Wednesday, March 23, 2011

Lies, damned lies, and BoingBoing; or why 42.03% is a dangerous number

A widely-read and widely-quoted post on BoingBoing said:

“Note 1: normal lifetime incidence in U.S. is about 42% (of developing any cancer). The increase in risk, i.e. 1 in 30 for 3,000 mSv (300 rem), changes the 42% to 42.03% chance of getting cancer in the exposed individual’s lifetime.”

1 in 30 = 3.3%, not 0.03%. Saying 42.03% understates the increased risk by two orders of magnitude. But BoingBoing trumpets the author’s credentials, highlights the sound bite which falsely trivializes the cancer increase, and ignores the numerous comments pointing out the error.

If you want to reassure me, you’re not going to do it by lying about the math.

Here’s another way of looking at it. Suppose you’re the Japanese government, trying to decide whether to expend the resources to evacuate 300,000 people from a hot zone where they’d receive a 3000 mSv cumulative dose over the next year. It might be nice to know whether the consequence of inaction is 10,000 additional people getting cancer or 90 additional people getting cancer. Math matters.

Tuesday, March 22, 2011

Last night at the PDF Hospital

Lisa checked a 6-page file into the PDF Hospital last night on behalf of a colleague. The complaint was a page that rendered very slowly and was impossible to modify.

Triage showed that the file was only about 2 megs, and the page in question only appeared to have a few images. Exploratory surgery confirmed that the problem was with the image components of the page by progressively removing the non-image components and still seeing the same symptoms.

This left 4 images to explore in depth. Each image was composed of a single background image, and 1024-16,384 copies of the foreground image mask and associated clipping path. The background image was larger than the foreground image mask, so it was easy to select and move out of the way. When we moved a copy of the foreground image mask, it disappeared because it was now outside the associated clipping path. But with the background image and one copy of the foreground image mask out of the way, we could select and delete the remaining image masks and clipping paths in the original image location. Once all of the clipping paths were deleted, the foreground image mask appeared again, and we verified through on-screen inspection and test prints that it was complete. We could then move the remaining background image and foreground image mask back into their original location without disturbing any of the item layering on the page at all.

Repeating this process on all 4 images resulted in a symptom-free page. We released the file back to its owner, and were left with the mystery of how you end up with thousands of identical copies of an image in a file. It’s almost certainly not the result of someone pressing a button 16,000 times.

The key is probably that the number of copies were powers of 2. If you are copying outside images and placing them into a file, the expected process is that when you select a new image to copy, that replaces the previously copied image in whatever bit of memory is being used as the temporary clipboard. Select A, place A, select B, place B. If, however, all previously placed images remain selected in memory along with the new image selected, then you’ll have exactly this sort of exponential result. If the brackets show the items previously placed, then you end up placing A, [A]B, [AAB]C, [AAAABBC]D, etc. You’ll have 16,384 copies of A happen after only placing 14 items. Fortunately, the damage was treatable in this instance. But it would be nice to know what software workflow is resulting in exponential pastes.

I’ve been thinking about opening the PDF Hospital for a while. We operate on PDF files all the time, and we’ve seen a lot of strange problems. Does your PDF file load or print really slowly, or give you PostScript errors? Do searches on the text not work? Do some pieces of text appear as different characters when printed than when they’re on screen? Do accents disappear, or capital W’s print as barcodes? Do some PDF viewers show little boxes in your PDF file? Do you want to move items around on the page, or make something larger or smaller? Do you want to remove or add items to the page? We deal with all of this frequently, so opening the PDF Hospital to outside business is the obvious next step. Not everything is treatable, but many problems are. The challenge is setting a price structure and managing expectations.

Monday, March 21, 2011

Help, I’m trapped in an elevator

No, seriously, we were trapped in an elevator on Saturday evening. Thank you to the Watertown fire department for getting us out. Apparently Schindler Elevator Corporation has repeatedly failed to fix the elevator, so people keep getting stuck in it.

It turns out that Massachusetts has no process for tracking elevator entrapments. The Department of Public Safety handles inspections, but they don’t take reports of entrapments. Neither does anyone else in the state. This means there’s no way to find out if Schindler Elevator Corporation is a terribly incompetent, dishonest, or dangerous company by tracking how many of the elevators they maintain fail repeatedly and compare that failure frequency to the failure frequency from other companies. It’s impossible to know whether this experience is an unusual occurrence linked to a single elevator that is supposedly maintained by Schindler Elevator Corporation, or a single incompetent repair technician working for Schindler Elevator Corporation, or a company-wide pattern of gross negligence. There’s also no way for the state to track whether elevator entrapments are linked to other sorts of elevator accidents, and no way for the state to decide whether elevator inspections should be about more than whether the alarm button makes a noise.

I like data. More than that, I like the idea of data, because I like thinking that decisions informed by data are, on average, better decisions. On the other hand, the building’s owner knows that they’re paying $4500 a year to Schindler Elevator Corporation, and they know that the elevator is failing constantly, and they know that Schindler keeps trying and failing to fix the elevator, and they know that people keep getting trapped in the elevator, and all of that is data. And the building owner’s decision is to keep paying Schindler, so clearly not all decisions informed by data are better decisions.

The plural of anecdote is not data. But the plural of silence is not data either. And since the data is unavailable, I’ll go with anecdote. Anecdotally, I don’t like Schindler Elevator Corporation right now. Because I don’t like being trapped in an elevator.

And no, I’m not retitling the post “Schindler’s Lift.”

Wednesday, March 16, 2011

Working my way up



For flooring, wide oak boards cut in curves and wandering paths to minimize wasted lumber by Bolefloor.

To put on that floor, an Art Nouveau rug from Sawbridge Studios, and perhaps one of their cherry and birdseye maple tables.

On second thought, maybe Lisa is right that the flooring should be on the wall, so you can see it properly.

And maybe that proverbial guidance counselor was right that I should go into a high-paying field, so I can actually afford this stuff.

Tuesday, March 15, 2011

Understanding the Fukushima news coverage

News coverage about the problems at the Fukushima nuclear reactors has been somewhat confused.

Radiation and radioactive are different terms. Radioactive substances emit radiation. A release of radiation is dangerous to those nearby. A release of radioactive substances is a wider problem, because those substances can then travel and emit radiation wherever they end up. For those of us who aren't near the Fukushima reactors, the key question is the volume and type of radioactive substances that are released into the environment, not the radiation levels near the plant.

A fuel rod within the reactor being exposed to the air does not mean that the fuel rod is exposed to the environment. There is air inside the containment structure. The concern with fuel rods at the Fukushima reactors being exposed to the air is that they then overheat, damage the cladding around the fuel rods, and make a core meltdown more likely.

A core meltdown is not, by itself, a widespread disaster. A meltdown that happens within an intact containment structure, as at Three Mile Island, doesn't mean that radioactive substances will escape. The key question with meltdowns at the Fukushima reactors is whether the containment structures stay intact.

Spent fuel rods are not safe. And the spent fuel pools at the Fukushima reactors are not inside containment structures the way that the reactors are (or used to be).

Using sea water for cooling is not just as good as using the purified water that is usually used for cooling. Purified water that becomes activated (turned radioactive) has a very short half-life. Sea water contains a lot of other substances that will stay radioactive a lot longer. Sea water can also damage pumps that are designed for fresh water, and the salt left behind may physically block continued access for the water needed to cool the fuel rods.

When radioactive substances escape into the outside air, it doesn't make a lot of difference whether they escape in a hydrogen explosion, in a steam explosion, in a fire, or in a steam release. What matters is the volume and type of radioactive substances that are then in the outside air, because they will gradually fall down to the ocean or the ground. The other key question that is unknown is how far the fallout will travel.

Wednesday, March 9, 2011

Choices

On its way to becoming a poster...

Monday, March 7, 2011