Where NWS locates tornadoes
This question has multiple answers. I'll try to cover as many as I can think of.
1.
Initially, at the location reported by the spotter. Given multiple spotter perspectives on the same tornado (both at different times in its lifespan and even at the same time from different angles), this accounts for the multiple reports of one tornado that appear in LSRs (and by extension, the SPC log). Individual WFOs may or may not have time to do much real-time filtering before they dump reports into LSRs, depending on staffing, other duties, and the ongoing weather. Local policies can differ too regarding LSRs. So it really can be a hodgepodge on many days. This is why the rough log has big fat "PRELIMINARY" tags plastered on it.
2.
Sometimes, at the location of the spotter instead of the location of the tornado. This happens by mistake, but it does occur sometimes. The solution is for the person taking the report to inquire about *both* the spotter's position and the direction and distance of the tornado he/she is observing.
3.
For tornadoes with tracks (as opposed to "brief touchdowns"), they are also often referenced at the point of greatest damage or noteworthiness (i.e., Hallam). This usually is done, still in preliminary mode, after multiple spotter reports have been reconciled at the WFO. Again, this process can occur anywhere from real-time to many days later.
4.
For Storm Data, paths are specified in county segments. This is because counties are how warnings are verified and (with just a few exceptions) WFO jurisdictional borders delineated.
5.
For the "final" SPC database, the SPC Warning Coordination Meteorologist combines county segments into whole tornado paths. This is why NCDC data (which is county segmented) will give you greater tornado counts for big outbreaks or multiple state/county border-crossers than the SPC data (available in SVRPLOT). It's also a labor intensive process at times, especially when one discovers a border-crossing path that "zigs" at that border, and needs some detective work and phone calls to be reconciled between two offices' tracks.
Given all that, and ~1000 tornadoes to account for per annum, it's not hard to understand why the SPC tornado database for any given year isn't finalized until a few months into the next year.
For more on segmented tornado data, see
http://www.spc.noaa.gov/faq/tornado/#segments
To map the SPC data, download SVRPLOT for Windows at
http://www.spc.noaa.gov/software/svrplot2/