I've read Gumshoe. I don't agree with it's premise. A zero-failure situation is nothing more than a railroad.
You may have read it, but I am not sure have understood the implications - reading doesn't necessarily show you how it operates in play.. Gumshoe doesn't remove the possibility of failure - it shifts where that failure may be.
In a typical old-style D&D scenario, the hard part is
finding the clue, but the content of the clue is blatant and makes what is going on bleedingly obvious - the puzzle is in finding information, but once you have it, you know exactly what's going on.
In Gumshoe, that is reversed. Getting the clue is easy, but the content of the clue does not automatically give you The Answer. You have to put several clues together, and interpret them correctly, to figure out what is going on.
Investigations are messy, sprawling, and never 100%.
Yep. And a good Gumshoe scenario is that. Remember, it isn't that the characters walk into a scene, and the GM just starts reading off clues. The players have to consider what it in the scene, and choose what skills to use to find information. They only get the clue if they have the right skill in the right place, and they interact with the scene appropriately. The only thing removed is the die roll.
For example, if they come into a scene with a bunch of forensics gear, but don't talk to the NPCs, they won't get information the NPCs have. In Gumshoe, you have to expect that the players will not find every single clue.
I go with a sandbox approach. Groups can fail. They can die.
"Sandbox" is orthogonal to the questions of failure, consequences, or character death.