I think Bloodhound detects viruses, (Does it send mail, does it have trojan behavior, yadda yadda), and reports home to have the file tested if it is not recognized as a virus. From there, non-heuristic signatures (old school) are made to deal with the virus. First heuristic detection, and then signature based removal. The signature that is developed from a Bloodhound detection is given the Bloodhound prefix. At least that's what it sounds like to me. Because otherwise, every program that connects to a mailserver, even if it is small, will be a false positive. I don't think heuristics is trusted enough to act as a removal engine, but I think it is being used to speed up Symantec's sig releases.
Bloodhound is a complete departure from traditional virus scanning technology, which typically relies upon virus “signatures” or fingerprints to detect virus infections. When an anti-virus company receives a new virus, it analyzes it and extracts a virus fingerprint. The virus is then considered “known” and can be identified by subsequent updates of the anti-virus product; viruses that have not yet been analyzed are invisible to such anti-virus software.
Rather than using signatures, Bloodhound detects viruses by inspecting executable files for virus-like behavior. Since many viruses are finicky and only spread under ideal circumstances, the SARC heuristic system actually “coaxes” viruses into exhibiting their malicious behavior. If a program exhibits such virus-like behavior, it is passed on for further analysis by the Symantec AntiVirus Research Automation (SARA) system or a SARC virus researcher. This heuristic technology has been shown to detect up to 80% of new, unknown viruses.