*Consistency:* If the prediction is correct the performance should be the same as the optimum offline algorithm
*Robust:* If the prediction is wrong the performance should not be worse than the best offline algorithm

# Lecture 11
1. **Consistency:** We see that if the $p$ is correct that this algorithm performs just as well as the optimum offline algorithm. Consistency is 1. **Robustness:** The optimum offline algorithm is to wait until day $S-F$. The worst case of the algorithm is $S$, the best case of the offline algorithm is $F$, hence we get a robustness of $\frac{S}{F}$.
2. **Consistency:** $1+\frac{k}{S}$. **Robustness:** $\frac{k+S}{k+F}$  
3. Assume we choose a different k'. 
4. $$t\leq (S-F)(1-\al)$$
# Lecture 12
# Lecture 14
x(1-x)-> (1-x)-x=