Sometime around 1870, the New York City Post Office established a new department, staffed by a small team of specialists. According to an 1871 profile in Harper’s Magazine, these postal workers spent their days scrutinizing what looked like “the foot-prints of a gigantic spider that had, after wading knee-deep in ink, retreated hastily” across envelopes and postcards. In reality, these would-be arachnologists were employed to make sense of the “miserable chirography” of city residents, whose poor penmanship was causing unacceptable delays in delivery.
One expert decipherer recalled working on a letter that had arrived back in New York after traveling hundreds of miles over four days, repeatedly rejected as illegible by clerks in regional post offices. He studied the chickenscratch for a full workday before finally cracking the code: it was addressed to Chappaqua, a city just thirty miles north. Taking its name from the informal term clerks reserved for these most challenging pieces of mail, the new outfit came to be known as the Bureau of Hards.
Delivering hards, no matter the cost, is a reflection of the US Post Office’s commitment to truly universal service—a radical vision of democratic communications infrastructure enshrined in the Postal Service Act of 1792. No matter the sender, the recipient, or the distance separating origin and destination, federal code stipulated that the Post Office must “bind the nation together.” As Alexis de Tocqueville put it in his 1835 treatise Democracy in America, the US mail system, unlike its European counterpart, “was organized so as to bring the same information to the door of the poor man’s cottage and to the gate of the palace.” To live up to this idealistic ethos, hards must be treated no differently than easies.
But, as the errant letter destined for Chappaqua demonstrates, universal mail service tends to be extremely laborious. Supplanting human postal workers—slow, error-prone, and wage-requiring—with nonhuman proxies has long been a prospect with considerable purchase for postal management. The first machines arrived in post offices in the 1870s, and it’s no coincidence that the first postal worker unions were formed then, too. By the turn of the twentieth century, the Post Office and its governmental overseers had set into motion an unceasing drive to maximize the role of machines and minimize the role of humans.
Today, the United States Postal Service possesses the third-largest information technology infrastructure in the world—a rarely cited superlative. Not often included in the discourse around Big Tech, the USPS controls a sophisticated and sprawling computer network, linking together over 30,000 facilities and nearly 10,000 pieces of automated machinery, shuttling 150 billion pieces of mail per year between 150 million delivery points: houses, businesses, and PO boxes from Utqiagvik, Alaska, to Key West, Florida. The number of mail pieces per postal worker, a rough measure of automation’s impact, has more than doubled since 1950.
Despite significant advances in postal technology made since the mid-twentieth century, however, the USPS remains the country’s largest public-sector employer. The majority of its workforce is stationed at the input or output stage of what is, in effect, an enormous circuit: clerks are responsible for getting mail into the mailstream, and letter-carriers handle mail once it has exited. No single mandate better captures the thrust of modern postal operations than that of realizing a fully automated mailstream capable of connecting clerk and carrier with zero intervention from humans along the way.
But a tiny fraction of the USPS’s half a million workers—about two-tenths of 1 percent—toil away in a modern day Bureau of Hards. Two miles south of the Salt Lake City International Airport, in a drab warehouse, these workers parse the squiggled and smudged addresses emblazoned on each piece of mail that has proven illegible to the advanced machine-readers deployed in processing plants across the country. The 1,100 workers on staff at the Remote Encoding Center (REC) tend to the “nonmachinable” scraps discarded by the Postal Service’s automated leviathan. They ensure that more than one billion pieces of mishandled, misdirected, and misidentified mail arrive at their destinations each year.
Hards have never been simply a technical problem in need of a technical solution. Rather, hardness is better understood as an index of the social and political conditions under which mail is delivered. Taken together, these two deciphering operations—the nineteenth-century Bureau of Hards and the twenty-first-century REC—become legible as something like the origin and destination of an arduous and ongoing struggle between postal management and postal workers over the question of technological change.
Workers stream into and out of the Salt Lake City REC at all hours, with shifts staggered to begin every 15 minutes, 24 hours per day, 365 days per year. Once inside, Data Conversion Operators, informally known as keyers, consult one of the many large monitors scattered around the facility displaying a clutter of acronyms and numbers. On the morning when I visited the REC last summer, this coded to-do list read “PARS 209K / RIPS 122K / PRES 28K,” indicating that just over 350,000 items were awaiting judgment from a discerning human eye.
At the center of the 77,000-square-foot facility is an elevated platform that keyers cheekily refer to as “air-traffic control.” From this vantage, managers consult a comically large array of computer screens with charts, spreadsheets, and data visualizations that track the incoming and outgoing flow of mail. What arrives at the REC is not the physical piece of mail itself, but rather a digital surrogate. Each time a sorting machine in a postal plant encounters an address it cannot match with one in the USPS’s database, it takes a snapshot and automatically sends it on a virtual detour to the REC. Keyers receive these images in large batches and make quick work of them; eyes quickly scanning for visual clues, fingers dexterously entering relevant address data.
Once the address data has been entered into the keying interface, the image is sent back to the plant, where the sorting machine applies a new barcode to the parcel, ensuring it remains legible to the subsequent sorting machinery it will encounter en route to its destination. This seemingly simple image-processing task depends upon a dense and dated patchwork of software and hardware linking the REC to more than 300 postal hubs across the country.
Acronyms abound at the REC, indicating the sheer number of image-processing systems in use. Different sorting machines, designed to handle different types of mail, send different types of data to the REC, using different software platforms designed over the last several decades. Some links in this complex chain have been designed in-house, but many have been farmed out to contractors, including Lockheed Martin and Siemens.
The goal of consolidating these many overlapping but incompatible systems has long been a high priority for keyers and management alike. RIPS, which launched in 2019, is the newest consolidation effort. Once completed, it will funnel the data from IPS, PICS, FICS, and PRES—four currently siloed systems that handle letters, packages, “flats” like magazines, and other postal paperwork—into a single keying interface.
In the past, different sections of the facility were hardwired to handle these different types of mail, creating a bustling, sometimes chaotic environment. Keyers would walk, or occasionally run, from one section to another as keying queues ebbed between parcel types throughout their shift. Endcaps on cubicles still display “No more cutting through aisles!” signs, despite the REC’s current library-like stillness. Much to management’s delight, less time spent on foot means more time spent keying.
Every keying station is stocked with a keyboard, a monitor, and three or four desktop towers, each dedicated to one of the many acronymic systems in use. Many of these towers are brand new, but run as virtual machines emulating legacy software platforms developed decades ago. In lieu of physically moving throughout the facility, keyers speed back and forth between different programs every several minutes—say, from APPS (packages) to PARS (change-of-address forms) and back to APPS—each time reentering their username and password.
The work of keying is almost unfathomably fast-paced. But it is graceful, not frenetic. Most keyers sport headphones, listening to music or podcasts as they nimbly flit through an unending sequence of rasterized black-and-white images, logging upwards of 10,000 keystrokes per hour. Some have elected to work at standing desks, a benefit associated with an intensive ergonomics program won by their union. To speed up the information exchange between processing plants and the REC, the images are heavily compressed, producing low-resolution depictions of physical objects that are, oftentimes, in far from pristine condition. Torn shipping labels, waterlogged envelopes, and smeared ink are common. This combination makes for an oddly compelling aesthetic, somewhere between the warped scanlines of artist Bruno Munari’s Xerografia and the lo-fi letterforms of mid-aughts reCAPTCHA puzzles.
What scant media coverage the REC has received has almost universally focused on the idiosyncrasies of bad handwriting. The thousands of letters addressed to SANTA, NORTH POLE sent each year by grade-schoolers still honing their penmanship is a recurring motif. One keyer reminisced about envelopes decorated with hand-drawn pin-ups, barbed wire, and skulls that inmates at a local prison used to send, and which he would occasionally receive for keying in the 1990s.
But according to several keyers I spoke with, the lion’s share of the five million items that pass through the REC each day are not handwritten addresses. Contemporary machine-readers, it turns out, can read handwriting with relative ease, leaving keyers to trudge through a bottomless pile of machine-printed detritus, much of it cheaply printed junk mail slung by mass marketeers.
The vast infrastructure required to affix problematic parcels with a packet of human-verified metadata—a thankless clean-up job, performed in the service of machines—is the culmination of a century and a half of technological change. This arc has not been one of linear progress, but rather one of tumult and negotiation, as postal workers from the late nineteenth century onward fought to retain autonomy in the face of encroaching machinery.
The first to arrive was the mechanical “canceller,” a device patented in 1876 by a pair of Boston inventors. The Post Office Department—not yet the United States Postal Service—contracted the production of one hundred cancellers, and allocated them to the nation’s busiest post offices to assist clerks in the slow-going process of manually voiding postage stamps to prevent repeated usage. The contraption’s hand crank rapidly fed letters through a pair of rollers, allowing clerks to cancel fifteen times more postage per hour than was possible by hand. Mechanical cancellers, like many of the technological novelties that would eventually make their way into the post offices, helped to set unprecedented expectations for postal worker productivity.
Innovations like this helped the Post Office keep up with a dizzying uptick in mail volume. Sending and receiving mail had become gradually more accessible during the nineteenth century, as a growing share of the public was now within spitting distance of a local post office, and a new policy offered rural delivery for no additional charge. A new class of senders had also entered the scene. Just a few years before the mechanical canceller was introduced, Montgomery Ward had sent its first mail-order catalog, using the Post’s unparalleled delivery network. Mail volume ballooned by a factor of fifty in the decades leading up to the twentieth century, and showed no signs of slowing down.
The rising demands of commercial mailers put even greater strain on postal operations in the first half of the twentieth century. The cumbersome work of sorting the mail was still performed using the peek-and-poke method: clerks would glance at and then manually deposit each item, one by one, into an appropriate cubbyhole. This outmoded process, developed in the eighteenth century, was plainly incapable of scaling up to the degree necessary. By midcentury, a full nickel’s worth of each six-cent stamp still went toward covering the labor costs of sorting the mail. With competition mounting from private upstarts like the quickly expanding United Parcel Service, the Postmaster General publicly committed the department to a “crash program of modernization and mechanization.”
In 1957, the Transorma, an impressive piece of Dutch engineering that mechanized the peek-and-poke, was brought stateside. Five clerks at a time would take their posts at terminals within the belly of this hulking, fifteen-ton apparatus. Letters were shuttled through the machine’s guts, briefly pausing in front of a clerk who would manually punch in a memorized code pegged to the letter’s destination. The machine would then whisk the letter away to a bin for subsequent processing. This new breed of machine enabled a five-fold increase in sorting productivity, but also gave management greater control over workflows.
By the 1960s, utility bills, catalogs, advertisements, invoices, receipts, and other forms of impersonal, bulk communication had come to account for more than 80 percent of all mail—clogging up the mailstream, but also providing a critical revenue stream. It was clear to postal management that, on their own, hardware innovations like the Transorma would not be enough. Accordingly, in 1963 the Post Office introduced what might be best understood as its first innovation in software: the five-digit ZIP code.
Above all, ZIP codes served as a new standardization protocol, transforming an unruly map into an efficient mosaic. Encoded in each five-digit string was a surfeit of data, helping to direct each parcel through a carefully delineated geographical hierarchy, from regional processing plant down to localized delivery zone. Not only did this numerical logic significantly simplify manual mail sorting, it also greased the skids of mechanization.
Homegrown alternatives to the Transorma were developed throughout the 1960s, designed specifically to take advantage of the new ZIP system, which theoretically enabled faster keying by workers. As installation expanded, sorting machines began to play the part of crucible for a brewing hostility between postal workers and management. By 1968, the Post had purchased 145 Multiple Position Letter Sorting Machines (MPLSM) designed by the Burroughs Corporation, famous for its hand-crank calculators. Unlike the Transorma, which advanced to the next letter only after the clerk had entered a code, the new American-built MPLSM made pacing a point of contention. Workers wanted to be able to advance the sorting machine themselves, letter by letter, allowing for more flexibility and precision in the keying process. Management wanted to program the machine to advance at automatic intervals, maximizing productivity and ensuring predictable throughput.
The distinction between operator-pacing and machine-pacing was the subject of considerable research: Which was more efficient? More sustainable? More cost effective? Consultants were hired to conduct extensive psychophysical studies, monitoring eye movements and keystrokes, fatigue and focus, hoping to determine the optimal balance between speed and accuracy. Despite the findings in these reports, management opted for machine-pacing, disregarding the ample evidence that this would greatly reduce overall efficiency and further degrade working conditions.
The tug-of-war over the MPLSM was not an isolated incident. Grievances over wages and working conditions—facilities were dated and deteriorating, the hours were getting longer, the productivity quotas higher—were piling up in postal facilities across the country. But disputes over the role of technology in particular helped to set the stage for, and ultimately played a starring role in, the most significant reshaping of the Postal Service in its history.
In the spring of 1970, several thousand disgruntled postal workers in New York City walked off the job. Over the next week, they were joined by over 200,000 of their colleagues around the country, forming the largest wildcat strike in American history, and bringing nationwide postal operations to a near standstill. This action put immense pressure on management and congress to come to the bargaining table. The hardscrabble negotiations that ensued between labor, management, and policymakers carved a new route for postal operations—a route leading directly to the REC.
These negotiations resulted in the 1970 Postal Reorganization Act (PRA), signed by President Richard Nixon, which minted the United States Postal Service. It earned postal unions the right to collectively bargain over wages, benefits, and working conditions for the first time, something expressly prohibited for the Post Office Department, which had been a part of the US Cabinet. This apparent win for organized labor, however, came at a cost, as the PRA also cemented a new ideological foundation undergirding all postal operations. It made manifest the decree of a federal commission assembled in 1967 and helmed by an ex-chairman of AT&T, one of the USPS’s key competitors in the private sector: “Today the Post Office is a business.”
The PRA renewed the Post’s commitment to provide “a basic and fundamental service,” but made clear that a balanced budget was of equal—perhaps, greater—importance. But in the eleventh hour of negotiation, an important caveat to this prioritizing of fiscal concerns was hashed out. Congressional representatives had attempted to slip in one last amendment stating that the “Postal Service shall promote modern and efficient operations and should refrain from [any activity] which restricts the use of new equipment or devices.” If accepted, this would significantly erode labor’s bargaining power by letting postal management make unilateral decisions about technological changes. But labor representatives refused to concede on this point. Tech, they maintained, must be bargainable.
After the amendment was struck down, the battles over technology continued with a renewed vigor, centered around the automated equipment that had begun to replace the mechanized machinery of the 1950s and ’60s. The key difference between the two paradigms lay in the question of who—or what—would be responsible for the work of actually reading the address line, a necessary first step before any sorting could commence. The promise of postal automation, which would require delegating the reading to machines, had long been undermined by optical-character recognition (OCR) technology’s failure to deliver on its own promise. OCR developers had a track record of lofty assurances about the efficacy of their machine-reading systems, stretching back to the early patents filed by AT&T in the 1920s. While the fanfare around OCR was clearly overblown, it had created a deluge of commercial research and investment into the technology in the 1950s.
The Post Office Department had begun experimenting with OCR in the late 1960s, but had quickly run aground. Throughout the 1970s, various attempts were made to enhance existing Multiple Position Letter Sorting Machines by integrating an OCR that could identify the bottom-most line of an address and read the five-digit ZIP code. All mail fed into this OCR first had to be examined and presorted by workers on site, as these initial iterations could only read a fraction of the most popular typefaces in use. They were also highly susceptible to paper jams, and handwritten mail remained especially elusive. Consultants determined that until 85 percent of mail could be accurately read by OCR, the mechanized MPLSM—and its waged, unionized operator—would be both more efficient and more cost effective than the automated alternative.
Automation would become viable in Reagan’s 1980s, which were a hotbed of innovation for both postal technology and management-labor relations. The decade opened with an unprecedented showing of strikebreaking force, as Reagan fired over 10,000 air-traffic controllers who were protesting over wages and working conditions. This sent a message to the American Postal Workers Union (APWU) who, weeks prior, had been on the verge of calling for another large-scale strike. Reagan’s airport intervention spooked the postal unions, who cancelled their pending strike authorization vote. The militant worker energy that came to define the 1960s and ’70s, giving rise to the PRA, continued to dissipate throughout the 1980s. With organized labor on the defensive, management saw an opportunity.
In 1982, the high-volume plant in Los Angeles was selected to pilot the Postal Service’s first multiline OCRs, which could automatically read entire addresses, not just ZIP codes like their single-line predecessors. To celebrate, the Postal Service also coined a new employment category. “Mail Processors,” who would monitor these new OCRs, were added to the employment hierarchy two rungs on the payscale below that of MPLSM operators. The APWU filed a grievance, alleging that this constituted unfair labor practices and went against the PRA provision about bargaining over new technology.
Reagan’s notoriously management-friendly National Labor Relations Board sided with the Postal Service, allowing the new employment category to stand. It turned out that the consultants had been wrong: it wasn’t the technical feat of bringing OCR up to 85 percent accuracy that made automation economically viable, but rather the Postal Service’s technocratic insight about how to redefine work and reduce wages.
This was a clever bit of politicking, but it improved neither the literacy rate of machines nor the USPS’s overall service outcomes. Over time, the optimistic narratives of innovation that dominated the 1980s began to wane. Progress was made in OCR, but not quickly enough to keep apace with the steady growth in mail volume. Two congressional reports published in the early nineties captured this sentiment: 1992’s “Automation Is Restraining But Not Reducing Costs” and 1995’s “Automation Is Taking Longer and Producing Less Than Expected.”
Hamstrung by successive waves of neoliberal policymaking, the twin values of service and innovation upon which the Post Office was founded had been rendered incompatible with one another. A regime of unrelenting austerity had motivated and justified a blind faith in the promise of automation—all the while undermining this promise. It was out of this failure that the Remote Encoding Center was born.
Nairn Higginson was a day-one hire at the Salt Lake City REC when it opened for business in 1994. He had responded to a job ad seeking keyers, which described the position as “long-term temporary” and “strictly transitional.” The gig paid more than double the federal minimum wage, and previous experience as a typist or computer technician was required.
Higginson recounted to me how frequently he and his colleagues had been advised by management over the years that their days were soon to be numbered. More than a quarter-century into his tenure at the USPS, he is now the REC’s Manager of Operations. Much remains the same since his days as a keyer, save for one notable difference: the hards have gotten considerably harder. “Those last few percentage points,” another veteran keyer hired in the mid-nineties noted, alluding to lingering OCR error rates, “take years and years and years.”
In the early 1990s, as a last-ditch effort aimed at bolstering the still imperfect machine-reader systems it had so heavily invested in over the decade prior, the USPS began to experiment with new “remote video encoding” technology. Rather than require on-site clerks to deal with each piece of mail rejected by the machine-readers, “remote encoding” provided an off-site, human backstop to augment the OCR systems. If successful, this fix would serve to prop up automated sorting operations until OCR technology had improved enough to finally make the remaining human cogs in the mailstream redundant, once and for all.
The USPS initially opted for part-time subcontractors located outside of expensive metropolitan hubs, rather than full-time career postal workers. Twenty-five of these remote encoding facilities were launched by private firms in 1992, with another two hundred slated to open in the following years. But the APWU intervened, claiming that this public-private subcontracting arrangement was a breach of the union’s collective bargaining agreement. This time, the Clinton-appointed National Labor Relations Board ruled in the union’s favor, bringing all remote encoding operations back in-house.
Salt Lake City was the inaugural outpost, and additional RECs in upstate New York, suburban Arkansas, and formerly industrialized East Pittsburgh soon joined the ranks. Many of the subcontracted workers hired and trained by private firms were recruited to join the unionized workforce of USPS keyers. On the eve of the twenty-first century, there were 30,000 keyers at 55 RECs, keying 25 billion images per year. RECs had been conceived under the pretense of imminent redundancy, but had proven surprisingly resilient.
When Higginson started as a keyer, each REC handled the nonmachinable parcels for only its regional processing plant. (Higginson once keyed a letter addressed to himself, sent by a friend with particularly inscrutable handwriting.) Today, however, the first of its kind is the last still standing: all mail that stumps the OCRs in every state and every territory flows through Utah. The barrios of Puerto Rico send a disproportionate share, several keyers told me, speculating that Postal Service OCRs had not been designed to account for the territory’s unique street address schema.
The long-presaged closures and consolidations had finally begun in the first decade of the 2000s. But the new century mirrored the old: tech had improved since the Postal Reorganization Act era, to be sure, but changes in the political climate were more decisive. The Postal Accountability and Enhancement Act, signed by President Bush in 2006, renewed and deepened the reign of postal austerity instituted by Nixon and Reagan.
Only ever a “temporary fix,” RECs were first on the chopping block. By 2007, all but eight were shuttered, and 80 percent of keyers had been laid off. Closures continued, and some seasoned keyers relocated more than 1,000 miles to continue keying at the final two remaining facilities—now deemed MegaRECs—in Salt Lake City and Wichita. The two, however, became one in 2014, when Wichita was decommissioned, and all national remote encoding operations were consolidated in Utah.
Today, only about one-third of the REC’s 1,500 beige cubicles are occupied at any given time, as budgets have continued to tighten and postal OCR has improved to read more than 99 percent of letters and about 85 percent of packages. As the overall stock of illegible mail continues to shrink, it follows that each remaining item sent to the REC is progressively more degraded and harder to read. “The quality of the images we get sent,” Higginson reported, “is getting worse and worse.” Incremental steps forward in the tech make for incremental steps backward at the REC.
Very Hard Tasks
The story of the REC—and its uncertain future—is a parable for what happens when a robust public service is systematically hollowed out by the dagger of neoliberalism. For the Postal Service, this dagger has often been hidden within the cloak of technological solutionism. But the fully automated mailstream, ever over-promised and under-delivered, may finally be materializing. In late 2019, the USPS announced a partnership with NVIDIA, the leading producer of the powerful computer chips that have catalyzed recent breakthroughs in artificial intelligence.
According to the press release, sorting machinery in more than two hundred USPS facilities will soon be enhanced with AI models designed by NVIDIA. The company’s image recognition systems have advanced so significantly in the past decade that they can now reliably distinguish a white wolf from a species of large white dog. Reading the mail, one might assume, should be a no-brainer. Indeed, OCR error rates have plummeted thanks to classification models like the ones developed by NVIDIA, and character recognition is now sometimes referred to by researchers as a “solved” problem. Perhaps the notoriously stubborn “last few percentage points”—the ones that have kept the Utah keyers busy for far longer than anyone expected—have finally met their match.
Curiously, however, the origins of these very same AI systems now installed in mail processing plants—known as “deep” neural networks—can be traced back to the USPS. In the late 1980s, a young researcher at AT&T named Yann LeCun began to experiment with neural networks using a dataset provided by the Postal Service. It contained about 9,000 images of individual digits, culled from handwritten ZIP codes. To this day, a modified version of the dataset is ubiquitous in computer science curriculums, serving as a benchmark for handwriting recognition systems. LeCun, now the head of AI at Facebook, expressed his gratitude to the Postal Service in an early published technical paper, citing the “very hard tasks” performed by USPS’s engineering department in preparing this dataset for use by AT&T researchers.
Just a few years before LeCun was given access to the formative ZIP code dataset, the USPS had been forced to acquiesce to AT&T on another front. Despite message volume increasing by a factor of ten in its first two years, the USPS’s ambitious E-COM program—a proto-email system—was discontinued after pushback from the private telecommunications industry. AT&T led the charge, claiming that it was unfair that the “Post Office is being encouraged to provide a kind of service” that “private industry is able to do.” That the Post Office could afford to invest in innovative and promising new technology, even when it was unprofitable, was an outrageous notion. Reagan’s Federal Communications Commission agreed, bringing to an untimely demise what may well have amounted to a publicly owned, state-of-the-art digital communications infrastructure.
The handwriting dataset and the “very hard tasks” required to produce it, as well as the preemptive gutting of E-COM, are just two particularly salient examples of an ongoing transfer of postal resources—both intellectual and infrastructural—from the public to the private sector. But providing a public subsidy for private enterprise was nothing new for the Postal Service. Long before Nixon and Reagan, the Post Office was forced to kowtow to the whims of commercial mailers, offering cut-rate postage and special delivery options. And more recently, the USPS expanded its delivery window to include Sundays for the first time in its history—except Sundays were reserved exclusively for Amazon packages, which provided a much-needed revenue stream for the fiscally precarious Postal Service. “We deliver Amazon packages until we drop dead,” read the headline of a 2018 USPS letter-carrier tell-all. Amazon, in effect, built its empire off the back of public postal infrastructure, all the while scaling up its own massive logistics operation, staffed not by public sector employees, but contractors and subcontractors.
Decades of austerity have driven the Postal Service into a state of submission, depending for its continued sustenance on the goodwill of the same private companies it helped get off the ground. Despite this, the USPS continues to serve as a keystone of democracy. Last fall, it executed an unprecedented vote-by-mail operation under the tutelage of a hostile Postmaster General. It continues to serve as a vital conduit of information for the more than 20 million Americans without internet access.
As postal history demonstrates, time and again, the work of overcoming hard problems like these is rarely just a technical achievement. After all, universal mail service means that the mail must be delivered, hardness notwithstanding, whether by humans, machines, or some combination thereof. And until the fully automated mailstream becomes a reality, some of the hardest problems facing the Postal Service will be solved by keyers at the Salt Lake City REC, five million times per day, one billion times per year.