Unbundling Freedom in the Sharing Economy

Deepa Das Acevedo[*]

Courts and scholars point to the sharing economy as proof that our labor and employment infrastructure is obsolete because it rests on a narrow and outmoded idea that only workers subjected to direct, personalized control by their employers need work-related protections and benefits. Since they diagnose the problem as being our system’s emphasis on control, these critics have long called for reducing or eliminating the primacy of the “control test” in classifying workers as either protected employees or unprotected independent contractors. Despite these persistent criticisms, however, the concept of control has been remarkably sticky in scholarly and judicial circles.

This Article argues that critics have misdiagnosed the reason why the control test is an unsatisfying method of classifying workers and dispensing work-related safeguards. Control-based analysis is faulty because it only captures one of the two conflicting ways in which workers, scholars, and decisionmakers think about freedom at work. One of these ways, freedom–as–non-interference, is adequately captured by the control test. The other, freedom–as–non-domination, is not. The tension between these two conceptions of freedom, both deeply entrenched in American culture, explains why the concept of control has been both “faulty” and “sticky” when it comes to worker classification.

Drawing on a first-of-its-kind body of ethnographic fieldwork among workers and policymakers across several sharing economy industries, this Article begins by showing how workers themselves conceptualize freedom as both non-interference and non-domination. It then goes on to show that both these conceptualizations of freedom also exist in case law and statutory law pertaining to work. In doing so, the Article demonstrates that there is no great divide between work law and work practices and that, if anything, the problem is that classification doctrine reflects and reinforces an irresolvable tension in the way lay and legal actors think about freedom at work.

Introduction

In 2015, the Northern District of California issued one of the most widely cited opinions on employment regulation in the sharing economy.[1] The dispute in Cotter v. Lyft was a relatively standard one—whether a company had misclassified a particular set of workers as independent contractors when in fact they ought to have been employees[2]—and the substance of the court’s analysis and ruling were similarly routine.[3] What made the Cotter opinion stand out, aside from its efforts to grapple with an unfamiliar economic space, was Judge Chhabria’s snappy summary of a widely recognized problem in worker classification law. Because the standard “control” based test for worker classification suggested that Lyft both did and did not control the drivers who operate on its platform (and that consequently the drivers could be either employees or independent contractors) “the jury,” Judge Chhabria observed in closing, “will be handed a square peg and asked to choose between two round holes.”[4]

The common law control test purports to distinguish employees from independent contractors on the grounds that employees enjoy less freedom in the “manner and means” of their work and thus merit a host of work-related safeguards.[5] Judge Chhabria argued that the misfit between legal categories and work practices stemmed from the fact that a “20th Century” test was being applied to a “21st Century problem” like the sharing economy[6]—but in truth, the control test seems to have always suffered badly from cubism­. Before Lyft drivers, for instance, there were FedEx drivers whose facial hair and sock color were dictated by FedEx and McDonald’s workers whose speech and hand motions were mandated by McDonald’s, all of whom were sometimes found to be independent contractors (and sometimes not).[7] Indeed, for virtually the entirety of its existence, the control test has proven unsatisfactory to courts and scholars—and, at least once, to Congress—because it is inefficient and often seems inaccurate.[8] But if control-based analysis is so deeply problematic, why has it also been so tough to get away from?

The concept of control has proven both “faulty” and “sticky” when it comes to worker classification because it captures one—but only one—of the two conflicting ways in which courts, scholars, and workers themselves think about freedom at work. One of these ways, which is more prominent as well as adequately captured by the control test’s “manner and means” analysis, is a classically liberal understanding of freedom as non-interference.[9] The second, less influential (but still widely visible) model is a thicker vision of freedom as non-domination.[10] These competing conceptions shine through with striking clarity in the sharing economy, but they are also apparent in other business contexts and clearly discernible from a sky–view analysis of our labor and employment infrastructure. The tension between these two visions of freedom is why scholarship and jurisprudence on worker classification has been filled with criticisms of the control test, yet unable to meaningfully move beyond it.

This paper makes two contributions to legal scholarship. First, it speaks to labor and employment law scholars by showing that control-based worker classifications, problematic as they may be, are linked to a particular vision of individual autonomy that is very compelling in America. While the consequences of courts’ reliance on control often seem perverse (as when innumerable workers are denied employee status because they are not directly or sufficiently controlled by their respective employers), the conception of personal freedom behind that analysis demands serious and careful treatment. Critics of the current classification regime do themselves no favors by trying to eliminate, supplant, or declaw control-based analysis. This is not so much because doing so constitutes the usual mistake involving babies and bathwater, but because critics fail to recognize that the baby and the bathwater are in some ways indistinguishable.

Second, this paper contributes to a broader conversation within the legal academy about the role of qualitative social science in the study of law. As I have argued elsewhere, the kinds of insights gleaned from ethnographic research are different from those facilitated by other empirical (and especially quantitative) forms of social science, but they are hardly incommensurable with the interests or the intellectual values of legal scholars.[11] Here, I use ethnographic fieldwork on the sharing economy as well as legal analysis of our labor and employment infrastructure to reveal the twin conceptions of freedom described above and to show why the tension between them creates problems for employee classification doctrine. The type of “cultivated attentiveness” that makes such a doctrinal critique possible is precisely why ethnographic research is a different but profoundly valuable mode of interdisciplinary legal scholarship.[12]

Part I begins with a brief overview of the origins of our classification system that highlights the centrality of freedom as an analytic rubric. I then detail the stakes of employee status as well as the criticisms that the existing system has provoked. Part II contains the ethnographic heart of the Article. Section II.A uses ethnographic research to demonstrate that workers in the sharing economy sometimes value independent contractor status and associate it with freedom-as-non-interference, while Section II.B shows that sharing economy workers and their advocates rely on a conception of freedom that is more akin to non-domination when they express concern about the lack of autonomy in this type of labor. Part III draws on case law and statutory law to demonstrate that these conflicting visions of freedom also exist in our labor and employment law infrastructure.

I.  Defining Work Relationships

The building blocks of work law are imports from elsewhere: the categories “employer” and “employee” arrived from the law of agency via vicarious liability, while the characterization of the employment relationship itself comes from contract law.[13] Labor and employment scholars have long bemoaned this lack of locally cultivated concepts (particularly as it relates to worker classification) because they believe that it creates a misfit at the foundation of our regulatory infrastructure.[14] And, they argue, the primary cause of that misfit is the importance of control in the allocation of the protections and benefits described below.[15]

But scholars and courts misdiagnose the precise nature of the problem: control is important, to be sure, and that importance derives from classification doctrine’s link with agency law—but control is really just a proxy for measuring worker freedom.[16] As the categories “employee” and “independent contractor” developed over the nineteenth century, placing individual workers into one bucket or the other was meant to reflect a sense that some workers were merely agents of their employers and not really free to act independently.[17] It followed that employers sometimes ought to be responsible for injuries caused or incurred by those workers (“employees”) who relinquished freedom in the performance of their tasks because, after all, it was the employers themselves who had dictated the “manner and means” in which tasks were to be performed.[18] As it turns out, however, “manner and means” analysis only captures one of the ways in which workers, scholars, judges, and even occasionally legislators have thought about what it means to be free at work.

That is where the trouble really lies—in the overlooked complexity of the concept of freedom. Unlike critiques that emphasize the narrowness of control-based analysis or the different goals of agency versus work law, an analysis that focuses on freedom can explain both the faultiness and the stickiness of the control test. Otherwise it becomes mystifying, as indeed it has been to generations of critics, why a test that dispenses the safeguards of employee status as inefficiently and stingily as the control test nonetheless retains such conceptual punch.[19]

A.  What’s in a Name?

The United States funnels an extraordinary range of protections and benefits through work relationships. Moreover, the vast majority of these safeguards, at both federal and state levels, are only available to workers who are categorized as employees.[20] Core safeguards linked to employee status include anti-discrimination and harassment protections based on protected categories like race, religion, sex, national origin, disability, and age (including the duty to accommodate, where applicable);[21] job protections for family and medical leave;[22] equal pay guarantees as between men and women;[23] minimum pay guarantees and rules about when over-time rates of pay should kick in;[24] fiduciary standards regarding health and retirement benefits (as well as the bulk of such benefits themselves);[25] workplace safety protections;[26] and, of course, protections for workers who engage in concerted or union activity.[27] All of these and more hinge on being the right kind of worker for the right kind of employer and even, sometimes, being in the right kind of industry.

To be sure, some of these safeguards, like unionization rights or workplace safety protections, are self-evidently related to work, although not necessarily to employee status. Others, like the imposition of certain fiduciary standards regarding the management of health and retirement benefits, have no necessary connection to work at all—as Americans partly began to experience under the Affordable Care Act.[28] This is not the place to rehash longstanding debates on the wisdom of tying safeguards to work in general; for better or worse, our social safety net is unlikely to become meaningfully detached from the work relationship in the foreseeable future.[29] Rather, it is simply worth noting that the range of benefits tied to work is largely co-extensive with the range of benefits tied to employee status because doing so explains why so many scholars (to say nothing of workers, worker advocates, and governmental actors across branches and jurisdictions) have been vexed by the issue of worker classification: many of the procedural and substantive safeguards that greatly contribute to a decent life are funneled through employee status.[30]

Despite the undeniable importance of worker classification, it is notoriously difficult to determine whether any individual worker is an employee or an independent contractor.[31] Statutes are of little help: several of the most significant federal acts contain delightfully circular language like “[t]he term ‘employee’ means an individual employed by an employer.”[32] In a 1992 case involving the Employee Retirement Income Security Act (ERISA), the Supreme Court held that such circular language reflected congressional reliance on the “common understanding . . . of the difference between an employee and an independent contractor” that in turn mandated judicial reliance on the common law control test.[33] Courts soon extended the holding to statutes beyond ERISA so that now the control test is the default for federal work law protections.[34]

Case law has also been of little help despite the fact that “the real work of identifying ‘employees’ . . . has always been in the courts.”[35] In the course of trying to implement the Fair Labor Standards Act’s (“FLSA”) broader definition of what it means to be an employer—to “suffer or permit to work,” rather than to control the means and manner of performance—some courts developed the “economic realities” test as an alternative to control-based analysis.[36] This new test was meant to expand the scope of analysis by considering workers’ economic dependence on their employers and functional (rather than merely nominal) employer control.[37] But the economic realities test has also come to draw criticism, partly because its multiple factors are open to divergent interpretations and partly because many of those factors bear a curiously strong resemblance to factors that are considered under the common law test.[38]

B.  What’s the Matter with Control?

The two criticisms of the economic realities test mirror the primary complaints about worker classification doctrine more broadly: it offers little guidance and it really always boils down to control-based analysis.[39] That tendency to revert to measuring and weighing control is in turn troubling because the control test seems prone to excluding workers with diminished freedom from the agreed upon suite of employment related safeguards.[40] Sometimes the exclusion simply occurs when workers are labeled independent contractors rather than employees of the companies they work for. This is the case with the decades-long litigation over FedEx delivery drivers, whom the Ninth Circuit described as being subjected to “exquisite” forms of control, notwithstanding FedEx’s claims that they were independent contractors.[41] But critics also fault the control test for narrowly construing who counts as an “employer” and thus absolving companies of their obligations under various statutes. This is the impetus behind efforts to hold franchisors like McDonald’s liable as joint employers of their franchisees’ direct employees.[42]

It is precisely this narrowness, critics argue, that led Congress to abandon the common law definition of “employee” when drafting the FLSA and to instead adopt the wider “suffer or permit” standard used in state child labor laws;[43] that led the Supreme Court in NLRB v. Hearst to argue for a “purposive” reading of the National Labor Relations Act, with its more expansive understanding of “employee” status;[44] and that led various federal courts to embrace an economic realities analysis that accounts for worker dependence.[45] A survey of worker classification literature brings such efforts to minimize or supplant control-based analysis into sharp relief, but it also reveals that courts, regulators, and even scholars continue to think about classification in terms of control.[46] Why?

One set of explanations argues that they keep returning to control because some particular feature of the law or its application pushes them to do so. The “feature” in question is often broad, like a failure to adapt given changing modes of production, or the fact that courts often engage in formalist rather than purposive analyses of employment statutes.[47] It is also often foundational to labor and employment law, like when scholars argue that our troubles arise from the inherent difficulty of distinguishing between the “contracting” and “producing” phases of employment relationships as the law essentially requires us to do.[48] Whatever the cause, the end result is that law fails to accurately regulate labor because it ties employee status to a kind of direct and active interference in worker autonomy.

Because law is the problem in these accounts, law is also the solution. Regulators should have different tests, different defaults, or different interpretive rubrics, so they can more accurately identify control in work relationships. In a pinch, they can draw on other areas of the law—antitrust is an emerging favorite—to mitigate the failings of the specific legal infrastructure governing work.[49] But arguing that some longstanding feature of “law” is responsible for failures to accurately understand and regulate some feature of “society” does not explain why even critics find it so difficult to let go of that feature, and it has the unfortunate side effect of reifying law as a thing that exists apart from and above the social world.

Conversely, another approach has been to say that courts, scholars, and even workers keep returning to control because they embrace the law’s narrow, formalist conception of freedom as non-interference and find it meaningful.[50] For example, some taxi drivers might prefer to be independent contractors because they genuinely feel this classification signals and enables greater autonomy than the category “employee.”[51] Likewise, courts might find independent contractor status to be both an accurate signal and effective source of “entrepreneurial opportunity.”[52] Since this type of argument also posits that legal categories are themselves historically contingent—say, in the way that the independent contractor classification became linked with entrepreneurship and freedom during the heyday of twentieth century neoliberalism—it is especially adept at acknowledging the mutually constitutive nature of law and society.[53] But precisely because it is so good at explaining why legal constructions of control gain social salience as well as how they are socially informed to begin with, this approach tells us little about why generations of scholars, workers, legislators, and judges have felt that control-based analysis just isn’t doing the trick.

In the end, we cannot explain the stickiness of control-based analysis without also accounting for the longstanding sense that it is inadequate. Likewise, we cannot explain control’s shortcomings without accounting for the fact that year after year, in court after court, and even for some of the workers whom it seemingly shortchanges, control and the categories it gives rise to continue to present a compelling vision of what it means to have or lose freedom at work. We can resolve both halves of the puzzle by understanding that our classification system has really been predicated on measuring freedom (not control), and that the common law control test captures one—but only one—of the two conflicting ways in which lay and legal actors think about freedom at work.

Because this is an exercise in chasing complicated concepts, it can be helpful to begin with the lived experiences of actual people and in environments that are relatively transparent. The ethnography that follows combines these advantages to reveal a tension in the way workers think about freedom that, in later sections, can also be seen in work law itself.

II.  Freedom at Work in the Sharing Economy

The “sharing economy” refers to a broad set of companies that use technology to offer products and services in a highly disaggregated, individualized way. Not all sharing economy companies actually present work regulation issues, which is why most scholarly and policy commentary (not to mention judicial and legislative engagement) has been concerned with a limited subset of this space. I refer to those companies that actually pose labor and employment law issues as “platforms” and I distinguish them from others within the broader sharing economy (that I have elsewhere referred to as “renters” and “swappers”) on the grounds that platforms do more than apply new technology to old practices and are more than virtual bulletin boards for third parties.[54] Rather, platforms actively participate in the transactions they give rise to and occasionally substitute themselves for government safeguards.[55] For this reason, when we talk about work regulation in the sharing economy, we are really just talking about platforms.

The ethnography presented in this Article was conducted largely but not exclusively in Philadelphia from 2016 to 2017. I participated in application and training processes for sharing economy companies, worked for a few of those companies, and engaged in distance-education classes for aspiring online workers.[56] I also observed online chat forums and discussion threads catering to platform workers at a national (and sometimes international) level.[57] In addition to these forms of participant-observation, I conducted semi-structured interviews or informal conversations with platform workers, worker advocates, policy analysts, and municipal officials.[58] Lastly, I benefited from the prior efforts of journalists, analysts, industry experts, and advocates who conducted their own research and whose work, whether published for a general audience or directly shared with me, complemented and enriched my own ethnography.[59] One of these interlocutors rightly observed that fine-grained, qualitative data is essential to the task of questioning the cohesive, statistics-based narratives put forward by platforms regarding the true nature of this work and what it means for our regulatory system.[60] This Article uses ethnography to further that goal.

A.  Freedom as Non-Interference

Workers suggest that the freedom afforded by platform labor is of three types: nobody tells me when to work, how to work, or how much I can earn. When stated this way, the autonomy-enhancing potential of the sharing economy is striking, as is the degree to which it contrasts with anything we might legally understand as employment. Although subsequent sections of this Article will show that this view of platform work rests on a narrow conception of freedom, there is no gainsaying the set of choices that platform workers can and do make as well as the real value of being able to describe your work using “I” statements.[61]

Consider Sam, a late 30s African-American man who drives for UberX (one of the company’s lower cost services) and occasionally works as a TaskRabbit tasker.[62] Sam has been driving for Uber ever since he lost his job last year as a technician for a major cable and internet provider, and he recently started doing some light furniture assembly and home repair work via TaskRabbit. He aims to drive around five hours per day and usually does this weekdays from 5–10 a.m. in order to catch the morning hospital and office crowds, and because it allows him to take care of his two children after school while his wife works. He never works weekends.

Driving is not Sam’s passion, but he enjoys chatting with passengers, setting his own schedule, and listening to his favorite music or playing games on his smartphone in between rides. He especially doesn’t miss the erratic schedules of his old technician job, which stressed him out and tied him up for most of the day, nor does he miss the local coordinator who (Sam feels) gave him especially rough timings because of personal animosity and who frequently criticized him for not completing jobs quickly enough. Moreover, he very much enjoys his TaskRabbit assembly work because it allows him to build tangible objects instead of just installing modems or fixing wires.

Sam does not know exactly how much he earns from his platform work. For a month before Christmas he also drove weekday afternoons, which got the family through the holidays but made it harder for him to estimate his average weekly take-home. He has not calculated his net income, but knows whether he earned less in a day than he spent on fuel that morning, he saves all his fuel receipts for tax purposes, and he figures that his TaskRabbit work involves no expenses at all except that new tool belt he bought two weeks ago.[63] When he first lost his job, Sam briefly thought about moving to an industry competitor or even getting a temporary retail or fast food job because of the hourly wage guarantee. In the end, however, he and his wife decided that their scheduling needs and his well-being pointed them towards platform work instead.

Sam’s story reads as one of intrepidity and relative convenience in the face of economic volatility, and indeed many platform workers (to say nothing of platform companies) articulate a similar, largely positive narrative of what it means to work in the sharing economy. But just as the glib presentation of platform work as easy and entrepreneurial hides far less pleasant realities, it also masks or trivializes the more meaningful positive aspects of this work.[64]

To begin with, setting one’s own schedule is about far more than mere convenience.[65] Indeed, many workers describe this aspect of their platform participation using language that smacks of empowerment and independence: “I never drive more than six hours a day” or “I only drive weekends.”[66] On the one hand, the ability to turn off an app or put a profile on hold allows workers to assert the primacy of their own priorities rather than allow them to play second fiddle to an employer’s timetable. On the other hand, platform work involves no reporting to supervisors—human or otherwise. There are no punch cards, no time sheets, and no bosses walking past desks. “I like Uber,” one driver said, “because [driving a] taxi is eight to eight, nine to nine.”[67] He added that he chose to drive full time rather than work in a casino like one of his brothers or in the family gas station business like another brother because he doesn’t like having to keep a schedule or deal with a boss.[68]

Likewise, not having a boss means more than having flexible schedules: it can also mean (within boundaries) the ability to embody flexible work styles to a degree where workers seem to be determining what the work is. An Uber driver can choose to put on some Hip Hop music in between passengers (or some high-decibel classical music with passengers in the car), just as she can decide to stay parked between rides instead of immediately driving to a high-density area or vice versa.[69] An Airbnb host can decide to allow guests to check themselves in (or not), to provide breakfast (or not), and to decide how many sets of linens are essential to successful hosting.[70] In other words, platform workers lack the over-the-shoulder supervision paradigmatic of traditional production or service jobs. Although this by itself does not make them unusual—taxi drivers and long-haul truckers are well–known examples[71]—it adds to the sense that they conduct their work lives in the relative absence of supervisory intervention.

Finally, just as there is no supervisor to set Sam’s schedule or tell him how to go about doing his work, there is also no supervisor rejecting his application for a raise or limiting his overtime hours. I do not mean to say that platform workers believe they can earn exceptionally large (much less unlimited) sums because they are only restricted by their own willingness to work or, more cynically, because they are only limited by their ability to game the platform’s algorithm. Some workers undoubtedly do still approach their platform labor with this attitude,[72] but they are likely rare; the conversation among workers and observers alike has come a long way from Uber’s childhood boasts of $90,000 annual incomes.[73]

Instead, when platform workers speak of their ability to control their own earning potential they tend to describe definite, but decidedly circumscribed goals: $200 per day, for example, or $200 per week, or enough to cover a particular expense like travel or car payments; the specificity of these goals represents a balance between their financial needs and their other priorities.[74] That is to say, the “freedom to earn” associated with the sharing economy is often understood as the freedom to set income ceilings rather than the freedom to break them, but it is nonetheless valuable to workers.

Platform workers are neither deluded nor unusual for valuing the freedom to make their own decisions regarding scheduling, work style, and earning potential—nor, for that matter, are they alone in associating this freedom with independent contractor classification. Non-white immigrant taxi workers in the Bay Area, for instance, have strongly preferred to be independent contractors rather than employees (to the point that San Francisco taxi drivers as a whole were unable to vote themselves into employee status when given the opportunity to do so by city government) because they attach symbolic and practical consequences to each classification.[75]

For these drivers, the often degrading experience of being a non-white taxi driver is somewhat mitigated by the knowledge that they do not work for anybody and by the cultural capital that this generates within their social circles.[76] More concretely, immigrant drivers worry that employee status will allow leasing officials to give free rein to their prejudices by enforcing specific dress and grooming requirements as well as by giving immigrant drivers the worst cars and schedules.[77] Drivers also (quite rightly) associate the universally despised practice of short-term or “day” leasing with employee status and worry that the problems of day leasing—such as long hours wasted in line waiting to be assigned a car, daily bribe payments to get better cars, and the generally demeaning treatment meted out by leasing officials—will only be exacerbated by a return to the classification scheme under which it was originally developed.[78]

As the worries of taxi drivers and the satisfactions of platform workers suggest, conceptualizing freedom at work as non-interference (“nobody tells me when to work, how to work, or how much I can earn”) is just that and no more: the freedom to not have another human being interfere in one’s work-related choices. Perhaps the salience of this limited vision of freedom means that it is uniquely disempowering to have a fellow person order us about.[79] This would certainly be in line with arguments that human action is uniquely effective at stopping the flow or impact of accountability—of acting as a “moral crumple zone” when essentially autonomous technologies malfunction—because it too emphasizes that our cultural and legal conceptions of responsibility are still centered on the individual agent.[80] Or, perhaps it suggests that interference is more demeaning when it is discrete and direct (like an order to report for work at 7 a.m), rather than when it is incremental, structural, or indirect (like when implicit bias in customer reviews eventually triggers deactivation from an app).[81] Regardless, the meaningfulness of this vision of freedom is real and it is reasonable, even if the underlying concept is exceptionally narrow and thus problematic. And because this understanding of freedom is explicitly defined by the presence or absence of control, it can make control-based categories powerfully meaningful for workers even when there is good reason to think those workers ultimately suffer as a result of a classification system based on control.

Even policy analysts and workers’ advocates sometimes find the idea of freedom-as-non-interference compelling, inasmuch as they sometimes use it to typologize platforms for the purposes of analyzing and responding to platform-related problems.[82] But separating “labor platforms” (like TaskRabbit and Rover) from “capital platforms” (most notably, Airbnb) is neither conceptually easy nor, if we are concerned with workers’ welfare, is it manifestly desirable. For instance, ridesharing is usually slotted into the “labor” category even though it requires a significant capital investment because ridesharing, like dog-walking or furniture assembly, involves personal services performed by the worker. But UberBLACK drivers can have commercial accounts listing multiple individuals who actually do the driving and who split earnings with the account holder;[83] conversely, many Airbnb hosts personally undertake part or all of the considerable labor involved in hosting.[84] The fact that most Uber drivers actually do the driving themselves while many Airbnb hosts outsource their maintenance work does not reflect anything intrinsic to the way the platforms work.

From a workers’ advocacy perspective, the suggestion that capital platforms are different—that they merely constitute a “side income” stream and are meaningfully less like employment than labor platforms—relies on the idea that a task is not “work” if someone has not directly forced you to undertake it at a given moment, in a given manner. Yet the same analysts and advocates who distinguish between labor and capital platforms often point out that algorithmic management techniques significantly restrict workers’ freedom (making them more like employees) even though the techniques generate few discrete, unavoidable demands and involve little direction by any human being.[85] Moreover, algorithmic management is hardly limited to a specific platform or even a specific kind of platform: Airbnb and Rover equally engage in algorithmic management when they rely on data-driven evaluation systems.[86] There may be other reasons to distinguish between “labor” and “capital” platforms, like understanding how class and race operate in the sharing economy, but understanding how to regulate work and how to safeguard workers’ rights are not self-evidently among them.[87]

This is not to say that any particular commentator wholly subscribes to a negative conception of freedom-as-non-interference—again, the central argument of this Article is that lay and legal actors do not wholly subscribe to any one way of thinking about what it means to be free at work (and neither does our work law). Rather, it simply goes to show that the idea of individual autonomy undergirding control-based analysis tends to inform analytic categories regardless of both one’s policy preferences and the role one inhabits with respect to the sharing economy.

There is also no question that the freedom described here is thin, not simply because it is contained in a few discrete and sometimes trivial types of decision-making, but also because those instances of decision-making are only free from interference if we draw a tight circle around what actually constitutes “interference.”[88] Platforms restrict workers’ choices by establishing cutoffs and penalties for various behaviors, including the rate at which workers accept client requests, the speed with which they accept them, and the rate at which they cancel accepted requests.[89] Some platforms establish fairly specific codes of conduct or limit the services a worker can offer unless she reaches an elite status.[90] And of course, Uber and Lyft seek to manipulate drivers’ personal choices regarding where and when to drive by using dynamic pricing.[91] The thinness of freedom-as-non-interference is such that the very things workers value—say, the ability to set earnings goals—are the means for platforms to shape worker behavior in ways that benefit platforms but not workers.[92]

Although scholars and policy analysts increasingly view these practices as restricting worker choice in meaningful ways, the practices are unevenly viewed as interferences in personal freedom by workers themselves and (with the equally uneven exception of ridesharing) are unrecognized as such by courts.[93] But that is precisely the point: platform control remains largely invisible because it does not follow the model of a X dictating to a Y on discrete matters at particular moments.[94] This does not mean that workers and observers feel that platform labor involves no loss of freedom—simply, as we will see, that the freedom they see being threatened is of a very different sort than the type of non-interference described above.

B.  Freedom as Non-Domination

Critics of the sharing economy argue that the liberating potential of this space has been drastically oversold and that platform labor generates uncertainty, conformity, and obsequiousness—all symptoms of domination—that are incompatible with freedom at work. According to this view, it does not matter so much that Uber, Rover, and Airbnb regularly discipline workers who cancel client requests, or that customers on these platforms discipline workers who do not meet their personal expectations—it is not the actual interference in worker choice that produces domination.[95] Rather, it is that platforms and customers have the power to discipline (and thus the power to restrict choice) if and when they feel like it and in essentially unpredictable or unknowable ways.[96] As a result of this unchecked authority, workers live in a state of uncertainty to which they respond by conforming their desires to the limits of the system and by behaving with obsequiousness towards those who dominate them.[97]

Take Alex, a retired Caucasian woman who rents her family’s spare bedroom via Airbnb and is also a Lyft driver.[98] After a few years of retirement, Alex and her husband decided to try homesharing because they felt they should do more to ensure their own financial stability during their golden years and because they wanted to help their struggling son with his daughter’s college tuition. Once they were more or less settled with their Airbnb listing, Alex decided to give ridesharing a try as well, mostly out of curiosity. Her husband does the cleaning and maintenance for their rental, while Alex manages their online profile and is primarily responsible for interacting with their clients.[99]

Alex has always considered herself a friendly, easy-going person so she was somewhat surprised to discover how invasive it initially felt to have strangers in her car and her home. She understood how some drivers could say that inviting strangers into their cars and making small talk for money made them feel like “pimps,” and although she didn’t feel quite as strongly, she was very aware of the fact that she had to provide a “service” while in her own home and her own car.[100]

Eventually, though, she started focusing on how to be a better host and driver. She would look carefully for mannerisms and word choices that indicated her clients’ moods, and experiment with different behaviors, questions, and phrasings to see what reactions she got.[101] As she got better at doing all of this, and as she started thinking of her work as providing something that people really needed, she started to enjoy herself more. One of her favorite tricks now is to ask for restaurant recommendations from her rideshare passengers but give restaurant recommendations to her homeshare guests—she doesn’t need to ask any more than her guests (who could always check Yelp) need to be told, but in each context her tactics seem to make clients feel good.

Still, Alex remains slightly worried about some aspects of her platform work. For instance, she is especially sensitive to the fact that guests might complain about food odors in her home since her husband is Indian and cooks often.[102] One of her very first guests did in fact note the smell in his review. It wasn’t a major criticism—he made a few other suggestions as well—but Alex is convinced that the smell was the reason he gave her a four–star rating. More experienced hosts whom she encountered on an online chat forum reassured her by saying that four stars was fine for an early review and even gave her tips on how to moderate food odors. Still, Alex is always a bit anxious when a guest walks in for the first time—and also when they walk out for the last time at the end of their stay.

She’s also a little concerned that one of her guests will have a problem with the fact that her family is interracial. She hasn’t faced any explicit instances of discrimination yet, but she has heard stories of frosty guests and inexplicably low reviews and is wary. Still, she knows that there is little she can do since Airbnb’s anti-discrimination policy—which she has read in painstaking detail—is more geared towards protecting guests than protecting hosts.[103] So she does her best to preempt the situation by being extra friendly and accommodating.

Food odors and prejudices against interracial couples are obviously not a problem in her work as a driver but Alex is worried that, now that she’s finally got the hang of it, Uber is going to switch to driverless cars and all the other rideshare companies, including Lyft, will follow suit. She’s actually really grown to like her driving: doctors rushing to work in the morning and businesspersons late for meetings are always so grateful that she feels like a hero, and thanks to Lyft’s tip feature they can and do express their gratitude.[104] There was a week back in December when the money was so good (and their granddaughter’s spring tuition was almost due) that she couldn’t stop driving and started to get worn down from the lack of sleep.[105]

Alex’s story does not read as one of terrible sadness or subordination to either her platforms or her customers, and it is not meant to. Yet the absence of freedom is a terrible thing—if Alex is not considerably unhappy and does not feel constantly thwarted in her work by being made to do things or obey orders that impinge upon her autonomy, why would we consider her unfree?[106]

In a word: uncertainty. Workers and observers are noting with heightened frequency and levels of eloquence that the experience of providing services in the sharing economy is marked by an overwhelming level of uncertainty—uncertainty regarding the requirements for gaining or maintaining access to the app, uncertainty regarding client expectations, uncertainty regarding situations where the two clash, and even uncertainty regarding how long the entire system will exist in its current form.

None of these uncertainties need actually produce bad outcomes for workers. The unknown platform requirements might be effortlessly met, clients might not have idiosyncratic expectations, the two sets of demands might never come into conflict, and (notwithstanding Alex’s fear of driverless cars) major changes might always hover just beyond the horizon.[107] Indeed, regulatory and technological hurdles make the worry about rideshare drivers training their own replacements less compelling than is sometimes suggested.[108] Nevertheless, it remains that the only thing that workers like Alex can be sure of is the power of the platform and of the client to constrain their options in any circumstance. It is an authority that is always there, waiting to be exercised, and the waiting creates uncertainty.[109]

To be sure there is uncertainty in all things and all jobs, and it is equally true that American work law has a peculiarly high tolerance for uncertainty as signaled by its embrace of at-will employment.[110] The problem is not so much that workers lack a script laying forth all the future twists and turns of their gigs, jobs, or working lives, but that the script they do have, bare bones as it is, is also at any time susceptible to alterations that are not of their choosing and may be intentionally hidden from them.

Uncertainty of this type constrains autonomy without ever rising to the level of direct interference because it makes it impossible for workers to exercise meaningful choice. When an Uber driver accepts every ride request she receives but is told she did not meet the acceptance rate requirement to qualify for a guaranteed hourly wage, she has no information with which to counter Uber’s assertion and, consequently, no way to judge her best future course of behavior.[111] Her safest bet is to continue accepting all possible ride requests in the hope that Uber’s assessment and her own assessment of her acceptance rate will eventually match up. Similarly, when a Fiverr worker does not know what will catapult her into the highest category of elite workers (“Top Rated Seller”) she has no way of choosing among an array of possible behaviors or business decisions in order to access the very real benefits that come with Top Rated Seller classification.[112]

Inasmuch as it forces workers to operate blindly, uncertainty also leads workers to literally and figuratively embody the domination they experience.[113] One paradigmatic way to respond to the unchecked authority of another is to curry favor with the authority figure so that she will minimize her interference in your choices.[114] Obsequiousness is neither necessarily painful nor gaudy: most of us like to please, and—as many rideshare drivers attest—sometimes silence itself is pleasing enough. What obsequiousness (even the silent variety) expresses, however, is the powerlessness of the person who embodies it. “Several [Uber] drivers said the best way to behave is like a servant,” noted one journalist, before going on to quote a driver in Sacramento who characterized her own role by saying that “‘[t]he servant anticipates needs, does them effortlessly, speaks when spoken to, and you don’t even notice they’re there.’”[115]

Though it may seem otherwise, obsequiousness is about more than exercising “natural” tact. Rather, it is a way of recognizing the particular social game being played and of obeying its rules.[116] Smiling to make others happy, curating conversations to boost others’ egos—the emotional labor performed by platform workers concretizes the power of clients and algorithms through the worker’s own physical gestures and speech patterns.[117] For instance, after a Lyft driver noted a sudden decrease of 0.1 in his rating, he started attempting to establish a conversational rapport with passengers “very quickly” because “[t]hat’s what they”—meaning both passengers and Lyft—“want. Accommodate and connect.”[118] Emotional labor is nothing new, of course, and even within the sharing economy, it is hardly limited to either rideshare drivers or verbal interactions.[119] But the sheer ubiquity of emotional labor makes it no less telling a sign of dependence on another’s will.

Because emotional labor and obsequiousness are unsuccessful where they are obviously insincere, workers must also at least partly conform their preferences to the constraints, actual or potential, that are placed upon them.[120] This too is neither necessarily painful nor outlandish. If one’s options are good ratings that produce a host of benefits—more income, elite statuses, perhaps an hourly wage guarantee—and bad ratings that produce their inverse—and perhaps deactivation—then it is natural to honestly prefer the former, and indeed, it is difficult to do otherwise.[121] Account after account of platform work emphasizes how seriously workers take such markers of success, not only because they translate into more desirable outcomes, but because they validate the decision to commit to platform labor itself. “I’ve got currently twelve excellent-service and nine great-conversation badges,” boasted one Uber driver (who is nonetheless looking for another job).[122] “It tells me where I’m at.”[123] But in the neo-republican account, while happiness—or at least, reduced frustration—may lie in learning to want what you have rather than having what you want, freedom emphatically does not.[124]

This is not to say that platform workers are automatons who feel what is wanted and want what is given (even though it can sometimes seem that way).[125] Critique and resistance abound, whether this consists of venting on chat forums, contesting client accounts of interactions, strategizing within the bounds of the system—as when workers try to game the algorithms that determine search results or that allot front page placement on their platforms’ websites—and it may even involve operating outside the bounds of the system itself—as when workers go “off app.”[126] But these acts of agency occur within a particular framework and using a particular set of idioms and intuitions; they are part and parcel of the system they ostensibly subvert. Consequently, while they might demonstrate dissatisfaction with the rules of the game, they do not fundamentally break with the game itself.[127]

When critics decry the imbalance of power between clients and platforms on the one hand and workers on the other, they are appealing to an understanding of freedom of non-domination rather than as non-interference. If only interference was at issue, the ability to ingratiate oneself with clients (and also with algorithms) and to thereby pursue one’s livelihood relatively unmolested would draw analogies to liberation, not servility. Likewise, if the sum total of freedom was the ability to avoid frustration, it would be enough to constantly recalibrate one’s preferences in light of new constraints and interferences. That neither the one nor the other is the case suggests that workers and observers also subscribe to a vision of freedom that the sharing economy does not enable. It is this failure that drives talk of regulatory dysfunction or breakdown with respect to platform work, and that underlies dissatisfaction with control-based classification even outside the platform context.

C.  Conflicting, Not Concentric Freedoms

Taken together, Sam and Alex reflect worker experiences across a range of platforms, but, more importantly, they embody two distinct ways of experiencing (or not experiencing) freedom at work. Sam’s ability to make choices about when he works, how he works, and how much he earns, as well as the absence of a supervisor monitoring his performance, all reasonably contribute to a sense that he is free because someone else is not directly forcing him to engage in or refrain from particular actions. Conversely, Alex’s worries about dealing with prejudicial or idiosyncratic clients, her reservations about having to behave obsequiously in her own car and home, and even her gradual enjoyment of the subservient role she had initially disliked, all signal her loss of freedom–as–non-domination. One is not simply broader than the other; rather, freedom–as–non-interference and freedom–as–non-domination are conflicting rather than concentric or additive concepts.

Non-interference is grounded in discrete, direct exercises of authority (“Pick up Joe Smith on Main Street, now!”). This is the kind of instruction someone like Sam is glad to be rid of, and labor and employment scholars rightly associate it with industrial and factory-based forms of labor that are a diminishing component of our work landscape. But the sort of freedom that someone like Alex is missing out on, freedom-as-non-domination, would not be captured by a broader “functional” understanding of authority because it is fundamentally distinct.

For example, even though Alex may not be given a specific order to “pick up Joe Smith,” she might know that if she does not pick up a large enough percentage of Joe, Jack, and John, she will be terminated from the app. The trouble is that Alex is not quite sure what that magic percentage is or whether she and Lyft will agree on when she’s met it, so Alex is not really free to reject any of those passengers. A broader definition of control would not solve her problem. Likewise, the constraints on her freedom do not arise from whether or not Lyft actually directs her to “pick up Joe on Main Street, now.” What would make Alex feel freer in the non-domination sense if she knew beforehand that picking up Joe and Jack would ensure that she met her daily acceptance rate.

This means that the difference between freedom-as–non-interference and freedom-as-non-domination is one of kind rather than degree. More importantly, it means that regulators cannot satisfy both conceptions of freedom by simply expanding the circle of “employees” or by understanding “control” more expansively. Control-based analysis is inherently tied to the idea that a worker’s freedom is impinged upon when an employer dictates the “how, when, and where” of her work—that is to say, the “means and manner” of performance. freedom-as-non-domination need have nothing to do with this sort of means and manner analysis. And because workers are not alone in valuing both types of freedom, our labor and employment statutes and case law reflect traces of freedom-as-non-interference as well as freedom-as-non-domination.

III.  Freedom(s) In, and Through, Work Law

This Part explores how the ethnography discussed above illuminates a fundamental tension in labor and employment law itself. It is one thing to say that platform workers and observers are genuinely attached to different visions of freedom at work, another thing to suggest that this dynamic can also be found in our work law, and yet a third thing to argue that the tension between these two conceptualizations of freedom explains our fixation—and our dissatisfaction—with control-based analysis. So far, I have only made the first of these claims, but in what follows I will make the second and third. These arguments necessarily take us from the fine-grained, ethnographic study of platform labor to the historical and doctrinal analysis of labor and employment law writ large.

A.  Non-Interference and Non-Domination in Work Law

Freedom–as–non-interference is undoubtedly more prominent within labor and employment law (and arguably beyond it) than freedom-as-non-domination. Few things convey the centrality of this way of thinking as well as the overall primacy of the common law control test, but we can pick out other, more specific instances where an understanding of freedom at work as non-interference shines through with especial clarity. Take the practice of upfront contractual specifications (“UCS”), in which detailed descriptions of the way work is to be done are included in independent contractor agreements, but are presented by companies and often understood by courts as evidence of the end product or service that is contracted for.[128] When courts read UCS clauses as detailing the “ends” rather than the “means” of performance, they understand them to support independent contractor classification because, in part, UCS obviates the need for human monitoring and scheduling.[129] In other words, some courts—like some platform workers—conceptualize freedom at work to be the absence of direct, interpersonal authority rather than as the absence of the power to exert such authority indirectly and without restriction.

For example, FedEx has successfully argued in several courts that the terms of its lengthy and non-negotiable Operating Agreement did not transform its drivers into employees because, among other things, the agreement “suggested a limited need or interest in real-time supervision.”[130] To be sure, FedEx’s position has been widely criticized and the judicial tide may have started to turn against the company on this issue,[131] but UCS is hardly limited to one company or even one industry.[132] Moreover, once we grant the paramount importance of identifying control over the means of performance, the frequently counterintuitive results in UCS cases become an intractable—because they are unavoidable—problem.[133]

We can see the importance of control and freedom-as-non-interference outside the realm of worker classification, too. Consider the practice of having workers contractually waive statutory protections like the ability to litigate rather than arbitrate future claims, or the ability to mount claims based on current or prior employment decisions in any forum.[134] The logic in doing so is that when waivers are signed under conditions that are not explicitly coercive, they represent choices made free of interference.[135] They may even be said to advance freedom, inasmuch as they empower workers to assess and realize the value of certain statutory protections by their own lights.

But of course, waivers are signed in situations where choice is severely constrained by asymmetrical knowledge and by asymmetrical power over goods like jobs and severance packages.[136] (There is also the very real concern that waivers defeat public interests even when they truly advance private ones, but “deregulat[ion] by contract” is an entirely different sort of objection to statutory waivers.[137]) Even where there is no interference of the stranger-in-a-back-alley variety, it requires single-minded focus on the bare act of assent to be able to say that the waiver itself expresses worker autonomy.[138] That assent is important, to be sure, and it meaningfully differentiates contractual waivers from agreements made under conditions of actual intimidation. But for our purposes, contractual waivers are interesting for their signaling value more than for their substantive effect: the fact that courts almost universally enforce waivers suggests that courts dealing with employment contracts—like lay actors dealing with consumer contracts—find a strictly non-interference model of free choice to be compelling in some circumstances, however thin that freedom might appear to critics.[139]

While work law is undoubtedly flush with examples of freedom-as-non-interference, it is also relatively easy to spot instances where individuals have tried—with varying success—to push the law toward a conception of freedom-as-non-domination.[140] Perhaps most strikingly, labor republicans of post-Civil War America directly drew on and refined the classical republican understanding of freedom as part of their efforts to advertise the “structural and personal domination to which a modern wage-laborer was subject.”[141] The open-ended authority of the labor contract, according to this new, more radical interpretation, merely replaced the unfreedom of slavery with the unfreedom of wage slavery.[142]

Nevertheless, republicanism petered out as the labor struggles of the late nineteenth–century segued into the Lochner era and later on as New Deal legislation sustained legislative and judicial onslaughts that revived the common law test and its control-based analysis.[143] The star surviving example of this is the FLSA’s definition of an employee as anyone who an employer “suffer[s] or permit[s] to work,” as well as its accompanying, judicially-created test, that purports to measure the “economic realities” of a work relationship rather than the quantum of control it involves.[144] The FLSA’s emphasis on knowledge and power along with the economic realities test’s complementary emphasis on dependence make the overarching—and not always actually exercised—authority of the employer the basis of worker classification.[145] As Section I.B noted, both the Act and the test are open to numerous criticisms, not least of which is that their vagueness produces judicial analysis suspiciously similar to what happens under the common law test. But it remains that the FLSA and the economic realities test represent concerted efforts to move away from a system that compensates only for the loss of freedom represented by direct interferences in a worker’s will.[146]

A similar conceptual move underlies the 2015 reassessment of the standard for determining “joint employer” relationships as well as the NLRB General Counsel’s investigation regarding McDonald’s liability for the working conditions of its franchisees’ employees. Together, Browning-Ferris[147] and the consolidated McDonald’s inquiry created uproar in the franchising world because they discarded an analytic framework that based employee classification on direct interference in worker autonomy as the basis for labor and employment protections.[148] Browning-Ferris announced that the NLRB (“the Board”) would henceforth only require a potential joint-employer “possess the authority to control employees’ terms and conditions of employment” rather than actually exercise such authority.[149] Likewise, it announced that the Board would acknowledge forms of control that were not directly and immediately exercised by the potential employer; instead, control exercised via an intermediary would count as well.[150] The McDonald’s investigation put this approach into practice (although it is worth noting that the General Counsel had initiated its inquiry well before a decision was issued in Browning-Ferris).[151]

In both instances, the Board abandoned a narrower understanding of freedom where only direct commands framed as commands are held to impinge a worker’s autonomy, and instead adopted a broader understanding of freedom in which the ability to elicit desired behavior warrants protection even if it is not exercised via direct command or remains unexercised altogether. The shift has been questionably successful: even though the Board confirmed its approach in subsequent cases,[152] Browning-Ferris was eventually overturned by Hy-Brand Industrial Contractors (which was itself later vacated).[153] But even though the new joint employer standard has been neither particularly impactful nor long lasting—in fact, especially because it has been neither impactful nor long lasting—it speaks to both the gravitational pull of control-based analysis and the dissatisfaction it occasionally produces.

B.  Tension, Confusion, or Failure?

Even if distinct visions of freedom-as-non-interference and as non-domination exist inside and outside work law, how can we be sure that the tension between them is responsible for critics’ seeming desire—and inability—to move beyond control-based analysis? Perhaps legislators are just responding to political pressure when they enact laws that promote control and non-interference in spite of its poor fit with the realities of work. Or, perhaps judicial actors are simply doing the best they can with vague laws and complicated facts, but their efforts also fit poorly with the realities of work. Bad statutes or bad case law—with “bad” meaning sinister, ill-conceived, archaic, or subject to internal contradiction (among other things)—are, overwhelmingly, the way labor and employment scholars have explained the continuing fixation and dissatisfaction with control-based analysis among themselves and decisionmakers.[154] Both of these explanations are undoubtedly part of the answer, but they are not, individually or even together, the whole answer. What’s more, considered by themselves, they paint a crude and starkly ungenerous view of all parties involved.

Take the idea that control-based analysis persists because labor and employment statutes overwhelmingly reflect the interests of elite actors (the “legislative failure” argument).[155] Political interests and constraints can certainly explain some of the stranger features of our classification system, including, for example, the exclusion of domestic and agricultural workers from the NLRA or that of tipped servers and farm workers from the FLSA.[156] Politics can even explain some of the back-and-forth between differing approaches to freedom at work, like the Taft-Hartley Act’s restrictions of the meaning of “employee” under the NLRA and Hearst.[157]

But “legislative failure,” though it offers some insights about our love‑hate relationship with control-based analysis, cannot explain that relationship on its own. Suggesting otherwise invites a kind of legal nihilism because it requires viewing law as nothing but a tabula rasa waiting to be written on by select actors. It also invites a kind of legal exceptionalism because it would mean that law has a singular power to change hearts as well as actions even among those whose interests it undermines. In other words, this explanation requires us to think of law as both acultural and as wholly constitutive of cultural traditions, and it calls on us to view any values discernible in law as little more than “glosses on property relations.”[158] To write this down is to demonstrate its impossibility. Beyond all this, “legislative failure” sadly underestimates the importance of courts in constructing, defending, and reformulating the building blocks of labor and employment law.[159]

The second argument (“implementation confusion”) corrects for that last shortcoming by putting the blame squarely on the way courts handle labor and employment cases. In one understandably popular version of this explanation, judges both perpetuate control-based analysis and occasionally undermine it because the relevant legal precedent is confusing and statutory guidance is in woefully short supply.[160] Another less frequently articulated version of this argument suggests that courts have repeatedly embraced a restricted vision of freedom-as-non-interference because it better aligns with employers’ interests and because, consciously or not, judges are predisposed to sympathize with employers.[161] When courts recognize thicker understandings of freedom at work—as in Hearst, for instance, or in the FedEx litigation, Browning-Ferris,[162] or Cotter[163]—it is because the realities of work (and the interests of workers) have managed to assert themselves despite these limitations.

“Implementation confusion” cannot offer an exhaustive solution to the puzzle of control-based analysis any more than “legislative failure” because it repeats the latter’s errors, albeit in more complex fashion. To the idea that the stickiness of control originates in legal infrastructure, “implementation confusion” adds a new type of law (adjudicatory outcomes, whether by courts or by agencies) and a new type of actor (courts, rather than just legislatures). This adds nuance to the legal nihilism and exceptionalism from earlier, but it does not fundamentally challenge the premise that law may construct social norms without also being constructed by them. Likewise, to the argument that the faultiness of control-based analysis stems from a class-inflected divide between law and reality, “implementation confusion” adds subtlety via implicit bias—“judges naturally think like employers”—rather than explicit preference. But of course, this merely casts lures to the legal realists inside many of us without doing justice to the sort of measured, multivariate analysis of judicial behavior pursued by many New Legal Realists themselves.[164]

Perhaps the most significant problem with these two explanations is that they do not really notice or explain the fact that the tension between non-interference and non-domination exists outside the law itself, in the vast realm of “society.” As Parts I and II showed, workers—and even some commentators—who might be expected to find a thicker vision of freedom at work uniquely compelling (and who often do find it compelling) also often think of freedom-as-non-interference. This matters. Since critics of control-based analysis fail to see that its faultiness and stickiness exist outside the law as well as within it, they reasonably view the tension in work law as emanating from a disjuncture between law and society. Once we see that this tension exists both inside and outside the law, it becomes impossible to think that a disjuncture between law and society—whether stemming from statutes and legislators or case law and judges—is all that lies behind it. Something else must also be at issue, something that does not reduce law to “glosses on property relations,” lawmakers to puppets (or puppeteers), or lay actors to dupes. The missing piece of the puzzle is that we—workers, scholars, and decision makers alike—have genuine commitments, visible in law and in everyday practice, to two different conceptions of freedom at work.

Conclusion

The tension between non-interference and non-domination that I have outlined here explains decisions like Cotter v. Lyft far better than any transformations in technology or employment practices. Precisely because it was thorough and measured, Judge Chhabria’s analysis exemplifies how work law uses a single concept (“control”) to try to capture a complex empirical phenomenon (“freedom”) as well as how the attempt often produces stalemates and confusion.

For instance, several of the factors Judge Chhabria considered link the concept of control to an understanding of freedom–as–non-interference: the “great flexibility” drivers enjoy regarding “when and how often to work;”[165] their ability to select “parts of San Francisco in which they accepted ride requests;”[166] and the “minimal contact with Lyft management” while working as drivers.[167] All of these factors interpret control to mean discrete or direct restraints on driver autonomy; because the restraints did not exist, control was also found to not exist.

Conversely, other factors considered by Judge Chhabria link control to an understanding of freedom–as–non-domination: the “right to penalize” that Lyft reserves to itself (whether or not that right is actually exercised);[168] the ambiguous standards on which such penalties can be based;[169] and, above all else, the power that comes from the ability to terminate at will.[170] All of these factors interpret control to mean a potential, and potentially unrestrained, ability to limit driver autonomy; because Lyft did indeed possess such an ability, it was found to enjoy control over its drivers.

Cotter makes clear that work law cares about how people experience freedom, that there are distinct ways to experience freedom, and that decisionmakers cannot effectively subsume these distinct visions under a single concept like control. When they do so, as Judge Chhabria was forced to do based on prevailing California law, they are left with little but a morass of crossed signals and repeat errors. So where do we go from here?

First, we acknowledge that freedom-as-non-interference is a relatively thin concept with a remarkably thick and rich tradition in the United States. This is true both among lay and legal actors, as well as in “law on the books” and “law in society.” It is simply not productive to dismiss a Lyft driver’s valuation of, say, scheduling flexibility on the grounds that it seems like a shallow sort of freedom.

Second, we recognize that middle-of-the-road attempts to fix classification doctrine by introducing new tests or new factors do not succeed. This is not because non-interference and non-domination are mutually exclusive in the abstract, but because, in practice, it is difficult to consistently identify losses of freedom when freedom means two distinct things. Decades of classification case law and scholarship attest to this fact. The failure of median approaches suggests that our options for improving classification doctrine lie at the extremes: either we enact piecemeal regulatory reforms that address specific aspects of work relationships but leave core conceptual issues as they are,[171] or we undertake the profoundly challenging task of regulating work relationships on the basis of something other than the amount of control and freedom they permit.[172] Piecemeal regulations need not be inconsequential, and conversely, systemic change may not be better or feasible, but either approach would depart from previous reform efforts by respecting and building on our dual conception of what it means to have freedom at work.

[*] *.. Sharswood Fellow, University of Pennsylvania Law School; A.B. 2006, Princeton; Ph.D. 2013, University of Chicago; J.D. 2016, University of Chicago. My thanks to Daniel Abebe, Brad Areheart, Shyam Balganesh, Stephanos Bibas, Richard Carlson, Miriam Cherry, Ryan Doerfler, Veena Dubal, Eric Feldman, Lee Fennell, Andrea Freeman, Camille Gear Rich, Michael Green, Dave Hoffman, Genevieve Lakier, Sophia Lee, Brian Leiter, Serena Mayeri, Sandy Mayson, César Rosado Marzán, Shayak Sarkar, Paul Secunda, Beth Simmons, Megan Stevenson, Lior Strahilevitz, Julia Tomasetti, Laura Weinrib, Tess Wilkinson-Ryan, Noah Zatz, Adnan Zulfiqar, and, as always, John Felipe Acevedo, for comments on various aspects or versions of this paper. Commentators at the 2016 and 2017 Law & Society Association meetings, 2017 AALS meeting, Third Labor Law Research Network Conference, Fifteenth Annual Marco Biagi Conference, Michigan Law Young Scholars Conference, 2017 Southeastern Association of Law Schools Conference, the University of Pennsylvania Law School, and the University of Chicago Law School also provided valuable feedback. This research was funded in part by the generous support of Dean Ted Ruger and the University of Pennsylvania Law School.

 [1]. See generally Cotter v. Lyft, Inc., 60 F. Supp. 3d 1067 (N.D. Cal. 2015). Cotter has been cited in at least twelve decisions and as many as thirty-seven law review articles in the three years following the decision. See, e.g., Bowerman v. Field Asset Servs., Inc, 242 F. Supp. 3d 910, 928 (N.D. Cal. 2017); Bekele v. Lyft, Inc., 199 F. Supp. 3d 284, 303 (D. Mass. 2016); Ryan Calo & Alex Rosenblat, The Taking Economy: Uber, Information & Power, 117 Colum. L. Rev. 1623, 1626 n.14 (2017).

 [2]. Cotter, 60 F. Supp. 3d at 1069.

 [3]. See id. (“We generally understand an employee to be someone who works under the direction of a supervisor, for an extended or indefinite period of time, with fairly regular hours, receiving most or all his income from that one employer . . . .”). See also Cmty. for Creative Non-Violence v. Reid, 490 U.S. 730, 751–52 (1989) (listing various factors that courts consider in making determinations of employee status); United States v. Silk, 331 U.S. 704, 716 (1947) (same); Restatement (Second) of Agency § 220(2) (Am. Law Inst. 1958) (same); Richard R. Carlson, Why the Law Still Can’t Tell an Employee When It Sees One and How It Ought to Stop Trying, 22 Berkeley J. Emp. & Lab. L. 295, 310 (2001) (same).

 [4]. Cotter, 60 F. Supp. 3d at 1081.

 [5]. Nationwide Mut. Ins. Co. v. Darden, 503 U.S. 318, 323 (1992) (citing Reid, 490 U.S. at 740). While there is no definitive articulation of the control test, I will refer to it in the singular for ease of reading and because all versions of the test emphasize the importance of employer control over the “manner and means” of performance. Accord Carlson, supra note 3, at 299.

 [6]. Cotter, 60 F. Supp. 3d at 1081. See also Senator Mark R. Warner, The American Dream Is Fading for Millions of Freelancers. Portable Benefits Could Save It, Senate.gov (July 5, 2017), https://www.warner.senate.gov/public/index.cfm/2017/7/the-american-dream-is-fading-for-millions-of-freelancers-portable-benefits-could-save-it (calling the United States’ approach of linking social safety benefits to employers a “20th-century approach” that is “failing workers in the 21st-century economy”).

 [7]. See Deepa Das Acevedo, Invisible Bosses for Invisible Workers, or Why the Sharing Economy Is Actually Minimally Disruptive, 2017 U. Chi. Legal F. 35, 50, 58 (discussing specific work practices that allow franchisors like McDonald’s to exercise control over their franchisees’ direct employees as well as some companies like FedEx to exercise control over their independent contractors). Compare Alexander v. FedEx Ground Package Sys., 765 F.3d 981, 997 (9th Cir. 2014) (finding FedEx drivers are employees), and Ochoa v. McDonald’s Corp., 133 F. Supp. 3d 1228, 1241 (N.D. Cal. 2015) (denying summary judgment for defendant McDonald’s on grounds that McDonald’s may be liable for labor code violations involving its franchisees’ employees under a theory of ostensible agency), with FedEx Home Delivery v. NLRB, 849 F.3d 1123, 1124 (D.C. Cir. 2017) (finding FedEx drivers are independent contractors), and Salazar v. McDonald’s Corp., No. 14-cv-02096-RS, 2016 U.S. Dist. LEXIS 108764, at *49–50 (N.D. Cal. Aug. 16, 2016) (finding that McDonald’s is not a joint employer of its franchisees’ employees).

 [8]. See infra notes 14, 15, 30, 31, 47 and accompanying text (discussing scholarship and jurisprudence critical of the existing classification regime and control based analysis). See also Bruce Goldstein et al., Enforcing Fair Labor Standards in the Modern American Sweatshop: Rediscovering the Statutory Definition of Employment, 46 UCLA L. Rev. 983, 1106 (1999) (arguing that in adopting the much broader “suffer or permit” standard of employee status instead of the control test, “Congress’s purpose was precisely to expand coverage under the [Fair Labor Standards Act] far beyond the common law”).

 [9]. Throughout this paper I will rely on Isaiah Berlin’s canonical definitions of positive and negative freedom and Philip Pettit’s interpretation of freedom as non-domination. Isaiah Berlin, Two Concepts of Liberty 7–8, 16 (1958); Philip Pettit, On the People’s Terms: A Republican Theory and Model of Democracy 1–18 (2012); Philip Pettit, Republicanism: A Theory of Freedom and Government 19–27 (1997). See also Alex Gourevitch, From Slavery to the Cooperative Commonwealth: Labor and Republican Liberty in the Nineteenth Century 7–17 (2015); Gerald M. Stevens, The Test of the Employment Relation, 38 Mich. L. Rev. 188, 197–98 (1939) (arguing that American courts have “turned for guidance to a more certain and indisputable principle” in which “control must have meant . . . actual interference and superintendence”).

 [10]. Non-domination is most concerned with the presence or absence of boundaries on external interferences rather than with the interferences themselves. For illustrations using the sharing economy, see infra Section II.C. See also Pettit, On the People’s Terms, supra note 9, at 1–18.

 [11]. Deepa Das Acevedo, Temples, Courts, and Dynamic Equilibrium in the Indian Constitution, 64 Am. J. Comp. L. 555, 560 (2016).

 [12]. Kaushik Sunder Rajan, Anthropological Fieldwork Methods 1 (2015) (unpublished syllabus) (emphasis in original) (on file with author). Ethnography is often described as a research method based on participant-observation, and this description is not wrong. But Kaushik Sunder Rajan offers a far more nuanced take on the ethnographic method and its value in his syllabus for a graduate-level methods class at the University of Chicago. “What makes good ethnography work . . . ,” Sunder Rajan writes, “is the fact that the ethnographer is capable of attending to things that her interlocutors might attend to differently (ignore, naturalize, fetishize, valorize, take for granted, etc.).” Id. Consequently, “the fundamental problem of fieldwork involves the cultivation of attentiveness.” Id.

 [13]. See Coppage v. Kansas, 236 U.S. 1, 17 (1915), noted in Alan Story, Employer Speech, Union Representation Elections, and the First Amendment, 16 Berkeley J. Emp. & Lab. L. 356, 406–07 n.264 (1995); Bower v. Peate, 1 Q.B.D. 321 (1876), noted in O.W. Holmes, Agency, 4 Harv. L. Rev. 345, 347 n.3 (1891) (discussing the proposition that a master/servant relationship is no different than other agency relationships inasmuch as the servant’s actions are attributable to the master).

 [14]. See, e.g., Brishen Rogers, Am. Constitution Soc’y, Redefining Employment for the Modern Economy 3 (2016); Franklin G. Snyder, The Pernicious Effect of Employment Relationships on the Law of Contracts, 10 Tex. Wesleyan L. Rev. 33, 36 (2003); Roscoe T. Steffen, Independent Contractor and the Good Life, 2 U. Chi. L. Rev. 501, 507 (1935); Noah D. Zatz, Beyond Misclassification: Tackling the Independent Contractor Problem Without Redefining Employment, 26 ABA J. Lab. & Emp. L. 279, 282 (2011).

 [15]. V.B. Dubal, Wage Slave or Entrepreneur?: Contesting the Dualism of Legal Worker Identities, 105 Calif. L. Rev. 65, 93 (2017); Julia Tomasetti, The Contracting/Producing Ambiguity and the Collapse of the Means/Ends Distinction in Employment, 66 S.C. L. Rev. 315, 356 (2014); Nancy E. Dowd, The Test of Employee Status: Economic Realities and Title VII, 26 Wm. & Mary L. Rev. 75, 86 (1984).

 [16]. On the centrality of “freedom” in American work law see Christopher L. Tomlins, Law, Labor, and Ideology in the Early American Republic, at xv (1993) (describing various areas of nineteenth century law as “tending during the period under study to move toward a representation of working life in voluntaristic terms . . . [an] empowering definition of individual freedom”) and Marion Crain, Work, Free Will and Law, 24 Employ. Resp. & Rts. J. 279, 280 (2012) (discussing competing meanings of “work” in the United States but stating that “[t]he dominant image of work in American law is as an exercise of free will”).

 [17]. See, e.g., Boswell v. Laird, 8 Cal. 469, 489 (1857) (holding that “[s]omething more than the mere right of selection, on the part of the principal, is essential” to a master and servant relationship). Boswell added that “[t]he relation between the parties was that of independent contractors” because the co-defendants “were engaged in an independent employment in the construction of a work which was entrusted entirely to their skill.” Id at 490, 494. For discussions of the development of “employee” as a category of free labor during the nineteenth century, see generally Karen Orren, Belated Feudalism: Labor, the Law, and Liberal Development in the United States (1991); Robert J. Steinfeld, The Invention of Free Labor: The Employment Relation in English & American Law and Culture, 1350–1870 (1991); Tomlins, supra note 16; Christopher L. Tomlins, Law and Power in the Employment Relationship, in Labor Law in America 71 (Christopher L. Tomlins & Andrew J. King, eds., 1992); John Fabian Witt, Rethinking the Nineteenth-Century Employment Contract, Again, 18 Law & Hist. Rev. 627 (2000).

 [18]. Gasal v. CHS Inc., 798 F. Supp. 2d 1007, 1013 (D.N.D. 2011) (“The rationale for the doctrine of respondeat superior is based on the employer’s right to control the employee’s conduct.”).

 [19]. Perhaps the best indication that the control test’s persistence is itself an object of puzzlement and anxiety for labor and employment scholars is the frequency with which they refer to its inadequacy, complexity, unfairness, and attempted replacement. See, e.g., Guy Davidov, The Reports of My Death are Greatly Exaggerated: ‘Employee’ as a Viable (Though Overly-Used) Legal Concept, in Boundaries and Frontiers of Labour Law: Goals and Means in the Regulation of Work 133 (Guy Davidov & Brian Langille eds., 2006); Marc Linder, Dependent and Independent Contractors in Recent U.S. Labor Law: An Ambiguous Dichotomy Rooted in Simulated Statutory Purposelessness, 21 Comp. Lab. L. & Pol’y J. 187, 190 (1999); Stevens, supra note 9, at 203; Tomasetti, supra note 15, at 317–18.

 [20]. See infra notes 22–27 and accompanying text.

 [21]. Civil Rights Act of 1964, 42 U.S.C. §§ 2000e-2–3 (2012). Note that the Civil Rights Act of 1866 does not limit its protection against racial discrimination in the formation of contracts (including employment contracts) to “employees.” 42 U.S.C. § 1981(a) (2012).

 [22]. Family and Medical Leave Act of 1993, 29 U.S.C. § 2611(4) (2012). Note that the FMLA adopts the FLSA’s “suffer or permit” definition of “employ.” Fair Labor Standards Act of 1938, 29 USC § 201, § 203(g) (2012).

 [23]. Equal Pay Act of 1963, 29 U.S.C. § 206(d) (2012).

 [24]. 29 USC § 201, §§ 203(d)–(e) (2012) (defining “employees” and “employers” to whom the Act applies); id. § 206 (establishing minimum wage provisions for employees); Portal to Portal Act of 1947, 29 U.S.C. §§ 251–262 (2012).

 [25]. Employee Retirement Income Security Act of 1974, 29 U.S.C. § 1002 (2012) (stating that ERISA only applies to “employees”). Note that ERISA does not require employers to offer any kind of health or retirement benefit plan at all—“it merely regulates retirement promises that are made.” Paul M. Secunda, The Behavioral Economic Case for Paternalistic Workplace Retirement Plans, 91 Ind. L.J. 505, 540 (2016). This is why “[t]he aggregate national retirement deficit number is currently estimated to be $4.13 trillion for all U.S. households where the head of household is between 25 and 64.” Id. at 507–08. Employer sponsored health and welfare plans were similarly voluntaristic; however, this began to change as a result of the “employer mandate” contained within the Patient Protection and Affordable Care Act, 42 U.S.C. §18001 (2012).

 [26]. Occupational Safety and Health Act of 1970, 29 U.S.C. §§ 651–52 (2012).

 [27]. National Labor Relations Act of 1935, 29 U.S.C. § 157 (2012).

 [28]. Although the Affordable Care Act did not entirely detach health insurance savings from work relationships, it created a system of “exchanges” on which individuals can purchase health insurance plans that are not contingent on employer sponsorship. King v. Burwell, 135 S. Ct. 2480, 2487 (2015) (“[T]he Act requires the creation of an “Exchange” in each State where people can shop for insurance, usually online.”) (citing 42 U.S.C. § 18031(b)(1)).

 [29]. See Alain Supiot, Beyond Employment: Changes in Work and the Future of Labour Law in Europe 26–57 (2001) (not rejecting the logic of tying safeguards to work even though one of its central conclusions was that the “employment relationship in its existing form has reached its limits”), noted in David Marsden, Introduction: Can the Right Employment Institutions Create Jobs?, in Labour Law and Social Insurance in the New Economy: A Debate on the Supiot Report 1, 3, 9 (David Marsden & Hugh Stephenson eds., 2001).

 [30]. Catherine K. Ruckelshaus, Labor’s Wage War, 35 Fordham Urb. L. J. 373, 378–83 (2008) (discussing the prevalence, costs to workers, and advantages to employers of misclassifying employees as independent contractors).

 [31]. See, e.g., NLRB v. United Ins. Co. of Am., 390 U.S. 254, 258 (1968); NLRB v. Hearst Publ’ns, Inc., 322 U.S. 111, 120–22 (1944), overruled in part by Nationwide Mut. Ins. Co., v. Darden, 503 U.S. 318 (1992); FedEx Home Delivery v. NLRB, 563 F.3d 492, 496–99 (D.C. Cir. 2009). Indeed, the Supreme Court of Mississippi’s frustration was such that it declared (in 1931) that,

[t]here have been many attempts to define precisely what is meant by the term “independent contractor;” but the variations in the wording of these attempts have resulted only in establishing the proposition that it is not possible within the limitations of language to lay down a concise definition that will furnish any universal formula, covering all cases.

Kisner v. Jackson, 132 So. 90, 91 (Miss. 1931).

 [32]. For this type of circular definition see, for example, Age Discrimination in Employment Act (ADEA) § 11(f), 29 U.S.C. § 630(f) (2012); Family and Medical Leave Act, 29 U.S.C. § 2611(3) (2006); Title VII, 42 U.S.C. § 2000e(f) (2012); Americans with Disability Act (ADA) § 101(4), 42 U.S.C. § 12111(4) (2012); Employee Retirement Income Security Act (ERISA) § 3, 29 U.S.C. § 1002(6) (2012).

 [33]. Nationwide Mut. Ins. Co. v. Darden, 503 U.S. 318, 325, 327 (citing Cmty. for Creative Non-Violence v. Reid, 490 U.S. 730 (1989) for the principle that “Congress means an agency law definition for ‘employee’ unless it clearly indicates otherwise” and stating that “[a]gency law principles comport . . . with our recent precedents and with the common understanding, reflected in those precedents, of the difference between an employee and an independent contractor”).

 [34]. See, e.g., Slingluff v. Occupational Safety & Health Review Comm’n, 425 F.3d 861, 867–68 (10th Cir. 2005) (applying Darden to the Occupational Safety and Health Act of 1970); Birchem v. Knights of Columbus, 116 F.3d 310, 312–13 (8th Cir. 1997) (applying Darden to the American with Disabilities Act of 1990); Frankel v. Bally, Inc., 987 F.2d 86, 89–90 (2d Cir. 1993) (applying Darden to the Age Discrimination in Employment Act of 1967).

 [35]. Carlson, supra note 3, at 298.

 [36]. See, e.g., West v. J.O. Stevenson, Inc., 164 F. Supp. 3d 751, 763 & n.6 (E.D.N.C. 2016) (noting that “in the labor relations context, the Fourth Circuit has instructed courts to examine only the economic realities of the employment relationship” because of “the more expansive definition of ‘employ’ used in the labor relations statutes” including the “FMLA or similarly-defined [FLSA]”).

 [37]. See id. See also U.S. v. Rosenwasser, 323 U.S. 360, 362–63 (1945) (noting that the FLSA definition of employment was created intentionally broad in order to fulfill the remedial purpose of the act); Zheng v. Liberty Apparel Co. Inc., 355 F.3d 61, 66 (2d Cir. 2003) (same).

 [38]. See, e.g., Lewis L. Maltby & David C. Yamada, Beyond “Economic Realities”: The Case for Amending Federal Employment Discrimination Laws to Include Independent Contractors, 38 B.C. L. Rev. 239, 249–50 (1997) (citing the pre-Darden case, Broussard v. L.H. Bossie, Inc., 789 F.2d 1158 (5th Cir. 1986), as an example of the continued importance of control even under the economic realities test).

 [39]. See id. at 249.

 [40]. See NLRB v. Hearst Publ’ns, Inc., 322 U.S. 111, 127 (1944) (“Unless the common-law tests are to be imported and made exclusively controlling, without regard to the statute’s purposes, it cannot be irrelevant that the particular workers in these cases are subject, as a matter of economic fact, to the evils the statute was designed to eradicate . . . .”). See also Keith Cunningham-Parmeter, From Amazon to Uber: Defining Employment in the Modern Economy, 96 B.U. L. Rev. 1673, 1677 (2016); Guy Davidov, The Three Axes of Employment Relationships: A Characterization of Workers in Need of Protection, 52 U. Toronto L.J. 357, 363–64 (2002); Linder, supra note 19, at 190.

 [41]. Alexander v. FedEx Ground Package Sys., 765 F.3d 981, 990–91 (citing Estrada v. FedEx Ground Package System, Inc., 64 Cal. Rptr. 3d 327, 336 (Ct. App. 2007), in which drivers subject to the same Operating Agreement as the Alexander plaintiffs were found to be employees based on “FedEx’s control over every exquisite detail of the drivers’ performance”).

 [42]. Ochoa v. McDonald’s Corp., 133 F. Supp. 3d 1228, 1239–40 (N.D. Cal. 2015) (allowing plaintiffs to move forward with their claim that McDonald’s is a joint employer alongside its franchisees on a theory of ostensible agency); Browning-Ferris Indus. of Cal., Inc., 362 N.L.R.B. No. 186, at 2 (2015) (revising the Board’s joint employer standard to require only “[r]eserved authority to control terms and conditions of employment” and that such control need not “be exercised directly and immediately”) (emphasis added).

 [43]. Goldstein et al., supra note 8, at 1094–1101 (arguing that Congress adopted the FLSA’s “suffer or permit” language from state child labor laws with the specific intention to ensure that the Act would cover workers not considered employees under the common law control test). Similarly, a 2016 Administrator’s Interpretation (AI) issued by the Obama Department of Labor specified that the Department would henceforth distinguish between “vertical” and “horizontal” joint employment and apply the “economic realities” test to determine vertical joint employer status instead of current FLSA regulations—a move that commentators immediately interpreted as reflecting “the agency’s longstanding priority to loosen joint employment standards.” Tammy McCutchen & Michael J. Lotito, DOL Issues Guidance on Joint Employment Under FLSA, Littler (Jan. 20, 2016), https://www.littler.com
/publication-press/publication/dol-issues-guidance-joint-employment-under-flsa. The AI was later withdrawn by the Trump Department of Labor. News Release, U.S. Dep’t of Labor,  US Secretary of Labor Withdraws Joint Employment, Independent Contractor Informal Guidance, Release No. 17-0807-NAT (June 7, 2017), https://www.dol.gov/newsroom/releases/opa/opa20170607.

 [44]. Dubal, supra note 15, at 85. A similar impulse led the Fourth Circuit to find that although Darden probably would not qualify as an employee under the control test, that outcome was inconsistent with ERISA’s purpose; consequently, they rejected it. Darden v. Nationwide Mut. Ins. Co., 796 F.2d 701, 705–06 (4th Cir. 1986), overruled in part by Nationwide Mut. Ins. Co. v. Darden, 503 U.S. 318 (1992).

 [45]. Davidov, supra note 40, at 367–68 (“Over the years, however, some courts—unsatisfied with the tests in their arsenal—have begun formulating a second test, this one aimed at the economic dependence of the worker.”).

 [46]. See, e.g., Cotter v. Lyft, Inc., 60 F. Supp. 3d 1067, 1075 (N.D. Cal. 2015); Cunningham-Parmeter, supra note 40, at 1677 (“scrutinizing the many sublayers of control” in order to outline “new methods for thinking about the concepts of ‘control’ and ‘employ’ that remain central to modem employment”).

 [47]. See, e.g., Katherine V. W. Stone, From Widgets to Digits: Employment Regulation for the Changing Workplace, at ix (2004) (discussing changes in work practices that demand new forms of regulation); Cynthia L. Estlund, The Ossification of American Labor Law, 102 Colum. L. Rev. 1527, 1530–32 (2002). On the importance of purposive statutory analysis, see, e.g., Brian A. Langille & Guy Davidov, Beyond Employees and Independent Contractors: A View From Canada, 21 Comp. Lab. L. & Pol’y J. 7, 12 (1999); Linder, supra note 19, at 187. See also Rachel Weiner & Lydia DePillis, How Congress Can Make Life Better for Uber Drivers and Bike Messengers, Wash. Post (June 3, 2015), http://wapo.st/1cytlM6?tid=ss_tw-bottom&utm_term=.3e485df68733 (quoting Sen. Mark Warner, D.-Virginia, as saying that “[the sharing economy] is a tidal change in the relationship between an individual and the workplace . . . . It’s stunning that nobody in Washington is talking about this”). It might even be that courts misrecognize the purpose of work law, in that they wrongly prioritize efficiency over other values like the ability to be free from subordination in a democratic society. Stephen F. Befort & John W. Budd, Invisible Hands, Invisible Objectives: Bringing Workplace Law and Public Policy into Focus 4–7 (2009) (making a similar argument with respect to work law’s devaluation of equity and voice); Brishen Rogers, Employment Rights in the Platform Economy: Getting Back to Basics, 10 Harv. L. & Pol’y Rev. 479, 500–05 (2016).

 [48]. Tomasetti, supra note 15, at 315.

 [49]. See, e.g., Sanjukta M. Paul, Uber as For-Profit Hiring Hall: A Price-Fixing Paradox and its Implications, 38 Berkeley J. Emp. & Lab. L. 233, 235 (2017); Dubal, supra note 15, at 123 (“Advocacy on behalf of taxi workers, for example, may involve engaging antitrust laws, regulatory laws, unfair competition laws, and even corporate laws.”). See also Ryan Calo & Alex Rosenblat, The Taking Economy: Uber, Information, and Power, 117 Colum. L. Rev. 1623, 1675 (2017) (advocating the use of consumer protection law). Intriguingly, both the antitrust and consumer protection arguments depend on taking seriously the platform argument that workers are consumers vis-à-vis the platforms through which they provide services.

 [50]. See Dubal, supra note 15, at 112 (drawing on ethnographic research among San Francisco taxi drivers for the observation that “though many [immigrant and non-white] drivers recognized the potential stability of being an employee, the status made them feel more out of control of their everyday lives”).

 [51]. Id. See also Yuval Feldman et al., What Workers Really Want: Voice, Unions, and Personal Contracts, 15 Emp. Rts. & Emp. Pol’y J. 237, 248 (2011) (“In other words, employees might gain more from personal contracts in terms of sense of influence and control, although in many cases they will have more limited bargaining power compared to unionized employees.”).

 [52]. FedEx Home Delivery, Inc. v. NLRB, 563 F.3d 492, 497 (D.C. Cir. 2009).

 [53]. Dubal, supra note 15, at 89–95 (tying the increasing importance of entrepreneurial opportunity in classifying workers to the rise of neoliberal ideology beginning in the 1970s).

 [54]. See Deepa Das Acevedo, Regulating Employment Relationships in the Sharing Economy, 20 Emp. Rts & Emp. Pol’y J. 1, 3–10 (2016) (arguing that platforms participate in the transactions they facilitate and substitute themselves for governmental safeguards, in contrast to “Renters” like Zipcar or “Swappers” like Couchsurfer).

 [55]. Id. It has become commonplace to preface any discussion of the sharing economy with the claim that there are no satisfactory taxonomies or definitions of this new space and by presenting a new taxonomy that can fill this definitional gap. See, e.g., Calo & Rosenblat, supra note 49, at 1466. But scholarship on the sharing economy is a few years old now and most commentators speaking from within scholarly or policy contexts have an understanding of what they and others are really interested in, even if that understanding is Potter Stewart-like in its articulation. See Jacobellis v. Ohio, 378 U.S. 184, 197 (1964) (Stewart, J., concurring).

 [56]. I applied for admission to the following sharing economy companies, not all of which are “platforms” according to the definition I use here and elsewhere, see supra note 54 and accompanying text: TaskRabbit, Instacart, Postmates, Rover, and Gigwalk. I was accepted by Instacart, Rover, and Gigwalk, and completed Instacart’s in-store training session, but did not work any shifts; I performed one “gig” on Gigwalk; and I established one client relationship (involving multiple visits) on Rover. I also viewed videos and completed online quizzes for the introductory course offered by Samaschool, an online provider of digital skills and internet-based work training that heavily incorporates platforms into its curriculum but does not exclusively focus on them. See Samaschool, http://www.samaschool.org (last visited Aug. 14, 2018).

 [57]. Scholars working on the sharing economy are increasingly mining online chat forums because of the relative difficulty in accessing sufficient numbers of platform workers for large-scale analysis and because opinions voiced on the forums have not been elicited for research purposes. See, e.g., Shu-Yi Oei & Diane M. Ring, The Tax Lives of Uber Drivers: Evidence from Internet Discussion Forums, 8 Colum. J. Tax L. 56, 66­–72 (2017) (using data from Reddit.com, Uberpeople.net, and the Intuit TurboTax AnswerXchange Forum and discussing methodological approaches to the use of online discussion forums as data sources); Alex Rosenblat & Luke Stark, Algorithmic Labor and Information Asymmetries: A Case Study of Uber’s Drivers, 10 Int’l J. Comm. 3758, 3760 (2016) (using data from five unnamed Uber driver chat forums).

 [58]. Although all the non-workers with whom I spoke are identified by name in this Article, I have anonymized all conversations with workers. This choice reflects a concern expressed to me by many workers about being identified while making critical comments regarding their platforms, even in the context of an academic research project. Indeed, I generally did not note down worker names in my field notes even though many of the workers I spoke with gave me their personal phone numbers for follow up conversations (I instead identified them by an interlocutor number, date, and location of interaction, much as they are referenced in footnotes below). I did record names for a few workers with whom I developed closer relationships, but our conversations continued based on an assumption of anonymity and are presented here accordingly.

 [59]. A few of these interlocutors, all of whom have been giving the issues surrounding platform labor careful thought for some time now and were generous enough to share their insights with me, include Kate Bahn, Todd Brogan, Harry Campbell, Ben Davis, Nicole DuPuis, Emily Guendelsberger, Kirk Hovenkotter, Michael McCall-Delgado, Jeremy Mohler, Mel Plaut, Alex Rosenblat, Becki Smith, and Katie Unger.

 [60]. Telephone Interview with Alex Rosenblat, Analyst, Data and Society (Aug. 11, 2016). See also Calo & Rosenblat, supra note 49, at 1634 (drawing on Rosenblat’s ethnography of Uber drivers).

 [61]. A survey of nearly 1,200 Uber and Lyft drivers revealed that 75.9 percent prefer to be independent contractors. Harry Campbell, 2018 Uber and Lyft Driver Survey Results, TheRideshareGuy (Feb. 26, 2018), https://therideshareguy.com/2018-uber-and-lyft-driver-survey-results-the-rideshare-guy.

 [62]. Sam is a composite figure based on my conversations with many platform workers. Sam’s demographic qualities are not intended to be representative of workers in any industry vertical or region: the few independent, large scale surveys we have of platform workers suggest that they are mostly young, white, and male. Still, his story reflects those of many of my interlocutors. On the profile of sharing economy workers, see, for example, Diana Farrell & Fiona Greig, JP Morgan Chase & Co. Inst., Paychecks, Paydays, and the Online Platform Economy: Big Data on Income Volatility 22 (2016); Andrew Jiang et al., The 2015 1099 Economy Workforce Report, Requests For Startups  30–39 (2015); Campbell, supra note 61.

 [63]. Many observers comment on the degree to which worker participation and satisfaction is dependent on a poor grasp of the financial realities of platform work or, at the very least, as a kind of bimodal distribution where the peaks represent considerable savvy and considerable ignorance. One researcher noted that many workers’ “financial logic is ‘whenever you’re making money you’re doing well’” but that “former professionals . . . are keeping spreadsheets at home.” Another felt that “even the folks who are making it work” fail to account for things like sick days, unexpected expenses, and vacation time as lost earning time. Telephone Interview with Alex Rosenblat, Analyst, Data and Society (Aug. 11, 2016); Telephone Interview with Katie Unger, Independent Labor Consultant (Aug. 8, 2016). My own fieldwork reflects the bimodal distribution of financial sophistication. See, e.g., Telephone Interview with TaskRabbit Tasker #1 (Sept. 12, 2016) (“I don’t do enough to pay any taxes on it, I’m exempt from taxes on it . . . because (a) I’m on a woman-owned business . . . and (b) I don’t make enough . . . to pay B&O taxes on it . . . I still have to claim the income but you know it’s nominal.”); Interview with Uber Driver #8, in Phila., Pa. (July 30, 2016) (“I’m supposed to be [tracking my expenses but] it’s too complicated . . . I’m gonna see after tax season how it worked out.”).

 [64]. Some of the more striking accounts of the realities of platform work include, Josh Dzieza, The Rating Game: How Uber and Its Peers Turned Us into Horrible Bosses, The Verge (Oct. 28, 2015), https://www.theverge.com/2015/10/28/9625968/rating-system-on-demand-economy-uber-olive-garden; Emily Guendelsberger, I Was an Undercover Uber Driver, My City Paper (May 7, 2015), http://mycitypaper.com/uberdriver; Sarah Kessler, Pixel and Dimed: On (Not) Getting By in the Gig Economy, Fast Company (Mar. 18, 2014), https://www.fastcompany.com/3027355/pixel-and-dimed-on-not-getting-by-in-the-gig-economy.

 [65]. Some commentators have suggested that platform workers enjoy even more freedom than the ability to decide when they will be available for work by opening a smartphone app or activating an online profile. In this view, even when workers are “active” on the platform, they can decide not to work because they have the option to decline client requests. However, as others have pointed out, this is simply the freedom we all possess to resist the demands of our jobs knowing that such resistance will produce disciplinary action or termination. Compare Seth D. Harris & Alan B. Kreuger, The Hamilton Project, A Proposal for Modernizing Labor Laws for Twenty-First-Century Work: The “Independent Worker” 9 (2015) (“Even if she does not turn off either app, she is not obligated to pick up any particular customer.”), with Ross Eisenbrey & Lawrence Mishel, Uber Business Model Does not Justify a New ‘Independent Worker’ Category, Econ. Pol’y Inst. (Mar. 17, 2016), https://www.epi.org/publication/uber-business-model-does-not-justify-a-new-independent-worker-category (“Uber drivers cannot keep their app on and monitoring potential riders and refuse to accept rides without incurring serious consequences, including being deactivated (i.e., fired) for having too low an ‘acceptance rate.’”). Note that Uber changed its official policy regarding deactivation in 2016 and that now, drivers should not be deactivated for low acceptance rates, but rather be subject to the smaller disciplinary measure of a temporary “time out” by being involuntarily logged out of the app. Harry Campbell, How to Take Advantage of Uber’s New Acceptance Rate Policy, TheRideshareGuy, (Aug. 5, 2016), https://therideshareguy.com/how-to-take-advantage-of-ubers-new-acceptance-rate-policy (describing the old acceptance rate policy, the new “time out” policy, and the benefits of being able to take longer to accept requests).

 [66]. Interview with Uber Driver #19, in Phila., Pa. (Aug. 14, 2016) (stating no more than 6 hours per day); Interview with Uber Driver #34, in Phila., Pa. (Sept. 3, 2016) (stating only weekends).

 [67]. Interview with Uber Driver #35, in Phila., Pa. (Sept. 3, 2016).

 [68]. Id.

 [69]. Interview with Uber Driver #42, in Phila., Pa. (Sept. 12, 2016) (“You leave, I put on some Spanish music . . . some hip hop . . . . I like it.”); Interview with Uber Driver #37, in Phila., Pa. (Sept. 4, 2016) (playing loud classical music during the ride); Interview with Uber Driver #57, in Phila., Pa. (Oct. 3, 2016) (waiting for twnty to thirty minutes, then targeting high-demand areas); Interview with Uber Driver #14, in Phila., Pa. (Aug. 6, 2016) (remaining parked in between rides).

 [70]. See generally What 2 Things Do You Wish You Knew Before You Started Hosting, Airbnbhostsforum.com (Sept. 27, 2016), https://airhostsforum.com/t/what-2-things-do-you-wish-you-knew-before-you-started-hosting/8133/37 (allowing users to recommend how to host).

 [71]. Marc Linder, Towards Universal Worker Coverage Under the National Labor Relations Act: Making Room for Uncontrolled Employees, Dependent Contractors, and Employee-Like Persons, 66 U. Det. L. Rev. 555, 556 (1989).

 [72]. See, e.g., Over/Uber, Comment to Advice to Addictive Personalities, UberPeople.net (Jan. 30, 2017), https://uberpeople.net/threads/advice-to-addictive-personalities.137449 (describing the “rush of thinking the ‘system’ or competing drivers have been outsmarted, the rider outwitted”).

 [73]. See Matt MacFarland, Uber’s Remarkable Growth Could End the Era of Poorly Paid Cab Drivers, Wash. Post (May 27, 2014), https://wapo.st/TOpLVu?tid=ss_tw-bottom&utm_term
=.282f4583ee7d  (“According to Uber, the median wage for an UberX driver working at least 40 hours a week in New York City is $90,766 a year.”).

 [74]. Interview with Uber Driver #34, in Phila., Pa. (Sept. 3, 2016) (describing a $200 per week goal); Interview with Uber Driver #51, in Phila., Pa. (Sept. 30, 2016) (describing a $200 per day goal); Interview with Lyft Driver #2, in Phila., Pa. (Mar. 22, 2017) (stating that he drives enough to cover his car payments).

 [75]. Dubal, supra note 15, at 69, 112–20. There may be parallel divisions in the rideshare context based on full-time versus part-time status rather than on race. See Campbell, supra note 61 (“When comparing full-time vs part-time drivers, we see a slight preference toward employee status from full-time drivers but a majority still want to be independent contractors.”).

 [76]. Dubal, supra note 15, at 117–20.

 [77]. Id. at 113–15.

 [78]. Id. at 111–14.

 [79]. Indeed, Pettit—quoting Kant, who was responding to Rousseau—says something like this when he observes that “[f]ind himself in what condition he will, the human being is dependent on many external things . . . But what is harder and more unnatural than this yoke of necessity is the subjection of one human being under the will of another. No misfortune can be more terrifying . . . .” Pettit, On the People’s Terms, supra note 9, at 44.

 [80]. M.C. Elish, Moral Crumple Zones: Cautionary Tales in Human-Robot Interaction 1–2 (We Robot, Working Paper Presented at We Robot Conference, 2016) (on file with author) (explaining “moral crumple zones” in the context of human-machine systems like airplane cockpit control and nuclear plant emergency protocols, and stating that “humans at the interface between customer and company are like sponges, soaking up the excess of emotions that flood the interaction but cannot be absorbed by faceless bureaucracy or an inanimate object”). Not having a human “crumple zone” is important for “employer” perceptions of self as well as lay or legal perceptions of the alleged employer. See, e.g., Lilly Irani, Difference and Dependence among Digital Workers: The Case of Amazon Mechanical Turk, 114 S. Atl. Q. 225, 226–27 (2015) (“The transformation of workers into a computational service . . . serves not only employers’ labor needs and financial interests but also their desire to maintain preferred identities; that is, rather than understanding themselves as managers of information factories, employers can continue to see themselves as much-celebrated programmers, entrepreneurs, and innovators.”).

 [81]. Noah Zatz, Beyond Misclassification: Gig Economy Discrimination Outside Employment Law, On Labor (Jan. 19, 2016), https://onlabor.org/beyond-misclassification-gig-economy-discrimination-outside-employment-law (discussing “the simmering concern about how customer feedback ratings may hard-wire discrimination into the supervisory techniques of gig economy platforms”).

 [82]. Farrell & Greig, supra note 62, at 5; Aaron Smith, Gig Work, Online Selling and Home Sharing, Pew Research Ctr. (Nov. 17, 2016), http://www.pewinternet.org/2016/11/17/gig-work-online-selling-and-home-sharing; Interview with Sarah Leberstein, Attorney, Nat’l Emp’t Law Project, in N.Y.C., N.Y. (July 28, 2016); Telephone Interview with Rebecca Smith, Deputy Director, Nat’l Emp’t Law Project (Aug. 30, 2016).

 [83]. Add Me to Another Driver-Partner’s Profile, Uber, https://help.uber.com/h/c736a696-3eac-422a-a7f7-7d61e532d7c8 (last visited Aug. 16, 2018).

 [84]. Indeed, the personal labor of Airbnb hosts has interesting implications for the acceptability of pink- and blue-collared work. Juliet B. Schor, Does the Sharing Economy Increase Inequality Within the Eighty Percent?: Findings from a Qualitative Study of Platform Providers, 10 Cambridge J. Regions, Econ. & Soc’y 263, 272–75 (2017) (discussing how platforms are getting white collar workers to perform pink- and blue-collar work by presenting the work as novel and technologically advanced). Pink collar work is work that is neither “white collar” (professional or managerial) nor “blue collar” (manual, whether skilled or unskilled); it is traditionally associated with clerical and secretarial office work, but often extends to other forms of personal service work that are similarly dominated by women. Emily Stoper, Women’s Work, Women’s Movement: Taking Stock, 515 Ann. Am. Acad. Pol. & Soc. Sci. 151, 156 (1991) (discussing pink collar jobs in the course of analyzing approaches to reducing the wage gap between men and women).

 [85]. Rebecca Smith & Sarah Leberstein, Nat’l Emp. L. Project, Rights On-Demand: Ensuring Workplace Standards and Worker Security in the On-Demand Economy 4 (2015).

 [86]. Lee et al., who first applied the term “algorithmic management” to platform work, describe it as a phenomenon enabled by “software algorithms that assume managerial functions and surrounding institutional devices that support algorithms in practice.” See Min Kyung Lee et al., Working with Machines: The Impact of Algorithmic and Data-Driven Management on Human Workers, 33 Ann. ACM Conf. on Hum. Factors in Computing Sys. 1603 (2015). They go on to study three “algorithmic features” of rideshare platforms: “passenger-driver assignment, the dynamic display of surge-priced areas, and the data-driven evaluation that uses acceptance rates and ratings.” Id at 1604. See generally Rosenblat & Stark, supra note 57 (seeming to expand this understanding of algorithmic management by also including minimum fares, rate cuts, and dispute resolution within its ambit). Regardless of one’s exact definition of algorithmic management, both “labor” and “capital” platforms clearly engage in it.

 [87]. Indeed, not all of the labor and employment problems considered by commentators who draw this distinction pertain to the regulation of work: sometimes, as with the JPMorgan report, labor and capital platforms are distinguished from one another because doing so tracks meaningful differences in who is participating, as well as how and why they are participating.

 [88]. I understand interference as the “removal, replacement or misrepresentation of one or more options” available to an individual. Pettit, On the People’s Terms, supra note 9, at 46. It is worth noting that Pettit views the attachment of a penalty as the “replacement” of one option with another that is different and less desirable (though perhaps not by much). Id. at 53. He also considers “misrepresentation” to include any action that “denies you the possibility of making a choice on the basis of a proper understanding of the options on offer” and lists “mesmerizing you with the prospect of extraordinary rewards” as one way of doing this. Id. at 55.

 [89]. Das Acevedo, supra note 7, at 42–43 (discussing these and other types of discipline established by platforms); Rosenblat & Stark, supra note 57, at 3761.

 [90]. Id. at 43–44 (discussing elite statuses). On codes of conduct, see Katie Benner, Airbnb Adopts Rules to Fight Discrimination By its Hosts, N.Y. Times (Sept. 8, 2016), https://nyti.ms/2cw2IsD, for a description of Airbnb’s new “community commitment” to non-discrimination that new and returning users must agree to before they can make bookings on the platform and Airbnb’s Nondiscrimination Policy: Our Commitment to Inclusion and Respect, Airbnb https://www.airbnb.com/help/article
/1405/airbnb-s-nondiscrimination-policy–our-commitment-to-inclusion-and-respect (last visited Aug. 16, 2018), for a description of Airbnb’s detailed non-discrimination policy—mostly geared towards hosts—that is powerfully reminiscent of Title VII and the ADA.

 [91]. See Rosenblat & Stark, supra note 57, at 3765–71 (discussing surge pricing). This is not to say that surge pricing succeeds in “mesmerizing” drivers into behaving according to the wishes of platforms—“don’t chase the surge” is standard advice to new drivers—but interference is not predicated on successful manipulation. Pettit, On the People’s Terms, supra note 9, at 55–56; Rosenblat & Stark, supra note 57, at 3766.

 [92]. Noam Scheiber, How Uber Uses Psychological Tricks to Push Its Drivers’ Buttons, N.Y. Times (Apr. 2, 2017), https://nyti.ms/2nMmDtc (discussing the use of earnings goals, among other things, to shape driver behavior “without giving off a whiff of coercion”).

 [93]. For instance, rideshare drivers complain frequently about acceptance rate and surge pricing policies, but seem to take less issue with elite status policies. See Lee et al., supra note 86, at 1603, 1608 (discussing disapproval of acceptance rate and surge pricing policies).

 [94]. An English employment tribunal poked fun at the interpretation of worker behavior as the result of good, but independently exercised business sense rather than as the product of a centralized authority, stating that “[t]he notion that Uber in London is a mosaic of 30,000 small businesses linked by a common ‘platform’ is to our minds faintly ridiculous.” Aslam v. Uber B.V., No. 2202550/2015, at 28 (Emp.’t Tribunals Oct. 28, 2016).

 [95]. In the neo-republican account, domination destroys freedom because it involves the power to restrict choice regardless of whether or not that power is actually exercised. To be sure, where domination leads to actual interference, the interference is itself also a manifestation of unfreedom. See, e.g., Pettit, On the People’s Terms, supra note 9, at 50.

 [96]. Much of the neo-republican literature refers to this state of affairs as vulnerability to the “arbitrary” will of another being, but Pettit rightly points out that modern English usage makes arbitrary a misleading choice of words: what matters isn’t the irrationality or unpredictability of the external will imposing itself upon you, but the fact that that will is uncontrolled by you in any meaningful sense. Compare Rogers, supra note 47, at 500 (using “arbitrary”), with Pettit, On the People’s Terms, supra note 9, at 58 (preferring “uncontrolled interference” to “arbitrary interference”).

 [97]. Pettit uses the terms “adaptation” and “ingratiation” instead of servility, but—at least in the employment context—these do not quite capture the sense of self-abnegation that I think worries critics. Pettit, On the People’s Terms, supra note 9, at 64–65.

 [98]. Like Sam, Alex is a composite figure.

 [99]. See Schor, supra note 84, at 272–74 (discussing the performance of pink and blue-collar tasks by white collar platform workers).

 [100]. One of the Uber drivers I interviewed said just this about his early experiences as a driver. He coped by switching off the app after completing just one ride per day until, after a few days, he felt he was sufficiently desensitized to handle more passengers without communicating his discomfort through his conversation. It took two more weeks for him to be comfortable driving full time. Interview with Uber Driver #24, in Phila., Pa. (Aug. 20, 2016).

 [101]. Another driver who emphasized his interests in psychology noted: “Sometimes somebody gets in your car and the best customer service you can provide is to say nothing . . . you can tell almost automatically . . . some people don’t have that gear but I know when to talk, when not to talk.” Interview with Uber Driver #26, in Phila., Pa. (Aug. 31, 2016).

 [102]. Food smells are a common topic of conversation on the Airbnb hosts’ forum. See, e.g., Guest Refuses to Stay Due to Smell?, Airbnbhostsforum.com (June 28, 2016), https://airhostsforum.com
/t/guest-refuses-to-stay-due-to-smell/5859; How Much Does Indian Food Smell Matter to Others?, Airbnbhostsforum.com (Oct. 11, 2016), https://airhostsforum.com/t/how-much-does-indian-food-smell-matter-to-others/8855.

 [103]. See Airbnb’s Nondiscrimination Policy, supra note 90.

 [104]. Workers—not just within the sharing economy—do often describe their efforts in somewhat heroic language. See, e.g., Dubal, supra note 15, at 119 (“As a taxi driver, I’m navigating San Francisco streets. My customer’s lives are in my hands as I take them from place to place. I am handling a weapon.”); Interview with Uber Driver #50, in Phila., Pa. (Sept. 19, 2016) (“I’m a superhero [because I get people to work].”). Uber has also added a tip feature in its app. Darrell Etherington, Uber Tipping Is Rolling Out to 121 U.S. and Canadian Markets Today, TechCrunch (July 6, 2017), https://techcrunch.com/2017/07/06/uber-tipping-is-rolling-out-to-121-u-s-and-canadian-cities-today.

 [105]. One of the drivers I spoke with expressed all these sentiments at once. Interview with Uber Driver #55, in Phila., Pa. (Oct. 1, 2016) (“I was Ubering so much I wasn’t getting much sleep . . . . Sometimes I do get a little angsty, I wanna get out there . . . you know, tuition’s due. The sad part [is that] I was thinking about retiring in four years . . . now Uber’s talking about driverless cars in four years so what I’m gonna do?”).

 [106]. To be clear, I am not using “unfree” in the sense that historians of nineteenth century labor law do. See, e.g., Robert J. Steinfeld, Coercion, Contract and Free Labor in the Nineteenth Century 33 (2001) (calling “unfreedom” the “coercion of labor through threats of corporal punishment or confinement”). Rather, I am using “unfreedom” to signal a less technical sense (albeit more aligned with agency law) that an individual’s actions are not her own to control.

 [107]. Changes need not only be of the job-negating variety. When TaskRabbit overhauled its system in 2014 from a bidding format to a selection and assignation format, it too effected systemic changes of the type being discussed here. Colleen Taylor, Through the Fire: What TaskRabbit Learned From Its Big Backlash, TechCrunch (Jan. 21, 2015), https://techcrunch.com/2015/01/21/through-the-fire-what-taskrabbit-learned-from-its-big-backlash.

 [108]. But cf. Calo & Rosenblat, supra note 49, at 1631. As rideshare expert Harry Campbell notes, technological obstacles to driverless cars are often both overestimated and underestimated. On the one hand, Tesla models are just a shade away from being able to do the work of navigating roadways. On the other hand, driving—especially driving for hire—involves far more than successfully decelerating or changing lanes. What happens when an Uber passenger vomits in a driverless vehicle? Or when a temporary roadblock is erected halfway down a one-way street? There will surely be technological fixes for these issues, but Campbell suggests that we are far from having them on hand. Moreover, he suggests that given the thickness of regulatory infrastructure in the United States, driverless transportation—whether in the form of individual cars or something analogous to Hyperloop—is likely to debut in a more flexible environment (with his personal pick being Dubai). Telephone Interview with Harry Campbell, TheRideshareGuy.com (Apr. 17, 2017).

 [109]. There are certainly strong connections between “uncertainty” on the one hand and “vulnerability” (as in the work of Martha Albertson Fineman) and “dependence” (as under the economic realities test) on the other hand. However, I have used uncertainty because it seems to best capture the source of platform workers’ vulnerability to and dependence on their platforms and consumers. See generally Sec’y Labor v. Lauritzen, 835 F.2d 1529, 1534 (7th Cir. 1987) (“For purposes of social welfare legislation, such as the FLSA, ‘employees are those who as a matter of economic reality are dependent upon the business to which they render service.’”) (citations omitted); Martha Albertson Fineman, The Vulnerable Subject and the Responsive State, 60 Emory L.J. 251 (2010) (arguing that vulnerability and dependence are inescapable aspects of the human condition and criticizing the extent to which they are overlooked in liberal law and policy analysis). Vulnerability has also been important to calls for a purposive approach to worker classification. See, e.g., Davidov, supra note 40, at 361.

 [110]. Over the course of the nineteenth century, the default rule for employment in the United States came to be that it exists “at will”—that is, so long (and only so long) as both parties are willing to uphold the contract. An employer does not have to offer cause or notice for terminating an employee, and vice versa. This principle, often called Wood’s Rule for its ostensible origins in Horace Gray Wood’s 1877 treatise Master and Servant, is usually captioned as allowing termination for “good reason, bad reason, or no reason at all.” See, e.g., Paul M. Secunda, Constitutional Employment Law: Zimmer’s Intuition on the Future of Employee Free Speech Law, 20 Emp. Rts & Emp. Pol’y J. 393, 405 (2016). The uniquely American nature of Wood’s Rule does not explain the dissatisfaction with control-based analysis I describe here because similar criticisms of control as an analytic rubric have been voiced in other contexts lacking at-will employment. See Jeremias Prassl, The Concept of the Employer 1–7 (2015) (discussing, primarily in the context of English law, the problem with control-based analysis that seeks to identify “employees”); Rachel Arnow-Richman, Just Notice: Re-Reforming Employment at Will, 58 UCLA L. Rev. 1, 48–57 (2010) (discussing the non-at-will systems in Canada, the UK, and various European countries); Langille & Davidov, supra note 47, at 15–16 (noting the importance of control to Canadian worker classification doctrine in a section titled “Our ‘Traditional’ Problem”).

 [111]. Calo & Rosenblat, supra note 49, at 1665.

 [112]. Fiverr’s Level System, Fiverr, https://www.fiverr.com/levels (last visited Aug. 16, 2018). Fiverr is an odd-job platform where clients can hire workers to perform a range of tasks, including anything from translating a document to designing a website. The platform began with the premise that all tasks would cost just $5, but it has since expanded its pricing approach into a complex system whereby workers gain the right to tailor the cost and nature of the services they offer as they climb Fiverr’s internally constructed hierarchy of elite service provider statuses. See Das Acevedo, supra note 7, 42–43 (discussing vetting and termination standards and the benefits that come with elite statuses across several platforms).

 [113]. See Noopur Raval & Paul Dourish, Standing Out From the Crowd: Emotional Labor, Body Labor, and Temporal Labor in Ridesharing, 19 ACM Conf. Comp.-Supported Cooperative Work & Soc. Computing 101 (“‘Pleasing the passenger’ is clearly an aspect of any ridesharing system, including traditional taxis, but it plays a much bigger role in crowd labor due to the specific intermediation of quantitative scores.”).

 [114]. Pettit, On the People’s Terms, supra note 9, at 64–67.

 [115]. Dzieza, supra note 64.

 [116]. Erving Goffman’s classic account of face-to-face encounters as “games” is particularly visible in the one-to-one or one-to-many interactions of the sharing economy. Goffman notes that in games like checkers or chess, the rules of the outside world are more or less suspended during the course of the interaction. Erving Goffman, Encounters: Two Studies in the Sociology of Interaction 26 (1961). When platform workers perform emotional labor, they are following their own internal “rules of the game” by ignoring gender and class stereotypes attached to pink and blue-collar tasks. Schor, supra note 84, at 272–74.

 [117]. Arlie Hochschild’s understanding of emotional labor remains definitive, as does her classic “flight attendant” illustration:

The flight attendant does physical labor when she pushes heavy meal carts through the aisles, and she does mental work when she prepares for and actually organizes emergency landings and evacuations. But in the course of doing this physical and mental labor, she is also doing something more, something I define as emotional labor. This labor requires one to induce or suppress feeling in order to sustain the outward countenance that produces the proper state of mind in others—in this case, the sense of being cared for in a convivial and safe place. This kind of labor calls for a coordination of mind and feeling, and it sometimes draws on a source of self that we honor as deep and integral to our individuality.

Arlie Russell Hochschild, The Managed Heart: Commercialization of Human Feeling 6–7 (1983).

 [118]. DRider85, Comment to Uber and Lyft Should Get Rid of 5 Star Ratings, UberPeople.net (Feb. 1, 2017), https://uberpeople.net/threads/uber-and-lyft-should-get-rid-of-5-star-ratings.137751.

 [119]. Written communication style—punctuation, vocabulary, and the use of politeness conventions, for example—reflects and shapes “real-world” dynamics. Naomi S. Baron and Rich Ling, Necessary Smileys and Useless Periods: Redefining Punctuation in Electronically-Mediated Communication, 45 Visible Language 46, 55 (2011) (observing that female study participants “were vocal about the importance of using emotion-tinged punctuation markers . . . both to express their ‘enthusiasm’ for the communications they were crafting as well as to soften messages that might otherwise seem overly direct”). When a TaskRabbit tasker texts immediately after receiving a client request with a series of short, enthusiastic questions liberally sprinkled with exclamation marks, she is not simply responding in a naturally excited or voluble way, she is trying to ingratiate herself with her client by demonstrating her interest in service as much as the Uber driver who unnecessarily asks her clients about restaurants.

  The centrality of emotional labor simply places platforms at the tail end of a very long line of service jobs in which the work to be done encompasses much more than the cleaning of a bathroom or the distribution of in-flight beverages, and where the affective dimensions of that work—and the display of pleasure in service—serve to refashion power disparities as good customer service. See generally Hochschild, supra note 117; Robin Leidner, Emotional Labor in Service Work, 561 Annals Am. Acad. Pol. & Soc. Sci. 81 (1999) (discussing McDonald’s fast food restaurants).

 [120]. In this vein, Hochschild distinguishes “surface acting”—in which “we deceive others about what we really feel, but we do not deceive ourselves”—from “deep acting”—in which “we make feigning easy by making it unnecessary.” Hochschild, supra note 117, at 33. She notes that “[t]he matter would be simpler and less alarming” if workers were “allowed to see and think as they like and required only to show feeling (surface acting) in institutionally approved ways” but that “[s]ome institutions have become very sophisticated in the techniques of deep acting; they suggest how to imagine and thus how to feel.” Id. at 49.

 [121]. Indeed, in one of the earliest first-hand journalistic accounts of platform work (and in the context of an article that was otherwise rather critical of the workings of ridesharing), Emily Guendelsberger observes that she was “weirdly proud” of the fact that she maintained a perfect five-star rating during her first few days as an Uber driver. Emily Guendelsberger, supra note 64.

 [122]. Scheiber, supra note 92.

 [123]. Id.

 [124]. Pettit, On the People’s Terms, supra note 9, at 65.

 [125]. See Scheiber, supra note 92 (describing various tools or approaches that Uber has developed using behavioral science insights to control driver desires as well as driver actions, including: having Uber employees adopt female personas when interacting with drivers, exploiting the “ludic loop” phenomenon by setting random, but concrete and constantly-shifting goals (e.g., “you’re $5.25 away from earning $330!”), offering work statistics and feedback in formats akin to video game scores to trigger competitive urges, and capitalizing on inertia and loss aversion by using automatic queuing features like “forward dispatch” to keep drivers on the road).

 [126]. Anthropologists have been thinking of resistance as more than self-conscious group protest for some time now. Sally Engle Merry, Resistance and the Cultural Power of Law, 29 Law & Soc’y Rev. 11, 15 (1995) (describing the movement toward understanding resistance as consisting of “subtle, unrecognized practices, such as foot-dragging, sabotage, subversive songs, and challenges to the law’s definition of personal problems in court”). See generally Rosenblat & Stark, supra note 57 (conducting a study of Uber driver chat forums that also speaks to the way drivers contest passenger and platform narratives). On “gaming” the system, see, for example, Nick Loper, How I Got on the Homepage of Fiverr and Earned $920 in 10 Days, Side Hustle Nation (Apr. 21, 2014), https://www.sidehustlenation.com
/fiverr-homepage-earned-920-in-10-days (describing several strategies for achieving a high-ranking on search results or a platform webpage). See also Advice to Addictive Personalities: Post of Over/Uber, UberPeople.net (Jan. 30, 2017), https://uberpeople.net/threads/advice-to-addictive-personalities
.137449 (describing the “rush of thinking the ‘system’ or competing drivers have been outsmarted, the rider outwitted”).

  Lastly, several commentators as well as several of my own interlocutors have described the practice of “going off app” by concluding an initial transaction or establishing repeat transactions directly between the worker and consumer. See, e.g., Kessler, supra note 64 (describing her initial guilt at accepting payment “off app,” but subsequent willingness to do so because of the unexpected difficulty in acquiring gigs); Conversation with Rover client #1, in Phila., Pa. (Jan. 19, 2017) (during which the client assumed we would schedule all sessions off the app); Interview with Uber Driver #25, in Phila., Pa. (Aug. 31, 2016) (describing the same practice involving ridesharing); Telephone Interview with Blow Me Worker #1 (Aug. 11, 2016) (describing how the stylist met new clients via the app and scheduled subsequent sessions with them directly).

 [127]. In other words, the uncertainty generated by the power dynamics of platform work—as well as the conformity and obsequiousness it prompts—constitute a “habitus,” or system of embodied tendencies that are “structured structures predisposed to function as structuring structures.” Pierre Bourdieu, Outline of a Theory of Practice 53 (1977).

 [128]. Tomasetti, supra note 15, at 366 (describing UCS as “the setting forth of detailed and extensive work rules in the written contract governing the work” and noting that it “is not just artifice disguising the alleged employer’s open-ended authority over production . . . but to some extent meaningfully directs the worker”).

 [129]. Id. at 368 (describing how many courts view upfront specification as removing the need for supervision rather than encoding supervision into the contract).

 [130]. Id. at 374 (discussing In re FedEx Ground Package Sys., Inc., 869 F. Supp. 2d 942 (N.D. Ind. 2012)).

 [131]. See, e.g., Alexander v. FedEx Ground Package Sys., 765 F.3d 981, 997 (9th Cir. 2014) (finding that FedEx drivers are employees).

 [132]. Tomasetti, supra note 15, at 368–76 (discussing several other examples of UCS analysis, including EEOC v. North Knox School Corp., 154 F.3d 744, 748 (7th Cir. 1998) (describing UCS analysis in a case involving school bus drivers); SIDA of Hawaii, Inc. v. NLRB, 512 F.2d 354, 358 (9th Cir. 1975) (containing similar analysis involving taxi drivers); and Brown v. Sears, Roebuck & Co., 31 Employee Benefit Cas. (BNA) 2467, 2470 (N.D. Ill. Sept. 24, 2003), aff’d, 125 F. App’x 44 (7th Cir. 2004) (containing similar analysis, but involving the employees of a Sears contractor that was responsible for selling and installing home improvements like roofs, gutters, and fences)).

 [133]. Tomasetti, supra note 15, at 325 (noting that the “UCS poses an intractable quandary for evaluating claims of control in employment status disputes and that the conundrum is rooted in the contradictory incorporation of master-servant authority into contract”). The results in FedEx cases where drivers are not found to be employees despite various forms of UCS-enabled authority seem counterintuitive because they clash with a vision of freedom as non-domination. Take two of the practices that FedEx presents in support of its position: drivers are free to subcontract their routes and they are not required to work specified hours. Alexander, 765 F.3d at 984–85. Both rightly suggest that FedEx does not engage in the kind of over-the-shoulder monitoring that would constitute violations of freedom-as-non-interference because the company is not telling a particular driver that she herself has to work these particular hours. Id. at 985. But drivers—and many commentators—feel that their freedom is nonetheless compromised for reasons that reflect an understanding of freedom as non-domination. For instance, drivers’ ability to subcontract is at the unlimited discretion of company officials, which raises familiar concerns like uncertainty and obsequiousness. Id. at 994. Likewise, drivers are not really free to work any hours they want because FedEx engages in highly sophisticated calculations to ensure that every driver works 9.5–11 hours daily, reports for pickup in the mornings, remains until all her packages are collected, and delivers certain packages at specific times—in other words, FedEx artificially and unilaterally limits drivers’ freedom ex ante so that they can “choose” to do exactly what FedEx desires. Id. at 984–85.

 [134]. See Eileen Silverstein, From Statute to Contract: The Law of the Employment Relationship Reconsidered, 18 Hofstra J. Lab. & Emp. L. 479, 482 (2001).

 [135]. Id. at 511.

 [136]. Waivers are generally assessed for indications that they were “knowing and voluntary,” and most courts make this determination using the capacious (and thus employer-friendly) “totality of the circumstances” standard. Id. at 484, 491 n.67. Congress did enact more stringent requirements for determining that a waiver was “knowing and voluntary” via the Older Workers Benefit Protection Act of 1990 (OWBPA), but most courts only apply the OWBPA’s seven-step analysis in cases involving the Age Discrimination in Employment Act. Id. at 488–92 & nn. 61–67.

 [137]. Id. at 493.

 [138]. And indeed, situations in which X changes the set of options available to Y (even by attaching a reward to one of the options) or situations in which X deceives or manipulates Y into picking a specific option are equally unlikely to constitute “free choice” under a vision of freedom as non-domination. Pettit, On the People’s Terms, supra note 9, at 50–56.

 [139]. See Silverstein, supra note 134, at 484; Tess Wilkinson-Ryan, Intuitive Formalism in Contract, 163 U. Pa. L. Rev. 2109, 2110, 2127 (2015) (stating that just as modern contract doctrine posits that “potentially enforceable deals (i.e., those that are supported by consideration and not illegal or unconscionable)” should be upheld “when the parties have objectively manifested assent,” “[t]he fact of assent seems, for the average consumer, to cleanse the transaction—to press the reset button, morally as well as legally” on the issue of enforceability).

 [140]. Gourevitch, supra note 9, at ch. 4. See also Carlson, supra note 3, at 310 (noting that “control over work was never the exclusive test of status for either respondeat superior or other statutory purposes”).

 [141]. Gourevitch, supra note 9, at 103.

 [142]. Id.

 [143]. See infra notes 155–56, 159–60.

 [144]. Fair Labor Standards Act of 1938, 29 USC § 201, §§ 203(d)–(e) (2012); Rutherford Food Corp. v. McComb, 331 U.S. 722, 727–30 (1947) (applying the economic realities test to the FLSA). Of course, the economic realities test was not originally limited to, or even articulated with respect to, the FLSA. See, e.g., United States v. Silk, 331 U.S. 704, 717–20 (1947) (elaborating the economic realities test in a case involving the Social Security Act); NLRB v. Hearst Publ’ns, Inc., 322 U.S. 111, 127–29 (1947) (articulating the rationale underlying the economic realities test in a case involving the NLRA);  Carlson, supra note 3, at 311–14 (discussing analysis consonant with the economic realities test in Lehigh Valley Coal v. Yensavage, 218 F. 547 (2d Cir. 1914).

 [145]. See, e.g., Katherine V. W. Stone, Legal Protections for Atypical Employees: Employment Law for Workers Without Workplaces and Employees Without Employers, 27 Berkeley J. Emp. & Lab. L. 251, 257­–59 (2006).

 [146]. Bruce Goldstein et al., Enforcing Fair Labor Standards in the Modern American Sweatshop: Rediscovering the Statutory Definition of Employment, 46 UCLA L. Rev. 983, 1043 (1999) (tracing the historical meaning of “suffer or permit” and arguing that, in the state child labor statutes from which it originates, the phrase meant that “when a business owner had the means to know and the power to prevent acts proscribed by the legislature, it made no difference whether the owner’s employee or an independent contractor engaged a minor child to perform the work”).

 [147]. Browning-Ferris Indus. of Cal., Inc., 362 N.L.R.B. No. 186 (2015).

 [148]. Erin Conway & Caroline Fichter, Surviving the Tempest: Franchisees in the Brave New World of Joint Employers and $15 Now, 35 Franchise L.J. 509, 515 (2016) (observing that the “change in the NLRB employer test is vehemently opposed by the International Franchise Association (IFA) and bemoaned as the end of franchising altogether by some”) (citations omitted); David J. Kaufmann et al., A Franchisor Is Not the Employer of its Franchisees or Their Employees, 34 Franchise L.J. 439, 448–52 (2015) (discussing the dangerous implications of Browning-Ferris and the McDonald’s inquiry on business-format franchisors).

 [149]. Browning-Ferris, 362 N.L.R.B. No. 186, at 2 (emphasis in original).

 [150]. Id.

 [151]. NLRB Office of the General Counsel Issues Consolidated Complaints Against McDonald’s Franchisees and Their Franchisor McDonald’s, USA, LLC as Joint Employers, NLRB.gov (Dec. 19, 2014), https://www.nlrb.gov/news-outreach/news-story/nlrb-office-general-counsel-issues-consolidated
-complaints-against.

 [152]. In 2016, the Board affirmed and built on its Browning-Ferris precedent by holding that solely employed workers and jointly employed workers need not obtain employer assent if they wish to form a single (and otherwise appropriate) bargaining unit. Miller & Anderson, Inc., 364 N.L.R.B. No. 39, at 13–14 (2016).

 [153]. See generally Hy-Brand Indus. Contractors, 365 N.L.R.B. No. 165 (2017). However, the NLRB later vacated Hy-Brand and in 2018 undertook an unusual notice-and-comment rulemaking process on the joint-employer standard. Board Vacates Hy-Brand Decision, NLRB.gov (Feb. 26, 2018), https://www.nlrb.gov/news-outreach/news-story/board-vacates-hy-brand-decision; NLRB Considering Rulemaking to Address Joint-Employer Standard, NLRB.gov (May 9, 2018), https://www.nlrb.gov
/news-outreach/news-story/nlrb-considering-rulemaking-address-joint-employer-standard.

 [154]. See infra notes 155, 159–60.

 [155]. See, e.g., Linder, Towards Universal Worker Coverage Under the National Labor Relations Act, supra note 71, at 555 (calling the Taft-Hartley amendments to the NLRA “[s]tatutory encouragement” for “[o]ne of the successful tactics employers have used in recent years to rid themselves of existing labor unions and to avoid collective bargaining”—namely the reclassification of employees as independent contractors).

 [156]. Sean Farhang & Ira Katznelson, The Southern Imposition: Congress and Labor in the New Deal and Fair Deal, 19 Stud. Am. Pol. Dev. 1, 12–15 (2005) (describing southern efforts to shield the remnants of antebellum work structures from the New Deal by carving out statutory exclusions for predominantly black labor); Marc Linder, Farm Workers and the Fair Labor Standards Act: Racial Discrimination in the New Deal, 65 Tex. L. Rev. 1335, 1336 (1987) (same). See also Vivien Hart, Minimum-Wage Policy and Constitutional Inequality: The Paradox of the Fair Labor Standards Act of 1938, 1 J. Pol’y Hist. 319, 337 (1989) (discussing the original version of the FLSA and observing that “because of the segregated, stereotyped, and regional structure of the job market, excluded workers were disproportionately women and from minorities”).

 [157]. NLRB v. Hearst Publ’ns, 322 U.S. 111 (1944); Carlson, supra note 3, at 322–24 (discussing Taft-Hartley as a response to Hearst); Dubal, supra note 15, at 82–84 & nn. 48–53 (discussing the importance of business interests in motivating the 80th Congress’s efforts to unravel New Deal legislation and referencing prior scholarship arguing the same).

 [158]. Clifford Geertz, Deep Play: Notes on the Balinese Cockfight, in The Interpretation of Cultures 412, 449 (1973).

 [159]. Carlson, supra note 3, at 298; Jeffrey M. Hirsch, Employee or Entrepreneur?, 68 Wash. & Lee L. Rev. 353, 360–61 (2011) (discussing the power of federal appellate courts, and especially of the D.C. Circuit, with respect to labor law generally and the NLRB in particular).

 [160]. See, e.g., Cunningham-Parmeter, supra note 40, at 1704 (“Many judges narrowly construe the meaning of control because no clear standard exists to outline the boundaries of employer-employee relationships.”); Tomasetti, supra note 15, at 336 & n.110 (citing various scholars who argue that one explanation for “the legal uncertainty regarding employment status is that the legal standards are hopelessly imprecise and unwieldy”).

 [161]. Charles W. McCurdy, The “Liberty of Contract” Regime in American Law, in The State and Freedom of Contract 161, 165 (Harry N. Scheiber ed., 1998) (“[N]o amount of thoughtful revisionism can erase the fact that the ‘principle of neutrality’ did not have a uniform operation . . . the freedom of contract decisions handed down by American courts beginning in the 1880s showed ‘a definite bias of policy’ against statutes favoring ‘the interest . . . of labor.’”); Matthew J. Lindsay, In Search of “Laissez-Faire Constitutionalism,” 123 Harv. L. Rev. F. 55, 55, 57 (2010) (describing the “progressive” critique of Lochner era jurisprudence as emphasizing the judiciary’s identification “with the nation’s capitalist class” and its “contempt for any effort to redistribute wealth or otherwise meddle with the private marketplace,” and juxtaposing this with a second critique, according to which “the Lochner era is best understood not as a politically motivated binge of judicial activism, but rather as a sincere and principled, if sometimes anachronistic effort” to distinguish valid economic legislation and invalid “class” legislation).

 [162]. Browning-Ferris Indus. of Cal., Inc., 362 N.L.R.B. No. 186 (2015).

 [163]. Cotter v. Lyft, Inc., 60 F. Supp. 3d 1067 (N.D. Cal. 2015).

 [164]. Scholars who might subscribe to the label “New Legal Realist” employ a variety of approaches. See, e.g., Bryant Garth & Elizabeth Mertz, Introduction: New Legal Realism at Ten Years and Beyond, 6 U.C. Irvine L. Rev. 121, 123 n.10 (2016) (describing a “big tent” New Legal Realism that extends beyond the quantitative and economics-oriented framework represented by Miles and Sunstein); Thomas J. Miles & Cass R. Sunstein, Introduction: The New Legal Realism, 75 U. Chi. L. Rev. 831, 831, 834 (2008) (describing the understanding of judicial personality using “testable” metrics as “[a] distinguishing feature of the New Legal Realism”).

 [165]. Cotter, 60 F. Supp. 3d at 1081.

 [166]. Id.

 [167]. Id. See also id. at 1078–79 (listing various behavioral controls imposed by Lyft, like “not to talk on the phone with a passenger present”).

 [168]. Id. at 1079.

 [169]. See id. (observing that “Lyft reserves the right to ‘investigate’ and ‘terminate’ drivers who have ‘behaved in a way which could be regarded as inappropriate’” and that termination might result from declining or accepting and then declining “too many” ride requests).

 [170]. Id. The “at-will” rule has interesting, shifting implications for the dual conceptions of freedom I describe here. At-will employment is characterized by uncertainty and potential power—formally enjoyed by both parties, but in practice mostly beneficial to the employer. When the at-will nature of a particular relationship is used to characterize it as one of employer-employee, the at-will rule is indexing a loss of freedom as non-domination. But when taken generally, the rule represents a valuation of freedom as non-interference: it is liberty-enabling because it eliminates all but the most minimal intrusions on an employee’s decision to work and an employer’s decision to offer work. On the practical effect of the at-will rule see, e.g., Jay M. Feinman, The Development of the Employment-at-Will Rule Revisited, 23 Ariz. St. L.J. 733, 736 (1991) (“Whatever its status as a formal presumption, Wood’s rule represented a signal to the courts to view skeptically employees’ evidence of contracts of long duration.”).

 [171]. Whether they are “piecemeal” or “systemic,” the approaches I have in mind would focus both on (1) reducing the importance of the “employer”–worker binary as a funneling mechanism for safeguards and (2) searching for an analytic rubric other than control and freedom with which to funnel those safeguards. In terms of piecemeal reforms, we might continue to distance health and retirement benefits from worker classification in the vein of the Affordable Care Act and various “portable benefits” plans. See supra note 28 and accompanying text; Deepa Das Acevedo, Addressing the Retirement Crisis with Shadow 401(k)s, 92 Notre Dame L. Rev. Online 38, 41& nn. 16–17 (2016). Or, we might say that since Uber creates public risk when it facilitates transportation, it has to account for that risk to the public by providing auto insurance or personal liability coverage to individual drivers. That kind of rule would not require use to measure the amount of control in an “employer”–worker binary, but it would still use that binary to channel work-related safeguards.

 [172]. Constructing a new analytic rubric for determining who gets work-related safeguards and obligations may not be an impossible task, but it is certainly a daunting one and outside the scope of this Article.