Let $(X_t)_{t\geq 0}$ be a continuous (or càdlàg), real-valued process, and define stopping times $$\tau_{s,a,b}=\inf~ [s,\infty)\cap\{t:X_t\notin (a,b)\}.$$ We can interpret $\tau_{s,a,b}$ as the first time after time $s$ that the process hits $a$ or $b$.

Suppose that for all $s,a,b$ we have:

$$\mathbb{E}[X_{\tau_{s,a,b}}|\mathcal{F}_s]\leq X_s$$

**Then is $X$ necessarily a local supermartingale?**

*Context:*

At first I thought that perhaps $X$ was necessarily a supermartingale, but Nate pointed out that there are *local* supermartingales with this property. For example,

$$ X_t = \begin{cases} W_{\min(\frac{t}{1-t},T)} &\text{for } 0 \le t < 1,\\ 1 &\text{for } 1 \le t < \infty, \end{cases}$$

where $(W_t)_{t\geq 0}$ is a Wiener process and $T=\inf\{t\geq 0:W_t=1\}$, seems to fit the bill.

*(The question was edited in response to Nate's comment.)*

I can solve the analogous problem in discrete time by induction, but don't know where to go from there, if indeed it's of any use.

Thank you.