Isbor, deaths lag infections by about 2 weeks, give or take.
Right now, the "heavy hit" areas pulled off successful distancing (like NYC). Their numbers are falling. The "less heavy hit" are making up for it, and have relatively explosive growth; but because we are adding, BIG NUMBER shrinking plus SMALL NUMBER exploding = BIG NUMBER shrinking slowly.
In the hard hit areas, the most vulnerable and exposed got burnt. The remaining vulnerable are hiding, because they where selected for hiding. Meanwhile, less serious cases who hid from ERs are getting tested, reducing the rate at which their numbers have fallen.
The future problem lies in either reignition of a hard hit area (NYC is about 25% infected, so it can repeat what happened before if you reopen), or more likely that areas that haven't been hard hit and haven't eliminated the virus continue to have numbers explode.
With deaths lagging infections by 2-3 weeks, and diagnosis lagging infection by a week, deaths are a 3+ week trailing indicator of things getting bad. And you have to pay attention to regional deaths, not national; national deaths are dominated by the areas the virus has done a burn-through wave of. The next areas likely to burn are ticking up from 1 death per week, to 1 death per day, to 2 deaths per day. With no effective distancing, and a 3 week lag, if an area has a doubling time of 3 days, that is 7 doublings "baked in"; there could already be enough people infected to be killing 100+ per day even if lockdown was perfect today.
OTOH, if their modest social distancing spreads doubling to 5 days, that is only 4 doublings, or 16x "baked in". A large swing, and leaves time for policy to change if there is an explosion of cases starting.
NYC burned brightly because it had a crazy spread rate of 2x every 2 days or something. By the time they hit 1 death/day, there where 10 doublings "backed in" (21/2 = 10.5), or a 1000-fold increase in deaths.
Background caution and spread out population gives time for a response to an exploding local epidemic. And it doesn't have to be perfect for this to be useful -- even if your measures result in Re over 1, so long as an epidemic doesn't sleep its way to 1000s of dead per day, you can choke it off.
Now, what happens if Re floats just above 1? You get a constant simmer and burn of dead people, but no spikes.