The Appeal and the Trap
Policy experiments are everywhere. They admit uncertainty and promise to learn and adjust, or even stop if a program fails. But in practice, many “experiments” are not designed in a way that can yield a decisive verdict. Instead, they often become a way to launch policies that could not pass as fully specified programs. I have advocated for experiments myself on the belief that government should take risks and then follow the evidence. The problem I had not taken seriously enough is that the politics of experimentation often undercut the experiment itself. Once a policy is implemented, it changes the landscape of interests and constraints in ways that make “learning” an unreliable guide to action. Proposing an experiment is a coalition-building strategy ex ante. Once passed, it becomes a renegotiation regime ex post.
Two Outputs, One Fast
Every policy experiment produces evidence and politics.
Evidence includes performance data, organizational learning, and market signals. Politics includes beneficiaries, intermediaries, compliance ecosystems, reputational stakes, opponents, and bureaucratic routines.
The politics almost always arrive before the evidence, meaning an experiment produces constituencies faster than it produces evidence. The moment an experiment launches, real people and organizations start reorganizing around the new regime. They invest, develop business models, and build compliance systems. Intermediaries emerge and bureaucratic routines form. All of this happens fast because the experiment authorizes real activity, and real activity demands real organizational responses.
Evidence, by contrast, is slow and often nuanced. Meaningful data takes time to accumulate. Causal inference is difficult, at least in part because programs are rarely designed in ways that facilitate causal analysis. And evidence rarely yields consensus. “Success” is inherently ambiguous when you never agreed on what it meant.
An additional problem is that evidence itself becomes endogenous to the politics. Each side funds, produces, and publicizes the data that supports its position. The experiment thus doesn’t produce a shared evidentiary record that both sides accept. It produces competing, internally coherent, records that support each side’s case. The problem is not insufficient data. It is a structural feature of experiments with multiple objectives.
Even if they did agree on an analysis, by the time evidence exists that might inform a decision the experiment itself has already shaped the politics of the program. The question is no longer “did this work?” but rather “who loses if we change course?” And the prospective loser is always an organized constituency that did not exist before the experiment started.
The longer the gap between the appearance of constituents and evidence, the more difficult it becomes to act on the evidence. Over time, the feasible set of actions narrows. At some point, ending the experiment is no longer politically feasible. Debates move on from terminating the experiment to bargaining over details, such as grandfathering, compensation, scope adjustments, and redesign. The experiment doesn’t end. It renegotiates.
“Experiment” as Coalition Building
Policy entrepreneurs choose to promote an experiment when opposition is likely to block a proposal for a permanent program. This is the key relationship between experiments and opposition. The experiment doesn’t just happen to attract opponents. Opposition is a key driver in making the proposal an experiment. If everyone liked the idea and agreed on a relevant endpoint and decision rule, there would be less need to call it an experiment.
Calling something an experiment also lets actors with incompatible objectives agree to the same proposal while meaning different things. The policy entrepreneur gets forward motion. Beneficiaries get a program. Skeptics get the promise of revisiting. Beneficiaries continue to call it experimental because it protects their access from being terminated. Even critics benefit from the framing because it gives them a standing claim and platform to demand changes. The experiment becomes a stable equilibrium that serves everyone’s interests except the interest in actually resolving the question.
Genuine uncertainty can make an experiment the right approach in principle. But even in honest cases, the political economy is the same. Once adopted, the experiment creates organized groups on every side of the question. Each side produces and publicizes evidence that supports its position, turning evaluation into recurring renegotiation rather than a path toward decision. It persists in a state of ongoing debates.
CBRS: An Experiment in Action
The FCC’s Citizens Broadband Radio Service (CBRS) is a good case study. CBRS is the FCC’s shared-use framework for the 3.55–3.7 GHz band, built around a three-tier priority structure (federal incumbents, auctioned Priority Access Licenses, and opportunistic General Authorized Access) with access and interference protection managed dynamically by Spectrum Access Systems.1
The proposal was a new way for commercial users to share spectrum with government users. It was born of a widely shared desire among commercial users to move some wireless spectrum from federal to commercial use. The Department of Defense and other agencies, who held the spectrum, have generally resisted pressure to make the spectrum available for commercial use. The proposed regime used sharing architecture conditioned on enforceable mechanisms, making civilian use of the band possible while retaining DOD priority use and without requiring the agency to clear the band.
While commercial users share the desire for additional usable spectrum, they do not agree on how that spectrum should be governed. Two major groups–cable and cellular wireless–had opposing ideas on the CBRS proposal.2 The cellular wireless interests argued that “short-term, non-renewable licenses” posed “the substantial risk” that “a number of potential bidders for PALs will simply stay away from this band,” and urged longer terms and renewal expectancy.3 By contrast, cable “share[d] the Commission’s view that small cell networks using the 3.5 GHz band … can be an important resource to help ‘address wireless coverage and capacity issues’….”4
An experiment provided a path forward, and seemed sensible on its face since nobody knew how well the institutional framework would work.
Commissioners embraced the experimental framing. FCC Commissioner McDowell called it an “educational approach,” saying he was “hopeful that this experiment in the 3.5 GHz range will promote additional investment and innovation.”5 Commissioner Pai asked “What works? Can our proposals be implemented in the real world? If at that point we decide to ratify the proof of concept….”6 And Commissioner O’Rielly explicitly noted an exit ramp, proclaiming that the band would be “one big experiment,” but “if it does not meet expectations, we are not precluded from altering it in the future.”7 Commissioner Rosenworcel pointed out that the regulatory process “can often be risk averse,” but “we can overcome this risk aversion if we experiment on a smaller scale, with ‘sandbox’ projects.”8
Framing the proposal as an experiment allowed nearly everyone to get something meaningful. The FCC was able to move some federally dominated spectrum into commercial use. Enterprise users got broad access without years of clearance proceedings. And large investors got a pathway to more predictable, financeable rights as the PAL design evolved. And while opponents did not win, they got a regime that was explicitly provisional and therefore continuously open to challenge and change, allowing them to live and fight another day.
So CBRS was framed as an experiment not merely because of technical uncertainty but because there was not enough consensus to adopt the framework as permanent. The experiment label made it possible to move forward over objections. But that means opposition is baked into the experiment’s DNA from the start. And unlike a real experiment, where the results would resolve the disagreement, CBRS’s “experiment” frame kept the disagreement alive and gave opponents permanent standing to relitigate the design.
This is exactly what happened. In 2018, the FCC adopted larger geographic license areas and longer license terms, framed as “promoting investment.” Supporters called it sensible iteration. Critics argued it tilted the regime toward carriers. The point is that the changes were renegotiations among groups the experiment had created, not adjustments triggered by experimental evidence.
These dynamics continue. In a February 2026 TPI panel, the FCC Chairman’s wireless advisor noted that the band “hasn’t really remained true to its original conception” and that the Commission “continued to iterate” across administrations. More than a decade after the experiment began, panelists representing different sides of the debate could not agree on what metrics should define success. Meanwhile, users who had invested under the existing rules filed comments warning the FCC not to change the rules under which they invested.
It is worth asking what we have learned after more than a decade of the CBRS “experiment,” and has any of that learning triggered a decision? We have, in fact, learned some things. We know that SAS-mediated sharing can work technically. We know that interference with incumbents has been minimal. We know that deployment has grown, though from a low base. But none of this information has settled the underlying dispute about whether the spectrum would generate more value under a different allocation. Each new data point gets absorbed into the pre-existing arguments. The experiment has produced information. It has not produced resolution. And that distinction is the whole point.
Scope and Limitations
This argument is not an ironclad law of policy.
First, some policy experiments do, in fact, end.
These cases tend to share common features.9 The experiments either never created a constituency strong enough to force renewal, or exit was bundled with a managed transition that re-homed beneficiaries and defused the distributional fight. In some cases, a hard sunset included in legislation was simply allowed to bite.
The dynamic is likely to be strongest when an experiment rapidly creates organized beneficiaries and opponents, when the policy becomes a platform with rights, routines, and intermediaries, and when the promised exit lacks binding transition terms. CBRS fits all three. A narrow procurement pilot with a statutory sunset and no installed base of reliance interests does not.
Second, not every contested policy becomes an experiment.
Many are simply enacted or defeated without anyone calling them experiments. Opposition and uncertainty appear to be necessary, but not sufficient, conditions for the experiment frame. What additional conditions make it attractive is a question I leave open, though I suspect the novelty of the proposed institutional design matters.
It’s Perpetual Negotiation, Not An Experiment
If policymakers want experiments to function more like actual experiments, they need to make credible commitments before constituencies form. These would involve decision rules, what ending the program might look like if that is where the evidence pointed, and who has the burden of proof. If the default is to continue unless someone proves failure, the experiment will almost certainly continue.
The problem is that it’s often impossible to make those commitments credible. The same political dynamics that make the “experiment” label attractive are the ones that make ex ante commitment devices hard to adopt. Decision rules, transition terms, and burden-of-proof assignments all require agreement on how to define success.
This is a version of the problem Williamson identified in his 1976 analysis of franchise bidding: parties entering a long-term arrangement under uncertainty cannot write complete contracts, and once relationship-specific investments are made, neither side can credibly commit to the original terms.10 Policy experiments work similarly. The “experiment” label functions as an incomplete contract, with the gaps filled not by courts or hierarchies but by recurring political bargaining among the constituencies the experiment has created. The greater the asset specificity, the less the original exit-ramp language matters.
We should still demand metrics and decision rules for experiments. But we should be more honest about what is happening. It’s not an experiment, but a nascent program financed as permanent and governed through periodic renegotiation. When someone proposes a policy experiment, the first question should always be, “what would cause the program to end?”
Footnotes
1 CBRS rules are codified in 47 C.F.R. Part 96. ↩︎
2 Many others submitted comments opposing or endorsing, but cable and wireless were the 800-pound gorillas in the fight. ↩︎
3 Comments of CTIA – The Wireless Association, GN Docket No. 12‑354 (filed Dec. 5, 2014), at 12–13. The filing warns that “short-term, non-renewable licenses” create “substantial risk” that bidders will avoid PALs. ↩︎
4 Comments of the National Cable & Telecommunications Association (NCTA), GN Docket No. 12‑354 (filed Dec. 5, 2014). ↩︎
5 Statement of Commissioner Robert M. McDowell, in Amendment of the Commission’s Rules with Regard to Commercial Operations in the 3550–3650 MHz Band, GN Docket No. 12‑354, 27 FCC Rcd 15594 (2012). ↩︎
6 Ajit Pai, Statement of Commissioner, Amendment of the Commission’s Rules with Regard to Commercial Operations in the 3550–3650 MHz Band, Notice of Proposed Rulemaking and Order, GN Docket No. 12‑354, FCC 12‑148, 27 FCC Rcd 15594 (2012). ↩︎
7 Statement of Commissioner Michael O’Rielly, FCC 14‑49 (Apr. 23, 2014) Further Notice, GN Docket 12‑354. ↩︎
8 Jessica Rosenworcel, written statement, Wireless Broadband and the Future of Spectrum Policy, Hearing before the Senate Committee on Commerce, Science, and Transportation, 114th Cong. (Feb. 2, 2016). ↩︎
9 Texas discontinued its Dual Eligible Integrated Care Demonstration Project on December 31, 2025, transitioning members to other Medicare‑Medicaid coverage arrangements (Texas HHS, “Dual Eligible Integrated Care Demonstration Project,” hhs.texas.gov). The USPTO allowed its After Final Consideration Pilot Program 2.0 to expire in December 2024 after repeatedly extending it since 2013, citing concerns about recovering administrative costs through user fees (“Extension and Termination of the After Final Consideration Pilot Program 2.0,” 89 Fed. Reg. 79,889, Oct. 1, 2024). Both cases fit the pattern: the Texas program bundled exit with a managed transition that re-homed beneficiaries, and the PTO pilot never created a constituency strong enough to force renewal. ↩︎
10 Williamson, Oliver E. (1976). Franchise Bidding for Natural Monopolies — In General and with Respect to CATV. Bell Journal of Economics, 7(1), 73–104. ↩︎
Scott Wallsten is President and Senior Fellow at the Technology Policy Institute and also a senior fellow at the Georgetown Center for Business and Public Policy. He is an economist with expertise in industrial organization and public policy, and his research focuses on competition, regulation, telecommunications, the economics of digitization, and technology policy. He was the economics director for the FCC's National Broadband Plan and has been a lecturer in Stanford University’s public policy program, director of communications policy studies and senior fellow at the Progress & Freedom Foundation, a senior fellow at the AEI – Brookings Joint Center for Regulatory Studies and a resident scholar at the American Enterprise Institute, an economist at The World Bank, a scholar at the Stanford Institute for Economic Policy Research, and a staff economist at the U.S. President’s Council of Economic Advisers. He holds a PhD in economics from Stanford University.
