1 Introduction

Technical failure remains one of the biggest threats to the legitimacy of technical organizations. There is nothing like a dramatic explosion, meltdown, or major service outage to put egg on the face of the organizations that administer these systems, undermining public confidence or sinking stock valuations accordingly (Chai et al., 2022; Downer, 2011; Perrow, 1999; Schmid, 2011). In addition to the safety, environmental, and other risk concerns involved, most sociotechnical organizations feel the imperative to minimize technical failure and thereby avoid the ensuing crisis of legitimacy (Chai et al., 2022; Lounsbury & Glynn, 2001). A wide range of literature on the causes and consequences of technical failure and on the appropriate management of risks serves as their guide. Famous analyses of disasters such as the Challenger, Columbia, Three Mile Island, or Fukushima generate cautionary tales alongside mechanisms for assessing risk (Perrow, 1999; Roberts, 1990; Sagan, 1995; Vaughan, 1996). Following certain dramatic failures, organizations take pains to cast problems as learning opportunities, rhetorically orienting audiences toward their continued credibility (Downer, 2007; Garud et al., 2014).

But a dramatic system collapse is only one way to delegitimate a technical organization. This paper describes another process, which we call The Resource Bind. We draw attention to an inherent duality in the symbolic meaning of the resources necessary to support a public-sector technical organization: both to expend judiciously as buffer against risk of failure, and to conserve in a strict audit culture. We show how, in a concerted effort to avoid technical failure, these resources (and their symbolic meanings) come into inherent and unavoidable conflict. In some cases, actors may withhold resources strategically from the organization, with the end result of reconfiguring a team in crisis and placing experts under strain. As members of these sociotechnical organizations confront the specter of technical failure while seeking to execute a public mandate, the impossibility of reconciling the competing meanings of their essential resources places them in an untenable bind.

Empirically, we reveal what happens to risk mitigation cultures and practices under the conditions of extreme audit culture. When a culture of “efficiency” and “austerity” dominates, the symbolic qualities of “more money” and “more time” are fraught with constraints. Our theoretical question concerns the causal order of technical failure and legitimacy crisis, particularly in situations where resources can be withheld or politicized to the strategic benefit of macro- and micro-political players. Prior studies listed above investigate legitimacy failures that follow technical failure, assuming that both are to be studiously avoided. In this paper we show how legitimacy failure and technical failure may not be arranged as expected under certain circumstances. For example, fear of technical failure may force a crisis of legitimacy absent technical catastrophes. This can then undermine technical organizations before their system is even launched. In other words, a legitimacy crisis can trigger technical failure rather than the other way around.

Our focus is two government agencies facing the starvation of resources that, for these technical actors, are essential ingredients for buffering against technical risk: time and money. While the dynamics we describe need not be limited to public sector organizations, these are fertile sites for investigating the Resource Bind in action. We describe two moments when powerful actors pushed these two organizations — NASA and the Census Bureau — toward what was perceived on the ground as assured technical failure. The goal was not to produce a misfired product, a meltdown, or an explosive accident. Instead, those actors who controlled the resources in the institutional field knowingly triggered a crisis of institutional credibility (DiMaggio, 1988). The resulting loss of political capital undermined a component of the administrative state and reconfigured a project and its players toward alternative aims. The story we tell is one where failure can be achieved not through a spectacular loss, but by pushing a technical organization to the brink via other means.

2 Background and Contribution

3 Ethnographic Setting and Methods

To illustrate this in action, we discuss cases gathered by our two authors, each of whom conducted ethnographic fieldwork within a U.S. federal scientific agency. First, we draw on boyd’s study of the 2020 U.S. Census. After completing background documentation, security checks, and trainings, and following two years of observing census planning preparations, boyd received permission to conduct research alongside civil servants: first on-site starting in 2019 at the Census Bureau Headquarters in Suitland, Maryland and then virtually in March of 2020 due to the Covid-19 pandemic. She had permission to attend meetings involving senior civil servants, project teams, and external coordination. boyd’s initial involvement in what qualified as “Year 7” of preparations enabled her to build relationships with both bureau staff and external stakeholders in advance of the intense phase of the census. The study involved over one thousand hours of direct observation of Census Bureau team meetings, as well as hundreds of one-on-one conversations with government employees and stakeholders. To supplement these observations, she also analyzed written communications and organizational documents and conducted 83 semi-structured interviews with those directly involved in the 2020 census. Fieldwork concluded at the end of 2022.

Next, we draw on Vertesi’s study of NASA’s mission to Jupiter’s moon, Europa Clipper. This was largely conducted at the Jet Propulsion Laboratory (JPL) in Pasadena, California: the primary contracting center in charge of building the spacecraft and coordinating among contracted component and instrumentation providers. Principal Investigators (PIs) led the planning and construction of scientific instruments at multiple scientific institutions, with their science team distributed across universities and research institutes in the United States. Beginning in 2011, Vertesi attended planning meetings for this prospective mission, which was eventually approved for funding and development in 2015. For 14 months between the summer of 2016 and fall of 2017, she had badged access to JPL and a carrell with a desk in the mission planning center, where they attended meetings and other mission activities regularly. She also flew to mission meetings or scientific conferences located in other institutions, visited affiliated institutions affiliated, and participated in online meetings related to mission development until the summer of 2019. Findings are based on meeting observation and attendance, backstage chatter with mission personnel, 40 semi-structured interviews, daily recorded fieldnotes, and numerous conversations, memos, and jottings.

These ethnographies took place across the Obama, Trump, and Biden Administrations, a period of considerable fiscal and political uncertainty for public sector projects in the United States. The bureaucrats and scientists employed in both sites are well accustomed to political shifts. Throughout their careers, they have been consistently squeezed to do more with less, even when such efforts render their agencies and efforts brittle. Yet the heightened austerity prompted by the financial collapse of 2009, a non-traditional approach to the management of government agencies led by Donald Trump, and the Covid-19 pandemic from 2020–2022 meant that the work of procuring time and money while maintaining legitimacy became increasingly precarious. Under the resource bind, the very tools that sociotechnical organizations might use to reduce risk were not external symbols of resilience and responsible risk management so much as indications of agency inefficiency and waste. In both sites, the ethnographers watched as our informants scrambled to hold their projects together while being edged off a cliff.

4 Case 1. Scheduling a Census

For the U.S. Census Bureau, time limits are absolute, predetermined, and statutorily dependent. The census’ temporal rhythm is enshrined in the Constitution and formalized through statute (Anderson, 2015). First completed in 1790, the census takes place every decade. Initially, Congress set the schedule anew each decade, but the Reapportionment Act of 1929 and Title 13 of the U.S. Code in 1954 codified a precise schedule defining when data are to be collected and delivered. By law, April 1 is “National Census Day”: the day by which enumeration must officially begin unless there is “reason of climatic or other conditions which would materially interfere.” Law also dictates that data representing the tabulation of the population must be delivered to the Secretary of Commerce (and thus the President) within eight months of the official April 1^st start date so as to reapportion the House of Representatives. For almost 100 years, the Census Bureau dutifully delivered data by December 31^st every 10^th year of each cycle.

Civil servants at the Census Bureau, who never consider delaying their statutory deadlines, wrap their lives and plans around executing the schedule accordingly. In an effort to reduce temporal risk, the Census Bureau strategically builds temporal buffer into its plans. Years before the census begins, the distributed units submit time estimates to planners tasked with creating a master schedule. Managers and other leaders add a certain degree of additional time as buffer along the way. Congress, meanwhile, chastises bureaucrats for these temporal buffers, lamenting the lack of efficiency. Under the resource bind, this temporal buffer is seen as a tool of resilience to the executing party, and a product of wastefulness to its stakeholders.

The 2020 census operational plan had significant temporal buffer built into its original schedule, but, in the end, it was not enough. The practical work of conducting a census during a global pandemic created significant delays. Political fights over and legal micromanagement of the schedule added additional complexity. And just as the census was shifting into full gear in March 2020, municipalities around the country began issuing lockdown orders. The bureau announced that it was pausing its field-based data collection operations. The team began revising the operational schedule and procedures to account for stay-at-home orders. The most important factor in shaping their re-plan was the statutorily imposed deadline for delivering the data essential for the Constitutional mandate of reapportionment.

To maintain temporal buffer, the Census Bureau immediately sought an extension of said deadline to account for delays in data collection and process changes that would become necessary to account for disruptions introduced by the pandemic. Only Congress has the ability to officially grant an extension. The Secretary of Commerce assured civil servants that the White House was supportive and was working with Congressional liaisons to get a formal extension. Confident in the political machinery, the Census Bureau published a revised operational plan predicated upon delivering the first apportionment product in April 2021 — four months late. The bureau redirected the timelines of hundreds of thousands of workers and thousands of processes.

But Congress never passed a bill to extend the deadline. Then, in July 2020, the White House announced that President Trump would veto any bill that delayed the delivery of the census data. The Secretary of Commerce told the Census Bureau to rework its schedule (again) in order to deliver the data by the statutory deadline of December 31. The bureau announced this replan on August 3. The new plan shortened the length of field operations and post-data collection processing, but the primary cuts were to the temporal buffer.

Internally, this was called “the happy path” because there was no longer any space in the schedule for things to go wrong. But many things went wrong. Disruptions due to the pandemic were ongoing, the 2020 hurricane and wildfire seasons upended fieldwork, and efforts to collect data about college students were regularly stymied. This alone was enough to undermine the “happy path.” Then the lawsuits began. A coalition of civil rights groups, local governments, and Tribal Nations sued the Census Bureau over its new schedule. Not only did the bureau need to devote significant resources to respond to the court case, but judges kept issuing orders that altered the schedule and workloads in ways that upended the system of people and processes. Eventually, the Supreme Court ruled that it had no standing to overturn the statutory deadline; this was up to Congress. The Census Bureau closed field operations two weeks later than its own estimate of the latest possible date in which data could be processed and delivered on time.

For the Census Bureau, a temporal buffer was essential to accommodate conditions that drastically impeded their work. The schedule delays directly affected the production work. At the same time, the political spectacle and legal fight challenged the legitimacy of the bureau’s work and the reputation of those involved in the system. Outside of public scrutiny, the bureaucrats focused on accommodating new threats (e.g., people being encouraged to fill out the census for others to avoid undercounts). However, they lacked the capacity and political savvy to curb the legitimacy attacks that emerged as external actors began to proclaim loudly that the census was fatally flawed.

As enumerators’ jobs ended, public attention to the census waned; those who needed the data were simply left with an impression that the quality was dreadful. For the internal team however, the hardest technical work was just beginning. Data processing begins when the data collection ends. This involves a network of teams who must coordinate to process, evaluate, link, merge, edit, de-duplicate, verify, and finalize the data. Each tightly coupled step involves a series of technical, procedural, and human dependencies, and relies on the data existing in an exact format. Data processing was initially projected to take five months, adjusted to six months (a.k.a. 182 days) under the extended Covid-19 plan to account for the pandemic. By the time that the courts had finished ruling on their schedule, the Census Bureau was expected to execute these tasks in 77 days. With no relief in the statutory deadline, the bureau was pushed to the brink. Executives concluded that they could do it in 84 days if every procedure was tightened as much as possible, and nothing else went wrong.

Things went wrong. Almost immediately, “technical anomalies” were discovered. The team had expected these; they created temporal buffer to accommodate them. After all, due to lack of funding earlier in the decade, the team was unable to test all of the processing tools. The team didn’t expect that dormant code brought back into view after a decade would run exactly as planned with new data streaming into the infrastructure for the first time in a decade. The “happy path” had no room for these technical anomalies. Temporally “in the red” due to Covid-related delays, under political pressure to speed up the work, and subject to legal dynamics where judges micromanaged the schedule, the bureau’s production teams saw disaster on the horizon even as they worked long hours to resolve emergent problems in situ.

No one inside or outside the Census Bureau was certain what would happen if the census data were not delivered by the statutory deadline of December 31, because it had never happened before. Consistent with the resource bind, we observed a combination of honor and professional responsibility that drove these civil servants to see the statutory deadlines as inflexible. However important the statutory deadline is, the identity that anchors the bureau’s statisticians is a devout commitment to data quality. As the situation was unfolding, numerous civil servants discussed quitting if they would be asked to publish shoddy data. Certainly, no census data are perfect, and the hard deadline has historically enabled a cultural process by which “good enough” does not get in the way of perfection. But the dynamics of the 2020 census positioned time and quality as oppositional goals. To complicate matters, the court cases had made visible how the bureau’s baseline for quality permitted imperfection in ways that outraged external stakeholders who expected unachievable perfection. This helped fuel an external legitimacy crisis.

After working around the clock for over a month and repeatedly encountering technical anomalies, civil servants decided to warn the White House that they would need 20 extra days. Only a week later, an additional problem was detected that would require even more time. Civil servants understood the political ramifications of this — and they were anxious about how the White House might respond. In short, the delays would mean that the data would be delivered to incoming President Biden instead of outgoing President Trump. Many executives at the bureau had been lambasted for being insubordinate; they feared being fired for doing their job. Representatives of the Trump Administration repeatedly told senior leaders that it was their job to deliver the data before the President left office on January 20, no matter what. The goal post had shifted — it was no longer about the statutory deadline, but the political one.

Civil servants did not try to delay the delivery of the data. To the contrary, they worked nights, weekends, and holidays trying to achieve what they understood to be impossible. In interviews afterwards, they described their fears and anxieties of failing to meet the expectations of the president. Senior leaders also recounted the political threats they faced from members of the Trump Administration along the way; they would publicly articulate some of them to journalists months later (Bazelon & Wines, 2021).

The situation came to a head on January 5, 2021 when the Director of the Census Bureau, a political appointment, requested that the Census Bureau deliver incomplete data. Civil servants were shocked by this request — which they saw as an attack on their professional responsibility — and used the whistleblowing processes to report malfeasance to the Office of the Inspector General (OIG). The next day’s insurrection at the Capitol made the issue moot; the Trump Administration shifted its focus elsewhere. However, two weeks later, the OIG issued a report; short afterwards, the Director chose to step down before his term was over.

Congress never changed the statutory deadline, but President Biden issued an Executive Order on the first day of his term that explicitly stated that high quality data was more important than the statutory deadline. This did not provide statutory relief, but it gave the Census Bureau political air cover. The bureau offered a new schedule based on where it was at with its data processing and quality checks. No one challenged this new schedule. The first tranche of data was delivered in April 2021, almost exactly as the initial Covid-19 replan predicted. However, the costs of the resource bind to internal morale, the quality of the data, and the reputation of the bureau far exceeded projections.

Bureau employees recognized early on that the Covid-19 pandemic would challenge their schedule, even with the temporal buffer that they had baked into the plan. Their temporal buffer was designed to guard against a range of unpredictable technical delays, including a pandemic. But it was no match for a resource bind. Litigation that stemmed from and contributed to the legitimacy crisis while also further limiting its buffering resource. As the court cases unfolded, journalists and the public — not to mention politicians, data users, and civil rights advocates — began questioning the validity of the data while attacking the reputation of civil servants. Civil servants felt attacked on all sides, even as they were desperately focusing on preventing a technical meltdown. The Census Bureau delivered its key product, but the legitimacy and civil authority of the operation was left in tatters.

5 Case 2. The Costs of Clipper

The census case shows how time can be subject to a resource bind. On NASA’s Europa Clipper mission, we observed similar dynamics around money — articulated as “budget” or “cost”. Sending satellites into near-Earth orbit is expensive enough, but sending spacecraft into the outer solar system places additional demands on propulsion, navigation, radiation shielding, communications, autonomous systems, and scientific instrumentation. Efforts to get past the asteroid belt on the cheap fail to do so for less than a billion dollars in today’s money, and each mission has unique requirements, posing challenges for an economy of scale. Historically speaking, NASA regularly faces the challenge of achieving its moonshot technical and scientific mandates while staying under cost projections at the same time.

Amid this background, the Clipper mission was born in a period of fiscal instability. Plans for investigating Jupiter’s icy, water-world moon were conceived in the mid-1990s and a small group of dedicated scientists returned to these plans in the early 2000s. A Europa expedition was one of two missions placed into a “finalist” category to be considered for the next big-ticket mission to the outer solar system. At point of selection the mission was projected to cost in the mid-$2bn range (excluding launch vehicle cost), although this was contested by an independent cost review which put the price-tag over $4bn. When the market crashed in late 2008 and a new fiscal reality settled over Washington, officials rescinded the green light and insisted the team go back to the drawing board. A sample return mission to Mars was their top priority, the first portion of which would be flown as the Perseverance rover in 2020. The Europa mission could only fly if it could be redesigned for less. NASA even solicited proposals for a mission to Europa that could be flown for under a billion dollars.

An alliance between the Jet Propulsion Laboratory (JPL)’s director and an enthusiast Congressman from Texas kept the project alive (Brown, 2021). The Congressman wrote the mission’s funding into a bill, requiring its launch in addition to the Mars Rover. This seemed a win for JPL: now they had two large missions in house, which meant money to keep the lights on and their experts employed. But at each request from Congress and NASA Headquarters, scientists and engineers had to squeeze their needs into an ever-smaller fiscal package while still meeting the technical challenges of getting to Jupiter, surviving its harsh radiation belt, and achieving basic science goals. JPL renamed their mission “Clipper” to recall the small, inexpensive shipping sailboats of the 19^th century, communicating their commitment to austerity. Meanwhile, other scientists and aerospace contractors complained that JPL was eating everyone else’s lunch and bitterly resented that — despite top billing in two decadal surveys in a row — in their view, the mission was moving forward thanks to a lab director’s professional relationship with a Congressman (an arrangement not unheard of in NASA history).

Mission planners experienced their resource bind as mixed messaging about expenditures in an environment focused on austerity. The “science definition team” who recommends a suite of detectors for a nascent craft recommended a payload of six instruments. But NASA officials gave the green light to nine instruments to satisfy top level scientific goals, with no concomitant increase in cost agreements. Any instrument that went over its budget would be threatened with removal. Selected team members called this a “Survivor” approach to science, referring to a television series in which individuals who are unable to keep up with a demanding set of activities — and maintain good relations with audience members throughout — are voted off the show. Soon, instrument PIs were exhorted to cut costs by cutting back on the size and scope of their science teams. But one PI, whose instrument was inexpensive precisely because they had few collaborators, was confused when asked to add more scientists to their team to assuage risks of insufficient staffing and expertise.

Practical attempts to maintain the mission and constrain costs resulted in overages. For example, the Congressional bill required that the mission be designed to accommodate one of three potential launch vehicles, promoting fierce competition especially between SpaceX’s Falcon Heavy and NASA’s Space Launch System (SLS). Neither had yet been completed, designed, or tested, but all Clipper technical design decisions had to be made in triplicate to fit each rocket’s unique payload faring and to satisfy this political requirement. This added cost and complexity. In another instance, NASA required all PIs to use specific cost estimation software in their proposals to standardize and produce trustworthy estimations. Several PIs confessed in interviews that they did not think this system adequately represented the cost of their instrument, with one admitting in an interview to purposefully underbidding to win the contract. NASA also required all proposals to include a preliminary statement of cost from contracting companies to fix the budget up front. But PIs found that contractors’ costs inflated from “projected” to “actuals” once their instrument was selected. By the time of its inception, the mission was already projected to run tens of millions of dollars over its projected cost.

Certain practices that emerged to contain costs jeopardized other elements of mission success. For instance, PIs were asked to meet early during planning to consider giving up certain aspects of their instruments, to offer elements up for redesign or review, to avoid full instrument removal. The project was also asked to track both cost and cost risk: a novel measure of how likely it was that an instrument’s cost would exceed its budget. The administrator at NASA Headquarters required PIs to agree to both a “cost cap” and a “cost trigger process.” If an instrument’s risk of going over their cap appeared to increase, this would “trigger” a series of reviews and oversight from Headquarters to produce course correction. Now both cost-actuals and cost-risk needed to be measured, monitored, and contained, and PIs needed to demonstrate that both were under control. There was no explicit discussion of whether the additional overhead of these imposed reviews or tracking would increase costs by increasing personnel hours.

Power played a key role in the surveillance of resources subject to this bind. Frank budget discussions were generally taboo. Numbers related to projected overruns were kept secret and fluctuated with some frequency. The Project Manager (PM) held regular meetings with each PI to go over their current and planned expenditures. Mistrust abounded as PIs came to believe that the PM kept his own budget indicating what he thought their instrument should cost, judging their requests or reports against his “shadow budget” to determine their risks of cost increase. They feared this made it likely they would be voted off the island.

Like the Census Bureau’s temporal buffer, NASA PMs are given some money to put aside, called “reserves,” to resolve problems that emerge during development without having to return to Congress to ask for more funds. Clipper PIs believed this money was supposed to be earmarked for changes leveraged upon their instrumentation due to design changes made by the central project office. But the Project Manager was disincentivized to release “reserves” to instrument PIs. Since “reserves” were a fixed amount to use throughout the lifetime of mission development, he did not want to spend them too early. Still, the mission’s three female PIs observed in interviews that the male PM was more likely to affably offer flexibility to some of their male PI colleagues with whom he had a prior relationship.

The resource bind was experienced as a dramatic tension in which PIs were held to the weighty requirement to demonstrate control over cost, while at the same time having little to no control over whether or not cost escalated in reality. More than one PI interviewed was reduced to tears or bursts of anger expressing that they had done everything they were supposed to do, they were on track for instrument testing, scope, and maturity, but due to factors outside their control their feet were being held to the fire. They were justifiably concerned if rumors about their costing woes would reach various higher up officials and affect their reputations, impacting the future viability of their instrument, science or leadership. The PIs were not the only ones under stress: the lab’s legitimacy was also on the line, stretched as it was for personnel under the development of two major missions at the same time.

The resource bind also offered certain actors the opportunity for strategic restructuring. In 2018, the above-mentioned “cost trigger process” was set off by the least expensive instrument, the magnetometer. To ensure correct readings of the magnetic field, this instrument is typically placed on a boom that extends from the body of the spacecraft. When the team switched from nuclear to solar power early in development, this had downstream effects on the heating of the boom and its cabling that impacted the magnetometer. Many ensuing problems could have been solved with a different kind of fiber-optic cable, directing reserves for procurement and putting additional experts on the problem. But under the resource bind, money was not forthcoming. The PM did not give out project reserves to the instrument PI. Her institution was already strapped for expert personnel due to the Mars mission under development. When other instrument PIs offered their own reserves to help procure the cable for her or find a cheaper version, the PM forbid it: they would each need their own reserves in case of future problems.

Other PIs leapt to her defense. This was disproportionate and unjust, they claimed. Since cost risk was a percentage measure pitting overrun against projections, an instrument pitched at $17 million incurring an additional $2 million looked like a large increase of cost risk on the chart, while a projected-$70 million instrument run by a male PI which regularly faced increases of far more than $2 million had not “hit” its trigger. But which cost mattered? The budget she had pitched was inflated at point of selection, and this differed again from her cost trigger point and projected cost including reserves. These many possible numbers made the problem and its cause open to interpretation. A series of reviews over several months saw the PI flown from California to Washington DC to present the ongoing situation to higher ups at NASA. In a few months, a PI at the same lab developing a Mars Rover instrument hit his trigger too and was called to reviews. It was whispered at the lab that Headquarters was determined to teach the lab a lesson about persistent cost overages. The question was which mission would bear the brunt.

A few months after the Europa-friendly Congressman lost his re-election in November of 2018, the magnetometer PI faced her final review. In an interview, she reported feeling good about her chances. She had figured out a way through the problem, had “constrained” (i.e. charted and figured out) the costs involved, negotiated a discount on the cable, and felt confident about the way forward. She hoped to convey that as PI she was taking full responsibility for her instrument’s new plan and had both cost and cost risk under control. Her superiors, meanwhile, took a less optimistic view of the numbers. She was removed as PI, her instrument was downgraded, and authority over its construction was assumed by the PM.

Just as the Census schedule crisis reconfigured political actors, this moment reconfigured the field. The revised magnetometer had only basic capabilities, compromising Clipper’s ability to deliver on the science goals that had sold it in the first place and leaving Europa scientists scrambling to figure out how to recover this essential data. The PM on Clipper kept his reserves and took further control over spacecraft redesign. Mars mission members interpreted the loss of the Clipper instrument as a shot across their own bow: fortunately for them, Perseverance cleared their review and kept their instrument. Headquarters representatives pointed to the magnetometer example as a lesson to their grantees to treat costs seriously. The descoped PI, meanwhile, faced reduced funding and loss of opportunities for the next several years in the scientific community. Five years later, the price of the replacement magnetometer is estimated at $80 million. As Lynne Zucker frequently reminds us (i.e. see Meyer & Zucker, 1989), the legitimacy game afoot may not have been about actual cost containment, even if its performance was important for the continuing status of its players.

Under the resource bind, demands for cost constraint repeatedly came into conflict with ongoing mechanisms for cost and risk control, stoking rising prices and rising blood pressure. This tension was especially keenly felt among PIs, who were under pressure to demonstrate efficiencies and savings while subject to rules that increased costs at the same time. PIs’ resulting perceived inability to manage budgets triggered legitimacy crises in the absence of any technical failure. Meanwhile, the bind presented opportunities for enacting power and restructuring towards its centralization. Micro-political players deployed personnel, reviews, “triggers” and other mechanisms to force individuals and institutions into a corner, preserving certain players at the cost of others. This reconfigured the mission and its capabilities, not the least through the reconfiguration of a key (and initially inexpensive) science instrument. Small wonder that for the scientists exploring Europa, orbital dynamics often appeared to be more accommodating than these otherwise intractable, unpredictable, and personally devastating interactions about cost.

6 Discussion

Neither project failed from a technical standpoint. The Census Bureau delivered a census, and Clipper is on track to launch in late 2024. However, near-failure looms large in both stories, alongside struggles to negotiate threats to legitimacy and play a broader array of games that have little to do with technical success. The participants we observed operated frenetically at what they saw as the brink of twin catastrophe: the potential inability to deliver on their technical and scientific requirements, and the potential inability to deliver on schedule or on budget. Caught in their respective resource binds, they experienced overwhelming pressure to fulfill both needs at the same time. As they labored to ensure both operational and organizational credibility, they confronted ever-present threats to their personal reputations, to the validity of their scientific findings, and to their institution’s status. They repeatedly experienced a feeling of being the proverbial frog in the slowly boiling water, aware that everything was on the brink of collapse, desperate to avoid it, and unable to avoid the increase in temperature. We believe that these conditions suggest new ways forward for linking the literatures on failure and institutional analysis in sociotechnical organizations.

7 Conclusion

Scholars of sociotechnical systems would do well to consider the institutional fields in which their organizations operate, the competing cultural valences of relevant resources, the political characteristics of restrained resource management, and the contradictory mechanisms of legitimacy between, in our case, technical authority and audit cultures. Examining situations where micro- or macro- actors push toward failure through the withholding of sociomaterial resources — and imbue requests for such resources with negative qualities of wastefulness and inefficiency — shows how these organizations end up with compromised systems and egg on their faces to boot. Such legitimacy crises ultimately impact an organization’s opportunity to appeal for further buffering resources in the management of their continuing sociotechnical systems. Continued funding and schedule opportunities are on the line should an institution face delegitimization in the view of political actors who define the system’s constraints.

Bringing an institutional eye to high-risk systems-in-formation during this historical moment in public sector work reveals previously unexposed factors that push toward failure in multiple ways. Decoupling technical failures from legitimacy failures enables us to observe those moments when systems are placed in an impossible bind. When an organization’s reputation bears the brunt of crises as fiscal and operational efforts are squeezed, this rearrangement may be to the benefit of a competing organization, or an agency with oversight situated among a broader institutional field of players. We suggest that legitimacy crises among sociotechnical organizations are far more commonplace and even politically expedient than we would otherwise assume, and do not always occur as a result of technical loss.

In many ways, legitimacy crisis without technical failure may appear easier to the public eye. The census did not fail at collecting data, and nothing blew up on the launchpad. There are also elements of these organizations that are, admittedly, uniquely extreme. For example, it is not possible to reduce risk through radically reducing the scope or scale of the product (Brown & Jones, 1998; Light, 2014; Miles & Snow, 1992) when these are federally mandated. Further, there was nothing entirely exogenous about the impacts in the cases we observed, as the scientists and civil servants we witnessed walked onto the floor of Congressional hearings to account for their activities and petition for resources directly. Given the active and highly sociopolitical mechanisms we witnessed, we are hesitant to view these stories in the light of ‘mere’ resource-dependency and more as a question of how actors manage competing logics associated with those resources in institutional fields. We witnessed how keeping up appearances and managing these competing forms of legitimacy produced continuing cascades of risks to sociotechnical system design. Even if a system manages to scrape by or avoid catastrophe, its sociotechnical organization accumulates blame and further resource loss over the “mismanagement” of their system. This makes their mission quite simply impossible.