The geodetic marks in the NGS database are adjusted. The horizontal control marks are horizontally adjusted and the vertical control marks are vertically adjusted. A few marks are both horizontally and vertically adjusted. Adjusted is a term used to indicate that a computer program has analyzed a set of survey data and recaluclated the positions of all the points in the survey. The recalculation process is called adjustment.
When you go searching for a geodetic mark, it is very important to note whether the mark is horizontally adjusted or vertically adjusted.
If the mark is horizontally adjusted, its adjusted horizontal coordinates will be rounded to 5 decimal places in the format dddmmss.sssss and accurate to that level. This is much more accurate than can be achieved by your handheld GPS receiver which has a precision of dddmmss.s and a similar level of accuracy.
If the mark is vertically adjusted, its horizontal coordinates can be up to a few hundred feet off, since, instead of being accurately measured and adjusted, they were probably scaled from a topographic map, perhaps even a 15-minute quad.
The concept of adjustment is an interesting mix of statistics, calculus, trigonometry, and matrix algebra. Imagine that you are to survey all the streets in a suburban area. The process of measuring with instruments has natural error in the process, both systematic (hopefully little to none of that!) and statistical error. You survey from corner to corner, going up and down all the streets. When you finish and go back to calculate results, you could start at any point in the survey, calculate the coordinates of each intersection on the way, and find that, at intersections you have calculated coordinates for before, you find you have now calculated a second different set of coordinates for the same survey station. If you start at a different point, you get different results. What you have is a matrix of 'opinions' of where each intersection is. The solution is to use an adjustment program. It simuiltaneously calculates the location coordinates of every intersection in the matrix to take into account all the 'opinions'. It uses what is called a least-squares best fit, a statistical concept that relates to the fact that statistical error grows like a square-root function (a curve). It thereby adjusts the locations of the intersections and then goes on to adjust the locations of the stations between intersections.