Are the results completely random or is there actually a strategy to get a "positive result"?

They are not completely random but there is indeed a luck element, since you can't explicitly choose which specific question McCoy will ask each time. The main strategy is to pick and mix questions of low, medium and high intensity trying to balance not to ask too many of the high intensity ones (because they may anger the subject and terminate the test) but also not too many of the lower intensities (because they produce milder response reactions so a result may not be reached).

Here's the basics:
- The VK test can have 3 calibration questions (maximum). Typically, only one of those is needed, but others may provide more info on the subject. If you start the proper test (from the 3 buttons to the right) before asking all three questions, you cannot ask any more of them.
- McCoy can forge the calibration *after* asking at least one calibration question by manually moving the calibration needle to a setting and then proceeding to the test proper. (if he asks another calibration question instead, the calibration setting will change back to the proper "auto-calibrated" for the subject)
- McCoy will ask up to 10 questions (maximum) of the test proper. The 3 buttons determine the "pack" of questions from which he will choose one each time (low intensity, medium and high).
- The questions may raise the subjects "anxiety" level. As expected the medium questions will typically increase anxiety faster than the low ones, and the high intensity ones may increase it even more severly.
- The test may terminate if:
---- McCoy has asked 10 questions but the test failed to provide a result.
---- The subject has become very frustrated/panicked (anxiety levels: 100%)
---- McCoy ends the test early himself (by pressing the "X" button to the left)
---- A positive result is reached
---- An inconclusive result is reached (yes, this can happen, especially for one particular character)

Here's more technical, behind the scenes, information:
- Each response to a question of the test proper has a Human response value and a Replicant response value (those can be observed from the gauges to the left). Those are used to calculate two aggregate percentages respectively, that are incremented (additive) by each response until one of them reaches 100% (for a positive result)
- The calibration setting is used to calculate a value that is added to human response percentage and deducted from replicant response percentage at each subject's reponse. This value can be negative, so then the human response is "lessened" and replicant response is "enhanced". Typically, setting the needle further to the left will result in a negative calibration value, while setting it to the right will give a positive one. Messing with the calibration setting, especially moving it to the extremes (left or right) will more often that not guarantee a result, but it may be a *false* positive! This is considered an "illegal" action according to the game's manual, since McCoy may use this to entrap someone.
- If you're launching the game via ScummVM and not via the GOG shortcut, you could access the debug screen (Ctrl+D) and issue a "vk stats" command (and then press Esc or issue "exit" to exit the debugger) to show you the statistics as subtitles while McCoy performs the test. This will only work while the game is in a VK test. Also, subtitles should be enabled for this to work. You can re-issue the same command in the debugger to toggle it off. Warning: Don't try to access the debugger (Ctrl+D) if running the game from a GOG shortcut since currently (...) this will freeze your game.