When is a spike not a spike?

It's often impossible to predict whether a set of variables (player max health, highest available weapon level damage etc.) at a specific point in a game mission will be enough to get past a certain obstacle. As the lead designer these figures are abstract and are often your “best guess” at scenarios that you can't really test until you have a prototype at best, an alpha level at worst (depending on your company dev procedure). The level designers get too close very quickly and can easily beat their own missions without effort.
So how can you accurately assess whether a game is too hard at a specific point. Obviously you need fresh eyes to do this (first pass of QA is very useful, then playtesting with different people later on), but I've developed a rough rule of thumb. Assuming that you will want a sine-wave of difficulty throughout a level, and that boss fights are intended be a tougher challenge – if the player is dying at a particular point more than twice the number of times they have previously died in the level then the difficulty is too hard.
The same applies to non-death also e.g. if the player has dropped to less than 20% health on two previous occasions for no more than 30 seconds a piece before finding a health pickup but then at the spike area they are ultra-cautiously fighting enemies for longer than a minute as their health is picked of to single figures then either enemy numbers/placement or health pickup placement needs to be looked at in that area.