The first thing everyone notices about Torres are the plastic castle pieces. Indeed using them to expand castles and the build towers for your knights to capture is the core of the game. Each turn, you are given a stack of tower blocks and 5 action points. Placing a block or moving a knight one space orthogonality, costs an action point. Adding a new knight to the board costs two action points. Three times in the game, your position is scored. For each castle, multiply the height the tallest tower you occupy by the area of the castle.
I'll talk about the other parts of the game in a bit, but the way knights and castles are restricted creates a wonderfully interactive and tense game. You can't merge castles and you can't build taller than the area of the base layer, so there will naturally be very large castles crowding out smaller ones. Since many people can occupy the same castle, the competition is often over space needed to build profitable towers. Building up requires building out so one player might spend actions and blocks giving another player space to build a rival tower. Knights can climb one level and fall any number of levels, so an important part of the game is blocking other players from climbing. But knights can also travel under castle blocks at no extra cost, which means the board gets effectively smaller as it's built up. Every turn demands thinking not only of your position, but how other players might use the board state you leave for them.
One extra wrinkle is the king token, which serves as a catch-up mechanic. The last player to move in the first turn and the lowest scoring player at the end of each scoring round may move the king to a new castle. If you have a knight in the same castle as the king on level 1/2/3 during scoring phase 1/2/3, you gain a bonus 5/10/15 points. It's not enough points to lose the game if you don't get them, but enough that it's usually worth spending a knight and some blocks to grab it. Whoever places the king has a tremendous advantage scoring the bonus.
Another wrinkle are the action cards. Each player has a deck of 10 action cards that give you special one-time abilities. For instance, one card allows a knight to jump two levels instead of one. That has the potential to negate someone else's plan, but only the one time. In the basic game, players spend an action point to randomly draw a card. The consensus seems to be the game is better if everyone starts with all 10 cards in their hand instead. That way nobody can luck into a better draw. Personally I don't mind the small amount of randomness. I still prefer giving players all their cards up front, however, because it's easier to teach and remember the cards exist. Failing to use action cards is a good way to fall behind.
There's also a deck of Master cards that add another layer of scoring by positioning knights in specific configurations. They are just a step too complicated for my tastes. Managing the scoring for castles is plenty without adding a bonus for, say, lining up all of your knights. It seems like a mechanic designed to spice up a game that's been played often by the same group of players. It's an interesting extra that would probably be sold as an expansion these days, but doesn't improve the game dramatically.
The action point system shines in Torres. Each turn requires stringing together a series of castle building and knight movement to (hopefully) leave you in a better situation than your opponents. These actions can't be interrupted, but it's common to wish you had an extra action or two to solidify the position. (Thankfully there are action cards that add one or two actions, if you chose to use them.) As a result, each turn feels like something important was accomplished, but there are still things left undone. It requires decisions be both tactical and strategic in order to succeed.
On the downside, these decisions tend to be taxing and mathy, especially later in the game. As more possibilities open up (more knights, more spots for taller towers and more freedom of movement from longer castle tunnel systems) the combination of actions spirals. With experience, players can spot better actions more quickly, but new players will need time to plan out their moves. It can be hard to visualize a turn without placing and moving pieces, so it's not unusual for someone to spend most of their actions before realizing their plan won't work. Some people might find this frustrating.
The game scales by decreasing the number of blocks allocated as more players are added. That means castles end up roughly the same size at the end of each scoring phase. Head-to-head games tend to be a bit more strategic as you can guess what the other player is planning. 3-4 players creates more tactical opportunities since the board state tends to change more between your turns. That also means pre-planning while other people play is less effective. So the more players, the more likely someone will wander off midgame to get a snack or something. I've had games end after the second scoring phase because people get bored or overwhelmed.
That said, Torres works well for people who enjoy concentrating on strategic and highly-interactive puzzles. It's in the same general class as Bosk or Power Grid. And that means a lot of quiet contemplation of the board state while working out the best way to play your turn. Some turns are especially tricky to game out and take even longer than normal. As a reward, players might find an unexpected solution that genuinely feels inspired. Torres delivers, but it cheats a little bit with the action cards. Without them, the game becomes an exercise in turn optimization.
Compare with Santorini, which feels like a spiritual successor to Torres, has even simpler rules and produces clever play with less downtime. It switches from action points to two defined actions (move then build), reduces the board to a smaller size, limits knights/workers to two and simplifies the victory condition. Then it adds in god powers that are like Torres' action cards that can be used each turn. Overall a crisper game that takes a fraction of the time. But it's not a game about building up so much as exploiting the mistakes of your opponents. The complexity that Torres adds means there's more room for people to come back from misplays. Each scoring round the castles are bigger and the rewards get better.
I'm not sure how to position Torres. It's a legitimately interesting game that I've enjoyed. But it really needs to be played by people ready for deep thinking and downtime between turns. The online implementation has revived the game for me since it's easier to find opponents. (And the downtime is a feature, not a bug!) But it's just not the same without the physical components that elevate (heh) the experience.