# News this Week

Science  11 Mar 2016:
Vol. 351, Issue 6278, pp. 1120
1. SCI COMMUN

# News at a glance

### Washington, D.C.

#### EPA nixes pesticide's approval

The Environmental Protection Agency (EPA) has announced its intent to cancel its conditional approval of flubendiamide, an insecticide used on more than 200 U.S. crops under several brand names, including Synapse. EPA granted Bayer CropScience conditional approval to sell the pesticide in 2008, pending further studies. It now says that the product breaks down into a more harmful chemical that can persist in the deep sections of freshwater streams and lakes, threatening benthic creatures. The company says EPA is relying on “overly conservative and unrealistic theoretical modeling” without real-world proof of harm, and has challenged the cancellation. While the two sides wait for an administrative hearing Bayer can continue to sell the chemical. Environmental groups and industry are closely watching the case, because flubendiamide is one of thousands of pesticides that have gained a green light while studies are pending, but the first for which EPA has pulled its approval.

### Alexandria, Virginia

#### Proper use of the p-value

In science's ongoing reproducibility crisis, misuse of the “p-value”—with 0.05 often considered a magical threshold for “statistical significance”—is particularly reviled. The American Statistical Association weighed in this week by releasing “six principles” on how to interpret p-values, marking the first time the organization has taken an official position on the matter. The principles were derived from more than a year of discussion by a consensus committee made up of 26 statistical experts. Among the new pro-tips: A p-value smaller than 0.05 doesn't necessarily mean that your hypothesis is true. And it doesn't say whether the results are due to chance alone, as is commonly believed. Also, the size of a p-value doesn't tell you anything about the significance of a result; even if the hypothesized effect is real, the effect could be so small it's barely worth publishing.

## Newsmakers

### Three Qs

Scott Halstead, 86, is one of the world's foremost authorities on mosquito-borne viruses, including dengue and chikungunya. Science spoke with Halstead, an investigator at the Uniformed Services University of the Health Sciences in Bethesda, Maryland, about the likely fate of the Zika virus. http://scim.ag/HalsteadZika

Q:Zika, chikungunya, and yellow fever seem to disappear for years and then return. Why?

A:The nature of these zoonotic diseases is that they involve primates. If you dug into it you'd probably find it has something to do with the weather and the fruiting of trees and the monkey populations. Herd immunity in humans must be very important, too, and population size. The herd immunity [threshold] for dengue, chikungunya, and Zika is about 80%; when 80% of the population is immune, transmission is being blocked four out of five times. That one out of five times is not enough to keep the disease going.

Q:How long is chikungunya going to stay around? Zika?

A:Five years, max. The only model I have is India. I watched when chikungunya went from Africa to India in 1963 and it disappeared from India in about 5 years. Zika's the same.

Q:By the time there's a Zika vaccine, you're suggesting Brazil's population may be largely immune.

A:Based on the observations that I've made and that anybody can see with their own eyes if they look at the data, this virus is just going to burn itself out. There's only a need for a vaccine in the acute emergency.

## Findings

### Mosquitoes overwinter in D.C.

The mosquito Aedes aegypti is a major vector for transmitting certain diseases, including dengue, chikungunya, and Zika. It has been in the United States for some 375 years, but its presence has been thought to be seasonal in states such as Maryland and Virginia, given its low tolerance for cold. However, a recent study, which examined the genotypes of 70 larval and adult A. aegypti mosquitoes collected each summer in Washington, D.C., from 2011 through 2014, suggests that a small population of A. aegypti mosquitoes are year-round residents of the city. The mosquitoes likely survived the winters by inhabiting humanmade underground spaces, the team reports in The American Journal of Tropical Medicine and Hygiene.

2. # Evidence on trial

1. Martin Enserink

Forensic science is reforming in the wake of a landmark report.

On 27 February, a court ordered the District of Columbia to pay $13.2 million to Santae Tribble, who spent 28 years in prison based on bogus science. After a taxi driver was murdered in Southeast Washington in 1978, a witness had seen the killer wearing a stocking mask. In a stocking found a block away, police found a hair that matched Tribble's “in all microscopic characteristics,” an analyst for the Federal Bureau of Investigation testified. Chances that it came from someone else were “one in 10 million,” a prosecutor told the jury. Tribble was convicted. But a DNA analysis 31 years later showed that the 13 hairs in the stocking came from three different people, none of them Tribble, and from a dog. His incarceration wrecked Tribble's life: The judge in this year's decision found that it contributed to severe depression, heroin addiction, and HIV and hepatitis infections, according to The Washington Post. His story is just one of many. Forensic hair analysts have systematically overstated their evidence for decades, the Department of Justice has found, landing hundreds of innocent people in jail and some on death row. Hair analysis is only one of many flawed forensic fields: A 2009 report from the National Research Council found that the analysis of many types of evidence—from footprints and tire tracks to bullet marks and blood splatters—lacks a solid foundation. Even DNA evidence, seen as the gold standard, can land innocent people in jail, now that new technologies can detect minuscule amounts of genetic material. Forensic analysts are trying to do better. Many fields are testing the accuracy of existing methods and developing new ones that are more science-based. Statisticians have embarked on an ambitious effort to express the strength of so-called pattern evidence, such as fingerprints, in a more scientific way. Meanwhile, some scientists are developing the forensic tools of tomorrow. Microbiologists are examining the possibility that the mix of bacteria living in and on the human body is so personal that it could help identify individuals. Computer scientists are helping to unmask criminals who use cryptocurrency, such as Bitcoin. Even hair has a forensic future: New analytical techniques stop short of identifying people but may provide reliable clues about a person's origins, history, or lifestyle. Given the history of forensics, new techniques will need to be validated more thoroughly than past methods were. And whether the methods are new or familiar, analysts, lawyers, and judges will all need to adopt a more scientific way of thinking. Bad forensic science has already wrecked too many lives. 3. # Sizing up the evidence 1. Kelly Servick Statisticians are on a mission to reverse a legacy of junk science in the courtroom. On a September afternoon in 2000, a man named Richard Green was shot and wounded in his neighborhood south of Boston. About a year later, police found a loaded pistol in the yard of a nearby house. A detective with the Boston Police Department fired the gun multiple times in a lab and compared the minute grooves and scratches that the firing pin and the interior of the gun left on its cartridge casings with those discovered on casings found at the crime scene. They matched, he would later say at a pretrial hearing, “to the exclusion of every other firearm in the world.” The detective's finding might have bolstered federal racketeering charges for two alleged gang members implicated in various crimes on that street. But the defendants' lawyers challenged its admissibility. The patterns on the cartridges from the lab weren't identical to those from the crime scene, they pointed out. So how could the detective be sure that the shots hadn't been fired from another gun? The short answer, if you ask any statistician, is that he couldn't. There was some unknown chance that a different gun struck a similar pattern. But for decades, forensic examiners have sometimes claimed in court that close but not identical ballistic markings could conclusively link evidence to a suspect—and judges and juries have trusted their expertise. Examiners have made similar statements for other forms of so-called pattern evidence, such as fingerprints, shoeprints, tire tracks, and bite marks. But such claims are ill-founded, a committee at the National Academy of Sciences (NAS) concluded in 2009. “No forensic method has been rigorously shown to have the capacity to consistently, and with a high degree of certainty, demonstrate a connection between evidence and a specific individual or source,” the panel wrote. In other words: Judges and juries were sometimes sending people to jail based on bogus science. The committee's report sent shockwaves through the legal system, and forensic science is now grinding toward reform. A series of expert working groups, assembled by the National Institute of Standards and Technology (NIST) and the Department of Justice, has begun to gather and endorse standards for collecting and evaluating different kinds of evidence. What is needed, says Constantine Gatsonis, a statistician at Brown University, who chaired the NAS committee, is statistical rigor. “When somebody tells you, ‘I think this is a match or not a match,’ they ought to tell you an estimate of the statistical uncertainty about it,” he says. Last May, NIST awarded$20 million to a team of about 30 statisticians and legal professionals to help develop tools for analyzing the strength of an apparent match. Called the Center for Statistics and Applications in Forensic Evidence (CSAFE), it will collaborate with NIST statisticians to develop statistical methods that describe how strongly a shoeprint in the dirt links the owner of a certain pair of sneakers to a crime scene, for example, or how many fingerprints other than the suspect's might have left a similar pattern on a murder weapon.

The group is staring down a problem of immense complexity. Pattern evidence has historically relied on the trained eyes and subjective judgments of human examiners, not on rigorous statistical analysis. It's not known how much variation exists in the world's population of shoes, guns, or fingerprints, or just how much similarity between two patterns is enough to suggest a common source. “I know some people think we are not going to be able to do this, [that] you cannot put a probability on some types of evidence,” says Alicia Carriquiry, a statistician at Iowa State University in Ames who heads CSAFE. “And they may be right, but we need to try.”

MANY FORENSIC DISCIPLINES have been plagued with high-profile errors. An ongoing review of the Federal Bureau of Investigation's (FBI's) microscopic hair comparisons, in which forensic scientists look for distinguishing features such as the thickness, texture, and pigment in a hair strand, has revealed erroneous statements in more than 90% of cases before 2000 in which FBI examiners gave testimony. Often, analysts said that hair could be associated with a specific person—which hair analysis cannot prove. At least five of the cases reviewed so far ended in convictions later reversed with DNA evidence.

The analysis of bite mark patterns has been shown to be so weak scientifically that a state commission in Texas recently recommended banning it from the courtroom. In one high-profile case, a man named Ray Krone was convicted of murder after prosecutors used bite marks on the victim's neck and breast to link Krone to the crime; he served 10 years in prison before DNA evidence showed that he was innocent.

Even more-established methods, such as fingerprint comparison, have faced criticism. Many fingerprint analysts use standard procedures to mark different levels of detail in a suspect's fingerprint and in a “latent print” left at a crime scene. But making a so-called individualization—a conclusion that the prints are from the same source—is “where it gets a little fuzzy,” says Glenn Langenburg, a forensic scientist and fingerprint examiner at the Minnesota Bureau of Criminal Apprehension in St. Paul. After examiners look at enough prints known to be from the same source and from different sources, “their brain gets calibrated” to some internal threshold of similarity, he says.

The fuzziness shows in their findings. One study of 169 fingerprint examiners found 7.5% false negatives—in which examiners concluded that two prints from the same person came from different people—and 0.1% false positives, where two prints were incorrectly said to be from the same source. When some of the examiners were retested on some of the same prints after 7 months, they repeated only about 90% of their exclusions and 89% of their individualizations.

Testing examiner accuracy using known samples can give the judge or jury a sense of general error rates in a field, but it can't describe the level of uncertainty around a specific piece of evidence. Right now, only DNA identification includes that measure of uncertainty. (DNA analyses are based on 13 genetic variants, or alleles, that are statistically independent, and known to vary widely among individuals.) Mixtures of genetic material from multiple people can complicate the analysis (see story, p. 1133), but DNA profiling is “a relatively easy statistical problem to solve,” says Nicholas Petraco, an applied mathematician at City University of New York's John Jay College of Criminal Justice in New York City. Pattern evidence doesn't operate under the same rules, he says. “What's an allele on a tool mark?”; “What's an allele on a hair or fiber?”

To estimate how frequently a given feature occurs in pattern evidence, researchers will need large databases. Carriquiry and her CSAFE colleagues will begin by exploring digital collections, such as images of bullet and casing marks assembled by NIST researchers, and one of the world's largest collections of crime scene shoeprints, kept by the Israeli police force. The team must also decide what aspects of an image are relevant for comparison. For example, sole patterns indicating the brand and model of a shoe may not be as informative for a comparison as acquired characteristics such as damage or wear patterns.

A large database and a set of rules for feature selection could then feed a statistical model that describes how unusual the set of similarities between two samples really is, relative to similarities between two randomly selected samples from the population. Ideally, says Carriquiry, the model would produce a “likelihood ratio.” That would allow an examiner to say, for example, that the similarities between two fingerprints are 10,000 times more likely to occur if they came from the same finger than if they came from different ones.

For fingerprints, that kind of assessment seems within reach. A model under development by forensic scientist Cedric Neumann and statistician Christopher Saunders at South Dakota State University in Brookings can estimate a likelihood ratio for prints once a trained examiner marks their similarities (see diagram, p. 1131). The approach still isn't quite ready for use in court, says Neumann, in part because its results vary too widely depending on which features an examiner selects as relevant. Tighter standards for examiners could resolve the problem, he says.

For other types of evidence, the approach may never work, some scientists say. For instance, a relevant database of shoeprints might not be practical, says Lesley Hammer, a forensic scientist in Anchorage, Alaska, who specializes in footwear and tire track analysis. The database would have to keep up with an ever-changing market of brand-name and counterfeit products, document distinctive features like wear or damage patterns, and possibly even account for regional variations in shoe frequency—the likelihood of a snow boot turning up in Hawaii versus North Dakota, for example.

WHAT STATISTICIANS manage to compute with their new models will have little value if forensic examiners, jurors, judges, and lawyers don't know how to interpret statistical claims. That's why CSAFE collaborator Brandon Garrett, a law professor at the University of Virginia in Charlottesville, has begun to study how jurors perceive a forensic examiner's testimony.

In a 2013 study, for instance, online participants had to rate the likelihood of a defendant's guilt in a hypothetical robbery based on different kinds of testimony from a fingerprint examiner. It didn't seem to matter whether they were simply told that a print at the scene “matched” or was “individualized” to the defendant, or whether the examiner offered further justification—the chance of an error is “so remote that it is considered to be a practical impossibility,” for example. In all those cases, jurors rated the likelihood of guilt at about 4.5 on a 7-point scale. “As a lawyer, I would have thought the specific wording would have mattered more than it did,” Garrett says. But if subjects were told that the print could have come from someone else, they seemed to discount the fingerprint evidence altogether.

“When Neumann and his colleagues tested out their fingerprint likelihood ratios on mock jurors, participants recognized that making an “identification” was fundamentally different from providing a probability statement. But they didn't seem to distinguish between a strong likelihood ratio (one in 100,000) and a weaker one (one in 1000) when estimating the probability that a suspect was the source of a print. Neumann suspects that numbers can still be useful for describing testimony, but that lawyers and cognitive psychologists will have to team up to figure out the best presentation.

The final decision about what kinds of statements jurors can ponder, though, is up to judges, who often confer with lawyers and forensic examiners to decide what evidence is admissible. CSAFE aims to reach all these players through a campaign to boost statistical literacy. Last week, the statisticians conducted training for practitioners across Florida crime labs at the Palm Beach County Sheriff's Office, and they intend to launch similar courses around the United States.

Some judges are already pretty savvy about statistics. In the Boston racketeering case, federal district court judge Nancy Gertner found the detective's conclusion that only one gun on the entire planet could have produced the imprints on the bullet cartridges “preposterous.” She believed the evidence should have been excluded completely. But Gertner—now a professor at Harvard University—feared that an appeals court would reverse that move, so she “reluctantly” ruled that the detective could describe ways in which the bullet casings looked similar, but not conclude that they came from the same pistol. Ultimately, a jury said there was no evidence of a racketeering operation; Gertner cleared the defendants of the more serious federal charges and their cases were moved to state court.

What's troubling, Gertner says, is that when judges accept junk science, an appeals court rarely overrules them. Attaching a numerical probability to evidence, as CSAFE hopes to do, “would certainly be interesting,” she says. But even a standard practice of critically evaluating evidence would be a step forward. “The pattern now is that the judges who care about these issues are enforcing them, and the judges who don't care about these issues are not.”

4. # How hair can reveal a history

1. Hanae Armitage,
2. Nala Rogers

Sophisticated methods are giving hair a new role in forensic analysis.

Forensic hair analysis has developed a bad reputation. The technique has relied on traits such as color, thickness, and curvature to link a suspect to a crime scene. But an ongoing reanalysis of old cases by the U.S. Justice Department found that analysts have often overstated their case in the courtroom; several people convicted based on a hair sample were later found to be innocent.

Now, sophisticated analytical techniques are giving hair a new role in forensics. The goal is no longer matching a suspect to a crime scene but using hair to infer physical characteristics or even the travel history of an unknown criminal or victim. Most hairs found at crime scenes don't have enough DNA in them for analysis; “doing a chemical analysis and trying to determine some trait about the individual … is really the only alternative,” says Glen Jackson, a forensic scientist at West Virginia University in Morgantown.

Keratin, the main component of human scalp hair, contains all 21 amino acids, but the ratios depend on the body's biochemistry and differ from person to person. Hydrolyzing the amino acids and measuring their quantities yields a profile that, when compared with a database, gives an indication of a person's sex, age, body mass index, and region of origin, Jackson says—although the accuracy varies by trait and more work is needed.

The ratios of isotopes—atoms of the same element that differ in the number of neutrons—in hair can also yield clues. The ratios of hydrogen and oxygen isotopes in drinking water vary from region to region and are captured in hair. As a result, isotopic analysis of hair can provide clues about where a person has been in the previous months—or years, if the hair is long enough. In 2008, a Utah company called Isoforensics in Salt Lake City discovered that “Saltair Sally,” an unidentified woman found dead in Utah in 2000, had repeatedly moved between the Pacific Northwest and the Salt Lake City area before she died—a clue that helped identify her in 2012. “People are coming to us and saying ‘Hey, I heard about this technique and I've got a cold case from 1976. Do you think it will help?’” says Isoforensics President Lesley Chesson.

5. # A trail of microbes

1. Kai Kupferschmidt

The unique mix of bacteria you leave behind wherever you go might be used to identify you.

One morning last summer, evolutionary biologist Jose Lopez was having coffee on the back porch of his house in Hollywood, Florida, when two burglars climbed in through a front window and did what home invaders usually do: They rifled through drawers, disconnected the TV to carry it off, and even opened the fridge to have a Coke.

This wasn't an ordinary break-in, however. The invaders were employees of the local sheriff's office, and the burglary was part of a science project. Later, forensics experts swooped in to swab down surfaces and handles in the house. In a laboratory at the University of Chicago (UC) in Illinois, colleagues of Lopez's isolated DNA from these swabs and fished out parts of one particular stretch named 16S rDNA—a sequence that is distinctive for almost every bacterial species. By amplifying and sequencing these bits, the scientists were able to identify hundreds of different species in each swab.

After eliminating the species left by the house's legitimate residents—including a cat—they were left with a new kind of physical evidence: the microbial DNA deposited by the thieves as they moved through the rooms. Because the mix of species and strains in a person's microbiome is highly individual, such molecular signatures might be used to place someone at a crime scene, says Jack Gilbert, a microbial ecologist at UC. And because the microbiome varies by gender, age, and lifestyle, the data could also be used to build up a picture of a suspect.

The field is in its infancy; so far, the only crime it has helped solve occurred on the hit TV show CSI: Miami. Some scientists are skeptical that microbial signatures are individual enough to be used as evidence in court. “I think we are very far from using the microbiome in forensic analysis. If it will ever be used,” says microbiologist Jacques Ravel of the University of Maryland, Baltimore. Others are more optimistic. “We have enough data to suggest it is well worth exploring,” says David Relman, who studies human and animal microbiomes at Stanford University in Palo Alto, California.

HOPES THAT MICROBIOMES could help solve crimes date from a study published in 2010 in the Proceedings of the National Academy of Sciences, which showed that bacterial DNA recovered from computer keyboards matched the microbiomes found on their owners' fingertips. The authors also sampled bacteria from nine computer mice and used the results to pick the owners out of a database of 270 microbiomes. “This was the first paper to demonstrate that people leave a highly unique and identifiable signature,” Gilbert says.

Adding to the microbiome's appeal for forensics, people shed bacteria constantly and indiscriminately. “You're shedding them from your face, spitting them out from your mouth, breathing them out though your nose,” Gilbert says. They aren't confined by gloves or clothes. “Your trousers or your pants are like a loose fish net material to bacteria. As soon as you sit down, your bottom or your vaginal microbiota is expelled onto that surface and it is actually reasonably persistent until the next person sits down,” Gilbert says. In a 2015 paper, scientists measured the airborne bacteria surrounding volunteers in a sanitized chamber and were able to identify some of the subjects by their microbial cloud.

All told, researchers have sampled microbiomes from thousands of people, mostly volunteers from Europe and the United States, and found that the relative abundances of microbial species are highly individual. Even identical twins can be told apart. One reason our microbiome is so highly personal is that it's partly determined by our genome and immune system. Chance plays a role as well. In the first 3 or 4 years of life, humans seem to pick up a unique set of bacteria from the vast natural diversity they're exposed to; that mix remains fairly stable all their lives, says Peer Bork, a computational biologist at the European Molecular Biology Laboratory in Heidelberg, Germany.

Yet the question is whether these signatures can be used to identify a person beyond a reasonable doubt, as human DNA evidence can. Bork thinks that will be possible, but not with the 16S rDNA technique alone, because it mostly picks up differences between bacterial species. Instead, he thinks investigators need to fully sequence all the DNA swabbed from a crime scene and from suspects' microbiomes, to detect small differences between strains of the same species. “You and I both have E. coli, but I have a slightly different strain than you,” Bork says. “We carry about a thousand different species in the microbiome. If we look at differences in all of them, we may well be able to distinguish 8 billion people.” So far, Bork's group has studied about 3500 individuals using this strategy; all of them were unique. But because it entails extensive sequencing, this approach comes with a higher price tag.

Another problem is that both individual bacteria and the composition of microbial communities change over time. The signature a murderer left at a crime scene a decade ago may not exactly match his microbial cloud today. Moving to a different country changes the microbiome as well, and antibiotics can radically alter it. Smart criminals might pop a few pills before they strike, “like criminals who used to burn their fingertips with acid and other unpleasant techniques,” says Rob Knight of the University of California, San Diego, one of the leaders of the new field.

To help solve crimes, forensic analysts would need the equivalent of a fingerprint library: a database of known microbiome profiles to which they can compare evidence from a crime scene. “We would need 8 billion stool or skin samples,” Bork says. “I'm not sure how practical that is.” But Gilbert doesn't think that's a big problem. He says such databases could be built up the same way they have been for fingerprints: gradually, starting with convicted criminals.

Even if microbiologists can't pinpoint one particular culprit, a microbiome left at a crime scene may hold important clues. Is the perpetrator a man or a woman? Does he smoke? Where does she live? “Unlike fibers or fragments of hair, the microbiome contains an awful lot of information,” Gilbert says. For instance, a small study he did with collaborators in Shanghai, China, found big differences between the skin microbiomes of people living in urban, suburban, and rural areas.

In the fake break-in, Gilbert compared the signatures of the two intruders with a database of a few thousand people that he has built up; based on the relative abundance of particular taxonomic groups, he predicted that one of the burglars had at least 10 alcoholic drinks a week and that the other was on migraine medication. He was right on both counts. “When I heard that, I was in shock,” says George Duncan, a DNA expert at the Broward County Sherriff's office who had organized the burglary. These kinds of leads could be very valuable to police, he says.

Bork notes, however, that “at the moment many of these associations are very shaky.” For instance, two groups have reported that the microbiomes of diabetes patients can be distinguished from that of nondiabetics. But a careful analysis by Bork's group, published in Nature in December 2015, showed that what scientists had detected was not a signature from the disease, but from the common diabetes drug metformin.

Gilbert is trying to improve his database by recruiting more people in Chicago and Florida with jobs or lifestyles that leave a particularly strong mark on the microbiome, such as bakers, farmers, vegetarians, and vegans. To help the field along further, Rhonda Roby of the J. Craig Venter Institute in Rockville, Maryland, has received a grant of more than $900,000 from the National Institute of Justice to build a microbiome database containing thousands or even tens of thousands of samples for the forensics community. One thing is for sure, Ravel says: Scientists will need to tread carefully. Junk science has landed innocent people in jail in the past; the last thing microbiologists should do is add another flawed technique to the forensic arsenal, he says. “You don't want to start accusing and messing up the lives of many people just because they have a microbiome similar to the one found at the crime scene.” 6. # Who dropped the bomb? 1. Richard Stone Postdetonation forensics may help provide answers if the nuclear nightmare becomes a reality. Last summer, an atomic bomb detonated in a city on the U.S. Eastern seaboard, killing tens of thousands and plunging the nation into despair. As first responders and the military grappled with the aftermath, elite teams of scientists raced to analyze the blast for clues to precisely what kind of bomb had gone of and who bore responsibility for the act. That was the premise of an exercise—the first of its kind—held in July and August 2015 to test a new network of sensors that would collect data during a surprise nuclear strike. The Mighty Saber simulation was a sobering acknowledgment of many experts' belief that an attack on U.S. soil is more likely than ever—yet tracing responsibility would be far harder than it was during the Cold War, when the chief threat was annihilation by the Soviet Union. “The scenario has changed,” says Thomas Cartledge, a nuclear engineer with the U.S. Defense Threat Reduction Agency (DTRA) in Fort Belvoir, Virginia. “Now, if you see a mushroom cloud go of in New York City, you won't know who did it, or what kind of weapon they used.” Possibilities include a warhead diverted from the U.S. arsenal or smuggled into the country by terrorists, or a bomb delivered by an enemy state such as North Korea, which has threatened to nuke the White House. The conceivable need to unmask a perpetrator and mount a response is propelling the emerging area of postdetonation forensics. “Someone's going to get the pointy end of the stick. You want to make sure the right entity gets it,” says Howard Hall, director of the Institute for Nuclear Security at the University of Tennessee, Knoxville. He and other nuclear detectives are devising new sensors, manufacturing artificial fallout to hone analytical techniques, and studying how the glass formed in the furnace of an atomic blast would vary depending on the nature of the bomb and the city where it detonated. The most likely nuclear terrorism scenario, experts say, is a bomb set of on a city street. Past experience offers only a sketchy picture of the resulting devastation. The atomic bombs the United States dropped on Hiroshima and Nagasaki in 1945 detonated about 500 meters above those cities. During the subsequent half-century, while the United States refined its atomic arsenal, nearly all tests were in the air or underground, not in citylike environments. Researchers did study fallout and how it forms, but they were seeking clues about how to prevent or alleviate radiation illness, not identify the perpetrator. “Scientists were not interested in figuring out what kind of device had detonated, because they already knew that,” says analytical chemist Michael Kristo, a nuclear forensics expert at Lawrence Livermore National Laboratory in California. Still, the testing program was a proving ground for postdetonation forensics. The U.S. national labs “put together some very good radiochemical procedures for analyzing debris,” says Hall, a radiochemist. Fallout is a mélange of the vaporized environment—soil and structures that were near the blast—laced with fission products (radioisotopes created when fissile materials like uranium or plutonium fission), activation products (radioisotopes formed when the blast radiation transmutes shielding and other bomb components), and residual nuclear material. The precise constituents vary according to a weapon's design—whether it's a simple gun-triggered uranium device, for example, or an intricate hydrogen bomb. “Each type of weapon has a distinct fingerprint,” says Michael Pochet, a U.S. Air Force electrical engineer detailed to DTRA. In plutonium bombs, for example, the fissile isotope is plutonium-239, made in nuclear reactors and extracted by reprocessing spent fuel, which contains a mix of plutonium isotopes and other actinides like americium. Detecting those nuclei indicates that the bomb's core was plutonium. Their proportions hold clues to the bomb's history, says Joel Ullom, a physicist at the U.S. National Institute of Standards and Technology in Boulder, Colorado, who, with colleagues at Los Alamos National Laboratory in New Mexico, has developed a superconducting sensor that speedily differentiates plutonium isotopes. The ratio between plutonium isotopes and americium-241, a decay product of plutonium-241, “can tell you the time since the plutonium was chemically purified,” Ullom says. Americium is removed during reprocessing, so as the freshly separated plutonium ages, americium starts accumulating again. Hall, meanwhile, is developing faster methods to analyze lanthanides, the 15 rare earth elements that, with the radioactive actinides, are key constituents of fallout. The mix of lanthanides and actinides reveals information about the weapon's shielding, for example, and the energy of the neutrons that bombarded it. He intends to fit his gas phase separation apparatus onto a “flyaway lab”: a skid that can be deployed quickly in the event of an attack. To ground-truth these analytical techniques, researchers at Livermore and other national labs are producing surrogate fallout representing different bomb types. The scientists have pressed into service the National Ignition Facility at Livermore, one of the world's most powerful lasers, which Kristo calls “a ready source” of neutrons at energies comparable to those produced in the deuterium-tritium fusion reactions that power a hydrogen bomb. Hall's team is cooking up another type of test sample for postdetonation forensics: artificial melt glass. The real thing forms when an atomic inferno instantly melts anything having the misfortune of being at ground zero. The glass varies with the explosion site, but different bomb specs also produce unique melt glasses, providing clues about what happened. Hall's group has developed a recipe book of melt glass for any geographic location based on a “witch's brew” of the bomb's fissile material and explosive yield, its detonation point, and the local geology and construction materials. The team reproduced trinitite, the greenhued glass left by the Trinity test, the first U.S. nuclear detonation, which took place in 1945 at the White Sands Missile Range in New Mexico. They have also baked up specimens for Houston, Texas, where the glass-dominated architecture would yield a grayish glass if nuked, and for New York City, whose iron-heavy construction leads to a darker, volcanic-looking glass. ATOMIC BLASTS ALSO UNLEASH an electromagnetic pulse—a blitzkrieg of gamma rays, x-rays, and radio waves that instantly fries most nearby electronics—as well as intense light, seismic waves, air pressure waves, and infrasound. All may provide information on the type of bomb and its origin. In the 1940s, scientists began designing sensors to capture these signals, first at White Sands and then primarily at the Nevada Test Site, where the United States detonated 928 bombs. Now, DTRA is leading a government-wide effort to upgrade those sensors and link them up in an array, called Discreet Oculus, which can be deployed in and around cities. “We've repurposed the sensors for an urban environment,” Cartledge says. That required devising algorithms to account for how cityscapes deflect or absorb various types of waves, for instance, and filtering out noise from sources such as subways, the vibrations of which could interfere with interpreting vibrations from the detonation. Mighty Saber set out to test the ability of Discreet Oculus to identify the type of bomb in a surprise attack. The exercise's premise was that a bomb had been diverted from the U.S. arsenal and detonated. “We pulled in weapon designers to see what those signals would be,” Pochet says. In late 2013, several dozen experts began ginning up a fallout profile and modeling how waves would propagate and attenuate in a real U.S. city. DTRA won't say which city it was; Cartledge refers to it as Gotham. “No city wants to know it was used as a model for a nuclear attack,” he says. Based on these models, DTRA sent data simulating what Discreet Oculus sensors would record during the explosion to the Air Force Technical Applications Center on Patrick Air Force Base in Florida, which distributed it to four teams of experts from the center and the U.S. national labs. “We said, ‘Here's the data, go and do your analyses’,” Cartledge says. The task was to identify the bomb, and time was of the essence. “In real life,” Pochet says, “we would be working against the clock, struggling to keep up with the news cycle.” The exercise ran for 25 days; all four teams figured it out, Cartledge says. He won't specify how quickly but says, “We need to be faster.” DTRA HAS ALREADY INSTALLED Discreet Oculus in several U.S. cities, where the arrays are undergoing testing. They are expected to be operational and transferred to the U.S. Air Force in 2018. DTRA has also begun working on a portable version called Minikin Echo that could be deployed for events like the Olympics. Although postdetonation forensics may well finger a bomb design, that knowledge by itself wouldn't always unmask the perp. A gun-triggered uranium bomb, for example, could be fashioned by any of a number of terrorist outfits with modest technological expertise, such as the Islamic State group, providing they can lay their hands on several kilograms of highly enriched uranium. That's “where intel comes in,” Hall says. But to have any chance of unraveling the details of a nuclear attack, investigators have to lay the scientific groundwork—while hoping it will never be needed. 7. # Whose voice is that? 1. Nala Rogers New research may make speaker recognition systems more accurate and more objective. On the night George Zimmerman fatally shot 17-year-old Trayvon Martin in Sanford, Florida, a 911 call captured the sound of someone screaming. But who? An expert for the prosecution testified it was Martin, begging for help in his last moments. But at a pretrial hearing, several scientists said the recording quality was too poor to tell. The call was not admitted as evidence. The case illustrates the problems in speaker recognition, a forensic field with a checkered history that is trying to find solid scientific ground. Police and lawyers “can't tell the difference between somebody who's deluded or who is a charlatan, and somebody who is actually doing solid scientific work,” says Geoffrey Stewart Morrison, an independent forensic scientist and former chair of the Forensic Acoustics Subcommittee of the Acoustical Society of America in Vancouver, Canada. In the 1960s, analysts began converting recordings into images using spectrograph machines and making subjective judgments about how similar they looked—a method once commonly used in courts but now widely discredited. More reliable alternatives have emerged. Signal processing engineers developed automated systems that typically measure the frequency components of speech every few milliseconds. Phonetics experts break up recordings based on individual sounds, then analyze the elements using statistical tests or their own judgments. Automated systems now work very well—some banks rely on them to identify their clients—but only if you clearly speak a standard sentence into a microphone. Comparing real-world samples is much more error-prone, says Hirotaka Nakasone, a senior scientist in the Federal Bureau of Inivestigation's voice recognition program who testified in the Trayvon Martin case. The same person will sound different during a bar fight versus speaking calmly in an interrogation room, and recording quality is often poor. That's why the admissibility of voice recognition systems in courts is contentious, although the systems are widely used in criminal investigations. To improve accuracy, scientists are studying how factors like inebriation, emotional state, and recording devices influence voice samples. They are also testing how well existing systems perform and developing standards for things like data selection and the presentation of results. For example, a panel chaired by Nakasone is working on standards for the U.S. National Institute of Standards and Technology. The field is moving away from subjective systems, Morrison says: “Automatic systems are more robust to cognitive bias and are more easily tested.” 8. # Clues from the ashes 1. Lizzie Wade* Were the bodies of 43 missing Mexican students burned at a dumpsite? Fire investigator José Torero says the science doesn't add up. On 12 July 2015, José Torero found himself standing in the municipal dump outside the town of Cocula in the Mexican state of Guerrero, tallying up everything he didn't see. Burn marks on the trees. Melted plastic. Anything that, to his trained eye, would indicate that 10 months before, the Cocula dump could have been the site of a massive fire that burned 43 bodies to ash. Torero, who was born in Peru and now teaches at the University of Queensland (UQ), St. Lucia, in Brisbane, Australia, had come to investigate a crime that shocked the world: the disappearance of 43 students from the Ayotzinapa Normal School, a rural teacher's college near Tixtla, Guerrero. According to Mexico's attorney general (AG), the crime culminated in the students' bodies being incinerated at the Cocula dump. But what Torero found—or rather, didn't find—at the alleged scene of the crime has threatened to unravel the government's story and has left the fate of the missing 43 even more mysterious than before. His investigation has also put a fresh spotlight on the forensic science of fire investigation, an area where Torero is seen as a world expert. Although tens of thousands of fires are examined yearly around the world for arson, few investigations employ state-of-the-art science. Many are done by firefighters who rely on their own experience with how fire behaves, rather than scientific studies, says John Lentini, an independent fire investigator based in Islamorada, Florida. Lentini himself helped raise major doubts about the evidence used to convict Cameron Todd Willingham, a man executed in Texas for setting the fire that killed his family. “The field is not very advanced,” Lentini says. Fire encompasses “biology, chemistry, heat transfer, fluid mechanics, chemical combustion, behavior of structures, behavior of materials. Scientifically speaking, it's a problem of enormous complexity,” Torero says. Few investigations take into consideration state-of-the-art research in each of these fields. Worse, Lentini says, many investigators set out trying to prove an established theory of the crime, rather than ruling out hypotheses with the help of models and experiments, as Torero does. TORERO GOT INTERESTED IN FIRE after he left Peru to study engineering at the University of California, Berkeley, where he met researchers working on fire safety problems for the International Space Station. “It was a combination of the NASA thing—it catches everybody's attention—and the fact that the problem in and of itself was incredibly complicated and unique.” During a postdoc with the European Space Agency, Torero's interests began to shift. “What really brought me into more ‘down-to-earth’ work,” he says, was a 1999 fire in the tunnel under Mont Blanc in the French Alps that killed 38 people. In its wake, Torero dedicated himself to the intersection of fire safety and engineering, studying how disastrous fires start and spread as well as engineering strategies to prevent them. After teaching at the University of Maryland, College Park, and the University of Edinburgh, he became the head of the School of Civil Engineering at UQ in 2012. When it comes to forensic work, Torero dedicates himself to cases “that have a significant social impact.” In the aftermath of the 9/11 attacks, he studied the structural weaknesses that allowed fire to bring down the Twin Towers in hopes of improving skyscraper design. In 2011, he investigated a fire that killed 81 inmates in Chile's San Miguel prison. Guards had seen smoke but had failed to open the padlock that kept prisoners trapped in the burning cell. “They were blamed for not taking action and not being able to rescue the inmates,” Torero remembers. After recreating the blaze in the laboratory and using computer models to understand its behavior, his team concluded that by the time the guards saw smoke, the padlock was too hot to open and the prisoners were already dead. “They would never have had the time,” Torero says. His conclusions refocused blame away from the individual guards and onto the overcrowded prison conditions that allowed the fire to endanger so many lives. Torero got involved in the Ayotzinapa case at the request of a group of five independent experts (known in Mexico as the Interdisciplinary Group of Independent Experts, or GIEI) convened by the Inter-American Commission on Human Rights to examine both the disappearances and how Mexico's AG has handled the investigation. The trouble began on the night of 26 September 2014, when students hijacked five commercial buses to transport them to a demonstration in Mexico City—an illegal but widely tolerated practice by students at Mexico's politically radical teachers' colleges. The students convened in the town of Iguala, where they came under gunfire by municipal and, allegedly, federal police. Some students escaped, others were killed while trying to flee, and 43 disappeared. According to the AG, the missing students were kidnapped by the Guerreros Unidos drug cartel with help from the local police and under orders from Iguala's mayor, who had family ties to the gang. They were then executed and their bodies incinerated in the Cocula dump in the early hours of 27 September 2014, the AG says. The executioners gathered some of the remains into trash bags and allegedly dumped them into a nearby river, where they were later recovered; other remains were found in the dump itself. The remains—mostly ash with a few bone fragments—were sent to a lab at the University of Innsbruck in Austria, where scientists have been able to make positive DNA identifications of two of the missing students. Still, doubts continued to swirl around the government's story, especially because no independent investigators were present when the remains were found, calling into question the chain of custody. The GIEI panel was asked to help resolve the doubts. “A LOT of the more difficult [fire] cases are highly politicized,” Torero says, and Ayotzinapa is no exception. “Politically, you need an answer, and you have to provide that answer now.” In the case of Ayotzinapa, arrested cartel members had confessed—under torture, the GIEI suspects—to burning 43 bodies in the Cocula dump. “The [AG's] entire investigation was driven to try to prove or create evidence that what the testimonies were saying was correct,” Torero says. For example, the government's report presents rocks found in the Cocula dump that had been cracked from heat as evidence supporting the confessions. The problem, Torero says, is that “I could have gotten the same cracked rock with a small fire, with an old fire, in a number of different ways”—none of which the AG ruled out. More things were amiss. Cartel members said they incinerated all 43 bodies at the same time on a pyre made of wood and tires. Past studies done with pig carcasses and human corpses revealed that when a body is burned on a pyre, the fat serves as fuel, but it doesn't provide enough heat to burn up all of the organic matter, Torero says. Yet the remains studied in Austria had virtually no organic matter left. (That's why the lab has been able to make only two DNA identifications so far.) “The only way you eliminate all that is if you have a source of heat that doesn't depend on the fat,” Torero says. Wood and tires could not have supplied so much heat: Torero calculated that the perpetrators would have needed to burn 20,000 to 40,000 kilograms of wood or 9000 to 18,000 tires to provide the necessary energy. Instead, the state of the remains “is typical of incineration in a furnace,” Torero says, such as those used in crematoriums. No crematoriums near Iguala have yet been investigated as possible crime scenes. Then Torero went to the Cocula dump himself. He saw some partially burned tires and melted plastic, but they were more in line with what he would expect to see after a series of small fires. Most telling, in his view, were the trees. Leaves burned off in a fire can grow back in 10 months, but when a big fire scorches a tree trunk, that scar never goes away; scientists can even see burn scars centuries later in tree rings. None of the trees bordering the dump showed such damage. TORERO ANNOUNCED his findings on 6 September 2015, when the GIEI released its full report: “The hypothesis that 43 bodies were burned in that dump is impossible.” The Argentine Forensic Anthropology Team, which is monitoring the Ayotzinapa investigation on behalf of the victims' families, recently released its own report supporting Torero's assessment. And Lentini believes Torero “almost certainly came to the right conclusion.” AG Arely Gómez did not comment on Torero's findings but did open a second forensic investigation of the Cocula site. Several international fire science experts contacted for this story declined to comment because they are participating in that new inquiry. The AG's office told Science that it expects to release the results in early April. Working 10 months after the fact with a questionable chain of custody, Torero knew he wasn't going to be able to reconstruct what happened at the dump. “It's like having three pieces of a 10,000-piece puzzle,” he says. Still, for some families it was valuable to have their doubts about the government's story confirmed. “We're poor, but we're not stupid,” a mother of one of the victims said at a press conference following Torero's announcement. “Our children weren't burned there!” For now, that's the only thing they can be sure about. • * in Mexico City 9. # The microbial death clock 1. Kai Kupferschmidt The succession of microbes breaking down a human cadaver may provide precise information about the time of death. When you die, a new life begins for the billions of microbes you carry with you. Unchecked by your immune system, waves of species start multiplying and breaking down your body. Microbes from the environment join in as well. Geneticist Jessica Metcalf of the University of Colorado, Boulder, hopes this macabre procession can provide a microbial clock that can help investigators tell the time of death more precisely than they can with current methods, which rely on body temperature, rigor mortis, and insects. Early in the decay, for instance, bacteria from the Moraxellaceae family and the genus Acinetobacter begin gorging on dying human cells. Soon after, the Rhizobiaceae family, often involved in breaking down nitrogen sources, takes over. The gases produced by these bacteria cause the body to bloat and eventually rupture, allowing oxygen in and giving aerobic species the upper hand. Microscopic worms also start to multiply, probably feasting on the bacterial biomass now covering the corpse. Metcalf first showed that she could use microbes, combined with a statistical model, to pinpoint the time of death of mice to within 3 days, even weeks after death. Then her team took samples from four human bodies at a so-called body farm, where cadavers are placed outside so that forensic scientists can study how they decompose. In a paper published in Science (8 January, p. 158), they reported that, again, the microbial dance was predictable enough to set a clock. “Over 25 days our error rate is about 2 to 4 days,” says Rob Knight of the University of California, San Diego, who is collaborating with Metcalf. In a large new project, the researchers will expose 36 bodies, three at each of three different body farms, in all four seasons. That will help them further calibrate their clock and tell them how it is affected by the environment. 10. # The Bitcoin busts 1. John Bohannon Its anonymity made Bitcoin popular among criminals. But even with cryptocurrency, researchers can follow the money. Bitcoin, the Internet currency beloved by computer scientists, libertarians, and criminals, is no longer invulnerable. As recently as 3 years ago, it seemed that anyone could buy or sell anything with Bitcoin and never be tracked, let alone busted if they broke the law. “It's totally anonymous,” was how one commenter put it in Bitcoin's forums in June 2013. “The FBI does not have a prayer of a chance of finding out who is who.” The Federal Bureau of Investigation (FBI) and other law enforcement begged to differ. Ross Ulbricht, the 31-year-old American who created Silk Road, a Bitcoin market facilitating the sale of$1 billion in illegal drugs, was sentenced to life in prison in February 2015. In March, the assets of 28-year-old Czech national Thomas Jiikovský were seized; he's suspected of laundering $40 million in stolen Bitcoins. Two more fell in September 2015: 33-year-old American Trendon Shavers pleaded guilty to running a$150 million Ponzi scheme—the first Bitcoin securities fraud case—and 30-year-old Frenchman Mark Karpelès was arrested and charged with fraud and embezzlement of $390 million from the now shuttered Bitcoin currency exchange Mt. Gox. The majority of Bitcoin users are law-abiding people motivated by privacy concerns or just curiosity. But Bitcoin's anonymity is also a powerful tool for financing crime: The virtual money can keep shady transactions secret. The paradox of cryptocurrency is that its associated data create a forensic trail that can suddenly make your entire financial history public information. Academic researchers helped create the encryption and software systems that make Bitcoin possible; many are now helping law enforcement nab criminals. These experts operate in a new field at the crossroads of computer science, economics, and forensics, says Sarah Meiklejohn, a computer scientist at University College London who co-chaired an annual workshop on financial cryptography in Barbados last month. “There aren't that many of us,” she notes. “We all know each other.” When Bitcoin first emerged, law enforcement officers were “panicking,” Meiklejohn says. “They thought these technologies were dangerous and made it harder for them to do their job.” But as the arrests and convictions have rolled in, “there's a steady shift toward seeing cryptocurrency as a tool for prosecuting crimes.” Even in the strange new world of Bitcoin, FBI Assistant General Counsel Brett Nigh said in September 2015, “investigators can follow the money.” UNLIKE MONEY ISSUED by governments, Bitcoin has no Federal Reserve, no gold-backing, no banks, no physical notes. Created in a 2008 academic paper by a still unknown person using the name Satoshi Nakamoto, Bitcoin “is an intellectual artifact,” says Patrick McDaniel, a computer scientist at Pennsylvania State University (Penn State), University Park. “It's the frontier of economics.” Strictly speaking, Bitcoins are nothing more than amounts associated with addresses, unique strings of letters and numbers. For example, “1Ez69SnzzmePmZX3WpEzMKTrcBF2gpNQ55” represents nearly 30,000 Bitcoins seized during the Silk Road bust—worth about$20 million at the time—that were auctioned off by the U.S. government on 1 July 2014.

Those Bitcoins have been split up and changed hands numerous times since then, and all of these transactions are public knowledge. The past and present ownership of every Bitcoin—in fact every 10-millionth of a Bitcoin—is dutifully recorded in the “blockchain,” an ever-growing public ledger shared across the Internet. What remains hidden are the true identities of the Bitcoin owners: Instead of submitting their names, users create a code that serves as their digital signature in the blockchain.

The job of keeping the system running and preventing cheating is left to a volunteer workforce known as Bitcoin miners. They crunch the numbers needed to verify every transaction. Added to this is an ever-growing math task known as “proof of work,” which keeps the miners honest. The calculations are so intense that miners use specialized computers that run hot enough to keep homes or even office buildings warm through the winter. The incentive for all this effort is built into Bitcoin itself. The act of verifying a 10-minute block of transactions generates 25 new Bitcoins for the miner. This is how Bitcoins are minted.

Just like any currency, Bitcoin's real-world value emerges as people trade it for goods, services, and other currencies. If you're not a miner, you can only get Bitcoins from someone who already has them. Companies have sprung up that sell Bitcoins—at a profitable rate—and provide ATM machines where you can convert them into cash. And of course, you can sell something in return for Bitcoins. As soon as both parties have digitally signed the transaction and it is recorded in the blockchain, the Bitcoins are yours.

As Science went to press, Bitcoin's market capitalization, a measure of the amount of money invested in it, stood at \$5.6 billion. That money is very safe from theft, as long as users never reveal their private keys, the long—and ideally, randomly generated—numbers used to generate a digital signature. But as soon as a Bitcoin is spent, the forensic trail begins.

BY 2013, millions of dollars' worth of Bitcoins were being swapped for illegal drugs and stolen identity data on Silk Road. Like a black market version of Amazon, it provided a sophisticated platform for buyers and sellers, including Bitcoin escrow accounts, a buyer feedback forum, and even a vendor reputation system. The merchandise was sent mostly through the normal postal system—the buyer sent the seller the mailing address as an encrypted message—and the site even provided helpful tips, such as how to vacuum-pack drugs.

Investigators quietly collected every shred of data from Silk Road—from the images and text describing drug products to the Bitcoin transactions that appear in the blockchain when the deals close. Ultimately, investigators needed to tie this string of evidence to one crucial, missing piece of data: the Internet Protocol (IP) addresses of the computers used by buyers or sellers.

The challenge is that the Bitcoin network is designed to blur the correspondence between transactions and IP addresses. All Bitcoin users are connected in a peer-to-peer network over the Internet. Data flow between their computers like gossip in a crowd, spreading quickly and redundantly until everyone has the information—with no one but the originator knowing who spoke first.

This system worked so well that it was carelessness, not any privacy flaws in Bitcoin, that led to the breakthrough in the investigation of Silk Road. When Ulbricht, the ringleader, was hiring help to expand his operation, he used the same pseudonym he had adopted years before to post announcements on illegal drug discussion forums; that and other moments of sloppiness made him a suspect. Once FBI tracked his IP address to a San Francisco, California, Internet cafe, they caught him in the act of logging into Silk Road as an administrator.

Other criminals could take solace in the fact that it was a slip-up; as long as you used Bitcoin carefully, your identity was protected behind the cryptographic wall. But now even that confidence is eroded.

Among the first researchers to find a crack in the wall were the husband-and-wife team of Philip and Diana Koshy. In 2014, as graduate students in McDaniel's lab at Penn State, they built their own version of the software that buyers and sellers use to take part in the Bitcoin network. It was especially designed to be inefficient, downloading a copy of every single packet of data transmitted by every computer in the Bitcoin network. “We wanted to see everything,” Philip Koshy says.

If the data flowing through the network were perfectly coordinated, with everyone's computer sending and receiving data as frequently as the rest, then it might be impossible to link Bitcoin addresses with IP addresses. But there is no top-down coordination of the Bitcoin network, and its flow is far from perfect. The Koshys noticed that sometimes a computer sent out information about only one transaction, meaning that the person at that IP address was the owner of that Bitcoin address. And sometimes a surge of transactions came from a single IP address—probably when the user was upgrading his or her Bitcoin client software. Those transactions held the key to a whole backlog of their Bitcoin addresses. Like unraveling a ball of string, once the Koshys isolated some of the addresses, others followed.

Ultimately, they were able to map IP addresses to more than 1000 Bitcoin addresses; they published their findings in the proceedings of an obscure cryptography conference. It is unusual for an academic paper to cause both The New York Times and the U.S. Department of Homeland Security to come calling. “It was crazy,” Philip Koshy says. Their technique has not yet appeared in the official record of a criminal case, but the Koshys say they have observed so-called fake nodes on the Bitcoin network associated with IP addresses in government data centers in Virginia, suggesting that investigators there are hoovering up the data packets for surveillance purposes too. (The pair has since left academia for tech industry jobs.)

AS CRIMINALS HAVE EVOLVED more sophisticated methods to use Bitcoin, researchers have followed apace. Meiklejohn—who says she regularly works with law enforcement but is “not comfortable discussing the details”.was one of the first researchers to explore Bitcoin “mixing” services. The basic idea is to protect the anonymity of transactions by swapping many people's Bitcoin stashes with each other, as in a shell game. The forensic trail shows the money going in but then goes cold because it is impossible to know which Bitcoins belong to whom on the other end. “So in principle, this is a solution to Bitcoin's anonymity problem,” Meiklejohn says.

But even mixing has weaknesses that forensic investigators can exploit. Soon after Silk Road shut down, someone with administrative access to one of the newly emerging black markets walked away with 90,000 Bitcoins from user escrow accounts. The thief tried to use a mixing service to launder the money, but wasn't patient enough to hide the tracks, Meiklejohn says. “It's difficult to push large amounts of Bitcoin through mixing services secretly. It's extremely noticeable no matter how you do it.” Thomas Jiikovský, the man under investigation by Czech police, is suspected to be the thief in question.

The beauty of Bitcoin, from a detective's point of view, is that the blockchain records all. “If you catch a dealer with drugs and cash on the street, you've caught them committing one crime,” Meiklejohn says. “But if you catch people using something like Silk Road, you've uncovered their whole criminal history,” she says. “It's like discovering their books.”

Exactly that scenario is playing out now. On 20 January of this year, 10 men were arrested in the Netherlands as part of an international raid on online illegal drug markets. The men were caught converting their Bitcoins into Euros in bank accounts using commercial Bitcoin services, and then withdrawing millions in cash from ATM machines. The trail of Bitcoin addresses allegedly links all that money to online illegal drug sales tracked by FBI and Interpol.

IF BITCOIN'S PRIVACY shortcomings drive users away, the currency will quickly lose its value. But the demand for financial privacy won't disappear, and new systems are already emerging. “I don't feel people have the right to know, unless disclosed, how much cash is in my wallet, just like I don't feel anyone should know what conversations I'm having with anyone else,” says Ryno Matthee, a software developer based in Somerset, South Africa.

Matthee is part of a team launching a new anonymous online market called Shadow this year, which will use its own cryptocurrency, ShadowCash. The goal is not to facilitate illegal transactions, Matthee says. It will be up to the users, who administer the system, to police it, he says, but to help prevent abuse, “we are going to try our best to filter out known keywords for drugs or worse.”

Shadow is far from the only Bitcoin competitor. Scores of alternative cryptocurrencies now exist. And some experts predict that one may finally go mainstream. Some banks already rely on a cryptocurrency called Ripple for settling large global money transfers. And the U.S. government “has been engaging with the cryptocurrency community and learning from them,” says Bill Gleim, head of machine learning at Coinalytics, a company based in Menlo Park, California.

Gleim believes the federal government will issue its own cryptocurrency, “maybe as soon as late 2016.” If so, it is likely to require users to verify their real-world identities. That could defeat the purpose of cryptocurrency in the eyes of privacy advocates and criminals. Or maybe not: In this technological game of cat and mouse, the next move may go to the criminals.

Correction (14 March 2016): “The Bitcoin busts” said that Bitcoin developer and investor Martti Malmi once boasted that Bitcoin is “totally anonymous” and that “the FBI does not have a prayer of a chance of finding out who is who.” Malmi tells Science he never said this, and that the quote was manufactured by a cyberbully.