Odd ODE15s behavior due to tspan argument

Can anyone explain why ode15s results are sensitive to the tspan input? I thought the solver just interpolated between time points at the tspan argument pointz. But running the below (in 2017a) gives odd behaviour for me. Specifically, even at similar times the results are markedly different across each example.

% oscillating decaying function
% same function deltaT=0.01
[T1,Y1]=ode15s(@(t,x) -0.06*exp(-0.03*t)+0.2*cos(t),0:0.01:100,7);
% same function deltaT=1
[T2,Y2]=ode15s(@(t,x) -0.06*exp(-0.03*t)+0.2*cos(t),0:1:100,7);
% same function deltaT=10
[T3,Y3]=ode15s(@(t,x) -0.06*exp(-0.03*t)+0.2*cos(t),0:10:100,7);
% same function deltaT=10
[T4,Y4]=ode15s(@(t,x) -0.06*exp(-0.03*t)+0.2*cos(t),[0 100],7);

Tal=0:0.1:100;
t=0:1:100;
%analytical solution
y=5+2*exp(-0.03*t)+0.2*sin(t);
figure(gcf)

%compare all
plot(T1,Y1,T2,Y2,T3,Y3,T4,Y4,t,y)

legend({'DT=0.01','DT=1','DT=10','start and end','Analytical'})
xlabel('Time')
ylabel('Y')
figure(gcf)

[Richard Allen]

Here is the output (from the MATLAB iOS App, but Mac R2017a is identical.

Richard