Yes, MAP is cumulative. Once MAP "discovers" a machine, it will try to inventory that machine every time MAP runs until MAP successfully collects 100% of the data object it expects to collect based on the inventory scenario chosen in the wizard. If you choose a new inventory scenario in subsequent runs, it is possible that machines previously flagged as 100% would be re-inventoried due to the new inventory scenario's data objects. In the status screen you see while the inventory engine is running, the line "Retrying from previous inventory" is the count of devices that have not reached 100% collection.
MAP has multiple ways it can discover the existence of a machine, such as querying AD DS, but it is best to stick to one method of discovery unless you know that the machines that will be discovered using another method aren't included in the first. One example of this would be Linux machines that are not listed in AD. When you choose multiple methods, the likelihood of duplicated discovery increases dramatically and just makes the whole process take longer.
If your AD is full of computer objects that no longer exist, MAP will discover them too, and even get some basic info about the machine from AD, but will obviously never be able to connect. This ends up cluttering up your reports and makes each inventory run take longer since MAP will try to connect to those machines every time. There are a number of useful articles out there that help with cleaning up stale Active Directory objects.
Please remember to click "Mark as Answer" on the post that helps you, and to click "Unmark as Answer" if a marked post does not actually answer your question. Please VOTE as HELPFUL if the post helps you. This can be beneficial to other community members reading the thread.