Randomizing stat upgrades in a video game

The conventional handling of technology upgrades in a video game is fairly simple:

• Item stats go up
• Cost of next upgrade goes up

The better games will open up new branches of gameplay possiblities, but let's leave that out of the discussion.

I've been playing a game where the tech upgrades often produce surprisingly large leaps in power. The game makes the player choose between replacing items with the newer technology (which costs a limited supply of money) or waiting for the next upgrade (which may be either a minor upgrade or a huge upgrade). These upgrades appear to be hardcoded. What if they were randomized?

It is common for modern roguelike games to have categories of upgrades. Upgrades only start with the classical D&D +1, +2, +3 system. Then there is a category of upgrades that upgrade one stat, a "rare" category that upgrades two stats, and a "legendary" category that upgrades three or more stats. The upgrade categories are usually color-coded for your convenience.

Let us consider a game where the player pours resources into research, and imagine possiblities of different outcomes.

• Complete failure. The scientists shamefully report that the plan didn't work and they have to go back to the drawing board. However, there is an increased chance of succeeding in the next research attempt. This was used in the DOS game Stellar Conquest 2469 (stelcon).
• Two steps forward, one step back. Two (or more) stats are upgraded, while one (or more) other stats are downgraded. This reflects the TANSTAAFL principle.
• Technology branch. A new branch of research is opened for this type of item. This may involve a downgrade of the item stats. However, because this is a new and unexplored branch of research, the cost of advancement is reduced. This can be considered a subcategory of "two steps forward, one step back".

In a game where different item choices require different gameplay styles, the optimal choice of items may become different for each time the game is played. Imagine a space combat game where missiles got cheap and powerful in one playthrough while their range and reloading speed improved in another playthrough. Imagine an RTS where in one playthrough your ranged units get increased range while in another playthough they get increased hit points. Imagine an RPG where in one playthrough your healer gets reduced cost and casting time of your healing spell while in another playthrough only the healing power improved significantly. The behavior of the game components may change just enough to make the player choose a different strategy, increasing replay value.

This is starting to sound like removing the upgrade tree from player control. My intended vision was for randomized upgrade paths resembling organic evolution, where each upgrade is the starting point for the next upgrade. Imagine that a crafter develops a new technique which requires more time and resources but produces a more resilient product; you can continue using the old technique or refine the new technique.

Counter-thought: the Dragon Warrior series randomized the values of upgraded stats when a player gained a level. It barely affected the game because the characters were different enough in their skillsets that the trivial differences in stats did not matter; the stat upgrades were guided by weights for each character class; and the randomness approached a mean average over enough levels. If the upgrades are expected to be different enough to affect gameplay when the game is replayed, these normalizing factors need to be overcome.