PaperlayerView exact revision on arXiv

Source-linked AI summary

The Source of Data Value: Consumer Disclosure, Access Design, and Platform Monetization

Hexiao Liu, Lijian Lu, Mingzheng Wang

arXiv:2610.22864v1cs.GTecon.GNmath.OCq-fin.EC

TL;DR

The paper asks how consumer disclosure rights affect sellers’ incentives to buy marketplace data and the platform’s incentive to provide it. A game-theoretic model compares governance regimes and finds that voluntary disclosure can remove defensive access value, while anonymous-price protection addresses a distinct consumer-welfare concern.

  • Problem

    The paper asks how consumers’ ability to decline a data use while continuing to purchase changes sellers’ willingness to pay for marketplace data services.

    Source: §1 · p. 2

  • Method

    A game-theoretic model examines a commissioned platform, two sellers, consumer disclosure and purchasing decisions, and sequential access offers under different collection regimes.

    Source: §2 · p. 8

  • Results

    Under voluntary disclosure, anonymous purchasing restores sellers’ no-data rejection payoffs and removes defensive fees; an enforceable no-data price cap protects consumers but can reduce platform profit versus uncapped disclosure.

    Source: §5 · pp. 27–28

  • Takeaways & Limitations

    Platforms should assess both bilateral-use and rejection outcomes, while regulators should distinguish equal-access rules from effective consumer control and monitor anonymous-channel pricing.

    Source: §5 · pp. 27–28

    The baseline assumes homogeneous privacy costs, perfect identification, symmetric fixed-position sellers, an exogenous commission, and no independent signals or direct nonprice benefits from disclosure.

    Source: §2.3 · p. 13

Abstract

from arXiv · show

Digital marketplaces increasingly earn revenue from transaction commissions and seller-facing data services. Sellers may purchase data to increase revenue or avoid competing with an informed rival. We examine how consumer disclosure rights shape these motives and the platform's incentive to provide data. We develop a game-theoretic model of a commissioned platform, two differentiated sellers, and privacy-sensitive consumers. Data generates value through bilateral sales-revenue gains and competition threats arising from asymmetric access. Under mandatory collection, the platform can monetize defensive adoption, even when bilateral personalization reduces seller revenue. Under voluntary disclosure, anonymous purchasing and strictly positive privacy costs make unilateral disclosure unattractive in the baseline. Exclusive access thus yields no usable information, restores sellers' no-data rejection payoffs, and eliminates defensive fees. Profitable provision therefore requires nonnegative bilateral revenue gains. Under full fee extraction, sellers retain their no-data net profits. This revenue alignment does not guarantee welfare gains: additional transactions may benefit consumers, whereas segmentation in an already covered market can raise anonymous prices and reduce welfare. Capping anonymous offers at their no-data benchmark protects individual consumers and limits positive-mass provision to the transaction-creation region. Extensions consider access choice, heterogeneous privacy costs, imperfect identification, and limited contractual credibility. Platforms should evaluate data services based on both bilateral-use and seller-rejection outcomes, and sellers should distinguish revenue-enhancing from defensive investments. Regulators should distinguish equal-access rules from effective consumer control and monitor pricing in anonymous channels.

1. Introduction

Digital marketplaces combine transaction commissions with seller-facing data services, raising how consumers’ ability to withhold information changes sellers’ willingness to pay. The paper models how disclosure rights alter both the revenue value of bilateral data use and the rejection payoffs behind defensive purchases.

1. Introduction

  • Consumer disclosure can change sellers’ rejection payoffs, so data-service adoption and fees alone do not reveal whether data improves bilateral revenue or funds defensive spending.The paper frames this distinction as relevant to platform investment, seller profitability, and personalization.Source: §1 · p. 2

1.1. Our Approach and Contributions

  • The model separates bilateral sales-revenue effects from competition-threat losses after rejection, which together shape platform access fees and commission income.Under mandatory collection, sufficiently large, enforceable rejection losses can support provision even when bilateral personalization reduces seller revenue.Source: §1.1 · p. 3
  • Under voluntary disclosure, consumers withhold information from a sole subscriber while retaining anonymous purchasing, restoring the rejecting seller’s no-data payoff and eliminating threat-based monetization.Provision then requires nonnegative bilateral seller-revenue gains; endogenous access choice leaves exclusive access with no incremental monetization value.Source: §1.1 · p. 3
  • Revenue alignment does not ensure consumer protection: segmentation can raise anonymous prices without adding transactions, while a no-data price cap preserves consumers’ purchasing option.With full fee extraction, sellers retain their no-data net profits; the cap can protect consumers while reducing platform profit relative to uncapped voluntary disclosure.Source: §1.1 · pp. 3–4

1.2. Literature Review

  • The paper’s contribution is to study data contracts in a commissioned intermediary where sellers depend on the marketplace for customer access and personalization.This complements work on personalized-pricing competition and consumer disclosure by linking those choices to platform access contracts and disagreement payoffs.Source: §1.2 · pp. 5–6

2. Model and The General Value of Platform Data

The section develops a commissioned-platform model in which data value reflects both bilateral revenue changes and the cost of rejecting access. It then examines how disclosure rules, seller adoption, and model scope shape provision and welfare.

2. Model and The General Value of Platform Data

  • The platform controls a seller-facing information service and compares NDC, MDC, and VDD, where consumers can decline data use while continuing anonymous purchases.The platform commits to offer order, fees, and access configurations contingent on seller decisions.Source: §2 · pp. 7–8

2.1. Seller Adoption and Platform Data Value

  • Under full implementation, data value combines bilateral revenue gains with threat-based fees; provision can pay despite Gg < 0, while full fee extraction leaves sellers at no-data net profits.The threat value comes from the revenue loss a seller faces in the exclusive-access rejection continuation.Source: §2 · p. 9, §2.1 · pp. 10–11
  • With limited commitment, data value depends on whether the platform honors a required rejection subsidy; greater credibility raises value when Tg ≥ 0 and lowers it otherwise.When Tg > 0, the first offeree pays weakly less and retains weakly more profit as credibility improves.Source: §2.1 · p. 11

2.2. Hotelling Microfoundation

  • The Hotelling microfoundation endogenizes revenue, disclosure, prices, and coverage, with positive privacy costs and equilibrium selection shaping unilateral disclosure and reported outcomes.The high-valuation segmentation outcome depends on the selected pricing equilibrium.Source: §2.2 · pp. 11–12

2.3. Scope, Extensions, and Limitations

  • The baseline assumes homogeneous privacy costs, perfect identification, bilateral equilibrium-path access, and full implementation; its thresholds and welfare rankings need not survive arbitrary extensions.The paper also excludes independent seller signals and direct nonprice benefits from disclosure.Source: §2.3 · p. 13

3. Product-Market Equilibrium

The section compares equilibrium outcomes under no data collection, mandatory collection, and voluntary disclosure, showing how consumer participation and disclosure rights shape monetization and welfare. Under voluntary disclosure, eliminating threat-based adoption does not by itself ensure welfare gains; protecting anonymous prices preserves consumers’ no-data options.

3.1. No Data Collection: Benchmark

  • Under NDC, all consumers enter, but when v < 1 uniform pricing leaves central consumers unserved, creating scope for data-enabled trade.For 1 ≤ v < 3/2 the market is covered but marginal participation constrains prices; for v ≥ 3/2 competition pins the uniform price at one.Source: §3.1 · p. 14

3.2. Mandatory Data Collection: Monetizing Competitive Vulnerability

  • Under MDC, the platform can monetize defensive adoption even though bilateral data use lowers seller revenue, because rejection losses can be larger still.The source of rejection losses depends on the setting: participation when v − c < 3/4, or personalized poaching when v − c ≥ 3/4.Source: §3.2 · pp. 14–15
  • MDC cannot expand coverage relative to NDC; when the platform monetizes, its profit weakly rises while coverage, seller profit, and social welfare weakly fall.Consumer surplus can rise only in the specified regions where stronger personalized-price competition dominates privacy and participation losses.Source: §3.2 · p. 15

3.3. Voluntary Disclosure: Eliminating Threat-Based Adoption

  • Under VDD, anonymous purchasing and positive privacy costs make unilateral disclosure unattractive, restoring the no-data rejection payoff and eliminating the threat effect.Provision requires a nonempty opt-in set and nonnegative bilateral seller-revenue gains.Source: §3.3 · pp. 15–16
  • With full fee extraction under VDD, sellers retain their NDC net profits; welfare gains over NDC mainly come from additional transactions, while segmentation in an already covered market can reduce welfare.In P3, higher anonymous prices and privacy losses lower consumer surplus and welfare despite nonnegative bilateral revenue gains; positive-revenue segmentation is not common to every equilibrium.Source: §3.3 · pp. 16, 18, 19
  • Capping anonymous offers at the NDC benchmark protects each consumer relative to NDC and confines positive-mass provision to transaction creation, but requires an enforceable benchmark price.The benchmark is determined inside the model; the analysis does not estimate it from marketplace observations or resolve practical implementation.Source: §3.3 · pp. 20–21

4. Extensions and Counterfactuals

The extensions preserve the distinction between bilateral revenue gains and defensive threat value: consumer control removes unilateral threat-based monetization, while endogenous access, privacy heterogeneity, and imperfect identification qualify the outcomes. Counterfactuals show when threat-supported provision is fragile and why conversion alone does not establish revenue gains.

4.1. Platform Choice of Access Structure

  • Under endogenous access, mandatory collection may favor exclusivity when it generates more sales revenue than bilateral personalization, but neither access arrangement raises sales revenue above no data.Under voluntary disclosure, unilateral exclusivity is ineffective because consumers refuse disclosure; provision then requires bilateral revenue gains.Source: §4.1 · pp. 21–22

4.2. Heterogeneous Privacy Cost

  • With heterogeneous positive privacy costs, voluntary disclosure still eliminates unilateral threat value; when both regimes provide, it weakly improves coverage, consumer surplus, and welfare relative to mandatory collection.Under mandatory collection, welfare remains below no data; bilateral comparisons depend on the selected continuation equilibria existing.Source: §4.2 · pp. 22–23

4.3. Imperfect Identifiation

  • For every positive identification probability, voluntary disclosure leaves consumers unwilling to disclose to a sole informed seller, preserving threat elimination under the reveal-or-no-signal specification.The result does not imply that all imperfect-information structures have the same effect.Source: §4.3 · pp. 23–24

4.4. Model-Based Counterfactuals

  • In the reported counterfactual, voluntary disclosure reaches full coverage while seller net profit remains at no-data levels; mandatory equal access creates no positive bilateral revenue gain and the platform withholds the service.The reported commission-based values for NDC, MDC, VDD, MEA, and VDD-IQ are 0.0000, −0.0332, 0.0067, 0.0000, and 0.0033, respectively; data-monetization values are 0.0000, 0.0383, 0.0379, 0.0000, and 0.0187.Source: §4.4 · pp. 24–25
  • A 20% purchase-rate lift paired with a 20% discount reduces revenue by 4%, so conversion lift alone can overstate the sales-revenue value of data.The zero-revenue-change boundary is ℓ∗(τ) = τ/(1 −τ).Source: §4.4 · p. 25
  • When GMDC < 0 < T MDC, mandatory-collection provision requires ψ ≥ −GMDC/[(1 −γ)T MDC], making threat-supported provision sensitive to available threat and commission.At baseline parameters, the threat share must exceed approximately 0.978; even with the full threat, data value turns negative when γ > 0.169.Source: §4.4 · pp. 25–26
  • Higher privacy costs need not raise the identification threshold for positive data value, and higher valuation need not facilitate monetization.At v = 0.73, c = 0.08, and γ = 0.15, platform and consumer outcomes can move differently as exact identification becomes more likely.Source: §4.4 · p. 26

5. Conclusion and Managerial Implications

The conclusion frames data-service value as both bilateral revenue change and the loss a seller avoids by refusing access. It recommends evaluating those counterfactuals separately and treating anonymous purchasing, equal access, and consumer disclosure rights as distinct policy levers.

Managerial implications

  • Evaluate data services against both bilateral-use revenue and the rival-only rejection outcome, since adoption can be defensive even when widespread use reduces product revenues.Assess conversion alongside gross revenue and retained profit after commissions and fees; deeper discounts can raise purchases while reducing revenue.Source: §5 · p. 27
  • Lower commissions strengthen extractable defensive payments by increasing sellers’ retained revenue at risk, while higher commissions increase the platform’s exposure to lost product revenue.Because commissions are exogenous in the model, the analysis does not identify an optimal commission policy.Source: §5 · p. 27
  • Voluntary disclosure can preserve anonymous purchasing, but only an enforceable cap at the no-data benchmark protects consumers from higher anonymous prices in an already covered market.The cap protects individual utility and welfare while preserving sellers’ no-data net profits; it may lower platform profit relative to uncapped voluntary disclosure.Source: §5 · p. 28
  • Equal-access rules and disclosure rights address different margins: equal access alone may leave defensive demand, whereas voluntary disclosure can support provision and full coverage in the reported counterfactual.Evaluate both instruments using revenue, retained profit, coverage, consumer surplus, and privacy losses rather than adoption or disclosure volume alone.Source: §5 · p. 28

Scope

  • The model isolates platform-controlled information, personalized pricing, seller adoption, and consumer disclosure, leaving heterogeneous sellers and independent first-party data for future research.Other proposed extensions include direct non-price benefits of disclosure and alternative information uses.Source: §5 · p. 29

Online Appendix to The Source of Data Value: Consumer Disclosure, Access Design,

The appendix contains baseline pricing-subgame equilibria, supplementary data-value analysis, proofs, and additional results and tables.

EC.1.1. MDC Pricing Subgame Equilibria

Under mandatory data collection, the equilibrium depends on which sellers adopt: bilateral personalization concentrates entry near the market center, while unilateral access can leave consumers out near the informed seller. The threat from unilateral access remains positive and can increase under an alternative equilibrium.

The “NN” Subgame

The “PP” Subgame

  • Under bilateral adoption, entry begins near the market center and expands toward the endpoints as v rises or privacy cost c falls.Personalized prices are highest near the endpoints and decline toward the center, so consumers enter where surplus covers c.Source: EC.1.1. MDC Pricing Subgame Equilibria · p. 32

The “PN” and “NP” Subgames

  • An alternative “PN” equilibrium has a smaller entry set and a threat effect that remains strictly positive and weakly exceeds the maximal-participation equilibrium’s.Selecting the maximal-participation equilibrium therefore gives a conservative measure of the threat from a rival’s data access.Source: EC.1.1. MDC Pricing Subgame Equilibria · p. 32
  • With unilateral adoption, partial coverage is concentrated toward the uninformed seller and expands toward the informed seller as v − c increases.The uninformed seller’s uniform offer constrains the informed seller’s surplus extraction and supports participation.Source: EC.1.1. MDC Pricing Subgame Equilibria · pp. 32–33

EC.1.2. VDD Pricing Subgame Equilibria

With voluntary disclosure, consumers can remain anonymous while purchasing, so unilateral seller adoption produces no usable data and leaves prices at the no-data outcome. When both sellers adopt, disclosure and personalized pricing depend on consumer valuation and privacy cost, with alternative equilibria affecting outcomes in a specified region.

The “NN” Subgame

The “PP” Subgame

  • With bilateral adoption, all consumers enter, while disclosure depends on valuation and privacy cost; disclosure expands toward the endpoints as v rises or c falls.Under the selected equilibrium, central consumers disclose when intensified price competition offsets privacy cost, while endpoint consumers are less inclined to disclose.Source: EC.1.2. VDD Pricing Subgame Equilibria · pp. 33–34
  • Alternative bilateral-adoption equilibria include full, empty, or partial disclosure in specified parameter regions, and the platform does not provide the service anticipating any of them.On P3, the alternative equilibrium weakly lowers platform profit and seller sales revenue, leaves seller profits unchanged, and strictly increases consumer surplus; all alternatives yield weakly lower seller sales revenues in the PP subgame.Source: EC.1.2. VDD Pricing Subgame Equilibria · p. 34

The “PN” and “NP” Subgames

EC.2.1. Credibility Microfoundation

The credibility extension makes the platform’s commitment to a subsidy depend on its incentive to honor it and the penalty for reneging. In the positive-threat region, higher implementation costs or commission weaken credibility and threat-based fee extraction, while a larger reneging penalty strengthens them.

Credibility endogenization

  • Credibility is the probability that the privately observed reneging penalty κ covers the platform’s continuation-profit loss from honoring the subsidy.The platform honors when κ ≥ ΔL_g, so η_g = 1 − Φκ(ΔL_g).Source: EC.2.1. Credibility Microfoundation · p. 34
  • When T_g > 0, the platform’s data value decreases with commission share γ, per-seller cost a, and fixed cost K, and weakly increases under an upward shift in the reneging-penalty distribution.The comparative statics concern the subsidy region.Source: EC.2.1. Credibility Microfoundation · p. 34
  • For T_g < 0, costs can raise data value only when their fee-extraction gains outweigh direct costs, and this case does not occur in an active service equilibrium under focal MDC and VDD.The effects depend on the stated thresholds and on γ’s direct and credibility effects.Source: EC.2.1. Credibility Microfoundation · pp. 34–35

EC.2.2. Conditional Bound on Threat Value

The threat effect is bounded by quality-weighted usable information available under unilateral access. Eliminating that information is sufficient to eliminate the threat, and is necessary when any positive usable information lowers the uninformed seller’s revenue below its no-data benchmark.

Threat bound and elimination

  • The threat effect satisfies 0 ≤ T_g ≤ Lβ_g, so β_g = 0 is sufficient to eliminate it.β_g measures quality-weighted usable unilateral information.Source: EC.2.2. Conditional Bound on Threat… · p. 35
  • If every β_g > 0 lowers the uninformed seller’s revenue below its no-data level, β_g = 0 is also necessary for threat elimination.The participation outcome must also return to the no-data benchmark; zero realized records alone may not remove a participation-driven threat.Source: EC.2.2. Conditional Bound on Threat… · p. 35

Appendix EC.3:

The appendix proofs characterize disclosure and pricing equilibria and derive conditions governing seller adoption, platform provision, and data value under mandatory and voluntary disclosure. They compute and compare seller, consumer-surplus, welfare, and platform-profit outcomes, including under anonymous-offer caps, exclusive access, heterogeneous privacy costs, and imperfect identification.

Proofs of Main Results

  • Without data, the symmetric price is v/2 for v < 1, v − 1/2 for 1 ≤ v < 3/2, and 1 for v ≥ 3/2.These are the equilibrium-price candidates selected by comparing covered- and uncovered-market pricing regions.Source: Proofs of Main Results · p. 36
  • Under MDC, both sellers adopt when the platform provides data, and provision occurs if and only if γ ≤ 1 + G^MDC/T^MDC.Because T^MDC > 0, the provision condition follows from the platform’s incremental profit.Source: Proofs of Main Results · p. 36
  • Under MDC monetization, market coverage does not exceed its no-data level, and social welfare falls throughout the provision regions.Consumer surplus can nevertheless rise for some parameter values: exactly when 1 < v < 3/2 and v − c > 1, or v ≥ 3/2 and c < 1/2.Source: Proofs of Main Results · pp. 36–37
  • Under VDD, unilateral access induces no disclosure, so defensive fees vanish and provision occurs exactly when G^VDD ≥ 0 and the selected opt-in set is nonempty.Provision occurs on the specified regions P1 ∪ P2 ∪ P3; outside them, the platform does not provide the service.Source: Proofs of Main Results · p. 37
  • When both regimes monetize data, VDD and MDC have the same opt-in sets and personalized prices on P1 ∪ P3, while VDD coverage equals min{D^MDC + D^NDC, 1}.The comparison differs on P2 when v > 3/2: VDD has only midpoint opt-in, whereas MDC can have a broader opt-in set.Source: Proofs of Main Results · p. 38
  • When both regimes provide data, VDD is at least as profitable for the platform if γ ≥ max{0, 1 − (r^PPV − r^PPM)/T^MDC}, and it yields higher consumer surplus and social welfare than MDC.The platform-profit comparison uses T^MDC > 0.Source: Proofs of Main Results · pp. 38–39
  • VDD yields a Pareto improvement over both NDC and MDC when 1/2 < v < 1 and (4v − 3 + √(2v^2 − 1))/4 ≤ c < v − 1/2.The proof excludes boundary cases where only the midpoint discloses and aggregate VDD outcomes equal NDC.Source: Proofs of Main Results · p. 39
  • Capping anonymous offers at the no-data benchmark preserves every consumer’s no-data purchase option, so consumer surplus and social welfare are weakly higher under the cap.With provision, the welfare gain is the consumer-surplus gain plus 2G^UP, which is nonnegative; without provision, the capped regime coincides with NDC.Source: Proofs of Main Results · pp. 39–40
  • Voluntary provision yields a Pareto improvement over both benchmarks only with positive-mass disclosure and γ ≥ γUM(v,c); positive-mass disclosure requires v < 1 and c < v − 1/2.Without positive-mass disclosure, voluntary provision reproduces NDC aggregate payoffs; below the γ threshold, the platform is worse off than under MDC.Source: Proofs of Main Results · p. 40
  • The platform weakly prefers exclusive access for seller 1 over seller 2, while exclusive VDD generates no usable information because no consumer discloses.Under the maintained provision convention, the platform therefore does not implement exclusive VDD.Source: Proofs of Main Results · p. 40
  • Exclusive MDC can be optimal only if v − c > (3 + √6)/4 > 1, yet lowers social welfare relative to NDC.The proofs also characterize parameter regions in which exclusive MDC raises consumer surplus above NDC.Source: Proofs of Main Results · p. 40
  • With heterogeneous privacy costs, bilateral MDC lowers each type’s revenue relative to NDC, weakly reduces coverage, and strictly lowers welfare whenever provided.These comparisons hold after weighting outcomes across privacy types.Source: Proofs of Main Results · p. 41
  • With heterogeneous privacy costs, unilateral VDD induces no disclosure and reproduces NDC outcomes; provision requires nonnegative bilateral revenue gains and a nonempty opt-in set.Under bilateral access, VDD weakly improves consumer surplus, coverage, and social welfare relative to MDC.Source: Proofs of Main Results · pp. 41–42
  • With imperfect identification, MDC is provided if and only if GMDC(q) + (1 − γ)T MDC(q) ≥ 0.Under bilateral VDD-IQ all consumers enter; under unilateral access no consumer discloses, and provision requires GVDD(q) ≥ 0 and a nonempty opt-in set.Source: Proofs of Main Results · p. 42

Proofs of Results in Appendix

  • In the baseline unilateral-MDC proof, maximal-participation equilibrium requires v − c ≥ 3/4 and makes the platform’s provision condition more demanding.In the bilateral no-data entry proof, entry occurs only when v − c ≥ 1/2.Source: Proofs of Results in Appendix · p. 43
  • Disclosure ranges from full or interval to singleton or empty as v and c vary; for v > 3/2 and c ≥ 1, interior disclosure is selected only when revenue and no-deviation conditions hold.On the selected interior equilibrium, the disclosure set is [1 − v + c, v − c].Source: Proofs of Results in Appendix · p. 45
  • In the alternative equilibria, cases (i) and (iii) change provision only at c = 1, with payoffs unchanged; for c > 1, both selections imply non-provision.On P3, the selected equilibrium provides data while an empty-disclosure equilibrium does not.Source: Proofs of Results in Appendix · p. 45
  • On P3, the selected equilibrium induces provision, while an alternative empty-disclosure equilibrium does not; seller profits are unchanged, and consumer surplus is strictly higher under the alternative.The selected equilibrium weakly raises platform profit and seller sales revenues relative to that alternative.Source: Proofs of Results in Appendix · p. 45
  • With exclusive access, every consumer prefers opting out by the privacy cost, so none discloses and all consumers enter.Opting out preserves the anonymous option to purchase or receive zero utility.Source: Proofs of Results in Appendix · pp. 45–46
  • When threat value T^g is positive, higher a or K lowers credibility and data value, while an upward shift in privacy-cost distribution κ weakly raises data value; when T^g is negative, the κ shift weakly lowers it.For T^g < 0, the effects of a and K depend on the stated density thresholds.Source: Proofs of Results in Appendix · p. 46
  • For MDC, the appendix derives conditional seller-profit, consumer-surplus, and welfare outcomes; whenever the platform does not provide data, the equilibrium coincides with NDC.The welfare calculation adds consumer surplus to the two sellers’ revenues because prices and platform payments are transfers.Source: Proofs of Results in Appendix · p. 46
  • For VDD, the appendix derives platform-profit and welfare expressions across provision regions; without provision, the equilibrium coincides with NDC.Conditional on provision, welfare equals consumer surplus plus twice seller revenue.Source: Proofs of Results in Appendix · p. 46
  • With identification probability q > 0, consumers disclose only when identified utility net of privacy cost is at least anonymous utility; global deviation checks establish a unique symmetric price and disclosure set.The equilibrium price has three branches, depending on the disclosure boundary relative to b_q and v/2.Source: Proofs of Results in Appendix · p. 47
Loading 2610.22864v1…