None of that has anything to do with sleight of hand though, nothing in the rules or my understanding of how lockpicking works would indicate that the noise level would change based on success or failure.
The rules don't need to spell
everything out, because the writers assume that the players and GM have an imagination. For instance, if you fail a stealth check, the rules don't say
how. But you probably wouldn't use the following:
Instead you'd say something like "I stepped on a twig." Even if a twig hadn't been previously established.
You are still adding the cook to the kitchen,
I also added the lock to the kitchen door, and the kitchen itself. Adding a cook is no different or strange.
I'm not and the original example didn't put a cook in the kitchen unless there was a failure. If the cook is always there why would they only scream because of someone is picking the lock but not a second later when the character opens the door after a success?
Because you did a bad enough job picking the lock that your actions alerted the cook. You didn't
You can, of course, still have the cook scream when someone enters the kitchen, even if they didn't hear the lock being picked. However, because the sound of the lock being very poorly picked didn't alert the cook, they're not paying undue attention to the door, leaving the PCs open to try to silence the cook in one way or another.
Take another example. The characters are in an not-so-abandoned crypt but they dealt with all of the undead denizens. The room is slowly filling the room with sand because they didn't completely disable a trap and I'm a fan of The Mummy movie so there's a time limit. There's an iron chest bolted to the floor but they fail the sleight of hand to pick the lock. Because of it's construction they could eventually break the chest open but the room will flood before then. What is the fail forward option? My option would be that they can't get it open in time, nothing happens.
The sand
is the fail-forward--the room is slowly filling with sand because they didn't
completely disable the trap. If they had a full success on the roll, thus completely disabling the trap, they wouldn't be dealing with the sand.
Also, the chest and trap are unconnected, beyond the fact that they're in the same area. Picking the lock, breaking the chest, ignoring the chest and moving on doesn't affect the trap in any way. It's not an Indiana Jones thing where they have to deal with the chest in a specific way to avoid spring the trap, after all.
The whole point was that they couldn't get into the event and failed a check to bluff their way past. Picking someone's pocket is fine, but it's a whole other set of actions and an alternative way to get in.
That's fine. That's still part of fail forward.
But nothing happened on a failure. Which is fine, I was asking what a fail-forward would look like.
The point of fail forward is to make sure the game doesn't stop moving, and especially so the GM always provides other options. Since there was still the long way around, the game didn't stop moving.
But let's assume that failure to find the passageway
would grind the game to a halt. OK: someone else enters the area. If the PCs run rather than fight, they may get herded to a potentially useful location. If they fight, they might find a key on one of the bodies that doesn't fit a known lock.
They had a chance to decipher the code, probably via an investigation check but failed. If they had figured out the code they could have quickly read it and learned the plans. There may potentially be other options of course depending on the capabilities of the characters.
Here's the problem. You gated the event behind a single die roll. This is what fail forward is designed to prevent.
So one thing to do is to decide how many pertinent bits of information are in the papers. Let's say there are five. On a successful roll, they get all five pieces. On a partial success, they get three. On a total failure, they get one. It doesn't have to be those exact numbers. Maybe a partial success is two pieces, or four.
To make it fairer, you could number the pieces of info and then ask them to choose the numbers.
With knowledge checks? Frequently there are multiple levels of knowledge but roll low enough and you don't get anything other than common knowledge.
Which is literally a type of partial success.