and the state of the record

Whether the white paper's findings were novel and useful is harder to answer cleanly. Standards for that judgment differ by field. The earliest high-quality reporting on the discrepancy between Japan's public narrative and its underlying data appeared in the New York Times, in a March 26 piece by Motoko Rich. That article was the first published work, to my knowledge, that moved beyond speculation to lay out facts incompatible with the prevailing consensus.

It connected dots that had not previously been joined in print: the intentional narrowing of testing, the tightening bed shortages in several cities, and the government-affiliated experts' conclusion that infections were growing exponentially in Osaka. The reporting was strong, well-sourced, and credible — and it shifted the conversation. NHK picked it up, Japanese reporters began pressing officials on testing adequacy and regional projections, and the tone of public discourse changed within days.

What newspapers typically do not do is make explicit, unsourced forecasts about the future. The Times piece offered none. Very few institutions whose job is to make predictions and act on them publicly behaved as though they assigned high probability to the correct picture in March. I will refrain from naming the organization most likely to have understood the near-term trajectory, for reasons that are probably obvious. History will sort out who saw what, when.

A cryptographic pre-registration

Pre-registering a conclusion means committing to it before the analysis is complete or the relevant events have occurred. My method was, in the jargon of security work, "dropping a hash on Twitter." To a security researcher, that has a precise meaning: I had a document in hand, demonstrated that I possessed it at a specific point in time by publishing its cryptographic digest without releasing its contents, and retained the ability to later prove that any publication was an exact, unedited match to that hash.

This is a standard practice in the security community, and scientists have used it for centuries. Isaac Newton employed an equivalent technique — using a cypher — to register his results without disclosing them.

How the Working Group formed

The Working Group's analysis was essentially complete by the early afternoon of Wednesday, March 25, though the writing continued for a bit. The group itself had formed only two days earlier, on the morning of Monday, March 23. It comprised four professionals living and working in Tokyo, from various career and national backgrounds. None of us had worked in medicine or epidemiology, and neither I nor, to my knowledge, the others had any prior expertise in viral outbreaks.

My own road to that Monday morning started in February, when I was occupied with ordinary life and only peripherally aware of the coronavirus. I had dismissed it as a problem for China and Iran. A friend corrected my calibration, and by early March I was deeply worried about my family in the United States. I began to doubt the official descriptions of Japan's situation, though I expected the local impact to be modest.

The exact moment I got concerned about Japan came on March 10, when reports from Italy made clear that the virus had nonlinear effects on healthcare systems, not just supply chains. Standing in Tokyo, I could picture the local system under comparable strain. What I did not do, at first, was act.

I assumed the people in positions to do something had already drawn the same conclusion. Surely epidemiologists understood exponential growth. Surely someone was working round the clock. Then, on the same day, two very intelligent people told me they believed the public consensus — that Japan was managing fine. I was stunned. I tried to update their understanding, assigning an 80% probability that Japan would follow its peer nations. And then I looked around more carefully and realized with a chill that perhaps many people in authority had not yet reached the conclusion I had.

I wrote a memo to myself on Sunday, March 22, to organize my thoughts. Its framing was simple: what would I expect to observe if Japan were in the middle of an uncontrolled epidemic, and did those observations match reality? They did. I was 90% confident in the conclusion and far less confident the memo would change anyone's opinion. So I hashed it and posted the digest on Twitter, without sharing the document itself.

Why not publish the memo outright?

I did consider publishing it immediately. But I did not think it would be instrumentally effective in saving lives, and I believed publishing could bring real risks — both a potential cost in lives from misguided panic and strong personal or professional sanctions.

My subjective confidence was 90%, but I was aware I might be miscalibrated. The odds that one non-expert, sleuthing in spare time, had outdone not merely individual experts but nearly the entire class of them, plus the authorities, seemed vanishingly small. Perhaps I was wrong.

So I reasoned through the two cases. If Japan did not have a present epidemic, there was no benefit to claiming it did, and some real risk of causing a panic or disrupting an orderly public health response. If Japan did have one, publishing speculation was unlikely to recalibrate experts who were presumably already working the problem; it might even delay better, more scientific analysis, and thus delay the response. There was also a live domestic political question about a long-standing economic strategy whose resolution I considered inevitable, and I worried that being publicly right about the epidemic would make me a convenient scapegoat for that unpopular outcome.

So I did what I have done a handful of times before — most often about Bitcoin exchanges behaving badly: I dropped a hash.

Turning a private warning into a public artifact

I saw a way to use the hash drop deliberately. After nearly 15 years of writing publicly, I have some standing in my community, and people read me as a congenital optimist. A coded, deniable message could say more than “I found something interesting about one of my hobbies.” It could say: I am convinced, in my bones, that the sky is actually falling, and you likely don’t know it yet. If you trust me and understand what I’m saying: get out the red binder.

If I was wrong, professional acquaintances might think less of me — an acceptable risk. If I was right, it would give some people time to prepare and create a record for later investigation. I thought some readers would pick up on the subtext of this tweet:

1) I am materially wrong about the most consequential thing I've had to have a view on in 15 years. You should probably degrade your estimate of my ability to think through complex problems.

2) We need a data point to couner "Nobody could possibly have seen this coming."

— Patrick McKenzie (@patio11) March 22, 2020

One reader got the message. He knew the idiom of dropping a hash on Twitter, had followed my work, and assumed the most consequential topic I might hold a surprising view on was covid-19. He urged me to publish immediately — to save lives by giving people time to prepare.

I explained why publishing my current findings could backfire, but that I might be able to work hard enough to produce something convincing, something that could accelerate policy decisions. He suggested working with a news organization. News outlets have more credibility with policy circles than private individuals do. They have resources to verify conclusions and strong editorial controls against imprudent publication. A reputation for that diligence lets them distribute trustworthy results across many policymakers quickly, in parallel — a speed advantage over serial briefings. And news organizations do not fear getting ahead of a story; they mostly see that as a win.

The idea had never occurred to me. I’d only considered two paths: publish in my personal capacity, or send a memo to… I didn’t actually know who would receive such a memo in the decision-making chain. As soon as he made the suggestion, the plan took shape: find like-minded people, write a stronger memo, run it past a medical researcher, and brief as many organizations as possible. Twenty minutes after that, in the dead of night, three people in Tokyo had agreed to a Monday morning kickoff meeting — over videoconferencing, of course.

A working group takes shape

That Monday we agreed on the goal (“accelerate the response effort, with the goal of saving lives”), a strategy (“leverage high-status organizations to surface internal conclusions we think probably exist, or generate evidence sufficient to bootstrap experts to those conclusions rapidly”), and a timeframe. We also added a fourth member.

We wanted our conclusions inside policy circles by Friday, because policymakers might take weekends off but the coronavirus does not. English-language reporting typically takes about half a day to reach domestic bilinguals before translation. That set a publication deadline of Thursday, and the assumption that media outlets would need our material by Wednesday to check it. Why write in English rather than Japanese? Partly because Japanese-language outlets were less likely to pick up this kind of story from our kind of sources, but mostly out of expedience — I write an order of magnitude faster in English than in Japanese, and that was an advantage worth keeping.

That left us roughly 54 hours: research, expert review, produce a credible artifact, and brief the right parties. Monday through Wednesday disappeared.

We worked through a dozen epidemiology papers. We reviewed two months of reporting in Japanese and English. We formed a hypothesis for why our conclusion was not obvious to existing efforts. We identified data sources and signals that looked under-examined or intriguing enough to chase. We ran numbers. We found a medical researcher to review the conclusions, incorporated their feedback, and wrote the memo — calling it a white paper because that sounded more credible.

Then we briefed a small set of organizations.

On Monday, almost by accident, we hit a major break: we found the document Governor Yoshimura had shown live on television that Saturday. Written by epidemiology experts, the document concluded that infections in Osaka and Hyogo had escaped containment and that exponential growth had already begun. After reading it, we moved to essentially 100% confidence that Japan had geographically distributed outbreaks — an official confirmation that excluded the possibility we were the only team to have reached the core result. That had always seemed likely, but it still came as a huge relief: the response effort would arrive sooner, and we only needed to help others appreciate the import of experts’ results, not convince them our own result was valid independently.

Our research was done, scrutinized as deeply as our networks permitted, and backed by official confirmation of the prevailing state of affairs. Then we started talking to the relevant parties. Some asked to remain unnamed; others may be open about speaking with us. The goal throughout was to accelerate their ability to act appropriately.

Did the white paper change anything?

Some organizations made strong, swift decisions after seeing the white paper, and in a few cases told us they used portions of it to shape their own work. Sooner or later, stakeholders across nearly every organization worldwide will likely ask how that organization performed during the pandemic. May every judgment be just and merciful.

On a sporting event by any other name

One theory starts from the observed timeline and infers that a great many human actions were driven by a sporting event. Treat that with suspicion: simple and narratively satisfying explanations for complex system behavior are usually wrong. I do not believe any decision was made to put a sporting event above public safety. I can easily believe the sporting event altered how a very complicated system operated. Those are very different claims.

Take me, a minor cog in that system. Nobody ever instructed me to put a sporting event first — nor would anyone who knows me accuse me of wanting to. Still, sontaku (忖度) describes what happened: intuiting the preferences of other parts of the system and acting to enable them without any explicit instruction from formal authorities. Some consider sontaku a uniquely Japanese trait. I see it as a Japanese label for an extremely common human behavior.

An American regulatory lawyer advising a client on risk is likely sontaku-ing — using a mental model of the regulator rather than asking the regulator directly. Being good at that guesswork is central to the job. And the client is probably sontaku-ing the lawyer in return. The lawyer does not need to order the client to abandon a bad course of action — has no authority to do so. The lawyer only needs to say, “It is my opinion as your attorney…” and sophisticated clients understand that phrasing as a non-order. They generally follow it, because getting quality non-orders is why they hire regulatory counsel in the first place.

That yields two possible interpretations: American regulatory lawyers are an inscrutable breed, or they’re human, doing what humans do in systems larger than themselves. This small cog sontaku-ed its way to understanding that the sporting event materially affected part of the calculus.

The world has notably few conspiracies. It has many intricate systems with complicated decision processes, internal data flows, and incentive structures for the people inside them — systems that occasionally produce results neither the system nor its actors would prefer.

Why the epidemic wasn’t spotted sooner

Japan’s coronavirus response, as described by a panel of experts, effectively had two playbooks. Plan A centered on aggressive cluster containment: testing capacity and manual contact-tracing effort were focused on rapidly drawing boundaries around known clusters, hospitalizing all diagnosed cases, and asking the public only for modest distancing measures such as avoiding large events in affected areas. Plan B was the harsher playbook adopted by most Western nations — variously called a lockdown, putting the economy into stasis, or strongly-suggested extraordinary social distancing — which nobody chooses unless they have to.

The design was a controlled transition. If explosive growth in infections was observed, Plan A would be abandoned for Plan B, with graduated intensities specified by predefined guidelines. Given that Plan A appeared to work, virtually everyone preferred it.

I have no reason to believe that the many departments implementing Plan A failed to carry out their duties to the best of their abilities. My hypothesis is simpler: it was many people’s job to execute Plan A, to collect data under its testing regime, and to report on its success. It was in many people’s interest for Plan A to work. But it may not have been anyone’s job to wake up every morning, assume Plan A was no longer viable, and search for proof that it had failed. I am substantially less confident in this hypothesis than in our earlier results — call it 60% that this is a major factor — but I offer it in the hopes that future researchers can ask the right questions.

The deeper issue may be that Plan A’s specified testing regime is accidentally incapable of telling you Plan A has failed. Both plans were substantially written years ago, before the biology of covid-19 was understood, and they may not have adapted quickly enough to what was learned in early 2020.

In particular, this non-expert believes the plans do not sufficiently allow for transmission by asymptomatic patients, and assume the supermajority of spread happens within clusters. Testing capacity is therefore concentrated “on and around” clusters, on symptomatic patients likely to have worse clinical experiences. By design, with the best of intentions, Plan A does not “waste” resources on asymptomatic patients or those outside clusters. If such patients happen to cause a cluster, that cluster will presumably be quickly detected and contained.

The failure mode: if you rigorously test people close to a cluster, you find approximately the correct number of symptomatic patients, hospitalize them quickly, and give them the best care. Everyone involved does their job and gets feedback that they are doing it correctly. Most patients get better. Plan A looks like it is succeeding, and considers itself succeeding, until it detects undeniable evidence of sustained community spread.

The perfect storm for this strategy is a pathogen capable of causing clusters but also of exponential growth via community transmission by asymptomatic patients. If cluster incidence is successfully reduced to the point that clusters don’t become undeniably common before tens of thousands of patients are well-distributed through the social graph, you end up with something not too unlike Japan in March 2020.

Open questions and hypotheses

Many doubted there was a coronavirus epidemic because the evidence should be extremely obvious. Three doubts stand out, with hypotheses as to why they pointed in the wrong direction. Experts will eventually have much better data and more reasoned conclusions.

“An uncontrolled epidemic should, within a month, result in many thousands of patients. Where are they?”

The asymptomatic ones are exactly where they would otherwise be on any given day in spring. Symptomatic patients are likely receiving care for a viral infection or shrugging it off because symptoms are mild. Those who seek care will likely be told to rest and drink liquids; most will do that and get better. A small number will be hospitalized with aggressive, high-quality care for pneumonia — but pneumonia is much more common than coronavirus and does not trigger a test under Plan A. Testing criteria require both severe symptoms and a link to a cluster, or a recent trip to a nation widely believed to have an epidemic.

“You can’t have an uncontrolled epidemic without a sharp increase in mortality. Where is it?”

An average of approximately 3,500 people pass away in Japan per day. Who is the first person to produce an Excel analysis on that number? What is their job title, and at what cadence do they refresh it? How many days does it take for data to flow through organizations and stabilize so the number is complete and reliable? What is the analyst’s threshold for surprise, and how quickly can they get that message to the departments running Plan A — and overcome those departments’ insistence that they’re running all the tests called for and getting results consistent with a well-managed public health issue?

I don’t know if that stylized analyst is a particular person, department, or process, but one system functions not dissimilarly. The National Institute of Infectious Diseases collects per-large-city stats on flu and pneumonia deaths. Data takes weeks to become available and quality is not uniform over the 21 cities covered. Tokyo’s numbers show excess deaths in late February relative to the Institute’s baseline — approximately 50 per week for two weeks, above the model’s statistical-variance threshold. Data for March is not yet available, and other cities either lack reports or do not consistently show excess deaths above threshold. Tokyo has reported approximately 75 coronavirus deaths total through this writing.

The March numbers for flu and pneumonia deaths are modeled to decrease as flu season winds down. If those numbers reflected primarily seasonal flu and typical pneumonia, April’s numbers should decline again. It will be very observable, in approximately six weeks, whether this happens. You may have a high-quality hypothesis as to what I believe those numbers will show.

“Why did it appear to work for so long?”

My leading hypothesis, again as a non-expert, is substantial path dependence in how coronavirus spreads over social networks. We’ve seen in cities across many countries that “bad luck” with one early patient at a wedding or funeral can spiral into hundreds of patients, each starting their own chain.

Early patients in Japan may have happened to not be central to the social graph. Most contacts got lucky, and it took a few more generations for the virus to reach the densely connected part of the network than in peer nations. Because coronavirus’ biology means a difference of a few generations makes a massive difference in scale and detectability — for a few days. Combine this with Plan A probably legitimately succeeding with regards to almost all the patients from almost all the early clusters. Against covid-19, “almost” buys you days but does not blunt the epidemic. We spend a day every day.

That plausibly gets you close to the observed experience of March without requiring notable malfeasance, theories that Japan is dissimilar to other Western nations, or a unique unobserved environmental X-factor.

When a fire alarm is the humane thing to build

The Working Group’s goal in publishing was not to announce a discovery for its own sake, but to demonstrate that a result could be produced early enough to be acted upon. The response to a novel pathogen is a race, and the prize for a faster, "good enough" answer is time — time that translates directly into decisions that can save lives. Showing that such a result is possible, even when imperfect, gives others permission to move with similar urgency rather than waiting for institutional sign-off.

This is the logic of a fire alarm. Its engineering purpose is not to inform people that a fire exists — many will already have perceived smoke. The alarm's real function is to provide unquestionable and immediate permission to evacuate, overriding the social instinct to look around, see others immobile, and conclude that raising an alarm would be an unwelcome disruption. That instinct has cost too many days this year. Too many people and organizations have waited for permission to act on evidence they already had. As the pandemic continues, similar decision points will recur — whether to stock supplies, change travel plans, or prepare for extraordinary measures. These decisions cannot wait for peer review or the history books.

Lessons from moving faster than usual

Reflecting on the work, several beliefs have shifted — things that would have surprised the author in January but now seem clear.

First, epistemic humility is invoked more than it should be. The idea that non-experts are "epistemic trespassers" who waste experts' time has a surface appeal, but in this case, believing that narrative cost precious days. No professional courtesy or fear of repercussion should delay a response. The aim must be to get a right enough answer and make a right enough decision — prudently, but boldly.

Second, speed is undervalued. Covid-19 does not sleep, and its doubling times are short — facts the universe offers regardless of institutional convenience. Organizations that can make decisions under uncertainty and iterate quickly will outperform those built for slower, more deliberate processes. This is a critical design consideration for any institution that becomes a systemic chokepoint during a crisis.

Third, Twitter proved essential. It flagged the coronavirus weeks before traditional sources and delivered real-time experience from peer nations. It connected four professionals in a large city, making them aware of each other's skills. It enabled the "hash-as-signal-flare" tactic, which put the right advice in the right hands days earlier than would otherwise have been possible.

Finally, software matters in ways that are easy to dismiss. The entire project was executed without the team ever sharing a room, using messaging apps, videoconferencing, and office suites that mostly run on phones. Much of this infrastructure did not exist in a usable form a decade ago, and none of it is marketed as a tool for advancing human knowledge. Yet it worked — flawlessly. The team spent less than 30 minutes total setting up infrastructure for four researchers, and the project's entire cost was under $2,000, funded by a personal credit card.

On truth and accountability

There is a distinction between the truth of a thing and the acknowledgment of that truth. Japan faced a coronavirus epidemic no later than early March; that was a fact, even before it was widely acknowledged. If any truth about this work is acknowledged, it should be a simple one: the author is a responsible professional with no relevant expertise or authority, who made some guesses during a year of widespread guessing. Some of those guesses may happen to align with official guidance. If the less convenient truths are quickly forgotten, that is acceptable. Should the guesses prove wrong, the consequences will be accepted. Actions were taken personally, with that responsibility in mind.

Team humanity

The pandemic implicates nearly everyone, in a world that is imperfect, divided, and hurting. The response will require comforting the afflicted, mourning those lost, and learning. It will be beaten with high science and with hand soap. The Working Group hopes its effort has been of some service. Until happier times, health and safety to all. Tokyo remains the greatest city in the world.