Hacker News new | comments | show | ask | jobs | submit login
Armchair air crash investigator (momav.me)
230 points by momavujisic on July 7, 2013 | hide | past | web | favorite | 150 comments

The accuracy of something like flighradar24 at low altitudes (where the difference between a landing and a crash is a matter of feet) is not high enough, and the data reporting interval is not small enough, to be used to do more than guess what happened. The article gives 3-4 data points per minute. You can't look at that data and draw any conclusions about the motion of an aircraft.

It's fun speculating, but please remember that the data from ADS-B is not designed to reflect the aircraft's motion over the time periods that matter in a crash. We have no idea what delays are present in the system, what delays flightradar's system adds in, and what the accuracy of the system is. It's very important to understand the limitations of any sensor data you analyse. It would be good to mention that in the article.

Edit: I notice that there's a little disclaimer at the bottom of the article saying that Flightradar24/flightaware data is "not 100% accurate", which is sort of covering the issue. However, when dealing with data like this it's important to remember that there's all sorts of ways data can be "accurate" or not. The uncertainty of the data itself, the frequency of data captured, the accuracy of the timestamps attached to the data and all sorts of other variables need to be considered when you're trying to understand the picture that sensor data is painting.

I would hope the difference between a landing and a crash is a little more than 'feet' otherwise we'd have a lot more crashes. Maybe the difference between a BAD landing and a crash is only feet :)

A typical airliner approach is a 3 degree angle touching down at 1000ft past the end of the runway. That means that they pass over the threshold at about 150ft. Let's say there's another 1000ft past that before things get problematic (at SFO it's considerably less, I believe), that means that the airliner passes over the dangerous bit at about 300ft when operating normally, and 300ft lower would cause a crash.

This isn't normally a problem, because detecting and ensuring that 300ft difference is not really hard.

Just a note about approaches and landings – most, if not all airline ops usually say that you should be ~50ft when passing above the threshold.

My simplistic analysis ignored the flare, thus why the real-world height would be lower. Thanks for pointing that out.

It goes without saying that my qualifications in aviation matters are only as an aviation enthusiast and for the full cause of the accident, we'll have to wait for the NTSB report (which has access to vital black box data and cockpit recordings).

However, I did give this accident some thought as well given the circumstances of the accident and came the pretty much the same final conclusion as the OP.

An airline pilot I once spoke to told me once that all the big jetliners use automated ILS with autopilot to automate the landings. The system is pretty accurate and reliable and can deliver landings as good as a well-trained and experienced pilot can.

When I saw the rather steep final approach the aircraft was making, I had suspected that the pilots were doing a manual landing. Now that we have information that the ILS was temporarily inoperational this appears to be what transpired and the pilots manually flew the final approach.

The precise reasons for why the manual approach ended up in this accident are still to be determined. One thing that I will point out though is that pilots for modern jetliners do relatively little actual manual flying of the plane. Modern airplanes have advanced FMS (Flight Management System) that allow for automatic flying from lift-off until landing with pilots overseeing the system. Given this lack of manual flying, there is a greater potential for human error when a pilot is required to actually do manual flying.

Pilots will actually very rarely use the autoland functionality (apart from when weather, e.g. low visibility, requires it, and also to ensure that they're current with the procedure). They'll use the autopilot for the initial part of the approach, but then when visual with the runway will often disconnect it in order to fly the actual landing manually (still following the indications on the flight director, which are provided by the localiser/glidescope signals).

Relevant link: http://www.freakonomics.com/2011/12/12/cockpit-confidential-...

"The touchdown had been a fully manual one, as the vast majority of touchdowns are."

Why is this? Can they do a better (in terms of passenger comfort) job landing manually?

I couldn't say in terms of comfort, but they need to be able to perform a manual landing (for example if certain equipment on board is not functioning correctly, or the airport is not equipped for autolanding) so it makes sense that they would take every opportunity to keep their skills current. There's probably also a sense of pride in wanting to carry out a smooth landing by hand - especially given that most of the climb/cruise/descent will likely have been controlled by the autopilot.

No, autoland is pretty good. But there are two factors that cause pilots to land manually most of the time: 1) pilots think they are better at handling last-minute unexpected contingencies like wind sheer or bird strikes (and they may well be right about that) and 2) pilots don't want to give the Powers that Be any additional data to suggest that they aren't actually needed to fly a plane at all. The fact of the matter is that there are no technological barriers to making aircraft completely autonomous, but pilots want the world to remain in denial about this for as long as possible for the sake of their job security. (FWIW, I'm an instrument-rated private pilot.)

> The fact of the matter is that there are no technological barriers to making aircraft completely autonomous, but pilots want the world to remain in denial about this for as long as possible for the sake of their job security.

Two words: anomalous conditions.

As we both know, pilots aren't needed in the vast majority of commercial flights. The plane is perfectly capable of taking off, navigating, avoiding collisions, and landing with minimal, mostly unskilled supervision.

We don't need pilots on planes... until something unexpected goes wrong, and we do.




That's true, but 1) anomalous conditions are very rare and 2) as often as not the pilot is the anomalous condition. Pilot error is the most common cause of aviation accidents.

EDIT: Notice that you had to go back 30 years to find your third example of a human pilot saving the day. And automation technology has improved a lot since then.

Most of the time when a pilot saves the day he just disconnects the autopilot and writes an incident report. There's been at least four inflight upsets due to autopilot or sensor malfunction in the past few years that I can recall. Some of these caused serious passenger injury. Latest one I heard about here: http://www.aibn.no/Luftfart/Rapporter/13-18 Another one: http://en.m.wikipedia.org/wiki/Qantas_Flight_72

Devil's-advocate: fine, OK, we still need pilots, for the time being. But do they actually need to be on the planes?

In the case of two recent crashes (AF447 and this one, assuming pilot error turns out to be the cause as seems likely), it's safe to say that a computer would've easily done a better job than the pilots.

Would be interesting to hear Capt. Sullenberger's take on the question.

If you're going to use remote pilots to deal with anomalous conditions, you'd better have an emergency electrical supply that'll keep your remote control running for however long it'll take to glide at the plane's maximum range. You'll also need a communication system that'll work over the ocean, in poor weather conditions, while the plane is having difficulty maintaining a stable attitude (good luck maintaining the alignment of a satellite dish when the autopilot disconnects due to a sudden upset).

On the other hand, if your onboard autopilot is in control of everything, you run the risk of having autopilot bugs causing an accident - indeed, a software issue in the airbus flight protections (which, unlike the autopilot, are always active unless a very severe system failure occurs, and can override the pilot's inputs) has caused an incident that resulted in passenger injuries: http://en.wikipedia.org/wiki/Qantas_Flight_72

While none of this is impossible, it's certain to be expensive, heavy, have its fair share of bugs at the start, and by quite difficult to get past the regulators. Not to mention passengers probably won't be very happy being on an unmanned plane, even if the system really was perfect. So it's easier in the end to just put pilots on the plane.

If you're going to use remote pilots to deal with anomalous conditions, you'd better have an emergency electrical supply that'll keep your remote control running for however long it'll take to glide at the plane's maximum range. You'll also need a communication system that'll work over the ocean, in poor weather conditions, while the plane is having difficulty maintaining a stable attitude (good luck maintaining the alignment of a satellite dish when the autopilot disconnects due to a sudden upset).

(Shrug) Sorry, but given the complexity of the rest of the aircraft's systems, none of these engineering tasks sound impossible. 40 years ago we did this stuff on the freakin' Moon.

Although yes, I tend to agree that any move in this direction amounts to fixing something that isn't broken. (In the "freakin' Moon" example, the human pilot still had to handle the tricky part.)

> 40 years ago we did this stuff on the freakin' Moon.


> In the "freakin' Moon" example, the human pilot still had to handle the tricky part.

Since you're apparently well aware that the Apollo missions were conducted with manual docking and landing, why bring it up?

Especially given the relative simplicity of "final approach" on the moon, given the lower gravity, zero wind speed, and slower pace, I don't understand what your point is.

Just pointing out that it's not valid to argue that this stuff is impossible, or even impractical.

What "stuff", though? The moon landings were conducted by a human brain in control of a (mostly[1]-)fully-functional machine. How does that translate to fully automated landing of an atmospheric aircraft already experiencing a malfunction?

[1] Actually, the Eagle did experience a computer malfunction that almost caused an abort to the landing before an engineer said they could ignore it. You want a computer to judge for itself whether it can still safely fulfill its mission while malfunctioning?

For further terror, consult the Apollo 12 lightning strike. When the computer is FUBAR, how do you propose it fix itself?

So, just to make sure I don't misrepresent your position, you're saying that the people who built and programmed this hardware ( http://www.space.com/16878-mars-rover-landing-sky-crane-guid... ) couldn't safely land a 777 in SFO (or, for that matter, an A320 in the Hudson) any day of the week. Is this the case?

What part of the skycrane experienced serious damage or malfunction that the software corrected for?

The whole point is we're not talking about when things go according to plan.

For that matter, how many times has the skycrane been used? What is its safety record? If you go ask the guys at JPL how many times out of 1000 they think it'll work perfectly, what will they say?

Incidentally, not cool moving goalposts.

Sigh. Yes, you're right. Nothing went wrong during the Curiosity landing, so I can therefore not use it as an example of successful flight automation.

You win this one, it looks like. GG.

Assuming we could make such a system secure and unhackable as well as unjammable, why would we want to? To save some money? I would rather pay a little more to have the pilot sit in the same airplane and have his life on the line as well.

I just find it fascinating how obsessed people seem with getting rid of pilots. You don't often hear the same discussion about train or bus drivers, or doctors for that matter, even though we are just as close to having automated robotic surgery and expert systems have been around for a long time.

You don't often hear the same discussion about train or bus drivers

Plenty of cities are getting rid of their subway drivers at least. And if many airport shuttle trains around the world are automated as well.

> I would rather pay a little more to have the pilot sit in the same airplane and have his life on the line as well.

I think I agree with you in general about this, but that's sort of a shitty and unpersuasive way of making that argument.

I guess latency is why they are on the plane.

Might it be possible to have a "house pilot" at each airport that could be patched in during landings or takeoffs? Latency could be made pretty negligible with a radio link only a few kilometers long at most.

That wouldn't do anything for patching in pilots for planes in the middle of the pacific, but I wonder how often that would be necessary.

Another big reason for piloted planes is, seriously now, who here would get on a plane knowing that it was being flown without a pilot.

If it's hard for most of us from varying technological backgrounds to say yes than it's going to be almost impossible to convince the more mature passengers.

This is one of the reasons why all planes look the same (they have the same configuration, low wing with engine below wing) even tho Airbus and Boeing keep churning out concepts. Would you get on a plane that looked even slightly funky?

> ...seriously now, who here would get on a plane knowing that it was being flown without a pilot.

Given that the most common cause of aircraft accidents (by quite some margin) is human error, I'll take my chances with the computer.

I don't find it likely that pilots hand-fly approaches to make a point. Isn't it more likely that this is codified in airline operation instructions?

As for completely autonomous, do you seriously think we have the technology available today to make an autopilot handle any conceivable situation without a human being present and ready to take over? (non-IR PP here)

It isn't "just to make a point". It's also to stay proficient, or to make the job less tedious. But whether to hand-fly an approach is always the pilot's discretion (as far as I know -- I'm a just a private pilot so I could be wrong about that).

And no, automation can't handle "every conceivable situation", but neither can humans. Furthermore, humans screw up more often than autopilots. Pilot error is currently the single biggest contributor to the overall accident rate.

Mind you, I'm not advocating fully autonomous aircraft. I like having a human in the loop, but that's in part because I am the human in the loop. It's far from clear that human pilots are a net win for safety.

Neither can humans, but humans can handle a lot more situations. We can take into account a lot of information and come up with creative solutions, while current computers will need a set of humans to come up with and program potential scenarios beforehand. I think we'll have computer level AI that will perform better in accidents eventually (AI is improving and our brains are not) but it a long way into the future, and not an imminent threat to pilot jobs as some would believe.

In the meantime, the solution to the human factor isn't to eliminate humans, but to improve training (which is already happening after AF447).

I like having human pilots primarily because I'm a programmer and I know how difficult it is to design robust computer systems. There's been one runway overrun and one serious in flight incident with passenger injury due to software design faults. Now try to design a system that makes sense of audio, video and smell in addition to the existing sensors, and not have it fail in some spectacular unforeseen way..

Actually, it is extremely rare for a human pilot to come up with a "creative solution" to an emergency. The vast majority of emergency responses are established procedures for which pilots train. Most of the time they're following a check list.

I can only think of a single example of a "creative" response to an in-flight emergency that actually helped, and that was UA232 in 1989.

Aircraft accidents are extremely rare in the first place, so that's a given. The question should rather be whether computers would do a better job than pilots in the same situation, and currently the answer is no. The actual flying of the airplane which autopilots do today is just a small part of the pilots job, and automating the rest of the tasks a pilot performs is non-trivial.

Are you a pilot? Because I am, and I'm telling you from firsthand experience that you're wrong. Except for takeoff and landing, there's next to nothing I have to do. And the only reason I have to do the landing is because my plane is a small GA aircraft without autothrottle or autoland technology.

Yes, I am a pilot, albeit not an instrument rated one. Are you telling me you don't do anything? Do you bring a book to read instead of monitoring the instruments? You don't talk to ATC, you never have to make a decision regarding a route deviation? You never take into account weather information and make a decision underway on whether to press on or to find another place to land?

No, of course I'm not saying that. I'm saying all those decisions could be automated, not that they have been. Except for ATC communications, all the information I use to make in-flight decisions is already available in digital form. All the engine parameters are digital. I get en-route METARS via XM. I have an WAAS GPS coupled to the autopilot. The only thing standing in the way of making my aircraft completely autonomous is a throttle actuator and the right software. And no, writing that software would not be trivial, but neither would it be impossible.

For normal operations, I think it's possible to make it _mostly_ autonomous today (TTS and voice recognition for ATC, and some heuristic metar/weather radar analysis might work). I don't think we have the technology today to make such a system safe enough to not have a human ready as a backup. And for an accident scenario, I think it's completely impossible today since we would need to integrate audio, video and smell sensors and AI software to rival humans in situational awareness. This would mean exceptionally complex software.

The way we have solved reliability in autopilots and FBW systems today is to make them as simple as possible and to give them sensible fallback modes (like the Airbus FBW removing stall protection when missing certain inputs), and even then we have had real life accidents because of programming errors or design errors. So if you think pilot automation is mainly a question of politics as you said earlier, I think you are being overly optimistic (which, of couse, is not uncommon in the software industry).

I think a better approach to removing pilots is to see the strengths in both computers and humans and design systems where the advantages of both are maximized.

If you are interested in reading more on software safety, http://sunnyday.mit.edu/ is a good starting point.

Is it standard procedure to land A320s in the Hudson? Prior to 2009, how likely do you think it would have been for someone to have programmed an autopilot with such a capability?

Do pilots typically glide unpowered 767s to landings at abandoned military airfields? (A result, incidentally, that could not be reproduced by other crews in simulators; are you sure it's the crews, and not inadequate programming? Still trust the computer?)

Are you willing to bet your life that the computer on a 737 that looked like this[1] could find its way to a safe landing?

As programmers, we should know better.

[1] http://en.wikipedia.org/wiki/File:Aloha_Airlines_Flight_243_...

> Is it standard procedure to land A320s in the Hudson?

Yes, ditching an airplane that has lost all of its engines is a standard emergency procedure.

> Prior to 2009, how likely do you think it would have been for someone to have programmed an autopilot with such a capability?

What difference does that make? I'm not saying it's a good idea to take pilots out of the cockpit right now, I'm just saying it's a lot more plausible than most people think. The main limiting factor is politics, not technology.

> Are you willing to bet your life that the computer on a 737 that looked like this[1] could find its way to a safe landing?

Sure, why not? Losing the top of the fuselage looks dramatic, but it probably doesn't change the flight characteristics all that much. Also, very good adaptive control algorithms exist that could almost certainly handle this.

Also, you're cherry-picking your anecdotes. There are plenty of examples of flights that would almost certainly have ended safely but for some stupid mistake the pilot made. Controlled flight into terrain accidents, for example, are much more common than heroic rescues, and they could be entirely eliminated if you took human pilots out of the loop.

> Yes, ditching an airplane that has lost all of its engines is a standard emergency procedure.

See my reply to krisoft. You're changing a specific situation into a general one. You can't just say "ditch the plane if you lose power", you have to anticipate every possible variable that may influence whether that is actually the correct course of action, and the manner in which it is carried out. And you have to do that before it ever happens.

> What difference does that make?

Unanticipated situations are unanticipated situations. We don't have AI. We haven't replicated the ability of a human being to adapt on-the-fly. There is no reason to believe we will in the near future.

> Controlled flight into terrain accidents [...] could be entirely eliminated if you took human pilots out of the loop.

The problem I have with this line of reasoning is that so much more could be done to prevent them even without taking the pilots out of the loop, and yet it's not. That does nothing to give me confidence that the right thing will be done when pilots ARE taken out of the loop.

Edit: What it comes down to is this. All you're ultimately doing is completely and irrevocably substituting the unalterable judgement of somebody in a completely different time, place, and circumstance for the adaptable judgement of the person on the plane. When you do that, what you're really saying is "I refuse to give people in a scenario I didn't think about the chance to survive". I can't accept that.

CFIT used to be common, but it's not anymore on commercial flights thanks to GPWS. There's not yet been an accident on an airplane equipped with EGPWS.

Emergency landing on water is "standard procedure". You might have the impression that it's something which was creatively invented by the pilot at a moments notice, but it's really not. The engineers designed the airplane with the capability, included advice and checklist in the operation manual, they even put a button labeled "ditch" on the dash! Can't really see why they could not make such an autopilot program.

You completely misunderstand. I wasn't talking about a generic "emergency landing on water".

A plane suddenly loses power in a highly urbanized area. It's very near multiple airports. Prior to 2009, I have no expectation that an autopilot would have been equipped to judge whether it should attempt to land in a heavily-trafficked river. (I further have no expectation that it would have been equipped to notice in the highly plausible scenario of one or more small boats being in the way.)

It's not a question of landing on water, it's a question of making the decision to do so and where.

Humans have the judgement gained by a lifetime of learning. Computers only have what it occurs to us to put into them while we're sitting safe and sound in our little offices pushing little buttons that don't threaten our lives.

I can't imagine anyone stepping onto an autonomous passenger aircraft for a long time, not until long after cars and smaller commercial aircraft are regularly driven using computers only. And even that seems like a stretch. I don't see how pilots use or non-use of some automated systems is going to influence popular opinion on that one way or another. But maybe there's a perception among pilots that it will.

The fact of the matter is that most commercial jetliners today are fully automated. It is only takeoff and landing that are still done by hand. No one ever hand-flies a jetliner during cruise except in dire emergencies. In fact, at high altitudes, where the plane is flying in the "coffin corner" of the flight envelope (http://en.wikipedia.org/wiki/Coffin_corner_(aviation)) hand flying can be extremely dangerous.

Getting in a Google self driving car is one thing, where you still have the ability to take over control or apply brakes yourself. But how many passengers would be willing to get on a fully self piloting aircraft? I think not too many, at least at first.

They are bound by regulations to do a certain number of landings every year.

I'm a widebody captain, maybe I can add some insight.

The handling of the aircraft is a small part of the job. Autoflight systems have been capable of flying and landing since the 1960s. It is a mature technology although it does need to be monitored by humans as equipment failures do occur. I am confident a system could be developed with sufficient fail safes to remove the requirement for monitoring, and a fully automatically flown aircraft could be built tomorrow although at a higher cost than the current monitored setup.

Even if an unmonitored autoflown aircraft was developed, judgement decisions about the operation of the aircraft still need to be taken, and these are best done by a human. Examples of decisions I make:

1. The best fuel consumption maybe achieved at a higher flight level, but I elect to stay low and burn into contingency because I've heard reports of turbulence at the higher level. I judge that I probably won't need the contingency fuel later based on previous experience of the route and destination. So I give my passengers a smooth ride, which helps repeat custom for the airline.

2. I experience radio crackle whilst in cloud and elect to change my routing as I suspect we are near a developing thunderstorm that may soon become active. I have to balance the extra fuel burn of a non-optimal route versus the risks of being caught in a thunderstorm which could destroy the aircraft.

I'm not an AI expert, but it seems to me these are value judgements that are best made by a human. So the fully autoflown aircraft will still need a human in the loop at some point.

It is possible to remotely site this human, and maybe make efficiency savings by a single human controlling many aircraft but this would require a bulletproof, near realtime datalink. Again, I'm not an expert in the field, but I would suggest this is, if not technically impossible, then a lot more expensive than the cost of employing human pilots. I imagine this datalink would also be of interest to people with ill intent- no fun when you get the airborne equivalent of a rickroll.

TL;DR: A human is required in the loop, not to handle the aircraft but to make decisions. It's cheaper to have them on-board than the cost of a datalink. It's cheaper to have them trained to take manual control than the cost of an autopilot sufficiently fail safe that it can be left unmonitored.

My father is a commercial airline pilot. He flew Boeing aircraft for 15 years and for the last 10 has been on a variety of Airbuses. He tells me that pilots very, very rarely use the auto land features and that he has never used it himself.

So the price of automation is eternal vigilance? Reminds me of the old argument that overuse of calculators means people have lost the ability to do mental arithmetic.

Here is graphical comparison of safe landing and Asiana crash: https://pbs.twimg.com/media/BOhIDCWCUAApHFV.jpg:large .

Also it appears that runway was modified: http://metabunk.org/sk/HL7742_777_Crash%2C_Korean_Asiana_Air...

Worth noting that with a modification like that, it looks like it actually gives extra space for landing, not less. So that piece probably isn't a factor at all.

Runway 28L at SFO is 10,800 feet long in its current configuration. The spec for the 777 is 5,800 feet of runway for landing[1]. There is obviously a fair amount of weight, wind dependency in determining the minimum required length, but the runway was almost 100% longer than was necessary.

[1] http://www.ana.co.jp/eng/aboutana/corporate/galleryclassi/19...

Interesting. I wonder what the Asiana's descent profile looks like compared to the average and some outliers of other safe 777 landings at SFO. Did you generate that image? If not, what was the source?

I saw another image online, which compared the descent profiles of the past week of landings for the same flight. It showed that the July 6th landing was very close to the normal descent profile, and the July 5th landing was actually the outlier -- it had an unusually shallow glide slope.

Unfortunately I can't find the image right now, but perhaps someone else knows where it is and can provide it?

like this link better than the OP which is fairly uninformed speculation ...

some video of the crash - kind of looks like it was skimming the waves in ground effect trying to go around


I stand by the parent having some useful data and analysis, and the OP being totally speculation, and in many respects uninformed.

- number 1, if you say 'loss of power', makes sense as a possible cause. if you say 'icing in the FOHE', it's like saying the murder was done with a yellow pistol. nothing points to that, no icing conditions (humidity/temperature), and of course not same engine as the previous plane crash attributed to that

- number 2, you can't be unaware of an ILS out of op, you wouldn't be cleared for the ILS approach, you wouldn't be using the ILS on visual approach, you would be cued there's no signal, you wouldn't hear the outer marker, etc. etc.

- number 3, auto-throttle, well again, if you say malfunction in power setting, OK could make sense, if you get as specific as 'confliction with autopilot/auto throttle' that's a purple pistol. Clearly not on cat3 autoland, you're basically saying pilot forgot to set throttle in the right configuration for visual approach and landing.

- number 4, pilot error/bad approach, can't argue with that. the story is what's interesting, eg undiagnosed walleye vision, ate the bad fish etc.

"- number 4, pilot error/bad approach, can't argue with that. the story is what's interesting"

Pilot induced oscillation on a really big scale. Coming in way too hot, slam down, whoops way over corrected, now coming in too low, whoops ran outta air and time to correct. Coming in way too hot, now are you better off trying to salvage or go around and get fired? Different nations airlines have differing policies on this...

Do you really risk getting fired for a go around? I mean, I can see that happening if you do it way more than other pilots, but just doing it once?

(For the record, I have no idea how these things works, it's just very surprising)

I think you might get a call from a chief pilot for an anomalous approach, whether you salvage it at last minute or go around. If there's a pattern of things not going by the book, would guess you get sent for 'retraining' before you get fired... airline version of big data FTW I guess...

There may be some parallels between this and the crash which happened over London in 1999, see http://en.wikipedia.org/wiki/Korean_Air_Cargo_Flight_8509

>no icing conditions

Icing in the fuel lines may be caused by fluid flowing through an orfice (pressure differential). All that is needed is water in the fuel.

water goes to the lowest point since it weighs more than fuel. then if it's not drained properly it gets sucked into the engine at the start of the flight. would be odd to discover water contamination after a long-haul flight.

I kind of assumed the FOHE was something that used external air, it wasn't water in the oil or the fuel. you descend through humid and coolish air ie not summer heat, water condenses and freezes as it's sucked through something from the pressure differential, or from actual icing conditions. but I could be wrong.

>water goes to the lowest point since it weighs more than fuel

While that's true, it takes time, and it won't get enough of the water out of the fuel[0]. Also, the aircraft must not be disturbed for the water to settle out. Operation demands for aircraft don't allow that. Even if they did, some water will remain in the fuel.

While the engines are operating, fuel is constantly circulating which disperses any water within the fuel. Airport fuel supplies at major airports are usually tightly controlled, but it's still be possible for water to get in.

>I kind of assumed the FOHE was something that used external air,

The FOHE is a Heat Exchanger, sort of like the radiator in your car, but instead of engine coolant/air the working fluids are Engine Oil/Jet Feul. Both the fuel and oil systems are closed systems, with one exception. As fuel is consumed in flight, the volume of that fuel must be displaced by outside air. Fuel tank to atmospheric pressure must never exceed some low differential pressure. This is one source of moisture ingress that must be dealt with, and can cause water to accumulate in tanks if left unchecked.

>you descend through humid and coolish air ie not summer heat,

Are you aware that the air temperature at altitude 35,000 ft is around -55C?

[0] http://www.petrolplaza.com/technology/articles/MiZlbiYxMDI3N...

edit: Here is a bunch of related photos if you're interested. http://englishrussia.com/2013/06/20/how-aircrafts-are-fueled...

From a reddit comment[1]:

"PPrune[2] also had a guy weighing in saying that when he flew for a Korean carrier he saw guys try to do banana-shaped approaches to get a smoother landing. Wouldn't be the first time the Koreans have pulled something like this."

[1] http://www.reddit.com/r/worldnews/comments/1hrewx/asiana_air...

[2] http://www.pprune.org

Poking around flighaware's website, I found data sources that could be used to make a better graph:




Nothing like URLs with obvious patterns!

I found picture on twitter, it is generated by FlightAware - live flight tracking (http://www.flightaware.com)

This is comparing one single safe landing to a crash landing. Does that gain us new insights? Not really. For that, we'd need the crash landing compared to the average landing, and also to other (safe and unsafe) extreme landings of the past.

And to get all Tufte of the graphics: Most of this is chart junk. The 3d view only serveres to hide information, and overlaying the paths (in 2d) would actually let us compare more accurately.

Average aircraft approach is a 3º glide slope. And usually all of them look exactly like that (only minor deviations due to the wind). If that chart is true is easy to see that they where high on the (non working) glidslope.

NY Daily News is reporting that VASI was down as well (http://www.nydailynews.com/news/national/expert-runway-guida...) Which appears to be due to someone incorrectly reading a post-crash NOTAM which is about the aids being gone due to the crash, so there's no particular reason to think VASI was down before the crash.

I really don't think I'd trust SF city government to run an airport, even under FAA regulations. Look at every other city service SF has, and imagine that quality applied to preventing hundred ton soda cans full of people and jet fuel from exploding.

How many of those other services are as highly federally regulated as flying an airplane, though? MUNI may suck, but MUNI doesn't need to answer to the feds.

Rail is federally regulated; it's part of why BART and Caltrain and HSR are so screwed up (passenger rail has to conform to some very very high crash standards to operate on the same network as freight rail, unlike in Europe and Asia -- so instead they built BART as a separate network and used essentially custom gauge, trainsets, etc. for everything.)

Admittedly the parts of Muni, BART, etc. which are the most screwed up are not the parts regulated much by the federal government, but at the very least SF is the organization picking the people to do things, and I'm wary of that. (although elected officials are also bad in SF; a Chief Law Enforcement Officer who beats his wife, etc.)

The short version is there are multiple things on a runway that emit light and I think they're being confused.

"Which appears to be due to someone incorrectly reading a post-crash NOTAM which is about the aids being gone due to the crash"

OK first of all VASI has been obsolete since I was a kid in the 90s, its all PAPI now. (edited to add, in the USA) Much like people still call the AWOS an ATIS because it does about the same thing. I'm not just picking nits, if you try researching this, you'll find the VASI has been outta service probably since the 90s, you want the PAPI. Its all the same anyway, white you're light, red you're dead. VASI is before my time but I'm told it was the same arrangement?

If you want NOTAMs you can just go to FAA's pilotweb, this link might work or you can search. Holy cow SFO has a lot of NOTAMS to read about.


!SFO 07/047 SFO RWY 10R/28L CLSD WEF 1307062309

!SFO 07/046 SFO RWY 28L PAPI OTS WEF 1307062219

The PAPI (precision approach whatever indicator or something) was marked out on the 6th at 2219 and the runways (not all listed above) formally closed at 2309. I think that is after the crash UTC time?

Now according to this NOTAM

!SFO 06/003 SFO RWY 28R ALS OTS WEF 1306011400-1308222359

They've been screwing around with the ALS lights for like 5 weeks now as per the daily news story. The ALS is mostly to light the place up at night, make sure you can line up on the correct runway (L or R) and most importantly in the USA this has the decision bar, if the weather is so poor that you can't see the decision bar, its too poor to continue the approach. I'm told the weather was beautiful during the crash, so I don't think the decision bar being rebuilt or whatever had much if anything to do with the crash.

Something I don't understand about ALS on an airport by an ocean (not exactly a problem where I live) is how they mount the decision bar and its little friends. Its going to be quite a distance from the end of the runway and some runways seem to go right up to the sea, so piers out in the ocean or something? If they hit the SFO ALS that would imply the plane would be well offshore under water.

The ALS, aside maybe from some wiring, is probably not any more or less messed up by the crash than anything else's power wiring. I could imagine a plane running off the runway into the PAPI and that would be about the end of that PAPI, or at least it'll have to be aligned.

It looks like they mount stuff on CA-92/San Mateo Bridge (http://www.flysfo.com/web/export/sites/default/download/abou...) and then from there along the approach. The main issue seems to be the two parallel runways. From when I took photos of the airfield, there were a lot of lights and other things sticking out of the bay on the approach path. Decision Bar is supposed to be 1000' off the threshold so that was probably it.

I'm not a pilot; once I have spare money and time (and enough to fly monthly to keep current), probably.

It's interesting reading about stuff like the "No Transgression Zone", though.

"once I have spare money and time"

I've been telling myself that for decades and never quite had both at the same time. Often a great excess of one or the other. By the time I finally have both, I'll probably fail my medical with my luck. I did do ground school + a bit more on my own, and I've got a couple hours in the air with an instructor. I do some semi-serious sim flying for fun, mostly the X-Plane. If I ever get serious I'll already be pretty good at navigation, flight planning, reading approach plates, E6B use, METAR decoding, NOTAM decoding, etc.

There's a remarkable number of free electronic E6B apps for most phones. And some paid ones that are marginally better. I find my tablet to be quite helpful when "flying" although I also use old fashioned mechanical E6B... Figure $10 for a paper one that'll last "awhile" and $30 for a lifetime metal one.

The frontier of flight sims right now is no one does NOTAM simulation. Why not shut down some ground stuff occasionally, just like you can shut down aircraft systems and screw around with weather? It may be there's a sim out there that does this that I don't know about.

I'm think in the Bay Area it's mostly a question of $10-15k and maybe 100 hours (flight + study time) to get the first license, and there are a decent number of rental choices.

My dream (well, the realistic one) is to get a CH-801 STOL kit plane with a diesel engine so I can live in Central Washington and get to I-5 in a reasonable amount of time, although weather probably prevents that often. That's only $150k or so, which would be saved in taxes and property cost differential several times over.

> I could imagine a plane running off the runway into the PAPI and that would be about the end of that PAPI, or at least it'll have to be aligned.

As best I can tell from the available photos, the reason the 28L PAPI is out is because the plane took out the three rightmost ones when it slid/bounced across the blast pad area.

I bet those other services aren't subject to frequent inspections by the FAA.

Sorry to go offtopic here, but the images seem to be used without any credits. The first one was taken by David Eun (https://path.com/p/1lwrZb)

I'm actually working right now to try to put some kind of overlay over the images for credit situations. My blog is still a WIP. Sorry!

You say that as if giving credit is enough. If you don't have a license/permission to use the image you're violating copyright and the correct response is to take it down.

Several of them might fall within the bounds of fair-use, such as the one about halfway down showing the debris trail, which is included in the context of a discussion of the debris trail. The large image at the top isn't used in a way that would make for a good fair-use argument, though.

They all fall under fair use since they were uploaded to social media (so the photographer is no longer the only rights holder and most social media sites assign rights to republish content).

That said, it is only fair and nice to credit all sources. The web relies on links and being able to source content back to origins.

Not just credit, but also permission.....

Well, the whole blog design is very gray-area 'theft' of Medium's design.

I don't know what Medium is, but ... a white page with a title on the top? This is just as much a 'theft' of the default Apache "It works!" page.

Design is about choices. In this case you can point out:

    1. top-left square logo, dark gray, slab serif, fixed on scroll
    3. bold, large sans-serif title in "off-black"
    4. large serif body text with extended line-height
    5. 100% width image on top
Despite the visual simplicity, there are a lot of variables in effect. It's definitely "inspired" by Medium.

(medium has a few different post layouts, but they share the same spirit: https://medium.com/lift-and-drag/5f803f1482e3 vs https://medium.com/lift-and-drag/51691e99279f)

Oh, I was actually thinking I was reading a post on Medium.

I think that the lack of author information in the top left is sufficiently different to not even call it gray area. And the lack of paragraph-level comments.

The logo and typography seem very similar, even to my untrained eye, though.

There is an excellent posting with graphs comparing the Asiana flight to a United 777 that landed 10 minutes before. The graphs compare height, speed and importantly energy. http://flyingprofessors.net/what-happened-to-asiana-airlines...

As most have hypothesized, it was an unstabilized approach and they should have done a go around (barring unknown technical/mechanical factors).

Looking at the graphs the plane was ahead of them by about 5-10 seconds. That usually doesn't turn out too well. Gotta be ahead of the plane, not the plane ahead of you. Usually a noob mistake (on the ground system failure, hard to fix) OR hopefully extensive extenuating circumstances (on the plane system failure, easy to fix).

I heard a reporter on TV reporting that the airplane was doing cartwheels on the runway, and I pictured the semi-truck being flipped over in Batman. And then I thought what the hell could have done that? Fishtailed makes more sense, thanks for the clarification.

There were a lot of reports like these that were complete assumption based on the fact that one witness said the airplane had flipped over. It actually hadn't.

That is not a cartwheel

Why, because the plane was not 100% vertical on the Z-axis?


Just like the comment that Michael Hastings' car crash 'sounded like a bomb' resulted in all sorts of assumptions that the car exploded.

I don't think airframes have quite the strength to flip end over end from contact with the ground, even at landing speeds. I can grab this pen rom my desk and throw it on the carpet so it flips vertically end over end. A 777 airframe might do something halfway between that and crumbling as it bounces.

CNN just released video footage they have of the actual crash.

The plane (from what I see) after crashing does a near 360-degree turn with the right wing high up – in fact you can see that the plane is almost nose down-tail up which almost makes it look like it was cartwheeling.

Very surprising revelation, I really did not expect an airframe to remain as intact as it did after going through that.

I looked at the video again and again, and ironically, if the jet had done a 360 on the ground it would have broken into pieces, but this jet did a airborne near-360 after the wingtip dug in and the jet became airborne again for an instant.

With cameras everywhere on the streets, I am baffled that there aren't cameras on the runways. Many, many airplane accidents on runways would be a lot easier to investigate, including, for example, the one where the SST caught fire.

There are runway cameras.

It looks like this was an unstabilized approach right down to the ground. It would be interesting to see what the company SOP is. Any safe airline has a no-fault go-around policy for unstabilized approaches, and you'll actually get in trouble if you continue with an unstabilized approach and you don't go around. From what pilots are saying on the message boards it seems that a lot of Asian airlines have pretty much the opposite policy - you get in trouble if you go around for any reason, and they would prefer you to salvage an unstabilized approach to save money.

pretty much looks the same to me - the guy stalled and when he finally, at the last second (actually - well past the last second), decided to go around, he did jerk up the nose and that increased the stall, so they fell. Reminds me about the AirFrance fall in the Atlantic some years ago - that guy was trying to gain altitude by jerking the nose up (i.e. increasing the "angle of attack") instead of "pressing the accelerator pedal".

If this is what really happened, I'll say the same thing I said then - it shocks me that airline pilots with thousands of flight hours can make the most elementary piloting mistake possible.

But I've read accident reports before, and it's rarely a good idea to speculate. I won't have an opinion about this until the official report is published.

Pilots are human, and they make mistakes. Good training, having two pilots on the flightdeck and having good SOPs should prevent anything like this ever happening. It looks like training and/or SOPs were severely lacking here.

As well as the "don't waste our company money doing a go-around" fuckwitted notion, people are saying that Asian airlines don't like doing visual approaches period. Therefore whenever the pilot is required to do a visual approach, he/she will be severely lacking in the skills to do it.

well, until report is out, we're here to speculate (or to read a good speculation if we're lucky :). From the Flightaware data the plane had 109kts at 100feet altitude before the seawall. The stall speed of 777 is about 110kts (its Vat is ~140, and it is 1.3 of the stall). And the speed was falling as the next recording is 85kts. While we can only guess about the original cause - pilot error or hardware, etc... - the stall seems to be there.

Are you sure that the Flightaware data is accurate enough? When flying a glider, I have attempted to make similar judegments from GPS log data when doing an outlanding, and I'd be similarly careful when using radar data from a website.

I don't want to be very negative here, but my point is I've made wrong calls about this when speculating in the past. So currently I'm opposed to doing it for reasons other than entertainment.

> for reasons other than entertainment.

well, i'd be surprised if there are people here for any other reasons :)

the flightaware data correlates very well with what is on video. Personally i spent a bunch of time on Coyote Pt. so i can mentally correlate the video with how it looks for other planes/landings at SFO. As we aren't ordained NTSB officers, we're free to make wrong calls :)

This is an excellent article. Like Air France 447, seems very plausible that pilot error did play some role. This reinforces my respect for the pilot of US Airways Flight 1549, Captain Sullenberger.

Curious as to why autopilot isn't being used to land planes yet. My understanding is that pilots can choose to use it to land, but only do so when visibility is very low. (ex: http://www.youtube.com/watch?v=DOFs-oa-bbc)

The autopilot requires the ILS to land the plane. If the ILS is inoperative, it can't autoland. However pilots do still land planes to keep their skills honed.

Presumably GPS and/or visual recognition system can be used to achieve the same, after all airstrip location is know in advance.

Presumably, the visual recognition system in this case was a human.

They mention that the ILS was off for repairs. Isn't that a major negative for doing autopilot landing (if indeed it's still allowed/possible without the external location based information).

Correct. You cannot Autoland without the ILS (both glide slope + localizer are required)

It's being used when needed. When not needed, pilots use the opportunity to practice landings. It's important to not get over-dependent on automation.

Here is my one minute as an armchair air crash investigator.

In recent times I've read some very strange incidents reports with regard to airline pilots.

One plane crashed on landing while coming in to final approach at twice the recommended speed.

Another was coming in for final approach some 500 feet below the allowable height and off the designated flight path.

These are examples of some very basic flying errors.

In years previous pilots had to do many hours in light aircraft, then move on to bigger aircraft and only after decades of flying experience did they make it onto the big jets.

But in todays cut price flying environment I suspect most pilots don’t do that much flying and as such they don’t know as much about flying as they should.

This one is going to be pilot error.

I don't really like when people post these sort of "arm-chair" hypotheses. As much as it's fun to try to figure things out, there is always the danger of people jumping to conclusions based on bad or incomplete data. It reminds me a lot of the reddit people immediately following the Boston Marathon bombings. It's definitely an interesting read, but I don't want to take it for much more than just idle speculation.

Idle speculation and having a bit of fun trying to play air crash investigator, is exactly what this article is. I hope that I was able to convey that this is about "what I think" happened and not "what happened".

Cellphone video shot by a guy, Fred Hayes, watching aircraft land yesterday at SFO shows the tail dragging in the water for quite a distance before striking the seawall.


I'm guessing that the pilot was praying for those engines to spool-up faster.

Some more relevant news just breaking now ... http://www.nydailynews.com/news/national/asiana-flight-214-p...

Does anyone know if they've released the information about where the people who died/were severely injured were sitting in the plane? I'd love to see a seating chart that shows the safest and least safe seats on a 777.

It really depends on the type of accident (skim Wikipedia on ones with survivors) ... and after looking at the CNN supplied video https://news.ycombinator.com/item?id=6003232 right now I'm thinking "Major miracle" and "If it's ain't Boeing, I ain't going", although I'm sure it's a lot more of the former than the latter.

"The final bit of information is interesting. It shows a positive rate of climb – 120 feet per minute to be exact, which for a jet airliner, is very slow. For reference, at takeoff the vertical speed is most often +1500 feet per minute. Another interesting bit of information is the aircraft’s airspeed; 85 kts – for a Boeing 777 that is (and I’m guessing here) probably way below stall speed."

That is actually expected. It is called the flare just before touchdown:


In a landing flare, you want to raise your negative V/S to near zero. A positive vertical speed means you're no longer descending, but ascending. Having a positive vertical speed during the landing flair is something you don't want to have as you're basically floating (and possibly climbing) above the runway and bleeding your airspeed which, in a couple seconds will most likely end in a hard landing as you've stalled while floating over the runway.

The positive rate of climb to me, indicates that they attempted to climb (go-around) but their low airspeed resulted in a stall.

It's important to remember that this data was pulled off Flightware which is basically aggregated data. and not from an accurate source such as the FDR. So my guess could be totally wrong.

It's been said, but I want to add weight to it: this is not at all expected. You never climb during a normal landing.

I'd wager that 120fpm climb reading to be due to a measurement error of some kind, not actually something that really happened.

Looking at the GPS coordinates, the second to last reading was ~200m before the seawall, while the last reading was around halfway down the runway.

The plane is far outside it's normal operating conditions by this point, it's possible the plane instruments actually experienced a brief moment of 'climb' while the plane was crashing, or it could be equipment malfunction.

However, from what I can tell, the plane never reached that far down the runway. And the gps coordinates have actually snapped to the middle of runway 19L, near where it intercepts 28L, suggesting that flight radar 24 are doing some post-processing to make the planes stick to runways better.

I wouldn't trust that last data-point at all, as it's definitely post-seawall.

Looking at the data, it's obvious that the vertical velocity figures are not anywhere near as accurate as "120" would indicate.

The data points are in the neighborhood of 15 seconds apart. There are 4 points labeled 2:25PM, 3 points labeled 2:26PM, and 5 points labeled 2:27PM.

The altitude data is all rounded to the nearest 100ft. Obviously this is coming from transponder data. The mode C transponder carried by an airliner reports pressure altitude back to the interrogating radar at 100ft intervals.

Vertical velocity is all multiples of 60fpm. I have no idea where that would be coming from. As far as I know, vertical velocity is not transmitted by either radar transponders nor the newer ADS-B position broadcasting equipment. Simply taking the derivative of reported altitude wouldn't produce these results, because with 100ft altitude intervals and 15s reporting intervals, that gives you a vertical velocity resolution of only 400fpm.

Let's skip that question for the moment. The raw altitude data does show an increase from 100ft to 200ft at the end. But what conclusion can we draw from this? Essentially nothing. Because the altitude is so quantized, this could in theory indicate anything between a 0ft and 200ft increase in altitude, non-inclusive. Of course, the instruments are not that precise, so in practice the range is even larger. This is especially true given that the instrument was in the middle of a fairly violent plane crash when it gave that last reading. From the video of the crash, it's clear the plane never got very high above the runway. Since the runway is only 13ft above sea level, and the plane didn't crash into the ocean, the amount of climb possible is very much limited. The transponder may well have increased in altitude a bit for this last reading, but if so, the alignment with the reported altitude increase is pretty much just coincidental.

Now back to the vertical velocity. Where did it come from? I can only speculate that they're applying some kind of curve-fitting or smoothing to transform the coarse altitude data into smoother vertical velocity figures. These numbers are therefore even less reliable than the altitude figures, and can probably just be ignored.

Just to briefly support that last bit, there are some obvious numerical anomalies in the vertical velocity figures earlier in the flight. At 9:38AM, the data source briefly switches from "FlightAware Transoceanic" to "Oakland Oceanic" and then back again. The two Oakland data points show an altitude of 37,000ft, while the FlightAware data shows 34,000ft. The vertical velocity for these four points is shown as 2280fpm, 9960fpm, -4320fpm, and -1800fpm. These changes obviously didn't actually happen (otherwise we'd have heard about people being flung about the cabin in mid-flight!) and are just some sort of reporting anomaly. The ridiculous vertical velocity figures (nearly 10,000fpm climb accompanied by a minor increase in horizontal speed!) are obviously just some sort of post-processing going crazy trying to fit the bad altitude data.

I've never flown a B777 or any heavy aircraft so I can't comment on their typical flares, but in a Cessna 172 the flare doesn't induce a positive rate of climb, it simply slows the vertical descent speed to something manageable for touch down, a positive vertical speed seems counterintuitive to me as a good flare. What I want to know is what his IAS (indicated air-speed) at the decision altitude.

Again the only thing I've flown are parachutes and some hours in a C172 and a C182.

It gives a pretty good estimate of the error involved in the descent rate reporting. Probably accurate +/- 200 FPM or so.

The airspeed encoder is probably from the same system that displays the pilot instruments, indicated airspeed IAS. I thought it odd that the analysis of conditions didn't think 125 and 98 were ridiculously low airspeeds, but claimed that an approach around 130 knots was a "fast approach". However what little I know about the 777 from the flightsim and real pilot community is that a typical approach is around 150-something slowing to a bit above 130 at flare. I've read some claims from actual heavy pilots that under really light conditions a Vref below 120 is theoretically possible.

None of us know the weight and balance calcs so its impossible to determine if the pilots were doing the correct thing, although it is possible to determine they were flying a flight path implying they thought they were very light indeed. Now why? Who knows. Maybe because they were right, they were light. Maybe because they were low-ish on fuel. Maybe they didn't trust their IAS instrument. Maybe their IAS was actually broken and comparing their visual approach with instruments, well, its just a really strong headwind today. Maybe they made a math error. Somebody overcontrolled?

He could have found this info by violating the copyright of numerous google-able flight sim forums, and forums with real genuine 777 pilots talking about their experience with the 777. Instead he violates the copyright of a picture which we've all already seen, oh well.

About the speed, you're correct – 130 kts for a 777 is not fast, I was mistaken. In fact the NTSB just said the target approach given by the pilots was to be 137 kts. I should have chose one of the higher airspeeds given but I did not want to give a high estimate. I am not familiar with 777 speeds and should of done some more research in that regard. I did however, write that I found the sub-100 speeds to be extremely slow, even suggesting that it's most likely below stall speed – which seems to be correct as the NTSB says the stick shaker went off some 4 seconds before impact.

I read sometimes you can get a Vref under 120, great density altitude, sea level, winter, no pax cargo and low on fuel, but yeah when the IAS is reporting 98 you're pretty much screwed under any conditions. Maybe it was already on the ground at that point, if it wasn't.. it going to be on the ground real soon...

I would imagine some of the flightsim people I know are flying the approach under identical conditions right now. Would be interesting to hear at what point they found it unrecoverable. 98 kts is unrecoverable, but I wonder when it went unrecoverable. One problem with sims is if you know how its supposed to turn out, that screws up your actual reaction if you sim it 20 times trying different things. All you need is a cockpit distraction to totally screw up an approach. Maybe a false alarm of some type at just the wrong moment.

I'm not sure we can believe the last data point - it's possible that's a real ascent, or it's possible it's caused by failing hardware as the plane crashes.

To the OP just curious -- are you a licensed pilot? Not to question you since you did say this was just speculation but just wondering. It's pretty well informed.

Is anyone else puking mad at those people carrying their luggage in the photo? I feel like getting off a burning plane should take priority over duty free purchases.

Thank you for your armchair analysis, but I will wait for the NTSB's report.

That seems needlessly snarky in response to some thoughts that are very clearly labeled as speculation.

It was snarky, and I expected the down votes. But, I think it needs to be said. Neither I, nor the OP, have enough information to make any conclusions. It is just a lot of observations, rumors, and opinions.

I love speculating about everything from government policy to interpersonal relationships. (It helps me validate my mental model of the world.) But, in this case, I know the NTSB currently has a better picture of the situation than anyone on the internet.

Why say it when it's already mentioned several times in the post? I'll wait for the report, in the meantime I'm interested in qualified speculation based on the available data.

indeed, which will take ~2 years

They'll make a preliminary, primarily "factual" report within days, or weeks. This will be based on the FDR, Quick Access Recorder, statements by crew, passengers and witnesses to the NTSB investigators.

Even though I'm a former pilot, I've not made any posts prior to this one, because I didn't want to speculate.

The recently released video (which I'm sure the NTSB already had) should be helpful as well (and appears to support the author's speculation).

Applications are open for YC Summer 2018

Guidelines | FAQ | Support | API | Security | Lists | Bookmarklet | Legal | Apply to YC | Contact