Bilattice offers grid logic puzzles you solve by shading cells in the style of Taiji, where you figure out the rules by experimenting. You'll have more fun in this style by playing Taiji before this, or indeed The Witness, Cosmic Express, Understand and Gridspech. Consider this when you're at the point of playing the good version of Pixoji or What the Witness?
If instead you come from a paper puzzle background in the Nikoli style, the core rule is the Light & Shadow variant of Nurikabe. You are best served in electronic form by the puzzles presented in the pzv/puzz.link editor/database, pedroworks, or the various versions of penpa/expo. Consider this when you want to face a rolling mix and match of styles that requires little browsing.
The learning the rules bit is supported by a curated manual inclusion of area counters, clue validators and automatic grid validation, which is an excellent approach. There's also the option to read the rule straight up that eventually shows up, so you're never hard stuck from ignorance. If you pass it up and eventually concede you need it though, you have to find which specific puzzle randomly contains the "spill the beans" button. The game forgets your solution when you're not looking at it, so experimentation across puzzles or sessions is obnoxious.
There's a couple of novel and interesting things eventually done here, one involving a relaxing of shading and one involving puzzle validation, but overall not every rule interacts in an deep, interesting way with every other. In design terms, some boxes in the interaction matrix are pretty sparse.
Most of the puzzles aren't for teaching though, but for breeding familiarity and eventually providing challenge. The puzzles themselves seem to be made in different ways, and not all are equally good. There's puzzles whose solving path is designed and you can see it, there's sequences of slightly changing aesthetically pleasing grids, and there's a lot of puzzles where it's very hard to see an articulate, human idea communicated. There's also a couple of options to solve computer generated puzzles under time constraint.
The map is astonishingly bad in several ways. Rather than spread out in 2D, it's a scrollable cone that drops into new screens every once in a while. This introduces a ton of clicking and dragging between functionally indestinguishable screens to find open puzzles in mid-game and makes it impossible to remember where a specific puzzle was. It's also thinly connected, so it makes it extra easy to get stuck. The linearity makes sense for tutorials, but most of the game is not that, and the difficulty is not tuned enough to assume subsequent puzzles would also be a problem.
The UI is a prank. There's absolutely no keyboard support. There's no undo. Where similar games offer cell colours, border marking, exhausted clues marking, freehand drawing, trial locks and so on this offers the ability to mark a cell as unshaded. The button named Main Menu opens the quit confirmation. The randomiser modes can be found in two different map screens, one of which looks like a puzzle and the other is just a button attached to the terrain opening a menu. The X in a square in the puzzle corner is not quit the puzzle, validate the puzzle, or a checkbox, it's "erase the shading". The 3x3 dot button is not a menu dropdown, it's map despite the map not being a grid. If you want to change how assistance works, you have to leave the puzzle, open a menu, and find the puzzle again instead; and you'll want to move back and forth between settings because you want different ones for tutorials and puzzles where single changes have very wide repercussions and would force you to slowly recalculate a bunch of stuff.
The UI slowly broke me. By slowly, I mean I went looking for a refund just after the 2 hour window. I'm resigned to finishing it now. Well, maybe not the integration puzzle. Your mileage may vary.
