Originally posted by milesy just thinking - do you agree that the two focus methods focussed at the same focus chart from t he same tripod etc should end up with the same focus point on the same lens??
The live view AF is extremely accurate on my K-7 for all my lenses, which it had better be, considering how long it takes. In broad strokes the idea that you should try to match this Live View AF precision by making microadjustments to the regular phase-detect autofocus that is active when the mirror is down SOUNDS good, except as stated in that article there are some drawbacks.
The most serious one is that the focusing behavior of your lens will almost certainly vary based on distance to your subject. Most focus charts and whatnot want you to keep your lens wide open and at its minimum focus distance to the chart. You can spend an hour or more tweaking to your heart's content only to find that though you can now focus like a champ at 10 inches, you're suddenly way off at 3 feet. Again as the article says, it's a bit of a compromise - choose a distance you think is actually representative of where you'll be using the lens at and make the adjustments to correspond to that.
The sad reality is that, in the Pentax system at least, traditional phase-detect autofocus is not nearly as accurate and consistent across a range of conditions (distance to subject and lighting) as the contrast-detect methods used during Live View. When the mirror is up, as it is in Live View, the camera's imaging sensor is exposed and the onboard computers have way more (and more relevant) information to play with.
Some people maintain (in their heads) a list of AF Microadjustment values for their lenses to use in different settings. For instance a +7 outside on sunny days, but a +3 indoors with the same lens. This is just the way the cookie crumbles.
Fear not though -- one day mirrors will be gone altogether and the computers will be so much faster that you'll get that scary accurate focus with all your lenses every time right off the imaging sensor with contrast-detect or whatever replaces it. Assuming the sensor is looking at what you think it is.