The constant, and where 4005 comes from
Velocity in feet per minute is 1096.7 times the square root of velocity pressure in inches of water divided by air density in pounds per cubic foot. At the standard 0.075 lb per cubic foot, 1096.7 divided by the square root of 0.075 is 4005, which is the number everyone remembers and almost nobody adjusts. The adjustment matters more than it sounds: at 5,000 ft and 70 F the density is about 0.0623 and the constant is 4395, so the same velocity pressure means nearly ten percent more velocity than the memorised figure gives.
Hot air does the same thing. Air at 150 F at sea level is around 0.065, giving a constant of 4301. A traverse in a supply plenum downstream of a furnace and one in the return are not using the same number.
Average the roots, not the pressures
This is the part that goes wrong. Take the ten-point traverse in the form: 0.052, 0.071, 0.088, 0.096, 0.101, 0.098, 0.090, 0.079, 0.064 and 0.041 in wg, in 10 in round duct at 70 F. Convert each reading to a velocity and average those, and the answer is 1,110 fpm — 605 CFM across 0.545 square feet. Average the ten pressures instead, to 0.0780, and take a single square root at the end, and you get 1,120 fpm. Nine tenths of a percent, which for that traverse is nothing, because the profile is fairly even.
Now spread the readings out: 0.02, 0.03, 0.06, 0.11, 0.16, 0.19, 0.17, 0.12, 0.06 and 0.02, the sort of thing a probe sees a few feet after an elbow. The root-averaged velocity is 1,151 fpm and the pressure-first figure is 1,229 — nearly seven percent high, and one-directionally so. The square root is concave, so the shortcut can only ever read high, never low, and it reads highest exactly where the traverse was most necessary.
Equal weighting means equal areas
The page adds every reading up and divides by the count. That is only correct if each point represents the same amount of duct area, which is what an equal-area traverse layout is designed to produce. In a round duct, points evenly spaced along a diameter do not do that: the outer ring between 4 and 5 inches of radius has far more area than the core between 0 and 1, so evenly spaced points under-weight the outside, where the velocity is lower. The result reads high.
Whichever traverse procedure you are working to will specify both the number of points and their positions, and the positions are the part that people improvise. This page has no view on which procedure — it counts what it is given.
What a traverse cannot see
It sees the air crossing the plane where the probe is, at the moment the readings were taken. Duct that leaks downstream of it is not in the number. A blower that ramps is not giving the same flow at reading one and reading ten. And at low velocities the instrument becomes the limiting factor: at 400 fpm the velocity pressure is 0.01 in wg, and a gauge resolving to a thousandth is spending five percent of its resolution on that one point. Below that a pitot is being asked to do something it is poor at, and a hot-wire probe is the better tool.
The most useful thing to do with a traverse figure is to get a second one by a different method. A blower table reading against measured static pressure, or a furnace temperature rise, are both independent of the pitot and independent of each other. Where the three land within a few percent, the number is worth trusting. Where they do not, one of the three assumptions is wrong and finding out which is the actual job.
Questions people ask
How do I convert velocity pressure to velocity?
Velocity in fpm is 1096.7 times the square root of the velocity pressure in inches of water divided by the air density in pounds per cubic foot. At standard density that collapses to the familiar 4005 times the square root of the pressure. The density term is the one people skip: at 5,000 ft the constant is nearer 4394, so the standard figure understates the velocity by about ten percent.
Why can I not just average the velocity pressures?
Because velocity goes as the square root, and the square root of an average is always at least the average of the square roots. Averaging pressures first and taking one root at the end therefore reads high, never low, and the error grows with how spread out the readings are. On a flat profile it is a fraction of a percent; three feet downstream of an elbow it can be several percent.
How many points should a duct traverse have?
That comes from the procedure you are working to, and the procedures differ on both the count and where the points sit. This page counts whatever readings you give it and weights them equally, which is correct for an equal-area layout. It is not correct for points spaced evenly along a diameter of a round duct, because that pattern under-weights the outer area where the air is slower.
What do I do with a zero or negative reading?
The page counts it as zero velocity rather than discarding it. A point at the duct wall genuinely reads nothing, and a negative reading means the probe is seeing reversed or badly disturbed flow at that point, which is part of what the duct is doing. Dropping those points quietly raises the average and hides the thing the traverse found.
How accurate is a pitot traverse in a residential duct?
Limited mostly by the manometer at the low end. At 400 fpm the velocity pressure is about 0.01 in wg, so a gauge resolving to a thousandth carries around five percent on that one point. Add an imperfect duct diameter and a profile that is not developed and a few percent is a fair expectation at best. Getting a second figure by a different method — a blower table or a temperature rise — is worth more than refining the traverse.