AWR vs. TKPROF debate

I am on a rather large team and we are trying to come up with a consistent 
strategy for performance testing. We want to come up with a way to gather 
metrics and measure performance and regiment it so that everyone does 
relatively the same thing. This is primarily a batch processing system written 
in pl/sql. 

I favor an approach of awrrpt.sql plus 10046 trace. We need the awrrpt.sql to 
gather bulk metrics during tests where we have alot of threads running to get 
cpu, physical IO, redo, etc... to put into our models. 

This does not provide enough granularity to measure individual sqls, so I also 
want us to gather 10046 traces. This is easy enough since we can gather them 
for each batch process instead of having to do it for 1 sql at a time. I want 
to keep the awrrpt and the 10046 traces in case we have a performance problem. 
We can then review our last well performing plan and compare them. I think it 
will make it easier to fix potential performance problems. This is easy enough 
and requires little effort. 

I am getting some push back on this. Some memberse of the team believe that the 
AWR and v$sql_plan views are mature enough where we can get the same 
information as we can with a tkprof. I have not looked at the AWR all that 
much. Is there any accuracy to this?

Other related posts:

  • » AWR vs. TKPROF debate