Sound policy choices need hard, factual evidence. But the constantly changing coronavirus pandemic evades reliable quantified data. Every number comes with a question mark, one so big it blots out the number.

Take a basic datum posted daily, total COVID-19 infections. But that’s only the number confirmed by testing, which is a fraction of those infected but asymptomatic. You chiefly get tested by showing symptoms. Testing all who wish it is spotty, and there aren’t nearly enough test kits. We may well have a vaccine before we have mass testing.

A positive test is highly useful for individuals, allowing physicians to prescribe treatment, including hospitalization. But such tests are useless as a measure of the virus’s spread in the general population. For that you need major random testing, something that does not interest the White House. Desperately wanting to re-open the economy, the Trump administration ignores grim reports and continually foresees an early turnaround. The worst was to be over by mid-April.

Truly random samples are hard to achieve. Many refuse to participate, especially in our mistrustful age. Testing the self-selected yields invalid results. Blood tests for antibodies, totally different from and administered long after diagnostic nose-swab tests, may capture how many were once infected and how widespread COVID infection was months earlier, too late for treatment. Some antibody tests already suggest infection rates are many times the “confirmed cases.” An incentive to get antibody testing: Thinking you’re immune lets you live more freely. One catch: You may not be immune.

A confined population where all undergo testing shows how fast coronavirus spreads. The carrier Theodore Roosevelt left Vietnam March 9 with a few crewmembers carrying the virus. The first case tested onboard March 22. By the time they reached Guam March 26, perhaps 100 tested positive, and the skipper sent out the distress calls that got him fired. Now, more than 1,000 of the 4,500-person crew have tested positive; one died. Crowded places — ships, nursing homes, meat plants, prisons — accelerate contagion manyfold.

In less-controlled settings, fractions based on guesswork denominators and numerators are worthless. For a death rate, the infection number is the denominator. But without random testing, that number (based on “confirmed” cases) is way too low. The numerator is the deaths, which, one might suppose, is known, but it also undercounts. Many, especially in the early weeks, were not listed as coronavirus victims. COVID can damage several organs, so many causes of death were listed as pneumonia, heart attack, kidney failure or stroke. Re-examining probable COVID deaths and adding them bumps up the totals.

A denominator that grows (more infections) lowers the death rate. But a numerator that grows (more deaths) boosts the fatality rate. With neither firm, estimates of fatalities vary from 6 percent of infections down to a more probable 0.5 percent. Even low percentage rates could mean a couple million Americans will die, much higher than the standard autumn flu.

One method is to note deaths in excess of a multiyear average of deaths, which becomes a baseline. By this measure, from March 1 to mid-April, 37,000 deaths can be attributed to COVID, perhaps twice that of official tabulations. Since then, “excess” deaths have soared even higher. A useful measure is the ratio of recovered to died, running in some places about 2:1. Increases in the recovered:died ratio would indicate that treatments are working.

Sweden’s lax rules aim for “herd immunity” — the point, expected to be around two-thirds, when spreading slows greatly. Sweden seeks to reach this by natural contagion with only limited distancing, but many die. Vaccines can also produce herd immunity — but more safely in controlled doses with attenuated viruses. Vaccines now in trials, however, could soon become ineffective against the quickly mutating coronavirus.

In mid-March, a British model predicted that, without mitigation such as social distancing, 1 million to 2.2 million Americans could die. In late March, another model predicted 100,000 to 240,000 U.S. deaths, a best-case scenario with lockdowns. Subsequent estimates fell to 81,000 and then 61,000. Counting “excess” deaths, we could soon reach 100,000. Roughly 2,000 die a day (likely rising), so we could top 240,000 this summer.

Media obsession with ventilator shortages was misfocused and fleeting. Ventilator shortages never materialized; redistributing and manufacturing them are impressive but do little to bring down the death rate. By the time a patient is on a ventilator, it’s probably too late; roughly half die.

Have we ever lived amid greater uncertainty? My daughters don’t know if midcoast summer camps will open — one reason they visit us. And they don’t know if schools will re-open in the fall. My son doesn’t know when supply chains for his furniture designs will resume. The stock market spooks on rumors. In our 80s, we wonder if we’ll have to self-isolate into next year.