Have you found circuits/progressions/chippers to be effective in translating to power-endurance on the bike/in other "endurance-based" activities?
The transferability of any "artificial" or "cross-training" exercise fluctuates between individuals and also according to where an athlete finds himself on the development path (seasonally and generally). I know that sounds like a cop-out. Happily it is also just the prelude.
Regarding the general nature of transferability: if the the energy system(s) addressed by both training and target objective are similar then - to a degree - the training transfers efficiently. Because rowing is your main interest we can look at the 2000m event: it lasts roughly six minutes, the contribution of energy is (depending on who you read/believe) is 70-80% aerobic and 20-30% anaerobic. A shorter race (1000m) might rely on 40-50% aerobic energy production and 50-60% anaerobic. To mimic these demands during artificial training ensures transferability in the context of energy systems. In this context circuits/progressions can be tuned to focus on 4-6 minute energy systems, and pacing rules may be applied to make sure the particular energy system is being used. However most individuals don't look at circuits in this way, instead going all-out for time (on shorter circuits) without regard for the desired outcome, which results in the anaerobic system being favored. Progressions, generally being longer, shift emphasis to a greater aerobic energy contribution. Proper pacing is generally voluntary when one knows the exercise will last in the 15-30 minute range.
By analyzing the target sport in the context of muscle contraction (intensity/peak force, and duration or rate of contraction) and also joint angles we can figure out which artificial movements to simulate the sport. If the sport features relatively long sub-maximal contraction durations followed by a period of relaxation it does not make sense to over-emphasize quick explosive efforts (though I wouldn't avoid them entirely) or to focus on maximal efforts or on sustained intensities. Again, most training is complementary so we don't reject any means except those proven to have negative consequences, i.e. lifting at certain percentages/reps/sets/rest causes one to gain un-useful mass so we don't do it. The joint angle issue should be respected as well: hiking uphill with a pack or dragging a sled transfers well to standing on the pedals on the bike (contraction duration factors into this as well). Weighted-hiking does not translate to seated pedaling at higher RPMs. As another example full one-leg squats (pistol) may transfer well to the seated pedaling on the bike but the risk of knee injury is great. A partial one-leg squat to a depth mimicking the maximum joint angle seen on the bike works just as well as a pistol without the risk.
So, for the artificial training to transfer, the movements and energy systems used in a circuit or progression should mimic or be very closely related to those used in the sport.
Overall, I prefer the IWT format to develop power-endurance characteristics useful for endurance and medium-endurance activities because the lifting is done under the best possible conditions (coming off the rest period), and we can adjust the duration of the free aerobic exercise intervals to target the energy system we choose. If you look through some of the IWT variations you will see we use mostly 2-min work intervals (accumulating a total of 12 minutes @ 90-95% MHR during a single workout). We also target pure power (explosive and CV power) by lifting heavier loads for fewer reps chased with 60-90 second "sprint" efforts during the aerobic exercise period and followed by longer rests.
The short answer is: if circuits/progressions are designed and used indiscriminately the transferability is general and limited. If the circuits/progressions are designed to produce a specific outcome the transferability can be quite good, and valuable.