Could a 3rd Hand on Belay Loop Save you from an Extended Rappel Failure?

Consider the common setup of a 3rd hand extended rappel ATC setup with a sling and prussik connected directly to your belay loop. Imagine that for some reason the sling fails or perhaps there is a mistake in the rappel setup, would the third hand arrest a potential fall? I realize this unlikely, and ideally, the climber would have fully weighted and tested their rappel setup before loading it.

Here's a short video I made for my own knowledge check. https://youtu.be/BvPVft6vE4A?t=520 Thanks for any comment or advice.

If I'm understanding correctly, you have two somewhat separate topics. (1) You want to know if the third-hand Prusik could save you if there's something wrong with the ATC setup. (2) The video shows you doing a standard rappel setup, and you're asking for comments on that. (The video does not actually show a test of question #1.)

Re #1: Yes, it's pretty likely that your Prusik would catch. Not 100% guaranteed, but it's likely. If you've done any significant amount of rappelling with this setup, then you know that the Prusik is pretty much always catching, and that's why you have to be tending it all the time in order to descend.

Re #2:--

Your setup looks fairly standard to me.

You're using a Prusik, which is what I do, but most people use an autoblock, precisely because the Prusik is so sticky that it makes for a slow, jerky rap.

What is the purpose of the knot in the blue and white sling? Are you just trying to shorten the sling to keep from having too much extension? Trying to make it so that the sling doesn't fail if a single strand of it breaks? It's generally a bad idea to tie knots in dyneema, although with this type of load it shouldn't matter. Realistically there is nothing that is going to break a dyneema sling, so making it redundant isn't necessary.

If you look in Accidents in North American Climbing every year, the big category of rappelling accidents is people rapping off the ends of the rope. The big safety precaution that you haven't mentioned is to tie stopper knots in the ends of the rope.

You need to be prepared for stuff that can go wrong. The rope might not reach the ground. The rope might get stuck when you try to pull it down. The ends might not be even like you thought they were. The rope can get snagged in brush. For some of these possibilities, depending on the situation, the solution may be to re-ascend the rope, so make sure you have that skill practiced up and that you're carrying the gear you need in order to do it.

Would you like observations on just the rappel set up or your whole video? The rappel set up looks fine. Consider

 for help in putting your rappel on(atc) that sometimes ropes will be already dropped down, and ropes will be heavy. So, you can go ahead and attach your third hand, then pull up some slack to help with putting on your atc. This could help with ease and smoothness.

Also, start with looking at the rap rings first before putting a rope through. That will help with making a decision first before setting up rappel and before you rappel.

While the prussik/autoblock should probably catch, it's not guaranteed. A friction hitch is considered a non-primary attachment. In your rappel setup the ATC is the primary attachment with you holding the brake strand. I have had autoblocks slip and slide when poorly made (this is why we test and check before committing to the rap) and some slippery dyneema klemheists slip on single strand ascensions. My point is a friction hitch is never guaranteed, but it'll probably work and I'd certainly chance it as a backup to my extended rap system totally failing vs no backup.

Related, this is why it is recommended to use a catastrophe knot or rolling clove hitch when ascending the rope on two friction hitches. I've had the top one fail (slippery dyneema on single strand) and it slipped down and started pushing the lower hitch down. Conveniently I had a clove hitch tied not too far away and my downward progress to my doom was very lovingly halted.

On the knot in the sling, I was taught by a guide to use this approach, presumably to build redundant loops, albeit at the cost of weaking the sling by knotting it.

Friction hitch Autblock backups do NOT catch reliably for all types of rappels device set up errors, particularly when a single strand of a two strand rappel is not clipped properly. (Thanks Mikey for the bit about prussiks below.)

http://staff.weber.edu/derekdebruin/research/rappelling.pdf

Also worth noting that the friction hitch should NOT be tended by grabbing the hitch, but instead by grabbing the rope above and tending the hitch with the bottom of the hand. Otherwise, the hitch is subject to "panic grab" and will most likely be rendered ineffective unless the rappeller loses consciousness.

Andrew C wrote:

On the knot in the sling, I was taught by a guide to use this approach, presumably to build redundant loops, albeit at the cost of weaking the sling by knotting it.

But, with Dyneema, “weakening” it with a knot isn’t necessarily catastrophic or really a bad thing. It’s just a fact of the matter type knowledge. A knot will weaken Dyneema, but Dyneema is pretty pretty strong.  

Derek DeBruin wrote:

Friction hitch backups do NOT catch reliably for all types of rappels device set up errors, particularly when a single strand of a two strand rappel is not clipped properly. 

http://staff.weber.edu/derekdebruin/research/rappelling.pdf

Also worth noting that the friction hitch should NOT be tended by grabbing the hitch, but instead by grabbing the rope above and tending the hitch with the bottom of the hand. Otherwise, the hitch is subject to "panic grab" and will most likely be rendered ineffective unless the rappeller loses consciousness.

That study is spot on! Thank you, answered all my questions.

If there is a higher chance of weighting the friction hitch independently or higher chance of having to go hands free (you still want a cat knot) I opt for a klemheist over an auto block.

Doesn’t really answer your question but just a general tip ig 

I've yet to be led astray by Derek nor do I expect to. I highly value everything he posts up here on MP and on insta. Thanks for the study, Derek, and for always chiming in.

Aaron Clifton wrote:

But, with Dyneema, “weakening” it with a knot isn’t necessarily catastrophic or really a bad thing. It’s just a fact of the matter type knowledge. A knot will weaken Dyneema, but Dyneema is pretty pretty strong.  

It can be catastrophic, and it's not just an incremental weakening of the material. The mechanism is that frictional heating between the strands inside the knot gets the dyneema hot enough that it melts. However, the lab tests that I've seen only seem to show that it happens either with large loads applied over a significant amount of stretching, or with factor-2 falls. Neither of those is possible here.

Andrew C wrote:

On the knot in the sling, I was taught by a guide to use this approach, presumably to build redundant loops, albeit at the cost of weaking the sling by knotting it.

If you're going to use a double-length sling for this, then knotting it is definitely a good idea because it makes it visually clear what you can actually clip into that will capture the loop. But it would work fine with a single-length sling and no knot.

Probably a more relevant thing to think about re materials is what you use for the Prusik cord. It can get very hot if you rappel quickly, and you don't want it to melt. There is a special cord sold for this purpose called a HollowBlock. Definitely don't use dyneema for the Prusik cord. You can use 6-7 mm nylon cord as well. The diameter of the cord is one of the factors that controls how sticky it is.

Ben Crowell wrote:

It can be catastrophic, and it's not just an incremental weakening of the material. The mechanism is that frictional heating between the strands inside the knot gets the dyneema hot enough that it melts. However, the lab tests that I've seen only seem to show that it happens either with large loads applied over a significant amount of stretching, or with factor-2 falls. Neither of those is possible here.

If you're going to use a double-length sling for this, then knotting it is definitely a good idea because it makes it visually clear what you can actually clip into that will capture the loop. But it would work fine with a single-length sling and no knot.

Probably a more relevant thing to think about re materials is what you use for the Prusik cord. It can get very hot if you rappel quickly, and you don't want it to melt. There is a special cord sold for this purpose called a HollowBlock. Definitely don't use dyneema for the Prusik cord. You can use 6-7 mm nylon cord as well. The diameter of the cord is one of the factors that controls how sticky it is.

Exactly. Different circumstance or a different time, yes. But, in the application of the op’s question, no.

It can be catastrophic under those considerations you mentioned (lab and factor 2), but when it comes to knotting Dyneema for a rappel, I would say a cut in Dyneema or extremely worn Dyneema is more likely to result in a catastrophic accident than a overhand or even a figure eight knot would.

Both right! Right? 

Derek DeBruin wrote:

Friction hitch backups do NOT catch reliably for all types of rappels device set up errors, particularly when a single strand of a two strand rappel is not clipped properly. 

http://staff.weber.edu/derekdebruin/research/rappelling.pdf

I think it should be noted that if you use a Prusik hitch instead of an Autoblock it does protect against missing one of the two strands.  I've tested this multiple times, including 5 min ago and the Prusik  Hitch (6mm tied cord) reliably caught the few times I tried it.  The Autoblock on the other hand failed (which is the hitch used in Derek's testing).  I only tried it on a couple of ropes that I had laying around so if definitely isn't conclusive but to say "friction hitches" don't protect for this failure method isn't necessarily correct.  Call me old school but I never bought in to the advantages of the Autoblock and continue to use a Prusik as my 3rd hand, which I probably use around 95% of the time I rap.

I think many attitudes about third-hand backups are dangerously complacent.  The backups are meant to save the day if the rappeller is rendered unconscious.  They are not meant for conscious releasing of the brake hand under panicky situations (eg swatting a copperhead per the video), and as far as I can tell from the anecdotes and testing I've heard about, there is a chance they will not work if some part of the extended rappel fails (the many variables involved in such situations makes it hard to evaluate conflicting test results).  From a safety perspective, as soon as someone says the backup "probably will work," I think the only good way to think about that is "yeah, but it won't work when I need it to."  

There are two aspects to the dangerous complacency I see. One is that the third hand is poorly installed and/or improperly managed, making it non-functional when it counts. Here's a random shot from the internet depicting a happy rappeller who has no idea that their backup isn't likely to do anything, not just in an emergency but also if the brake hand is released to swat a mosquito. I see setups like this at the crag periodically.

The second aspect is the deprecation of the mission-critical role of the brake hand.  In the days before autoblocks, it was drilled into the rappeller's psyche that releasing the brake hand meant death. Those days, and the life-saving attitudes that went with them, are long gone. Now you have a backup and can let go whenever you want. Here's an example of where that attitude can lead.  We can muse about whether the autoblock was too loose or improperly tied, or whether the inability to release the knot-managing hand in a panic situation kept the knot from deploying, but the real problem is that the rappeller swung just a bit to the right and let go with the brake hand to stop the swing.

I don't know if it is possible, but bringing back the mantra that we NEVER LET GO OF THE BRAKE HAND, because our autoblock may not catch, is ultimately going to save some lives and/or prevent some injuries. Of course, proper installation and management of the knot, as well as a robust pre-rappel test while still tethered in, is also something that shouldn't be skipped. With regard to knot management, note Derek's recommendation to have the managing hand on the rope just above the knot rather than actually grasping the knot is critical to having the knot function in panic situations.

rgold wrote:

I think many attitudes about third-hand backups are dangerously complacent.  The backups are meant to save the day if the rappeller is rendered unconscious.  They are not meant for conscious releasing of the brake hand under panicky situations (eg swatting a copperhead per the video), and as far as I can tell from the anecdotes and testing I've heard about, there is a chance they will not work if some part of the extended rappel fails (the many variables involved in such situations makes it hard to evaluate conflicting test results).  From a safety perspective, as soon as someone says the backup "probably will work," I think the only good way to think about that is "yeah, but it won't work when I need it to."  

Well, if p is the probability that the primary system will fail, and q is the probability that the backup with fail, the the probability that both will fail is pq. When both p and q are pretty darn small, this is a winning game. Even if q is not so incredibly small, like 0.01 or something, the benefit of multiplying by q is a huge benefit.

There are two aspects to the dangerous complacency I see. One is that the third hand is poorly installed and/or improperly managed, making it non-functional when it counts. Here's a random shot from the internet depicting a happy rappeller who has no idea that their backup isn't likely to do anything, not just in an emergency but also if the brake hand is released to swat a mosquito. I see setups like this at the crag periodically.

I don't see anything terribly wrong with this setup. I can't see the friction hitch very well, because this is just a photo with limited resolution. It looks like maybe it isn't very well dressed, but the person doing this is probably a better judge than we are as to whether that hitch is functioning or not. I know that there are failure modes for the biner clipped to the leg loop, but I'm not aware of any real-world evidence that this setup actually causes accidents. It may cause q to be 0.0002 rather than 0.0001, but the reality is that pq is still super small. What this woman is doing looks so safe to me (pq is so small) that it's probably safer than crossing a street near my house where the speed limit is 50 mph.

It's possible that if this woman took her hand off the brake strand, the friction hitch would hit her ATC and jam it. This would be an inconvenience, not a yer-gonne-die.

When you say, "their backup isn't likely to do anything," what are you basing this on? Were you there? Do you know whether she tested her setup's ability to brake before untying a tether from her anchor? To me, "isn't likely to do anything" implies q>0.5. Is that what you mean?

Ben Crowell wrote:

Well, if p is the probability that the primary system will fail, and q is the probability that the backup with fail, the the probability that both will fail is pq. When both p and q are pretty darn small, this is a winning game. Even if q is not so incredibly small, like 0.01 or something, the benefit of multiplying by q is a huge benefit.

I don't see anything terribly wrong with this setup. I can't see the friction hitch very well, because this is just a photo with limited resolution. It looks like maybe it isn't very well dressed, but the person doing this is probably a better judge than we are as to whether that hitch is functioning or not. I know that there are failure modes for the biner clipped to the leg loop, but I'm not aware of any real-world evidence that this setup actually causes accidents. It may cause q to be 0.0002 rather than 0.0001, but the reality is that pq is still super small. What this woman is doing looks so safe to me (pq is so small) that it's probably safer than crossing a street near my house where the speed limit is 50 mph.

It's possible that if this woman took her hand off the brake strand, the friction hitch would hit her ATC and jam it. This would be an inconvenience, not a yer-gonne-die.

When you say, "their backup isn't likely to do anything," what are you basing this on? Were you there? Do you know whether she tested her setup's ability to brake before untying a tether from her anchor? To me, "isn't likely to do anything" implies q>0.5. Is that what you mean?

You’re entering circle jerk territory at this point. The fact of the matter is the technique is antiquated and people have died from complications stemming from the use of this method. For example a couple peoples poor positioning of the hitch on the leg loop undid their leg loop buckle causing the hitch to fall out, others have been flipped over and hit their head, and others still have gotten the device jammed in their belay device. Sure feel free to use the method, but extending the rap and hitching to your belay loop is just as easy so why are we sitting here arguing about the efficacy of an antiquated and inferior method 

Mikey Schaefer wrote:

I think it should be noted that if you use a Prusik hitch instead of an Autoblock it does protect against missing one of the two strands.  I've tested this multiple times, including 5 min ago and the Prusik  Hitch (6mm tied cord) reliably caught the few times I tried it.  The Autoblock on the other hand failed (which is the hitch used in Derek's testing).  I only tried it on a couple of ropes that I had laying around so if definitely isn't conclusive but to say "friction hitches" don't protect for this failure method isn't necessarily correct.  Call me old school but I never bought in to the advantages of the Autoblock and continue to use a Prusik as my 3rd hand, which I probably use around 95% of the time I rap.

That's a very good point. I edited my post above to say autoblocks specifically. Thanks.

As for being old school, you're not that much older than me, so I'm not sure what that might mean for me. Haha.

Ben Crowell wrote:

Well, if p is the probability that the primary system will fail, and q is the probability that the backup with fail, the the probability that both will fail is pq. When both p and q are pretty darn small, this is a winning game. Even if q is not so incredibly small, like 0.01 or something, the benefit of multiplying by q is a huge benefit.

I don't see anything terribly wrong with this setup. I can't see the friction hitch very well, because this is just a photo with limited resolution. It looks like maybe it isn't very well dressed, but the person doing this is probably a better judge than we are as to whether that hitch is functioning or not. I know that there are failure modes for the biner clipped to the leg loop, but I'm not aware of any real-world evidence that this setup actually causes accidents. It may cause q to be 0.0002 rather than 0.0001, but the reality is that pq is still super small. What this woman is doing looks so safe to me (pq is so small) that it's probably safer than crossing a street near my house where the speed limit is 50 mph.

It's possible that if this woman took her hand off the brake strand, the friction hitch would hit her ATC and jam it. This would be an inconvenience, not a yer-gonne-die.

When you say, "their backup isn't likely to do anything," what are you basing this on? Were you there? Do you know whether she tested her setup's ability to brake before untying a tether from her anchor? To me, "isn't likely to do anything" implies q>0.5. Is that what you mean?

That backup is so poorly tied I can see it despite the low resolution. It certainly does not resemble a well-dressed hitch. The issue isn't that the friction hitch gets jammed in the ATC (annoying, but solvable), it's that it gets tended when it hits the ATC, leaving it wide open. And a wide open friction hitch entirely defeats the point of having one at all. At which point, you're likely better off not using one instead of engendering a false sense of security in a backup that will certainly not work.

The probability compared to other activities is irrelevant as this is domain specific, unless we should start comparing rappelling and all other climbing activities to driving in every conversation. What's being discussed is a pretty low-cost way to gain additional security. Sure, you could drive your car without a seatbelt, but it's easy enough to put it on, so why not? And if you're gonna put the seatbelt on, it wouldn't make any sense to render it ineffective while you're at it.

As for the actual math, pq is the resultant probability only if both events are independent. However, in this case they are conditional such that the human actor in the system can simultaneously mess up both the rappelling and the backup through panic grab. Keep in mind this is exactly the thing we do want to be concerned with--worst case errors. Most of the time people don't have accidents climbing, even if they do things poorly. But the residual risk is exactly why climbing is risky, so to actually achieve lower risk, we want to identify and mitigate (or even eliminate) edge case problems. It's the same reason you test factor 2 falls--they're unlikely, but if they do happen, that's the thing that could actually rip an anchor out of the wall. 

Therefore, once you've mastered keeping your hand on the rope and other basics, risk is reduced on the margin. That's how it is for many complex problems.

Ben Crowell wrote:

Well, if p is the probability that the primary system will fail, and q is the probability that the backup with fail, the the probability that both will fail is pq. When both p and q are pretty darn small, this is a winning game. Even if q is not so incredibly small, like 0.01 or something, the benefit of multiplying by q is a huge benefit.

I don't see anything terribly wrong with this setup. I can't see the friction hitch very well, because this is just a photo with limited resolution. It looks like maybe it isn't very well dressed, but the person doing this is probably a better judge than we are as to whether that hitch is functioning or not. I know that there are failure modes for the biner clipped to the leg loop, but I'm not aware of any real-world evidence that this setup actually causes accidents. It may cause q to be 0.0002 rather than 0.0001, but the reality is that pq is still super small. What this woman is doing looks so safe to me (pq is so small) that it's probably safer than crossing a street near my house where the speed limit is 50 mph.

It's possible that if this woman took her hand off the brake strand, the friction hitch would hit her ATC and jam it. This would be an inconvenience, not a yer-gonne-die.

When you say, "their backup isn't likely to do anything," what are you basing this on? Were you there? Do you know whether she tested her setup's ability to brake before untying a tether from her anchor? To me, "isn't likely to do anything" implies q>0.5. Is that what you mean?yi

Oy.  My message is to (1) encourage folks to pay attention to the details of setting up and managing an autoblock and (2) not to expect it to do things it isn't well-suited to doing. Some quibbling with  p's and q's is an unfortunate distraction.  Unless you like such techno-wanking, I suggest reading no further.

The pq calculation is wrong.mathematically as Derek pointed out, but even if the events were independent, which they certainly are not, it's the wrong computation for making judgments, and represents one of the problems climbers face generally when it comes to risk evaluation. The problem is that we pay a lot of attention to forestalling very low probability events. Look at all the effort that goes into anchor construction and rigging, in spite of the fact that the kinds of falls that would test an anchor are very rare. Multiplying the probability involving a failure that matters by a tiny additional p-value representing how likely a failure-causing event happens leads to exactly the kind of reasoning you've advanced---these events are too unlikely to be of much concern.

We want to know the likelihood that the medicine works IF you get sick, and should not confound that with the probability of getting sick, as the pq computation does. So the probability we actually care about is a conditional probability, namely the probability that the friction knot will hold given any aspect of the system fails, eg brake hand release or misthreading of the device or whatever. That was my point, and you mischaracterized it as just a system failure (which is what the OP asked about). This conditional probability might be the q you've mentioned, but in any case pq is the wrong number.

Getting to your question, "when you say, 'their backup isn't likely to do anything,' what are you basing this on? Were you there? Do you know whether she tested her setup's ability to brake before untying a tether from her anchor? To me, 'isn't likely to do anything' implies q>0.5. Is that what you mean?"

The short answer is yes, it's exactly what I mean.  If you look at that setup (and no, the resolution isn't even remotely a problem) and don't see at least three serious problems, then I guess we have nothing to discuss.  But you've also misunderstood the role of the picture as a generic warning, and instead are taking it as a particular incident about which we have only some visual information.  No, I wasn't there. No, I don't know if she tested it. It doesn't matter, because as a generic example it is highly problematic.  Out of curiosity and skepticism, I've played around with such backups, well and poorly dressed,  and I'd say, from my experience, that the probability that an autoblock tied like the one in the picture is going to fail if she lets go of the brake hand is indeed greater than 0.5.  So when I said it "isn't likely to do anything," I meant it.  

But trying to assign probabilities when we have no idea what they are cloaks ignorance with empty precision. If the probability of failure turned out to be 0.25, for example, I would consider that totally unacceptable. My language is perhaps at fault when I said her backup isn't likely to do anything, not realizing that someone would want to quantify "likely" when I meant it emphatically.

OP - if you put the third hand on FIRST (autoblock or prussik) you can verify much easier that it is done right and will catch. It will also take the weight off of the rope making it easier to attach your belay device and I recommend incorporating it into your routine. Finally there is another method possible where the third hand could go above the belay device if you are concerned about going off the ends (which should have knots anyway). It is not popular on this continent though. 

rgold wrote:

Oy.  My message is to (1) encourage folks to pay attention to the details of setting up and managing an autoblock and (2) not to expect it to do things it isn't well-suited to doing. Some quibbling with  p's and q's is an unfortunate distraction.  Unless you like such techno-wanking, I suggest reading no further.

The pq calculation is wrong.mathematically as Derek pointed out, but even if the events were independent, which they certainly are not, it's the wrong computation for making judgments, and represents one of the problems climbers face generally when it comes to risk evaluation. The problem is that we pay a lot of attention to forestalling very low probability events. Look at all the effort that goes into anchor construction and rigging, in spite of the fact that the kinds of falls that would test an anchor are very rare. Multiplying the probability involving a failure that matters by a tiny additional p-value representing how likely a failure-causing event happens leads to exactly the kind of reasoning you've advanced---these events are too unlikely to be of much concern.

We want to know the likelihood that the medicine works IF you get sick, and should not confound that with the probability of getting sick, as the pq computation does. So the probability we actually care about is a conditional probability, namely the probability that the friction knot will hold given any aspect of the system fails, eg brake hand release or misthreading of the device or whatever. That was my point, and you mischaracterized it as just a system failure (which is what the OP asked about). This conditional probability might be the q you've mentioned, but in any case pq is the wrong number.

Getting to your question, "when you say, 'their backup isn't likely to do anything,' what are you basing this on? Were you there? Do you know whether she tested her setup's ability to brake before untying a tether from her anchor? To me, 'isn't likely to do anything' implies q>0.5. Is that what you mean?"

The short answer is yes, it's exactly what I mean.  If you look at that setup (and no, the resolution isn't even remotely a problem) and don't see at least three serious problems, then I guess we have nothing to discuss.  But you've also misunderstood the role of the picture as a generic warning, and instead are taking it as a particular incident about which we have only some visual information.  No, I wasn't there. No, I don't know if she tested it. It doesn't matter, because as a generic example it is highly problematic.  Out of curiosity and skepticism, I've played around with such backups, well and poorly dressed,  and I'd say, from my experience, that the probability that an autoblock tied like the one in the picture is going to fail if she lets go of the brake hand is indeed greater than 0.5.  So when I said it "isn't likely to do anything," I meant it.  

But trying to assign probabilities when we have no idea what they are cloaks ignorance with empty precision. If the probability of failure turned out to be 0.25, for example, I would consider that totally unacceptable. My language is perhaps at fault when I said her backup isn't likely to do anything, not realizing that someone would want to quantify "likely" when I meant it emphatically.

Rgold with the mic drop