Before everyone gets all hot and bothered there is a reason why this DMM Sentinel carabiner failed at 21kN (for those who don't know the Sentinel is rated at 24kN).
The carabiner in question was the hardware component in a test conducted by Treetools in conjunction with splicing guru Drew Bristow.
Thoughtplanter Andy Neverman and Dave Darbyshire from Asplundh were also present for the test.
The DMM Sentinel screw gate was used to hold the configuration together (and it was expendable).
Everyone assumed the 'soft' elements in the system would give way first - not the hardware.
We were all wrong! As it turned out the carabiner was seriously side loaded in this configuration. The 'side-loading' issue only presented itself once the system had pressure applied.
Apparently the top bar DMM Sentinel is too narrow to accommodate the Fishermans knot (in the Arrow Frog) and the two spliced legs of the 8mm Armor-Prus e2e.
At first glance there appeared to be plenty of room on the top bar to accommodate all the three legs.
But once the pressure went on the two legs of the hitch slid to the top corner of the carabiner, loading it at that point (there is a slight upward angle on the bar leading to the top corner of the carabiner). The Fishermans knot was forced in the opposite direction to a point above the gate.
From the two opposing corners of the carabiner the pressure quickly transferred itself to the weakest point, the thin bar section before the nose ball.
The nose ball snapped off at this point and the carabiner bar was ripped open in a split second at 21kN. The Fishermans knot is almost 'welded' to the carabiner bar at the top of the nose above the break.
This doesn't make the Sentinel a bad carabiner - just one configured incorrectly (we now know).
If there is a lesson to be learned in this test, its the fact that 'side loading' does occur even when it appears not to be the case. All three legs of rope looked to sit comfortably on the carabiner bar and there was no apparent side loading at all (when it was originally set).
The 'soft' elements in this system remained intact throughout the test.
We later broke the same Armor-Prus e2e (used in the first test) at 14kN.