Set in Stone
Why a Fixed Base Produces More Innovation Than a Flexible One
There is a question that sits underneath most arguments about protocol design, and almost nobody asks it directly: where does the competition go?
Competition does not disappear. It relocates. If you build a system whose foundations can be altered, then altering the foundations becomes a competitive strategy, and rational firms will invest in it. If you build a system whose foundations cannot be altered, that strategy is closed, and the same firms must spend the same money doing something else. The something else is engineering. The something else is what we normally call innovation.
This is not a moral argument about who is virtuous. It is an argument about incentives and the direction they point. The most productive way to see it is not through ideology but through a domain where the experiment has already been run at enormous scale for thirty years: telecommunications standardisation.
The 3GPP case
The Third Generation Partnership Project produces the specifications on which mobile telecommunications runs. It is not a small operation. The specification corpus runs to thousands of documents, organised into Technical Specifications and Technical Reports, grouped by functional domain — radio access, core network, security, service requirements, terminals, and so on. Every mobile handset, base station, and modem in the world implements some subset of it.
The critical property of this corpus is not that it is large. It is that it is frozen in versions.
3GPP works in Releases. Release 15 gave us the first 5G specifications. Release 16 extended them. Release 17 extended them further. Each Release goes through stages, and at a defined point it is frozen: functional freeze, then stage-3 freeze, then the protocol code freeze. After that point, the content of that Release does not change. Corrections are issued, but the architecture is settled. A device built to Release 15 does not stop working when Release 17 arrives, and a network implementing Release 17 continues to serve Release 15 devices.
This has an obvious consequence that is easy to overlook because it is so ordinary: nobody can lobby to change Release 15. There is no committee you can capture, no working group you can flood with delegates, no vote you can win that will retroactively alter what Release 15 says. The document is what it is. Your equipment either implements it or does not.
The result is a peculiar kind of freedom. Because the base is settled, everything interesting happens elsewhere. Firms compete on implementation quality, on power efficiency, on beamforming performance, on chipset integration, on cost, on the applications that sit above the radio layer. They compete on what they can do within the specification, not on what the specification should say. The specification is the ground, not the prize.
I want to be careful here, because the picture is not perfectly clean and I will come back to the parts that are dirty. But the basic dynamic is real and it is measurable.
What the patent evidence actually shows
Standardisation in this domain runs on Standard Essential Patents. A firm that believes it holds a patent necessarily infringed by anyone implementing the standard declares that patent to ETSI, and accepts an obligation to license on Fair, Reasonable and Non-Discriminatory terms. The declaration is the point at which private intellectual property meets the public specification.
There is now a substantial empirical literature on what these portfolios look like and where they come from. Bekkers, Verspagen and Smits documented the GSM case decades ago and showed how concentration of essential IPR shaped the market structure that followed. Bekkers and West traced strategic patenting through UMTS. Bekkers, Bongard and Nuvolari examined what determines essential patent claims. Baron and Pohlmann built the mapping between standards and declarations. Brachtendorf, Gaessler and Harhoff took a semantic approach to asking which declared patents are truly essential. Lemley and Simcoe asked the same question from the legal side.
Reading across this literature, one pattern recurs, and it is worth stating plainly because it cuts against the intuition most people bring to the subject.
Accumulated technical knowledge predicts standardisation position better than accumulated legal ownership does.
The number of patents a firm holds is a weak signal. What predicts whether a firm’s technology ends up embedded in a specification is whether the firm has spent years working in that technical domain — building the depreciated stock of understanding that lets its engineers see, before anyone else does, what the next specification will need and how to build something that satisfies it elegantly. Ownership follows capability. It does not substitute for it.
There is a stronger version of the claim that I want to flag rather than assert, because the evidence for it is thinner than its proponents suggest. The stronger version is that ownership and knowledge are not merely differently weighted but actively substitutive — that beyond a threshold, additional patents held by a firm that already possesses deep domain knowledge contribute nothing, and may correlate with a worse standardisation position rather than a better one. If you already know the field, buying more paper adds nothing. If you do not, buying paper does not fix your problem.
I find that story plausible and I think it is probably right. I also think the published estimates supporting it are, at present, too fragile to lean on. The samples are small — sometimes a few dozen firms. The interaction terms that carry the result are frequently estimated without controls. The dependent variables are often composites that turn out, on inspection, to be dominated by raw portfolio size. That is a criticism of the measurement, not of the idea. But an idea this useful deserves better evidence than it currently has, and I would rather say so than pretend otherwise.
What is not in dispute is the first-order finding. In a domain where the rules are fixed, the returns accrue to those who understand the domain, and not primarily to those who own claims over it.
That is exactly what you would predict if the analysis at the top of this essay is correct.
The counterfactual: what happens when the base can move
Now consider the alternative arrangement. Suppose the specification is not frozen. Suppose it can be amended, and suppose the amendment process is open to whoever shows up with resources.
Every firm now faces a portfolio decision. It can invest in engineering — building better implementations of the current specification. Or it can invest in influence — shaping what the next specification says, so that its existing assets become essential and its competitors’ assets become obsolete.
The second investment has a return profile that the first cannot match. Engineering returns are incremental and contested; every competitor is doing the same thing, and margins compress. Influence returns are step-functions. If you can get a clause into the specification that only your technology satisfies, you have not improved your position by ten per cent. You have made your competitors’ entire investment worthless overnight.
Under those conditions, the rational firm shifts capital toward influence. So does every other rational firm. The aggregate result is what Gordon Tullock described and Anne Krueger quantified: a society, or an industry, in which enormous resources are consumed in the contest over the rules, and the resources so consumed produce nothing. Tullock’s insight was that the waste is not the transfer itself but the expenditure on capturing it. Krueger’s contribution was to show that in some economies the dissipation ran to several per cent of national income. Buchanan, Tollison and Tullock later assembled the general theory.
Douglass North built the corresponding positive account. Institutions — the formal and informal rules of the game — determine the payoff structure that organisations face, and organisations invest in whatever the payoff structure rewards. Where the rules are stable and enforcement is credible, organisations invest in productivity, because productivity is the only available route to advantage. Where the rules are contestable, organisations invest in contesting them, because that route is both available and cheaper. North’s point was not that some societies happen to be more entrepreneurial than others. It was that the institutional structure selects for one kind of activity or the other, and the selection is close to deterministic.
Mancur Olson added the temporal dimension: coalitions organised around rule-capture accumulate over time, they do not spontaneously dissolve, and their accumulation progressively strangles the productive economy underneath them. Systems that permit rule-capture do not merely tolerate it. They ratchet toward it.
Apply this to a technical protocol and the prediction is precise. A protocol whose rules can be changed will, over time, develop a governance apparatus. That apparatus will become the primary competitive arena. Firms will staff it. Positions on it will become valuable. Participation will require resources that small entrants do not have. The protocol will develop a politics, and the politics will consume the attention that would otherwise have gone into building things.
And here is the part that the advocates of flexible governance never confront. Mutability does not distribute power. It concentrates it. A fixed rule is equally binding on everyone, and equally knowable by everyone, including the smallest participant with no lawyers and no delegation. A mutable rule is knowable only to those who are in the room where it is mutated. Flexibility is sold as democratic and functions as oligarchic, because the capacity to exploit flexibility is distributed far more unequally than the capacity to read a fixed document.
Fixed is not frozen: the architecture that makes it work
The obvious objection is that a fixed base means a dead system. Requirements change. Technology improves. A specification set in stone in 1995 would be useless now.
This objection confuses two different things: the base and the system built on it. The answer to it is one of the most important results in the history of computer science, and it was published in 1984.
Saltzer, Reed and Clark’s end-to-end argument holds that functions requiring knowledge of the application belong at the endpoints, not in the middle of the network. The core should do one thing — move packets — and do it without knowing or caring what the packets mean. Every function that depends on application semantics should be implemented by the communicating endpoints themselves.
The consequence is that the core never needs to change in order for the system to acquire new capabilities. The web did not require a new Internet Protocol. Neither did email, streaming video, voice over IP, instant messaging, file sharing, or any of the other things nobody imagined in 1984. Each was built by someone at an endpoint who did not need permission, did not need to convene a working group, and did not need to win a vote. They needed a stable substrate and the ability to write software.
Steve Deering later described the resulting shape as an hourglass: a wide range of applications on top, a wide range of physical media underneath, and a single narrow waist in the middle through which everything passes. The waist is narrow and it is fixed. That is precisely why the top and bottom can be wide. Every attempt to widen the waist — to add capability to the middle — has narrowed the top, because it converts something that endpoints could decide freely into something that must be negotiated centrally.
This is the design principle, stated generally:
A fixed, minimal, universally-implemented base is what makes unbounded variation possible above it. Not despite the fixity. Because of it.
3GPP achieves the same result through additive versioning. Release 17 does not modify Release 15; it adds to it. The corpus grows by accretion, and every implementation built against any past Release remains valid. That is why an eighteen-year documentation record exists in a form that can actually be reasoned about — the older documents were never overwritten.
Compare this to a system where the base is revised in place. There, the past is not a foundation. It is a liability. Everything built on the old base becomes technical debt the moment the base moves, and the cost of that obsolescence falls on whoever was building rather than on whoever was legislating. You could hardly design a better mechanism for punishing construction and rewarding politics.
The Bitcoin case, stated plainly
I have been circling this and it is time to land it, because the argument applies to distributed ledger protocols more exactly than to anything else I know.
A ledger protocol has a base layer that defines what a transaction is, what a script may do, how blocks are formed and validated, and what the economic rules are. Everything else — every application, every overlay, every token system, every commercial arrangement — is built on top of that.
There are two available designs.
In the first, the base is set and does not change. Script is fully enabled, block capacity is unbounded by protocol, and the rules that existed at the outset are the rules that exist now. There is no upgrade mechanism because there is nothing to upgrade. Under this design, a developer who wants to build something new does so at the edge: writes the software, deploys it, and competes on whether it is good. There is no committee to persuade, because there is no committee. There is no governance, because there is nothing left to govern. The absence of governance is not a deficiency to be remedied — it is the entire point. Governance exists only where the rules are in question, and the rules are not in question.
In the second, the base is treated as a living document, revised through a process controlled by whoever maintains the reference implementation. Capacity is constrained by policy rather than by economics, functionality removed from the original design stays removed, and new functionality is added through the same process that removed it. Under this design, a developer who wants to build something the base does not currently permit has two options: build it anyway on top of an inadequate base, or engage with the process to change the base. The second is frequently the cheaper route. So the process becomes the arena, and the arena develops the exact pathologies Tullock described: proposal wars, faction formation, the accumulation of veto-holders, and the emergence of a class of participants whose expertise is procedural rather than technical.
The word for a failed governance dispute in this domain is fork. Forks are not accidents. They are the predictable output of a system in which the rules are contestable and the contest has no terminal decision procedure. A system with a fixed base cannot fork over rule disputes, because there are no rule disputes to fork over. It is not that disputes are resolved well. It is that they never arise.
I want to state the strongest version of the opposing case, because it exists and deserves an answer. The case is that a fixed base cannot correct its own errors — that if the original design contains a genuine flaw, immutability entrenches the flaw permanently, and the cost of that entrenchment may exceed all the rent-seeking losses avoided. This is a serious argument and it is not obviously wrong. My answer is twofold. First, it proves too much: the same reasoning would justify a mutable Internet Protocol, and the alternative history in which IP was continuously revised is not one in which we got a better internet — it is one in which we got a slower one with more gatekeepers. Second, it misidentifies where flaws are best repaired. A base-layer flaw that can be worked around at the edges should be worked around at the edges, because that solution is voluntary, competitive, and reversible, whereas a base-layer amendment is compulsory, monopolistic, and permanent. The bar for touching the base should be set at the level of flaws that cannot be addressed any other way — and once you set the bar there honestly, you discover that almost nothing clears it.
The empirical evidence from standardisation supports this. Firms in 3GPP do not lack for capability to innovate because they cannot rewrite Release 15. They innovate ferociously within it, and the returns go to the ones with the deepest domain knowledge. That is the mechanism working as designed.
Depth, breadth, and where the edge actually is
There is a further refinement in the standardisation literature that maps neatly onto edge innovation, and it is worth drawing out because it tells you something about strategy rather than merely about structure.
Firms operating on a fixed specification face a choice about where to concentrate. They can go deep — accumulating position within a small number of functional domains, becoming the acknowledged authority in radio access or in security, exploiting a knowledge base they already possess. Or they can go broad — spreading across many domains, exploring adjacent areas where their existing understanding gives them a partial advantage.
This is March’s exploitation–exploration distinction, applied to a technical architecture. And what makes it possible is precisely that the architecture is stable. The functional domains are defined by the specification. A firm can commit to being the best in the world at one of them because the domain will still exist next year, defined the same way, with the same boundaries. Its investment in depth is protected not by any legal instrument but by the simple fact that the ground beneath it does not move.
Take that stability away and depth becomes irrational. If the domain boundaries themselves are up for negotiation, then specialising is a bet on the outcome of a political process rather than on your own competence. The rational response is to stay shallow and diversified, hedging across whatever the committee might decide — or to stop building and start lobbying. Instability does not merely redirect investment from engineering to politics. It redirects engineering itself from deep to shallow, and shallow engineering is where you get systems that work in demonstrations and fail in deployment.
The edge, then, is not a vague gesture at “applications.” It is a structured space, and its structure is inherited from the fixed base. A stable base does not just permit edge innovation. It organises it, by defining coherent domains that specialists can own.
The dirty part: where fixed systems leak
I said I would come back to what is unclean in the 3GPP picture, and intellectual honesty requires it, because the failure mode is instructive.
SEP declarations are self-made. A firm decides for itself that a patent is essential, tells ETSI, and ETSI records the declaration. ETSI does not adjudicate essentiality. Nobody does, absent litigation. The consequence is systematic over-declaration: declaring is cheap, the downside is negligible, and a larger declared portfolio improves your position in licensing negotiations regardless of whether the patents in it are truly essential.
The European Commission’s pilot study on essentiality assessment — the JRC work published in 2020 — examined this directly and confirmed what practitioners had long assumed: a meaningful share of declared SEPs are not, in fact, essential. Brachtendorf, Gaessler and Harhoff reached a compatible conclusion by semantic analysis.
Here is why this matters for the argument rather than against it. The specification itself is fixed and honest. The measurement layer sitting beside it is neither. And the measurement layer is exactly where the rent-seeking went once it was excluded from the specification. You cannot capture Release 17, so you inflate your declared position relative to Release 17 instead.
The lesson is not that fixed systems fail. It is that rent-seeking is conserved, and it will find whatever unverified interface you leave open. Freezing the protocol closes the largest opening. It does not close all of them. Any measurement, registration, or attestation layer attached to a fixed base must be verifiable, or it becomes the new arena — and the returns to gaming it will attract exactly the talent that the fixed base was supposed to redirect into engineering.
The design principle extends accordingly. Freeze the base. Then make sure that everything the base is measured by is checkable by anyone, without permission, from the record itself. A ledger that records what actually happened, immutably and publicly, is not merely a payment system. It is the verification layer that the SEP regime lacks — which is a rather larger observation than it first appears, and one I intend to develop separately.
What follows
The argument compresses to four claims.
Competition is conserved. It moves to whichever margin is contestable. If you leave the rules contestable, that is where it goes, and the resources it consumes there produce nothing.
A fixed base is what makes edge innovation possible, because it is what makes specialisation rational and permission unnecessary. The end-to-end argument and the hourglass are not accidents of internet history. They are the general solution to this problem, discovered once and applicable everywhere.
Mutability concentrates power rather than distributing it, because the ability to exploit a changeable rule is far less evenly distributed than the ability to read a fixed one. Every argument for flexible governance is, in its operational effect, an argument for governance by those who can afford to attend.
And the returns in a fixed system accrue to knowledge rather than to claims. This is the part that ought to be encouraging to anyone who actually builds things. Where the ground does not move, you cannot buy a position — you have to earn one by understanding the domain better than your competitors do. The standardisation evidence points this way consistently, and it is the strongest practical case for immutability that exists, because it says that a fixed system is not merely more efficient in aggregate but more favourable to competence in particular.
Set the base in stone. Then go and build something at the edge, where the work is.