Most of the time the answer is not that sleep restriction failed. It is that nobody told you what to do in week three.
Every article about this explains how to set the window. Almost none explain how to adjust it, which is the part a clinician spends the actual appointments on.
First: has the protocol actually been run?
Before touching anything, two checks. In clinic these account for most stalls.
Has the rising time genuinely held - every day, including weekends? Not "mostly". A Saturday lie-in shifts the whole schedule and takes days to unwind, which is why Sunday and Monday nights are so reliably bad. If the wake time has drifted even twice a week, the protocol has not been run, and adjusting the window will not fix that.
Has the window been widened early? Three good nights arrive, going to bed thirty minutes earlier feels like a fair reward, and it dilutes exactly the pressure that produced the good nights. This is so common it is almost a rule.
If either is true, that is your answer. Reset and run it clean for two weeks before concluding anything.
The decision rules
Assuming it has been run properly, everything now keys off one number: sleep efficiency - the share of your time in bed that you were actually asleep, averaged across the last seven nights. The diary is what produces it, which is why the diary is not optional.
Weekly, not nightly. One bad night means nothing; a week is a signal.
85% or above: widen. Add about 15 minutes, always by moving the bedtime earlier - never by moving the rising time later. Then hold for a full week before deciding again.
Roughly 80% to 85%: hold. Change nothing. Run another week at the same window. This is the range where most people widen anyway, and it is the single most common way a program quietly stops working.
Below 80%: consider tightening by about 15 minutes - with a floor. Never below five and a half hours, and there is a case for stopping at six. Below that you are trading sleep debt for a number, and daytime sleepiness stops being a side effect and starts being a safety issue.
The reason it is always the bedtime that moves is that the rising time is the anchor for your body clock. Move it and you lose the one fixed point holding the schedule together.
How long is a plateau supposed to last?
Longer than people expect.
Trials run this over four to eight weeks, and change usually appears somewhere in weeks two to four - usually, not always. A flat week or two in the middle is ordinary and is not information.
The honest checkpoint is week six, not week three. Judging it earlier is how people quit during the part where it is working and has not surfaced yet.
What "working" looks like before it feels like working
People look at total hours slept and conclude nothing is happening. Total hours is the wrong number to watch - in the pooled trial data it is the one measure that does not reliably improve, and the improvement it did show was not statistically significant.
