A number of top-flight economists with whom I usually agree, most notably Paul Krugman, assure me that the U.S. job market is at full employment, meaning … actually, it’s not easy to describe what we mean by full employment. The easiest explanation is that that everyone who wants to work and is reasonably able to do so has a job that provides them the hours of work they want.

Yet, even at 3.9 percent unemployment, there are 6.3 million people unsuccessfully seeking work, and another 5 million who are working part-time but want full-time work. What’s up with that? It’s possible the discrepancy is due to these folks having personal limitations that make them un- or under-employable (surely, some persons of color in those numbers face employer discrimination). In other words, full employment doesn’t mean a jobless rate of zero. It means we’re left with only the structurally unemployed (those not helped by even very low unemployment) and the frictionally unemployed (people between jobs).

But the concept is ambiguous. Not only is it hard to separate out the structural from the cyclical, but there’s some evidence that strong cyclical effects — a persistently low unemployment rate — can pull people who are thought to be structurally out of work back into the job market.

Luckily, we can get around this head-scratching with the definition that Dean Baker and I used in a book on the benefits of full employment. When an economy is at full employment, any extra demand shows up not as more jobs but as wage or price inflation.

Cue the famous water glass analogy. Think of the labor market’s capacity as a glass and labor demand — jobs that need workers to fill them — as water pouring into the glass. Once the glass is filled with water, any more just spills over. Once the available labor supply has been fully tapped, any extra demand doesn’t create more jobs or higher real wages, but higher inflation.

Does that describe the current U.S. job market?

Not really. Wage growth has been famously tepid — that mystery was the point of Krugman’s post. And while inflation is picking up a bit, it has been very low for a very long time, unresponsive to falling unemployment, and no one is arguing that it is “spilling over the glass” (spiraling up in response to a full-capacity economy).

In fact, the full employment conversation seems pretty muddled for one simple reason: Economists don’t know how large the glass is.

To be clear, if Paul’s wrong — and he wisely admits he may be — I doubt he’s off by much, but neither he nor I, nor anyone else, knows how much. I deal with this uncertainty through the squirrelly phrase: “We’re closing in on full employment.” I think that’s true, but if it implies I know the ultimate destination on which we’re closing in, then it, too, is misleading.

One bit of evidence everyone throws around in this discussion is the fact that the employment rate for “prime-age” guys (25 to 54, making my 62-year-old self well past prime) has yet to climb back to its pre-recession peak, so maybe there’s more room in the glass than we thought. But Paul and others (e.g., economist Jason Furman) argue that based on the long-term negative trend in this series (see figure below), we shouldn’t expect them to get back to their peak.

That may be right, but again, we just don’t know. The figure shows the trend decline in employment for these guys (and since 2000, the pattern for prime-age women has been similar), but it also shows their strong, cyclical rebound. Before I accept the Krugman/Furman conjecture that prime-age employment rates are topped out, I want to see the end of this trend flatten out, which, if you look closely, has been the case in past cycles, prior to downturns. If anything, it looks to have accelerated a bit in recent months. Surely, this should give the “we’re-back-on-trend” advocates some pause.

Of course, we must consider the risk invoked by pouring more water in a glass that might be full, i.e., higher inflation. By itself, that wouldn’t be worrisome, given how low inflation has been. A bigger risk is the de-anchoring of inflationary expectations: What happens when people believe the Fed has given up on or cannot enforce its 2 percent inflation target. Then, when inflation goes up, people don’t view its rise is a temporary spike. They think it’s more or less permanent and expect faster price growth in the future. Though the likelihood of such de-anchoring seems low, it would be costly problem, as the Fed would likely accelerate its schedule of interest-rate hikes.

Still, even Paul agrees that policy-wise, we should act as if there’s more slack than he thinks there is, because the “the costs of [Fed] tightening when the economy still has room to grow are much bigger than those of waiting and discovering that we’ve overshot a bit.”

Bringing this back to the wage mystery, I’ve maintained that, along with slow productivity growth, wage growth is constrained by decades of hammering workers’ bargaining power. Counteracting that deeply embedded problem will take not just low unemployment, but really low unemployment, sustained for a number of years.

This generates a testable hypothesis. If the unemployment rate continues to fall, say to 3.5 percent or even lower, as some forecasters believe will occur, that’s when we’ll see wages for middle-class and low-income workers take off. Depending on the trajectory of unemployment, we’ll see if I’m right by later this year or next.

At the end of the day, our best move is to take degree-of-slack assessments with a large grain of salt, and assume that there’s room to run until obvious price pressures develop and/or the prime-age employment rate clearly tops out. When it comes to knowing whether we’re at full employment, I’d go with the old Wittgenstein line: “That of which we do not know, we must not speak.”