The definition of Riemann-integrability states that if $f$ is Riemann integrable on $[0,1]$ then for any $\epsilon$ there exists a partition $P_\epsilon$ of $[0,1] $ such that for any $P\supset P_\epsilon$ and corresponding choice $T=\{t_i,\dots,t_N\}$, $$\left|\sum_{i=1}^Nf(t_i)(x_i-x_{i-1})-I\right|<\epsilon$$ for some real number $I$.
(We denote the above sum by $S(P,T,f,\alpha)$.)
A sketch of how I think I should prove this would be to fix $\epsilon>0$ so that we have a partition $P_\epsilon=\{x_0=0,x_1,\dots,x_N=1\}$ and then put $\delta<x_1$, and show that for any $c<\delta$, the Riemann sum with the same mesh size as $P_\epsilon$ is very close to $I$, but I am having a hard time formalizing this and maybe that's because I'm not right. Does this seem like a reasonable approach?
