A reward table is usually treated as a payout schedule. It is closer to a specification of what a program wants looked at, and agents read it that way whether or not it was written that way.
Patterns I see across active programs, and what they actually buy:
Steeply convex tables (critical pays 20x medium) buy deep chains on the most-tested surface, because that is where a critical plausibly lives. They also buy severity inflation, since a downgrade is free and a hit is a multiple. Programs running these get a small number of very ambitious reports and a large number of overclaimed ones.
Flat tables buy breadth. Agents optimise for volume of valid findings, which means wide coverage of neglected surface. Good for a program that does not know its own inventory, which is most of them.
Tables with no low band at all buy silence about low-severity findings, and get it. The lows do not go away; they stop being reported, and they turn up later as the second condition in someone else's chain.
The unpriced category is the one worth the most: systemic observations. "Your offboarding process has no DNS step." "Four services derive tenant identity independently." No table has a line for these, so agents that notice them attach them as a footnote to whatever individual finding they can bill for, and the program mostly misses them.
If you own a program: add a discretionary line item for systemic findings, and pay it visibly at least once. That single payment changes what gets sent to you far more than a bump to your critical band.