At a sober press briefing in the White House last week, members of President Trump's coronavirus task force unveiled data supporting the need to continue the national effort to limit the spread of the virus.

Even while maintaining policies aimed at limiting person-to-person contact, the administration projected between 100,000 and 240,000 Americans would die of covid-19, the disease caused by the virus. One slide, using data from the Institute for Health Metrics and Evaluation at the University of Washington, showed a predicted peak in the daily death toll from the disease arriving in the middle of April.

Anthony S. Fauci, a key member of the task force, made an important point about those projections the day before.

“Models are as good as the assumptions you put into them, and as we get more data, then you put it in and that might change,” he said. That point was soon reinforced, with the IHME estimate shifting upward soon after the White House briefing.

Late Tuesday night, however, the IHME estimate shifted in the other direction. While the model last week projected nearly 94,000 deaths by late summer, its new estimate puts the toll by August at 60,400 — a decline of 26 percent from the model’s previous estimate.

What's more, the day with the most projected deaths was shifted from over 3,100 deaths on April 16 to 2,200 deaths on April 12.

This is unequivocally good news, but it carries with it several caveats.

The first is that the accuracy of the IHME model already has been called into question by officials fighting the virus. In Colorado, IHME’s estimates indicate the state’s peak daily death toll already passed — an assertion public health officials in the state believe inaccurate.

“We are confident our peak has not hit,” the head of the Colorado Department of Public Health and Environment told reporters this week, according to the Denver Post.

The Post reported last week that Washington, D.C., is relying on a separate model offering a starkly different picture for how the virus will affect the district. At the time, IHME saw the peak number of deaths arriving in mid-April. The D.C. model, developed by Penn Medicine, estimates the peak will come in late June.

That variance is a function of the difficulty of modeling the pandemic, something FiveThirtyEight explored last month. Models should get more accurate as the actual peak approaches — though identifying when the peak has arrived is itself tricky, predictive models aside.

In part, that’s because the United States probably won’t see one peak but a series of peaks spaced out both by geography and time.

In New York, hardest hit by covid-19 infections, encouraging signs show the number of new infections declining, even as the number of deaths each day peaks. (This, experts have reminded Americans repeatedly, is to be expected: the lag time between infection and death means the death toll trails the infection rate.) Data in the state, though, still offers an unclear picture.

William Hanage, an epidemiologist at the Harvard T.H. Chan School of Public Health, said there was evidence the outbreak in New York is slowing in response to aggressive efforts last month to stall the spread of the virus. The decrease in the rate of new hospitalizations is evidence to that end.

“This does not mean an ‘inflection point,’ because they are still going up,” Hanage said in an email. “It is not at all clear how long it will take to turn it around.”

More importantly, even if New York's peak has passed, other places are just beginning to see the worst effects of the virus.

“The pandemic is only just getting its boots on in other places,” Hanage said. He added he remains “incredibly anxious” about smaller communities around the nation, where the virus can take hold and trigger new hot spots.

“They can be hit as hard or harder than urban areas, because they lack resources to deal and probably won’t be testing until too late,” he said.

Marynia Kolak, assistant director of health informatics at the University of Chicago’s Center for Spatial Data Science, who is part of a team studying county-level data on the outbreak, agreed that localized outbreaks remain a problem.

“It would be great if things suddenly get better, but from all the data we have, it suggest we’re just beginning to approach the peak for several regions of the country,” Kolak said.

“It’s all happening region by region,” she added. “A lot of parts of the country have just started to expand as hotspots in the past one or two weeks."

While the IHME model tries to estimate the peak in each state, there’s another constraint to keep in mind: it looks forward for only a few months. There is a chance the coronavirus might recede over the summer and resurge in fall, echoing the pattern we see with the seasonal flu. The IHME’s estimate of 60,000 deaths takes us only to August and excludes any potential deaths that might occur in the fall and winter if the virus continues to spread without effective treatments or a vaccine.

The model also depends on the social distancing efforts encouraged by the White House of staying in place through May. It doesn’t tackle what happens after that point. If federal and local leaders ease restrictions without other interventions in place, it could result in a resurgence of infections — and deaths.

Perhaps the most important thing to keep in mind about the shift in the IHME model is the warning Fauci offered last week. The model depends on the data and if the data change, so will the model. The downward shift in the estimated death toll reflects positive changes in the data states are reporting about the effects of the virus. Shifts in the opposite direction can mean the model’s prediction gets grimmer.

“For people to get lax now is the worst case scenario,” Kolak said of the need to maintain the current containment efforts. “If anything, leaders should assure people they are doing the right thing, that they are doing this for the community."

“In public health, we want everyone to think we’ve overreacted a few weeks from now,” she added. “We don’t want to start preparing mass graves."

Brady Dennis contributed to this report.