diff --git a/constants/24a.md b/constants/24a.md
index 1e8d3cd..c0bcf1e 100644
--- a/constants/24a.md
+++ b/constants/24a.md
@@ -31,7 +31,7 @@ Since it is not known whether $C_{24}$ is finite, results are typically stated a
 
 | Bound | Reference | Comments |
 | ----- | --------- | -------- |
-| $n$ | Trivial | Since $\|\|A_{\ast j}\|\|_{2}\le 1$ implies $\lvert a_{ij}\rvert\le 1$, we have $\|\|Ax\|\|_{\infty}\le n$ for every $x\in\{-1,1\}^{n}$. |
+| $n$ | Trivial | Since $\|\|A\_{\ast j}\|\|\_{2}\le 1$ implies $\lvert a_{ij}\rvert\le 1$, we have $\|\|Ax\|\|_{\infty}\le n$ for every $x\in\{-1,1\}^{n}$. |
 | $O(\log n)$ | [Bec1981], [Spe1985], [Glu1989] | Partial-coloring/entropy-method bounds yield $O(\log n)$ discrepancy for Komlós-type instances.  |
 | $O(\sqrt{\log n})$ | [Ban1998] | Banaszczyk’s vector-balancing theorem (via Gaussian measure) gives the first $o(\log n)$ bound.  |
 | $O(\sqrt{\log n})$ (poly-time) | [BDG2019] | Polynomial-time algorithm matching Banaszczyk’s existential bound up to constants.  |
@@ -72,4 +72,4 @@ Since it is not known whether $C_{24}$ is finite, results are typically stated a
 
 # Acknowledgements
 
-Prepared with assistance from ChatGPT 5.2 Pro.
\ No newline at end of file
+Prepared with assistance from ChatGPT 5.2 Pro.