Question: strange numerical integration

Using the following integrand

.71428491807021e-1*(1/2-1/2*erf(.50253255206611e-2*xi*2^(1/2)))/Pi^(1/2)*
exp(-1/2*(.14285698361405*xi-5.6995353453900)^2)*2^(1/2)*(1/2+1/2*
erf(1.1171516020037*(5.4091420603773+.44756682898115*xi)*2^(1/2)))

I get -0 for Int( %, xi= -99.496042185589..infinity): evalf(%,14);

which certainly is false. But if I cut off at a reasonable upper bound
(say exp(...) <= 1E-16) I get what I expect (up to rounding errors).

I consider that as a bug and wonder whether it is in the NAG library
or through Maple calling it - any explanation?

Edited to add: `Maple 10.02, IBM INTEL NT, Nov 8 2005 Build ID 208934`, on Win ME
Please Wait...