Digital Economies

The framing under test is that digital markets follow different economic rules. It is half right in a way that matters. Digital markets do have distinctive structural features — multi-sided pricing, near-zero replication cost for information, nonrival data, and interactions whose value rises with the number of participants. Those features are real and the theory describing them is good.

The strong version of the claim is the part that has been tested. It says these features produce outcomes standard economics cannot handle, above all inevitable tipping to a single winner. Every time someone has measured that against a credible counterfactual, it has come back weaker than asserted. Combining the two largest pet-sitting platforms left consumers, on average, no better off than with two competing ones. A second entrant into dockless bikesharing made the incumbent better off — more trips, more bikes, higher revenue per trip, a wider network. And an incentivised experiment on social-app valuations suggests the standard method for defining markets systematically overstates network-good market power. In all three the mechanism that defeats tipping is the same: users can be on both platforms at once.

Two gaps are stated here rather than buried, because they bound what this brief can claim. No court document could be fetched in this pass — not a filing, not an opinion, not a remedy — so nothing below is a characterisation of what any court found in the proceedings against Google, Amazon or Apple. What follows is the economics literature bearing on the same questions, which is a different and weaker thing. And the app-store commission argument is under-evidenced on both sides: the fetched theory reaches a conclusion favourable to platforms, no fetched empirical work benchmarks the commission level against anything, and the pass-through and entry studies that would settle it could not be reached.

Established Start with the three designs that measure network effects instead of asserting them, because they are the strongest thing in this subject and they point the same way. Farronato, Fong and Fradkin study the merger of the two largest pet-sitting platforms and find that consumers are on average not substantially better off with a single combined platform than with two separate competing ones: users of the acquirer gained through network effects, users of the acquired platform did worse, and the authors emphasise that platform differentiation remains valuable even where network effects exist. Established Cao, Jin, Weng and Zhou study competitive entry into dockless bikesharing and find that when a second firm entered, the market-expansion effect on new users dominated the market-stealing effect on old users: the incumbent served more trips, invested in more bikes, raised revenue per trip, improved utilisation and built a wider and more evenly distributed network. Published in the RAND Journal of Economics. Entry made the incumbent better off, in a textbook network-effects market. That is the opposite of the winner-take-all prediction.

Frontier The third design attacks the measurement rather than the outcome, and it cuts against the plaintiffs. Bursztyn, Gentzkow, Jiménez-Durán, Leonard, Milojević and Roth run incentivised experiments comparing valuations of alternative social apps under a collective ban of one platform against an individual deactivation. Valuations of alternatives are substantially higher under the collective scenario, and the authors conclude that individual-level substitution estimates can be an unreliable guide to market definition for network goods. Frontier Every antitrust market-definition exercise, and every hypothetical-monopolist test, rests on individual substitution. If individual substitution systematically understates collective substitution for network goods, then standard market definition systematically overstates network-good market power — a plaintiff-unfriendly result from authors with no evident platform interest, which is why it is weighted up. Speculative It is a 2025 design and it has not been replicated.

Established The pricing theory, by contrast, is well corroborated, and its principal expositor is an interested party. Evans and Schmalensee's canonical statement is that two-sided platforms serve distinct customer groups and facilitate value-creating interactions between them, with the distinctive empirical prediction being asymmetric pricing: platforms “may find it profit-maximizing to charge prices for one customer group that are below marginal cost or even negative”, a skewed pattern that is “prevalent, although not universal” in platform industries. Established Evans discloses that he “works for a large number of companies that are multi-sided platforms” on antitrust matters, with publicly named clients including two of the largest. The two-sided-markets framework has been the single most successful defensive doctrine in platform antitrust and its most prolific expositor is a paid platform advocate. Frontier That does not make the economics wrong — the asymmetric-pricing prediction is independently corroborated — but the normative extension, that market definition, merger analysis and monopolisation all require different treatment, is a legal-doctrinal argument rather than an empirical finding, and this pass found no fetched source testing it.

Established Zero marginal cost is the weakest of the framing's structural claims, and the disciplined version of the framework does not make it. Goldfarb and Tucker's statement is that digital technology reduces five costs — search, replication, transportation, tracking and verification — and replication is one of the five. Established Three fetched results show where the strong claim fails. AI inference: foundation models carry significant economies of scale and scope in compute, data and talent, and a model that costs money per query is a utility rather than a software good. Physical delivery: Jo, Matsumura and Weinstein, using Japanese data with historical catalogue sales as an instrument, find e-commerce's reduction in price dispersion raised welfare by 0.3% and cut variety-adjusted prices by 0.9% on average, larger in cities with highly educated populations. Frontier For a technology reported to have abolished the cost of distribution, a 0.9% price effect is a strong indication that the binding costs — logistics, returns, last mile — never went to zero. Verification: its presence as a separate cost category is the framework's own acknowledgement that digital markets spend real resources establishing that the counterparty and the good are what they claim, and that spend does not fall to zero either.

Established The single most policy-relevant fetched result discriminates between remedies that are usually discussed as equivalent. Allcott, Castillo, Gentzkow, Musolff and Salz ran a randomised field experiment separating defaults, habit and inattention, and perceived quality in web search. Active choice barely moved the challenger's share — requiring users to pick a search engine did very little on its own. Users paid to try the challenger updated positively about its quality and many preferred to continue. Users defaulted into the challenger largely did not switch back, consistent with inattention. The counterfactual: eliminating demand frictions roughly doubles the challenger's market share. The authors conclude that exposure to alternatives is more effective than data-sharing mandates. Established Choice screens do little; forced exposure does a lot; data mandates do less than either. Those three remedies appear side by side in every regulatory discussion of search, and one randomised experiment separates them.

Established Self-preferencing has been measured, and its regulatory response has been measured too. Farronato, Fradkin and MacKay find Amazon-branded products ranked higher than observably similar products, with the preferential treatment amounting to 30–60% of the prominence granted to sponsored products. Frontier That is a measurement of ranking behaviour; the abstract does not establish consumer harm and this brief does not assert it. Frontier Waldfogel examines 22 Amazon domains around the September 2023 designation as a gatekeeper under the European Digital Markets Act: before, Amazon's own products ranked roughly 24 positions higher than comparable rivals; after designation the differential fell from about 30 positions to about 20, and the change appeared in Europe and in other jurisdictions. Frontier This is the closest thing in the corpus to a clean regulatory natural experiment, and its lesson is not about network effects at all: designation changed conduct, partially, and extraterritorially. It is a before-and-after design around a known date, with the usual concurrent-events caveats.

Established The attention evidence is the best-supported instance of the framing actually holding, and its window is a real limitation. Boik, Greenstein and Prince use clickstream data from thousands of US households 2008–2013 across amount, location and intensity of attention. Their headline is a paradox: vast and expected changes in where attention goes — away from chat and news, toward video and social media — alongside remarkable stability in how much attention is allocated, with persistent income-based elasticity and complete stability in how attention disperses across and within sites. New offerings redirect where people go without altering overall attention patterns, which the authors characterise as challenging standard models of attention allocation. Frontier If the aggregate attention budget is fixed, attention markets are zero-sum in a way ordinary product markets are not, and platform growth is necessarily at another platform's expense. That is a genuinely different economic rule. Frontier The measurement stops in 2013, before the smartphone-native attention economy matured, and nobody in the fetched corpus has redone it.

Frontier On data, the theory explains why every attempt to price it has struggled. Jones and Tonetti's core result is that data is nonrival — a person's location history, medical records and driving data can be used by any number of firms simultaneously — which creates increasing returns and makes the property-rights allocation decisive. Their central finding is that firm ownership creates hoarding incentives: firms restrict circulation fearing creative destruction, which is socially inefficient given nonrivalry. They conclude consumer ownership is superior. Frontier A nonrival good has no scarcity-based price; its value depends entirely on the buyer's complementary assets and on how many other buyers already hold it, and it falls as it circulates — socially desirable under this model, commercially fatal for any data marketplace. The economics predicts thin data markets, and they are thin. Established The natural experiment confirms the relational reading: Aridor, Che and Salz find the European privacy regulation's opt-in requirement produced a 12.5% drop in intermediary-observed consumers, while the remaining consumers became trackable for longer and their average value to advertisers rose, partially offsetting the losses. Privacy-conscious users opting out made those who stayed more valuable — a privacy externality with a sign nobody intended.

Frontier Do network effects reliably produce tipping? The strong claim is theory plus casual observation of concentrated markets; against it stand a merger where combination left consumers no better off, an entry episode where the incumbent improved, and a measurement result suggesting the standard method overstates market power for network goods. Frontier This is the most confidently asserted claim in the subject and among the least supported when measured. What would settle it is more measurement against credible counterfactuals in markets where tipping is claimed rather than markets where it can be studied — which is a sample-selection problem nobody has solved.

Frontier Is the operative variable multihoming rather than network effects? Across the bikesharing and pet-sitting results the mechanism defeating tipping is the same: users can be on both platforms. Frontier If that generalises, the policy question is not whether network effects exist but what raises the cost of multihoming — which is a question about interoperability, data portability, exclusivity clauses and contract terms, and not about network effects as such. Speculative This is this brief's synthesis rather than a claim any fetched source makes, and it is the reframing the evidence most supports.

Frontier Does AI restore utility economics to a sector that had escaped them? Foundation models carry very large fixed costs and non-trivial marginal costs, producing a tendency to concentration, risk of tipping, and concerns about vertical integration. Frontier If inference has real marginal cost, AI markets look like capital-intensive utilities rather than like software — different pricing, different entry conditions, different antitrust. Speculative It is the same hypothesis AI Driven Productivity holds from the productivity side, where its consequence is that rents can accrue to a few providers while measured aggregate productivity stays flat. Each brief holds one end; neither develops both.

Established What actually sustains platform position — defaults, inattention, quality, or the network? The randomised search experiment is the best-identified result in the subject and it points at defaults and inattention: active choice barely moved share, paid exposure changed beliefs about quality, and users defaulted into the challenger largely stayed. Frontier Eliminating demand frictions roughly doubles the challenger's share, which means roughly half the incumbent's position is attributable to frictions rather than to the network or to quality. Speculative Whether that generalises beyond search is untested, and search is the market where defaults are most heavily contracted for.

Frontier Does regulation change conduct, and does it stop at the border? The gatekeeper-designation evidence says yes to the first and no to the second: a self-preferencing differential falling from about 30 positions to about 20, in Europe and elsewhere. Speculative The extraterritorial spillover is the striking part and the least understood — it suggests either a global ranking system too costly to fork by jurisdiction, or anticipatory compliance, and the two have very different implications for how much regulatory capacity a jurisdiction needs to matter.

Frontier Has the attention budget stayed fixed? The stability result runs to 2013. Everything usually described as the attention economy — the mature smartphone, the algorithmic feed, short-form video — postdates it. Established The best measurement of the subject's most distinctive claim stops before the phenomenon it is used to describe. What would settle it is the same measurement for 2013 to the present, and nobody in the fetched corpus has attempted it.

Frontier Can data be priced at all, and would consumer ownership deliver the modelled optimum? The nonrivalry result predicts thin markets and context-dependent value; the privacy natural experiment demonstrates that one user's choice changes another user's price. Handwave The step from consumer ownership is optimal in the model to give consumers property rights in data assumes away the entire transaction-cost problem: the welfare result requires a functioning market in consumer data consent, and none exists. Speculative The corporate data-valuation exercises, the data-dividend proposals and the national statistical attempts to capitalise data in the accounts could not be fetched and are named without citation; what the fetched corpus establishes is the theoretical reason they should be expected to struggle.

Frontier Is the enforcement story economics or political economy? Lancieri, Posner and Zingales document the decline of US antitrust enforcement since the 1960s and argue it “did not reflect a popular demand for weaker enforcement or any other kind of democratic sanction”, was “engineered by unelected regulators and judges”, and that academic ideas mattered “except where they coincided with the interests of big business, which appears to have exercised influence behind the scenes”. Speculative The documentation rests on new datasets; the counterfactual claim about what a democratic process would have produced is an assertion about an unobserved world.

Speculative And the fringe reading, stated because the fetched evidence is compatible with it and the literature is unwilling to write it down. On this view “different rules” is a rhetorical device rather than an economic finding: every distinctive digital feature — multi-sidedness, network effects, nonrivalry, low replication cost — is a known case within standard industrial organisation, and the new-rules framing has functioned principally as an argument for exemption from ordinary antitrust. Frontier Its supporting evidence is the disclosure attached to the framework's leading expositor and the repeated finding that measured network effects come in weaker than asserted. It is stated as the reading the evidence is most compatible with, not as the reading this brief adopts.

Established The first bottleneck is that this pass could fetch no court, regulator or government document. Filings, opinions, remedies and designation decisions were all unreachable, which means nothing in this brief is a characterisation of a holding in the proceedings against Google, Amazon or Apple. Established The single most important legal artefact for the two-sided-markets framework — the US Supreme Court decision that made two-sided market definition doctrinally load-bearing — could not be verified and is named without citation. Frontier The economics literature bears on the same questions and is a different and weaker thing, and this brief keeps the two apart deliberately rather than letting the second stand in for the first.

Frontier Second: market definition for network goods may be systematically biased, and the direction of the bias is inconvenient for everyone. If individual substitution understates collective substitution, standard methods overstate network-good market power — which favours defendants. Speculative But the same result implies that measured switching costs and measured network effects in the academic literature are also mis-estimated, which destabilises the empirical basis of both sides. A measurement problem that helps one side of a legal argument and undermines the shared evidence base of both is the hardest kind to absorb.

Established Third: the app-store commission debate is under-evidenced on both sides, and this brief declines to resolve it. Gans' analysis reaches three conclusions under specific modelling assumptions: unregulated app commissions are set at a level that maximises consumer surplus; eliminating commissions would raise device prices rather than benefit consumers; and requiring a menu of options for how device makers receive developer subsidies would constrain commissions. Frontier That is a theoretical result and the abstract does not benchmark any observed rate against a competitive level. Established The empirical literature that would — pass-through of commission changes to app prices, developer entry and exit around rate cuts, and the natural experiments created by anti-steering injunctions — could not be fetched. The two-sided pricing result says charging one side above cost and subsidising the other is the predicted pattern, not evidence of abuse in itself; what theory does not give you is the level, and the level is the entire dispute.

Frontier Fourth: the attention measurement stops in 2013. The claim that attention is a fixed budget is the best-supported instance of the framing holding, and it rests on clickstream data ending before the smartphone-native attention economy existed. Speculative Whether the budget expanded, contracted or held is not merely unknown but unattempted in the fetched corpus, and it is a straightforwardly measurable quantity with the same instrument.

Frontier And fifth: the free-goods and attention literatures measure two sides of the same transaction and have never been reconciled. Welfare-based accounting values free digital services by what consumers would give up; attention accounting treats attention as the price they actually pay. Speculative No fetched source attempts the reconciliation, and until someone does, an estimate that one large social network alone contributes 0.05 to 0.11 percentage points of annual welfare-adjusted growth sits alongside a fixed attention budget with no accounting identity connecting them.

Established Nothing on this map produces a result this brief waits on, and no typed depends-on edge is claimed. What it waits on is one market condition and two institutional capacities, recorded as typed requirements in section 13.

Frontier The first is a cost structure that permits more than a handful of model providers. If inference retains a real marginal cost and scale economies in compute, data and talent dominate, the market structure prediction is concentration rather than diffusion, and the antitrust apparatus built for software firms with near-zero replication cost is aimed at the wrong economics. Speculative That is a market condition, not a policy choice, and the productivity consequence of it — large value creation with flat measured aggregate productivity — is developed in AI Driven Productivity.

Frontier The second is a market-definition method that works for network goods. Every hypothetical-monopolist test rests on individual substitution, and the one experiment comparing individual with collective substitution finds them substantially different for a network good. Established Until a method exists that handles the difference, market-power findings for platforms are being produced by an instrument with a measured bias of known direction and unknown size. That is an institutional capacity question — competition agencies commission methods — rather than a research frontier.

Handwave The third is a functioning market in consumer data consent. The welfare argument for consumer ownership of data is a clean theoretical result whose policy conclusion depends entirely on consumers being able to trade consent at low cost to many buyers at once. Frontier No such market exists, the transaction costs are the whole problem, and the privacy natural experiment demonstrates that the value of one person's data depends on what everyone else chose — which is a negative externality inside the very transaction the proposal requires people to make individually.

Established The highest-value experiment is the one already invented and run once: replicate the collective-versus-individual valuation design on other network goods. If the gap between individual and collective substitution recurs for messaging, marketplaces and operating systems, the standard market-definition instrument has a measured bias and competition authorities need a correction factor. Frontier If it does not recur, the result is specific to one social platform and the doctrinal implication evaporates. Either outcome is worth more than another round of argument about whether network effects are strong.

Established Second: redo the attention measurement for 2013 to the present. The instrument is household clickstream data, it exists, and the claim it would test — that the aggregate attention budget is fixed — is the best-supported instance of the framing holding. Frontier An expanding budget would retire the zero-sum reading of attention markets; a stable one would extend the subject's single most distinctive finding across the era it is invoked to describe.

Frontier Third: exploit anti-steering injunctions and commission rate changes as natural experiments. App-store commission schedules have changed on dated, published schedules, small-developer rates have been cut, and steering restrictions have been lifted in specific jurisdictions on specific dates. Established Pass-through to app prices, developer entry and exit, and consumer switching are all measurable around those dates. The commission debate is currently conducted with more theory than data on both sides, and the data are generated by regulation nobody designed as an experiment but which functions as one.

Frontier Fourth: separate the two explanations for the extraterritorial compliance spillover. A self-preferencing differential fell in Europe and elsewhere after one jurisdiction's designation. Whether that is a global ranking system too costly to fork or anticipatory compliance with expected regulation elsewhere is testable by looking at whether the change tracks jurisdictions with pending legislation. Speculative The answer determines how much regulatory capacity a jurisdiction needs before its rules bind globally, which is the most consequential open question for every smaller regulator.

Frontier Fifth: test multihoming cost directly. If multihoming is the escape valve, then interventions that lower its cost — account portability, cross-platform identity, bans on exclusivity — should reduce concentration more than interventions aimed at network effects. Speculative No fetched source tests a multihoming-cost intervention against a network-effects intervention in the same market, and the two are routinely proposed as if they were the same remedy.

Established And sixth, the cheapest: apply the search-remedy design to other default-mediated markets. One randomised experiment separated choice screens, forced exposure and data mandates in search and found them to differ enormously in effect. Frontier The same three remedies are proposed for browsers, app stores, assistants and payment defaults, on no evidence at all about their relative efficacy in those markets.

Frontier The design variable the evidence keeps pointing at is the cost of being on two platforms at once. In both markets where tipping failed to occur, users multihomed. That makes exclusivity clauses, account portability, identity re-establishment cost, cross-platform reputation transfer and the friction of maintaining two profiles the operative engineering parameters. Speculative They are also the parameters least addressed by remedies aimed at network effects, which typically mandate data sharing or interoperability of the incumbent's own service rather than lowering the cost of holding a second account.

Established The search experiment specifies a remedy hierarchy that is otherwise argued about on intuition. Requiring a choice does little because the friction is inattention rather than ignorance. Paying users to try an alternative changes their beliefs about quality and many continue. Defaults, once switched, largely stick. Frontier The engineering implication is that a remedy must move usage rather than presenting a decision, and that data-sharing mandates — the remedy with the most administrative machinery attached — do less than either of the other two.

Established Ranking self-preference is a measurable, tunable quantity and it has been tuned under regulatory pressure. Own-brand products ranked roughly 24 positions above comparable rivals; after gatekeeper designation the differential fell from about 30 positions to about 20, globally rather than only in the regulated jurisdiction. Frontier A partial, global, voluntary-looking adjustment is what a firm does when the cheapest compliance route is to change one parameter in one system everywhere. That is an argument for regulating the parameter rather than the outcome, and it is the only piece of the corpus where a rule visibly moved a number.

Frontier And the cost structure that matters most for the next decade is inference, not replication. A per-query marginal cost changes deployment, pricing and entry: it rewards scale in compute procurement, penalises free tiers, and makes vertical integration into hardware a structural rather than a strategic choice. Speculative If that is where digital markets are heading, the distinctive feature of the next era is closer to a utility's than to a software firm's, and the doctrinal apparatus assembled around near-zero replication cost will be aimed at a receding phenomenon.

Within this map: AI Driven Productivity, which holds the same cost-structure hypothesis from the productivity side; Future Labour Markets, which owns the platform-work evidence that sits inside these market structures; Innovation Ecosystems, on the public instruments aimed at the firms these markets concentrate; Digital Citizenship and Civic Technology, which carry the consent and participation questions data ownership runs into; Future Legal Systems, where the doctrinal argument this brief deliberately does not enter is assessed; AI Governance, on regulating the model providers; and Digital Constitutional Systems, on the rights architecture underneath data property claims.

Outside it: industrial organisation and the theory of multi-sided platforms; antitrust law and the economics of market definition; the economics of privacy and information; measurement economics, which supplies the free-goods valuation methods; and competition-agency practice, which is where the market-definition instrument this brief questions is actually used.

Established The first institutional fact is a sourcing one and it bounds the whole brief. No court, regulator or government document could be fetched in this pass. Nothing here claims what any court found, and the decision that made two-sided market definition doctrinally load-bearing in US antitrust is named without citation because it could not be verified. Established Stating that plainly is more useful than a hedged paraphrase, because a hedged paraphrase of a holding is how a brief acquires a legal claim it cannot support.

Established The second is that the field's most influential framework is expounded by a disclosed interested party, and the disclosure is in the paper. Evans works for a large number of multi-sided platforms on antitrust matters. The asymmetric-pricing prediction is independently corroborated and this brief carries it. Frontier Every normative extension of the framework into doctrine should be read with the disclosure attached — not because the economics is wrong, but because the framework's value as a defence is the reason it was developed so thoroughly on that side and not on the other.

Frontier The third is that the direction of enforcement has been documented as an elite rather than a democratic outcome. The decline of US antitrust enforcement since the 1960s is argued to have been engineered by unelected regulators and judges, with academic ideas mattering where they coincided with business interests. Speculative The counterfactual about what a democratic process would have produced is unobservable. The documentation is new-dataset work and the counterfactual is an assertion, and the paper is routinely cited as though both halves had the same standing.

Established And the fourth is that regulatory designation is now a measurable instrument with an extraterritorial reach nobody legislated for. One jurisdiction's gatekeeper designation moved a ranking parameter in that jurisdiction and elsewhere. Frontier That gives a large regulator global influence over conduct without any corresponding accountability to the people whose markets it changed, which is a governance problem rather than an economic one and belongs alongside Global Cooperation Models.

Established The privacy externality has a sign almost nobody intended and it complicates the standard consent framing. When an opt-in requirement removed 12.5% of observed consumers, the ones who stayed became trackable for longer and more valuable to advertisers, partially offsetting the intermediary's loss. One person's privacy choice raises the price of another person's data. Frontier That makes individual consent a poor instrument for a collective outcome: the people who most value privacy exit, and their exit concentrates value on the people who did not or could not.

Handwave Consumer ownership of data is welfare-superior in a model and is not yet a policy. The theoretical result is clean and the conclusion follows from nonrivalry and hoarding. The step from there to granting property rights assumes away the transaction-cost problem entirely: it requires people to price a good whose value is relational, contextual and dependent on how many others have already sold it. Frontier The economics that makes consumer ownership optimal is the same economics that makes the market it needs impossible to run.

Frontier If the attention budget is fixed, attention competition is genuinely zero-sum, and the ethics of that are different from ordinary product competition. A product market can grow; an attention market can only be reallocated. Speculative That makes every design improvement that captures more attention a transfer from another use of a person's time rather than a creation of value — and the measurement supporting it stops in 2013, before the designs most accused of this existed. The strongest ethical claim in this subject rests on the dataset with the largest gap behind it.

Frontier And there is an uncomfortable symmetry in the measurement findings. The market-definition result cuts against antitrust plaintiffs; the search-remedy result cuts against the remedies regulators most favour; the entry and merger results cut against the tipping story that motivates intervention. Established None of these comes from authors with a platform interest, which is exactly why they matter. Speculative An honest reading is that the case for intervening in digital markets is currently stronger on distributional and governance grounds than on the structural-inevitability grounds usually offered for it, and that substituting one for the other is what has made the argument so hard to settle.

Established The terminal position is that digital markets have different features and, wherever anyone has managed to identify their effect, produce outcomes closer to ordinary differentiated-product competition than the theory advertised. Network effects exist and are measurable. Measured against credible counterfactuals they have been too weak to produce the structural results attributed to them: a merger of the two largest platforms in a market left consumers on average no better off, and entry into a textbook network-effects market made the incumbent better off. Frontier The framing is not refuted; the strong version of it is unsupported on the occasions it has been tested.

Frontier The change of question is from structure to friction. What the fetched evidence keeps identifying is not a law of digital markets but a set of frictions with measurable magnitudes: multihoming cost, default stickiness, inattention, verification cost, and a per-query cost that never went away. Established Roughly half of one search incumbent's position is attributable to demand frictions rather than to the network or to quality. Speculative A policy apparatus aimed at frictions would look very different from one aimed at structure — smaller, more empirical, and much less doctrinally satisfying to either side.

Frontier The long-run structural question is whether the next era resembles the last one at all. The doctrinal apparatus was built around near-zero replication cost. If the dominant digital good becomes one with large fixed costs and a real per-query marginal cost, the prediction is a capital-intensive, concentrated, utility-like industry, and the tools assembled over thirty years are aimed at a receding phenomenon. Speculative That is a hypothesis, and it is the same one this brief hands to AI Driven Productivity from the other side.

Speculative And the largest civilisational question here is one the corpus barely touches: whether a fixed attention budget makes a whole class of value creation illusory. If the aggregate quantity of human attention is stable and only its allocation moves, then a century of improvements in capturing it produce reallocation rather than surplus, and the welfare accounting that values free digital goods is measuring a transfer. Frontier The counter is that people can prefer the reallocation, which is exactly what the free-goods valuations measure. Handwave The two literatures measure opposite sides of one transaction, nobody has reconciled them, and until someone does, the most basic question about the digital economy — whether it has made people better off — has two credible answers with no accounting identity between them.

These horizons track regulatory dates, measurement opportunities and one cost curve rather than technology:

• 10 yr: Frontier Gatekeeper designation has produced the corpus's one clean regulatory natural experiment and will produce more, on dated designations across more firms and more obligations. Established Expect the extraterritorial spillover question to be answerable inside this window, because it needs only ranking data and designation dates. Frontier Expect the app-store empirics to arrive from anti-steering injunctions and rate changes rather than from anyone designing a study. Speculative Expect the attention measurement not to be redone, because the clickstream panels that supported it were commercial assets and the incentive to publish from them has weakened.
• 25 yr: Speculative If inference cost stays material, digital market structure converges on utility economics and the antitrust apparatus re-specifies around capacity, procurement and interconnection rather than around network effects. Frontier If it collapses toward zero, the current doctrine remains apt and the concentration concerns are about data and talent rather than about compute. Handwave Which of those happens is a semiconductor and energy question, and forecasting it from an economics brief is not a measurement.
• 50 yr: Speculative At this horizon the durable object is the multihoming margin. Every episode in the fetched corpus where tipping failed to occur, it failed because users held two accounts, and every technology that raises or lowers that cost is therefore a market-structure intervention whether or not anyone intended it. Handwave Extrapolating a specific structure from that is a story; extrapolating that the parameter keeps mattering is about as safe as anything in this subject gets.
• 100 / 250+ yr: Handwave Beyond useful forecasting. The subject is sixty years old at most, its best measurements span single decades, and its most distinctive finding rests on a five-year window that ended in 2013. Handwave A century-scale claim about digital economies would be a genre exercise, and this brief has none to offer.

• Depends on Nothing on this map. No result another brief produces is a precondition for this one: the measurement problems here are properties of the instruments competition authorities use and of the documents this pass could not reach, rather than missing science. No typed depends-on edge is claimed.
• Requires (not on this map) One market condition and two institutional capacities, none of them a research result this map produces. First, an inference cost structure that permits more than a few model providers: foundation models carry significant economies of scale and scope in compute, data and talent alongside a non-trivial marginal cost per query, and if that persists the structural prediction is a capital-intensive, concentrated, utility-like industry rather than the near-zero-replication-cost software market the whole doctrinal apparatus was built around. Second, a market-definition method that works for network goods: every hypothetical-monopolist test rests on individual substitution, and the one incentivised experiment comparing individual with collective substitution finds valuations of alternatives substantially higher under a collective scenario, implying that standard methods overstate network-good market power by an amount of known direction and unknown size. Competition agencies commission methods; this is a capacity question rather than a research frontier, and until it is answered every platform market-power finding is produced by an instrument with a measured bias. Third, a workable market in consumer data consent: the welfare case for consumer ownership of data follows cleanly from nonrivalry and firm hoarding incentives, and its policy conclusion depends entirely on people being able to trade consent at low cost to many buyers simultaneously. No such market exists, the transaction costs are the whole problem, and the privacy natural experiment shows the value of one person's data depends on what everyone else chose — an externality sitting inside the very transaction the proposal asks each person to make alone.
• Enables In principle the market structures assessed here condition every digital deployment on this map — who can enter, what a service costs, and who captures the value. No typed enabling edge is claimed, and the reason is the brief's own finding: the structural results the framing predicts have not been reliably observed, so there is no established structural fact for another brief to depend on.
• Adjacent Industrial organisation and the theory of multi-sided platforms; antitrust law and the economics of market definition; the economics of privacy and information; measurement economics, which supplies the free-goods valuation methods; competition-agency practice, where the questioned instrument is used; and within this map AI Driven Productivity, Future Labour Markets and Digital Citizenship.

Frontier “Network effects make digital markets tip to a single winner.” This is the most confidently asserted claim in the subject and among the least supported when measured. A merger of the two largest platforms in a market left consumers on average not substantially better off than two competing ones, with the acquired platform's users worse off and platform differentiation retaining value. Established Competitive entry into dockless bikesharing — a textbook network-effects market — produced market expansion dominating market stealing, with the incumbent serving more trips, investing in more bikes, raising revenue per trip and building a wider network. Frontier Entry made the incumbent better off. In both cases the mechanism is multihoming, which means the operative policy question is what raises the cost of being on two platforms, not whether network effects exist.

Frontier “Digital goods have zero marginal cost.” True of pure information replication and transmission, which is a narrower domain than “digital markets”. The disciplined framework says five costs fall — search, replication, transportation, tracking and verification — and replication is one of five. Established It fails for AI inference, which costs money per query and looks like a utility; for physical fulfilment, where e-commerce's measured effect is a 0.3% welfare gain from reduced price dispersion and a 0.9% average fall in variety-adjusted prices; and for verification, an irreducible cost of establishing that the counterparty and the good are what they claim. Frontier A 0.9% price effect from a technology said to have abolished distribution costs is the clearest available evidence that the binding costs never went to zero.

Established “This brief tells you what the courts found.” It does not, and the boundary is deliberate. No court, regulator or government document could be fetched in this pass — no filing, no opinion, no remedy, no designation decision. The Supreme Court decision that made two-sided market definition doctrinally load-bearing in US antitrust is named without citation for exactly that reason. Established What is presented here is the economics literature bearing on the same questions, which is a different and weaker thing. Anyone editing a claim about a holding into this page would be adding something the sources do not support.

Frontier “The 30% app-store commission has been shown to be supracompetitive.” It has not, and neither has the opposite. The fetched theory argues that unregulated commissions are set at a consumer-surplus-maximising level, that eliminating them would raise device prices, and that requiring a menu of subsidy options would constrain them — under specific modelling assumptions, without benchmarking any observed rate. Established The empirical work that would settle it — pass-through, developer entry and exit, anti-steering natural experiments — could not be fetched. Frontier The debate is under-evidenced on both sides, and the two-sided pricing result tells you the pattern of charging one side and subsidising the other is predicted, not that any particular level is justified.

Established “Choice screens will fix search competition.” The one randomised field experiment that tested it found active choice barely moved the challenger's share. Paying users to try the alternative changed their beliefs about its quality and many continued; users defaulted into it largely did not switch back; and eliminating demand frictions altogether roughly doubles the challenger's share. Established The authors' conclusion is that exposure to alternatives is more effective than data-sharing mandates. Three remedies routinely discussed as equivalent differ enormously, and the one with the most administrative machinery attached does the least.

Frontier “Data is the new oil and can be priced as an asset.” Nonrivalry is precisely why it cannot. A nonrival good has no scarcity-based price; its value depends on the buyer's complementary assets and on how many others already hold it, and falls as it circulates — which is socially desirable and commercially fatal. Established The privacy natural experiment demonstrates the relational property directly: an opt-in requirement removed 12.5% of observed consumers and made the remainder trackable for longer and more valuable, partially offsetting the loss. Frontier The economics predicts thin data markets, and no fetched source reports a successful data-pricing regime. The corporate valuation exercises, data-dividend proposals and statistical attempts to capitalise data in the national accounts could not be fetched and are named without citation.

Handwave “Give consumers property rights in their data and welfare improves.” The model's welfare result is real and follows from nonrivalry plus firm hoarding incentives. The policy conclusion requires frictionless consent markets, and the step from one to the other assumes away the entire transaction-cost problem. Frontier It also runs into the externality the same corpus measures: your privacy choice changes my data's price, which makes an individually traded right a poor instrument for a collective outcome.

Speculative “Digital markets follow different rules.” The fringe reading of the whole subject is that this has functioned as a rhetorical device rather than an economic finding — that multi-sidedness, network effects, nonrivalry and low replication cost are all known cases within standard industrial organisation, and that the new-rules framing has principally served as an argument for exemption from ordinary antitrust. Frontier Its evidence is the disclosure attached to the framework's leading expositor and the repeated finding that measured network effects come in weaker than asserted. Established The best-supported counterexample is attention: if the aggregate budget really is fixed and only allocation moves, attention markets are zero-sum in a way ordinary product markets are not, and that is a genuinely different rule. Frontier It rests on a clickstream measurement that ends in 2013, before the attention economy it is invoked to describe.