Some sports, such as tennis, use a complicated points system (point, game, set, match; with deuces and tie-breaks) for what would otherwise be an extremely simple and monotonous sport. The main reason, I suppose, is to make the sport more stressful, by making some parts of the match more crucial than others. It should be possible to make it a lot more 'stressful' mathematically.
Mathematical scenario:
There is an event, played by 2 players, worth 1 point. The probability of winning the entire match, if they (a) win (b) lose the next point is calculated. The difference between these values is the 'stress' value of that point. The sum of all the stress values is called the 'total stress' value of the match, and this number divided by the number of points played is the 'average stress'. For players of equal skill, and stress-handling, the probability of either player winning any given point is $\frac12$. The 'skill' of a player is a randomly generated number from $0\%$ to $100\%$, and the probability of a player winning a point is given by $\frac{Skill(own)}{Skill(own)+Skill(opponent)}$
Your job is to create a points system, however simple or complicated, using any kind of conditions, limits, etc, that produces the maximum possible:
(a) Total stress
(b) Average stress
for a match. You are allowed to create your own definitions of 'point', 'set' and so on, however, the 'points' I am referring to is a single valid serve (double faults to be ignored) with a single winner.
The only constraint is that the total number of points played should be between $1000$ and $1500$ for at least $90\%$ of the scenarios (9 in every 10 matches should not have the players play less than $1000$ or more than $1500$ points).
What you are missing is called "competitive balance" in the economic literature. You need a set of rules to assure it, otherwise, you will have a major leaguer play a t-ball player.
Since players are randomly generated, you need a rule system that begins to systematically sort players of similar rank.
There are two distinct issues here. The first surrounds the mean, the second the variance. Two t-ball players are unlikely to meet a stress condition because their natural variance is high. Two professional players are likely to face the stress conditions you are seeking because the mean rank will be highly similar and the variability will be very small.
Your rules do not seem to allow learning. Random systems will have grave difficulties in creating a stress system because it would be like playing competitive Roulette. While 90% of the scenarios have to have between 1000 and 1500 points, you could play low-level qualifying matches with few points. You could use something similar to a Swiss System tournament with an Elo system to rank. This is prebuilt so you can adapt it to your needs.
If sorting is not allowed, then handicapping is your alternative. This takes away any advantage. If you want to maximize stress, then you want to include rules to minimize the probability of a statistical run. The handicapping will normalize the odds, you will want to create some favoritism against the leader to minimize the variability.