A piece of ecological science fiction with obvious similarities to the James Cameron Avatar, Changes relies on a contrast between an idyllic setting and violence and destruction. Some fairly nasty behaviour is required to make much progress in the plot: (Spoiler - click to show)a space explorer crashed on an alien world, you must kill a succession of alien creatures in order to steal their bodies and abilities, enabling you to return to your ship.
The immediate attraction of Changes is environmental; it is set in a good-sized map full of attractively-described locations, pleasant to explore and absorb. (Given how well-suited IF is to environment-focused games, it's surprising that so few exist, so this was pretty refreshing in its own right.) The writing is strong enough to serve as an immediate draw. At the larger scale, it has a consistent, overarching set of puzzle goals that are readily grasped, are deeply tied into the world, themes and plot, and do a good job of directing short-term motivation.
It's at the intermediate scale that Changes stumbles: between the immediate experience of setting and prose and the grand arc of the puzzle sequence, a player has to figure out the shape of individual puzzles and get them to work. Here, the extensive map becomes a drawback: it's not always very clear which problems can be tackled at any given time, and even when you know what to do the execution can be fairly frustrating. A good deal of effort has been made to provide clues, but these often appear long before they can be usefully acted upon and don't show up again. Experimentation isn't always as well-rewarded as it might be.
In some ways, it's tempting to think of Changes as a belated artefact from around the tail-end of the Middle School period, something to be shelved alongside The Edifice and Babel, intended to be played over multiple sittings, likely to stump the player for considerable periods of time. (Tending to support this: the backstory is doled out through amnesia-recovery.) The game might have been served better by that model, perhaps; at any rate, the two-hour Comp doesn't seem to be its optimal environment.