During the most recent Ice Age, from 100,000 12,000 years ago, the ice probably did not advance south of the Bears Paw Mountains; but at some earlier time, in a more intense ice age, it surely did.  At some time the ice reached the foot of the Highwood Mountains, and may have engulfed the entire north and northeast sides of the mountains.  This would have completely cut off all the drainage channels of the Missouri and other rivers flowing from the northern Rocky Mountains.  The rivers may have subsided to a mere trickle, never completely filling the lake dammed behind the ice.  But at some time the climate was sure to warm, and the rivers would flow and create a huge lake.  Old shorelines indicate that the resulting Lake Great Falls rose eventually to a level of about 3850 ft (1173 m).  This vast lake was one of the largest in North America.  It would have continued to rise until it found a way out.  The easiest way out was along the edge of the glaciers pushing up against the Highwood Mountains.  Once that gap was breached (M) the torrent must have carried away everything: blocks of ice, boulders, and soil.  This may not have been the event that formed most of the Shonkin Sag, but it certainly was an episode when the Sag was subjected to rapid erosion.
Lake Great Falls at its maximum elevation would have contained as much as 500 to 800 cubic miles of water (2000 to 3000 cubic km).  This would have produced a tremendous pressure at the point where it breeched the ice barrier.  The overflow channel at the town of Highwood is about 250 feet deep (76 m); and lies several hundred feet higher than the Shonkin Sag ten miles downstream.  This would have caused such a tremendous flow that the overflow channel may have been gouged out in a time measured in weeks or months.  That would have quickly lowered the lake to about 3500 feet (1066 m).  The volume of water lost in that short time would have been as much as 200 cubic miles (800 cubic km).  The stream of water must have filled the overflow channel and the Shonkin Sag almost to the brim, eroding away the sides to produce the "U" shape we see today.
A further piece of evidence that the present-day channel of the Missouri was blocked by ice is the diversion of the outflow in a curious dry channel at P, also called, confusingly, the "Big Sag."  For a short distance there are two parallel valleys of about the same size, several miles apart, in one of which flows the Missouri, and in the other of which there is now only an insignificant stream.
Apparently Lake Great Falls rose and overflowed at least two times, probably in two different Ice Ages.  On the hills above the present Highwood overflow channel are the eroded traces of an earlier channel, several hundred feet higher.  This has been mostly erased by subsequent erosion, except in the vicinity of the Dry Falls.  One wall of an overflow channel still stands far above the Shonkin Sag; and the Dry Falls plunge several hundred feet to the floor of the Sag.
The detail map below (with contours indicated at 3200 ft (975 m) and
at 3850 ft (1173 m)) shows the first spillway channel (Q), which was
opened when Lake Great Falls was at its highest level.  That
channel must have been short lived, for is is not very deeply incised,
and is now cut off by several streams that have cut small valleys
across it.  It did, however, persist long enough to form the
gigantic Dry Falls, where the water plunged nearly 250 feet (76 m)
before joining the lower channel of the Shonkin Sag.  The lower
overflow channel (R) must have persisted for a longer time, and been
the main spillway for Lake Great Falls.  That channel may have been
active many times as the lake rose and fell with the changing ice age
A peculiar feature of the Shonkin Sag is that there are several low gaps in the northeast walls, connected to relatively shallow channels shown as dashed lines (S). This raises the possibility that the outflow was not always constrained by the glaciers, and the torrent spilling out of the Lake Great Falls overflowed the walls of the Sag.  Again, there is no way to be certain of the sequence of events; those minor channels have been greatly eroded by ice and later streams.
One misconception that has been stated in several places is that
the Missouri may have been diverted far to the south of the channels
shown here.  The evidence is in several relatively broad valleys
near the town of Denton, Montana.  A diversion through those
valleys does not seem feasible, because they lie on a ridge (T)
that is higher than the level of the spillway channel (R).
A more reasonable explanation for those broad valleys is that
they are prexisting valleys that were later bisected by Arrow
Creek (See the pre-glacial alignment of
Next, the lake drains to an equilibrium level