Reputation Economies

This is the most speculative slot in its category, and that is a reason to map it more fully rather than more cautiously. It is about reputation as an economic object — its price, its inflation, its counterfeiting, its transferability — and about the claim that reputation can function as a medium of exchange. It sits with two others on this map where a measurement has come to stand in for the thing it names: in Human Flourishing the thing is wellbeing, in Human Development Metrics it is development, and here it is trustworthiness. In all three the number acquired institutions the thing never had; here, uniquely, the number acquired a market.

Established “Medium of exchange” is a technical phrase, and the framing has to be read as four separable claims, because they are supported to wildly different degrees and public argument slides between them. (1) Reputation is a signal that changes behaviour — demonstrated repeatedly in field experiments, with effect sizes. (2) Reputation carries a price, capitalised into transaction terms — measured directly, at 8.1 per cent in the cleanest experiment. (3) Reputation is a store of value that persists and accumulates — true in narrow settings and degrading in most, because the scales inflate. (4) Reputation is a medium of exchange, the thing one gives up to obtain a good — largely untested, and where reputation is transferable, it is transferable by purchase.

Frontier The gap between (2) and (4) is the whole subject, and it is not a matter of degree. A price is not a currency. A seller with a good rating gets a higher price; that is reputation being rewarded by money, not reputation replacing money. The framing asserts the fourth claim; the evidence supports the first two, thins at the third, and largely fails at the fourth.

Established Start with the theorem, because it is what makes the framing a serious hypothesis rather than a metaphor. Kocherlakota's Money Is Memory works in a random-matching environment without commitment and defines memory as costless access to the full histories of all agents with whom one has had direct or indirect past contact. Two results follow: any allocation feasible with money is feasible with memory alone — money's function is replicated by an imaginary balance sheet in which producing raises your claim and consuming lowers it — and memory sometimes achieves allocations money cannot, because money requires producers to receive higher future utility than non-producers while memory can support symmetric allocations. The conclusion is that money is “a primitive form of memory”, technologically unnecessary where complete records exist. Frontier The assumptions are where the argument lives. The theorem requires records that are costless, complete, universally accessible, unforgeable and non-transferable, covering the entire chain of direct and indirect contact. Speculative Read that way, every empirical section below is a measurement of how far a real reputation system falls short of one of those five conditions: inflation attacks completeness of information content, fake reviews attack unforgeability, credit invisibility attacks universality, and name-trading attacks non-transferability.

Established The premium is real and it is small. Resnick, Zeckhauser, Swanson and Lockwood ran a controlled experiment in which an established high-reputation eBay dealer sold matched pairs of vintage postcard lots under his own long-standing identity and under new identities he also controlled. The measured difference in buyers' willingness to pay was 8.1 per cent of the selling price. Established A subsidiary experiment produced the more surprising result: one or two negative feedbacks on a new seller did not affect willingness to pay at all. That is the price of reputation, measured cleanly, and it is a modifier on a money price rather than a substitute for one.

Established The information content of a platform score is far lower than its numbers imply, and the best evidence comes from inside the platform. Nosko and Tadelis, with eBay data and eBay cooperation, report that eBay's percent-positive measure has a mean of 99.3 per cent and a median of 100 per cent; that about 65 per cent of buyers leave any feedback; and that only 0.39 per cent of transactions received negative feedback while more than 1 per cent had a dispute ticket opened, with silent dissatisfaction — the share of post-transaction messages containing language indicating a poor experience — running above 3.3 per cent. Their constructed effective percent-positive, which treats a silent buyer as informative, has a mean of 0.64 and a median of 0.67 against a public score whose median is 1.00. Established Their field experiment ran from 14 December 2011 to 2 January 2012 on about 10 per cent of eBay's US traffic, reordering search to favour high-effective-score sellers for a treatment group of 1,258,455 users against a control group of 11,486,810: the measured effect on 180-day return probability was +0.00112, about 0.1 percentage points — statistically significant and small. One author was at eBay Research Labs, and a finding that the platform's headline reputation metric is nearly uninformative runs against the platform's interest, which raises its weight. Frontier Their own recommendation abandons reputation as a displayed currency: put the quality signal inside the search ranking rather than on the page. That is reputation as an internal control variable, deliberately hidden from the people it describes — the opposite of a reputation economy.

Established Reputation inflation is the best-documented pathology in the field, and the private-versus-public comparison is what makes it decisive. Filippas, Horton and Golden, on a large online labour marketplace over 2007–2016 with 1.34 million transactions in the 2014–2016 window alone and corroborating data from four further peer-to-peer platforms: the share of transactions receiving a perfect five-star rating rose from 33 per cent to 85 per cent, average scores rose from 3.74 to 4.85, and ratings below three stars essentially disappeared. Established When employers gave private feedback invisible to the worker, 15 per cent expressed negative sentiment while only 4 per cent gave a public rating of three stars or below in the same period; among those privately saying “definitely not” would they hire again, 28.4 per cent publicly assigned four or more stars; and over the same period private scores fell 3.5 percentage points while public scores rose. Using written feedback text as an independent satisfaction measure, the authors attribute 67.7 per cent of the observed score increase to inflation — text-predicted scores reached 4.25 stars against an actual 4.86 by 2016. Established The mechanism is a ratchet: as some raters give the maximum unconditionally, the social cost to any individual rater of giving an honest lower score rises, because a merely-good rating becomes a signal of dissatisfaction. The consequence is pooling, and pooling at 85 per cent destroys the monotonic relationship between satisfaction and score.

Established The same terminal state arrives on a different platform by a different mechanism. Fradkin, Grewal and Holtz, on 3,295,188 listing-week observations covering 198,743 listings and 137,687 hosts from July 2014 to November 2015, find 91.6 per cent of listings rated at least 4.5 stars with an average review rating of 4.7. Airbnb's July 2014 redesign — a 14-day window with simultaneous reveal — was explicitly intended to remove retaliatory reviewing. Frontier The paper does not measure the change in review rates or negative-review frequency across that policy shift, restricting to post-change data precisely to avoid the contaminated period, so the redesign's effect is not established here. Frontier Read the three together: the two most-studied reputation systems on earth and a third all converge on near-unanimous positive ratings, by different mechanisms, and all three platforms responded by reducing the informational role of the visible score rather than repairing it.

Established Reputation is already purchasable and the price is known. He, Hollenbeck and Proserpio tracked the open market for fake Amazon reviews run through Facebook groups: about 23 active groups observed daily from March to October 2020, averaging 16,000 members and 568 posts per group per day, and approximately 1,500 unique products soliciting reviews between October 2019 and June 2020, with 84 per cent of identifiable sellers based in Shenzhen or Guangzhou. The unit price: the reviewer is refunded the product price plus PayPal fees of 2.9 per cent and sales tax of about 6.6 per cent, plus a cash commission averaging $6.24 and reaching $15. Established Short-run returns are real: ratings rise about 0.11 stars within two weeks, weekly review counts roughly double, log sales rank falls by 0.260, and search position improves by roughly 20 places; in a difference-in-differences using Amazon's March 2020 review purge as a shock, treated products gained 10 per cent more cumulative reviews and a 45.5 per cent greater reduction in sales rank than controls.

Established The long run is the important half and it reads like a monetary system under attack. After a campaign ends, average ratings fall back below their pre-campaign level, sales fall significantly (log sales rank +0.246), and the share of one-star reviews rises from about 8 per cent to over 15 per cent, exceeding 20 per cent for young products. Roughly 43 per cent of the purchased reviews are eventually deleted by the platform, at a median lag of 53 days and a mean above 100, while improved keyword rank persists about two months. Frontier Those are the economics of counterfeiting a currency: a short-lived claim on real resources that decays as the underlying quality asserts itself, an issuer who eventually voids about half the fraudulent units, and a lag between issue and voiding in which the profit lives. That is evidence for the monetary analogy in the framing and evidence against the desirability of what it proposes.

Established Credit scoring is the strongest case the framing has, and it is where exclusion is best measured. It is the only reputation system at national scale that reliably converts a behavioural history into access to real resources on stated terms. The CFPB's Credit Invisibles, on December 2010 data: 26 million US adults — 11 per cent — had no credit record at all, and a further 19 million (8.3 per cent) had records too thin or too stale to score, split 9.9 million insufficient and 9.6 million stale. Nearly one adult in five was outside the system. Established The distribution is not neutral: credit invisibility ran 28.9 per cent in low-income neighbourhoods against 3.6 per cent in upper-income ones, with unscored rates of 16.3 against 4.9; and by group, Black 14.8 per cent invisible and 13.0 unscored, Hispanic 15.8 and 11.9, White 9.4 and 6.7, Asian 9.8 and 7.3, with the differences emerging young and persisting.

Established Alternative data closes part of that gap, and the result is unusually clean. Berg, Burg, Gombovic and Puri, on about 250,000 purchases at a German e-commerce retailer where customers paid after shipment: mobile-phone users defaulted at 2.14 per cent against 0.74 for desktop; Android 1.79 against iOS 1.07; a premium email provider 0.51 against 1.45–1.96 for free services; purchases between midnight and 6am 1.97 against a 0.94 average; an email address containing a typographical error 5.09; and having one's own name in the email address was associated with a 30 per cent lower default probability. Discriminatory power by area under the curve: credit bureau score alone 68.3 per cent, digital footprint alone 69.6, both combined 73.6. Ten variables scraped from a checkout session outperformed a national credit bureau, and for customers the bureau could not score at all the digital footprint reached AUC 72.2, matching its performance on scorable customers; after adoption, default rates fell about 50 per cent for customers with low credit scores while acceptance held. Frontier That is the sharpest single result for the framing's optimistic reading: a reputation signal assembled from behavioural exhaust, at near-zero cost, extending credit to people the formal system could not evaluate.

Established And the fairness result runs the other way, on the intensive margin. Fuster, Goldsmith-Pinkham, Ramadorai and Walther, on 9.37 million conventional fixed-rate US mortgages from 2009–2013 with an equilibrium analysis on a 100,000-loan 2011 subset: moving from a nonlinear logit to a random forest raised the acceptance rate from 89.8 to 92.0 per cent and slightly reduced cross-group dispersion in acceptance (0.024 against 0.026). But equilibrium interest rates came out at 4.55 per cent for Asian, 4.59 for White non-Hispanic, 4.64 for Hispanic and 4.64 for Black borrowers, with cross-group rate disparity rising 23 per cent under the machine-learning model, and within-group rate dispersion doubling for Black borrowers, standard deviation 0.40 to 0.80, against roughly 1.6 times for White non-Hispanic and Asian borrowers. Removing race from the inputs barely degraded predictive performance, because the model reconstructs it from correlated variables. Established The finding to carry is uncomfortable and precise: a better reputation instrument admits more people and prices them further apart, and the widening falls hardest on the groups with the weakest reputational records.

Established Citation is the reputation system that most resembles a currency, and its counterfeiting rate is measurable. It is accumulated, denominated, portable with the person between institutions, and exchanged for jobs, grants and salary. Ioannidis, Boyack and Baas's standardised author database ranks scientists on a composite of six metrics covering the top 100,000 across all fields plus everyone in the top 2 per cent of their subdiscipline. The self-citation sensitivity result: 4.9 per cent of scientists in the top 2 per cent for career-long impact drop below that threshold when self-citations are excluded, and 0.01 per cent — fifteen people — fall below the top 10 per cent. Frontier Both sides of that matter. The currency is mostly robust to the most obvious self-dealing — 95 per cent of top-2 per cent standing survives removing self-citation entirely — and there is a thin tail whose standing is an artefact of their own citations, which is what a counterfeit looks like in a citation currency. Established The proposed alternative currency has weak convergent validity: a meta-analysis of 914 correlation coefficients from 111 studies across 13 altmetric indicators found Mendeley, usage metrics, ResearchGate, Twitter and peer ratings most strongly associated with citations, while Facebook, Wikipedia, Google+, blogs, news and Reddit were only weakly associated, with no variable consistently moderating the relationship — and the authors conclude that the weakly associated channels are measuring a different construct.

Established Now the correction, because the most-cited real-world instance of a reputation economy is described wrongly in specific, fixable ways. There is no unified national citizen score in China, by explicit policy. MERICS, reading the Chinese policy documents, states that there will not be a unified social credit score rating individual behaviour and that an all-encompassing scoring system was not part of the original plan; Daum, working from the primary texts, finds the documents focus more on information collection, consolidation and aggregation than on scoring, and envisage many different ratings for different uses. The Unified Social Credit Code frequently reported as a score is an identification number — a national ID for individuals, an assigned number for organisations — used to associate records. Established What exists is a compliance-file and blacklist architecture aimed mainly at companies. The regulatory focus splits companies 73.3 per cent, government entities 13.3, individuals 10.3, social organisations 3.3, and annual blacklisting rates run 1–2 per cent of companies, 0.15–0.3 per cent of citizens, and under 0.1 per cent of government entities.

Established The one consequential individual list is the courts' judgment-defaulters list — people and organisations with final judgments against them who can comply and do not. Sanctions include denial of air tickets, high-speed rail, private schooling and certain entertainment spending, with approximately 11 million ticket purchases denied, and removal available by performing on the judgment. That is a targeted enforcement mechanism attached to a court finding, not a behaviour score. Established And Sesame Credit is not the social credit system and never became a licensed credit bureau. In 2015 the People's Bank of China authorised eight companies including Sesame Credit to experiment with individual credit reporting; all eight were subsequently denied personal credit-investigation licences, on conflict-of-interest grounds and doubts about the creditworthiness indicators used, and were instead made equal shareholders in a shared database of non-bank online lending information. Daum's characterisation is “a Loyalty Rewards Program with fantastic swagger rather than a real credit rating”: voluntary enrolment, benefits limited to deposit waivers and affiliate discounts, no negative consequence for a low score. Established The corporate system is sector-specific rather than unified — tax authorities rate on an A-B-M-C-D scale, customs uses certified, general-credit and discredited categories, with redlists, blacklists and close-watch lists and publicity periods from three months to three years — and the Draft Social Credit Construction Law has been pending since November 2022, while an April 2026 implementation plan for comprehensive enterprise credit evaluation signals movement toward unification.

Frontier The corrected picture is more interesting than the myth. The most ambitious state attempt at reputation infrastructure deliberately did not build a general-purpose reputation currency, kept individual exposure to about a quarter of a per cent a year, tied its sharpest sanctions to an existing legal finding, and refused licences to the private scores that came closest to a transferable personal reputation asset. The strongest available real-world data point does not show a reputation economy being built. It shows a state with the capacity to build one choosing a narrower design.

Established Which brings the brief to its crux, and the crux has a formal answer that is thirty years old and runs against the framing. Tadelis's What's in a Name? Reputation as a Tradeable Asset models firms identified only by a name carrying a history, where clients cannot observe changes of ownership. Proposition 1: names with good histories are traded in every equilibrium. Even starting from no trade, a firm with a successful history earns more revenue, so a new agent can profit by secretly buying the name; there is no equilibrium in which reputations stay with the people who earned them once transfer is possible. Proposition 3: there is no equilibrium in which only good types buy good names. A maintenance effect makes good types value an established name more because they can sustain it; a start-up effect makes bad types value it more because good types can build their own name and bad types cannot. If good types monopolised purchases the start-up effect would dominate and bad types would outbid them, so pooling is forced. Established The answer to the crux is therefore: no, not without an observability condition. Reputation can be transferable, or it can be a clean signal of the current holder's type. It cannot generally be both. Frontier And the model has an empirical shadow already in this brief: the market for fake Amazon reviews is a market for unobservable transfer of reputation, at $6.24 plus a refund, exhibiting exactly the predicted dynamics. The theory said bad types would buy good names. They do, and the price is known.

Speculative The most interesting open question in this subject is the one nobody has built and no fetched model analyses: transferable-but-provenance-visible reputation. Tadelis's result depends on transfer being unobservable. If every transfer is recorded and visible, the name carries its provenance, and what is bought is a history explicitly labelled as someone else's — which is not what the adverse-selection argument rules out. Frontier A public ledger provides exactly that observability, and the leading blockchain proposal declines to use it. The soulbound-token design chooses non-transferability rather than transparent transferability, so the one escape clause in the governing theorem has never been tested by the technology best placed to test it. Speculative Under that design reputation would behave like a used capital good with a service history rather than like a currency or a personal attribute, and the open questions are sharp: does a provenance-labelled name still command a premium; does the premium decay with each transfer; and does a market in labelled histories produce pooling anyway through some channel the model does not contain.

Frontier Question two: do reputation systems inevitably inflate to uselessness? Three platforms, three mechanisms, one terminal state — 33 to 85 per cent perfect ratings on a labour marketplace by social ratchet, 91.6 per cent at 4.5 stars or above on Airbnb under retaliation pressure, a 99.3 per cent mean positive on eBay. Frontier The pattern is established; the inevitability is not. It would be falsified by a large system that has held a dispersed distribution for a decade, and no such system appears in anything fetched here. Speculative Question three, its corollary: is inflation fixable by mechanism design? Simultaneous reveal, private feedback channels and moving quality into the ranking all had measurable effects, and none restored a dispersed public scale.

Frontier Question four: is the solution to hide the score? The eBay researchers' own recommendation is to use the quality signal inside search ranking rather than displaying it, and that is now standard platform practice. It works, and it abolishes the reputation economy in the process: an internal quality variable is not a currency, because nobody can hold it or spend it. Speculative Question five: must reputation be non-transferable to be meaningful? This is the soulbound-token thesis, and Tadelis's adverse-selection result supports its underlying logic. There is no deployment, and its authors say so themselves.

Frontier Question six: is a reputation economy a surveillance economy by construction? The theoretical point is solid — the theorem's own precondition is records of all direct and indirect contact, which is total observation — and the digital-footprint result, in which ten checkout variables outperform a national credit bureau, shows how little collection is needed to build a working score. Speculative The empirical claim that any deployed reputation currency has required mass surveillance is not supported, for the trivial reason that none has been deployed. Speculative Question seven: will China build the thing it is said to have built? The defensible version of the claim points at the Draft Social Credit Construction Law pending since November 2022 and an April 2026 implementation plan for comprehensive enterprise credit evaluation, both pointing toward integration. That is a forecast about legislation, not a description of a system.

Frontier Question eight: is credit scoring already the answer, and is the debate over? For it: national scale, real access consequences, and alternative data reaching the unscorable at AUC 72.2. Against it: 26 million people outside the system entirely, and a better instrument widening within-group price dispersion for the groups with the weakest records. Speculative Question nine: is uniqueness of persons the binding constraint? That is the implicit premise of most blockchain identity work, and everything in the inflation and counterfeiting record argues against it — uniqueness was never the failure mode in any measured system in this brief. Speculative Question ten, the oldest position: does reputation only ever work inside a bounded community with shared norms and exit costs? It predicts that every attempt at a universal reputation currency degrades to inflation or surveillance, which is consistent with every finding here, and it is unfalsifiable as stated because “bounded community” is not operationalised.

Handwave And question eleven, the strongest dissolving objection, stated because a map that omits it is incomplete: the whole framing may be a category error. On this view reputation is a belief, and beliefs cannot be exchanged, only caused; a reputation economy is therefore always a system that exchanges records, and records are documents, which behave like documents. Handwave Nobody has proposed a way to test this against the memory theorem, which is why it carries the weakest flag in this brief despite being the most economical explanation of why every empirical section ends in the same place.

Established The first bottleneck is that the visible score stops carrying information long before anyone notices. A median public score of 100 per cent against an effective measure with a median of 0.67 is not a small calibration error; it is two different quantities on the same page. Established The inflation ratchet is a social mechanism, not a design bug — once enough raters give the maximum unconditionally, an honest lower score becomes an act of hostility, and the cost falls on the rater. No mechanism in the record has reversed it.

Established The second is unforgeability, and it has a price. A reputation unit costs a refund plus 2.9 per cent in payment fees, about 6.6 per cent in sales tax and a cash commission averaging $6.24. The issuer voids roughly 43 per cent of the counterfeit units at a median lag of 53 days. Frontier A monetary system with those parameters would be described as badly defended, and the lag rather than the detection rate is the binding problem: the profit lives entirely inside it.

Established The third is universality. Nearly one US adult in five was invisible or unscorable, with invisibility running 28.9 per cent in low-income neighbourhoods against 3.6 in upper-income ones. Frontier A currency a fifth of the population cannot hold is not a currency; and the instrument that fixes the coverage — behavioural exhaust at AUC 72.2 for the previously unscorable — widens price dispersion for the same populations, so the two failure modes trade against each other rather than resolving.

Frontier The fourth is that no reputation system in the record has a unit of account or a clearing mechanism. Stars, percentiles, h-indices and blacklist membership are not denominated in anything, cannot be netted between parties, and have no settlement. Speculative Every proposal that treats reputation as money skips this, and it is where the analogy actually breaks: the fake-review market prices reputation in dollars, which is the opposite of reputation functioning as money.

Established Nothing on this map produces a result this brief waits on. It waits on one missing formal result and two institutional facts, recorded as typed requirements below. The formal result is a model of transferable reputation under observable transfer — the escape clause in the governing theorem, which no fetched work analyses. The institutional facts are a transfer registry that makes reputation provenance visible and a platform willing to publish its internal quality score beside its public one.

Frontier What it waits on from research is unusually concrete, because the data already exists inside firms. The private-feedback comparison that proved inflation was possible only because one platform collected private ratings; the effective-percent-positive measure existed only because one team had transaction-level dispute and message data. Speculative Every decisive result in this brief came from inside a platform, which means the research frontier is a data-access question rather than a methodological one.

Speculative Two areas are empty rather than thin. No formal or empirical work analyses provenance-visible transfer. And no study measures whether a reputation system has ever held a dispersed rating distribution over a decade — the observation that would settle whether inflation is a tendency or a law.

Speculative The experiment that would move this subject most is the one nobody has run: build a market in provenance-labelled reputation. Permit transfer, record every transfer publicly, display the chain of custody, and measure whether a labelled name still commands a premium, how fast the premium decays across transfers, and whether pooling appears anyway. Frontier This is the direct test of the observability assumption on which the governing theorem's negative result depends, and a ledger-based platform could run it in one product cycle.

Established Second, and cheap: publish the private score beside the public one. Any platform holding both already knows the size of its own inflation — on the one marketplace where the comparison was made, 15 per cent private negative against 4 per cent public at three stars or below, and 28.4 per cent of the privately damning giving four stars or more. Frontier Doing it publicly, on a second platform, would establish whether 67.7 per cent inflation is a general figure or one marketplace's number.

Frontier Third: measure the counterfeit lag rather than the detection rate. Detection at 43 per cent with a median 53-day lag is a system where the arbitrage is temporal. A randomised variation in review-verification latency would price how much of the fake-review market is a function of speed alone. Speculative Fourth: run the alternative-data fairness experiment forward. The mortgage result shows acceptance and dispersion moving in opposite directions; a lender adopting behavioural-exhaust scoring under a pre-registered fairness protocol would establish whether the tradeoff is intrinsic or a property of one model class.

Speculative Fifth: test whether anything restores a dispersed public scale. Forced-choice ranking, quota-constrained ratings, decaying scores, comparative rather than absolute judgement — none of these has been tried at scale in the fetched record, and the platform response has uniformly been to hide the score rather than repair it. Frontier A single platform trying repair rather than concealment would be the first real test of hypothesis (4) in section 3.

Established The engineering requirement implied by the theorem is stated precisely enough to audit against, and no built system meets it. Records must be costless, complete, universally accessible, unforgeable and non-transferable, across all direct and indirect contact. Established Real systems fail each in a measured way: information content collapses (effective score median 0.67 against a public median of 1.00), unforgeability fails at $6.24 a unit, universality fails for 26 million US adults, and non-transferability fails as a matter of equilibrium rather than enforcement.

Frontier The most developed engineering proposal is the soulbound token — a non-transferable, issuer-revocable credential held by a wallet. The argument is precisely this brief's crux: transferable credentials permit reputation laundering and Sybil attacks, and transferability divorces the achievement from the person. Proposed applications include provenance staking for creative work, uncollateralised lending against a constellation of credentials, social recovery of lost wallets by community consent, Sybil-resistant governance through correlation discounting of voting power, and data-provenance-tracked AI. Established The authors' own epistemic status should be quoted rather than softened: the work consists of “sketches” requiring “extensive red teaming”, many proposals are “more suggestive than fully prescriptive”, the correlation-discounting formula is “a first pass for experimentation” with acknowledged vulnerabilities, and no deployed system demonstrates these mechanisms at scale. It is a design agenda by people who would build it.

Established The largest deployed identity layer supplies uniqueness, not reputation, and its constraint is regulatory rather than technical. World Network reported 10 million verified humans as of January 2025 via iris scanning. The measurement that matters is its legal record: Chile's Supreme Court upheld an order to delete a minor's World ID and all associated records within 30 days after a scan taken without parental consent; German regulators found GDPR violations; and the Dominican Republic ordered a suspension of operations on data-privacy grounds. Speculative Proof-of-personhood establishes that you are one person, not that you are trustworthy, and every proposal in this space assumes uniqueness is the hard part. Frontier The evidence in this brief says the hard part is keeping the content informative once the identity exists, which no blockchain design in the record addresses.

Frontier The engineering gap that would matter most is a transfer registry. If reputation is going to be transferable in fact — and the fake-review market demonstrates that it already is — then the choice is between unobservable transfer, which the theorem says forces pooling, and recorded transfer with visible provenance, which nobody has built. Speculative A ledger is the obvious substrate and the leading ledger proposal declined it. That is an engineering decision, not a constraint, and it is reversible.

The boundary with the funding-models slot is by function rather than by object, and it needs stating because citation appears in both. Scientific Funding Models owns funder allocation — what a score buys in grant dollars, and the concentration that follows from it. This brief owns citation and altmetrics as reputation currencies: that citation is accumulated, denominated, portable with the person between institutions, exchangeable for posts and salary, and counterfeitable at a measurable rate, with 4.9 per cent of top-2 per cent standing failing on removal of self-citation and fifteen people falling below the top decile. Funding-model design is not claimed here. Established Scientific Governance Models separately owns the peer-review score as an allocation and gatekeeping rule and its reliability; the one structural import from it is its checkpoint-versus-judgement finding, used here as a lens, since every platform intervention that worked was a rule applied to every item — screening at acceptance, simultaneous reveal, moving quality into the ranking — while attempts to improve individual raters' judgement are absent from the measured record. That is an application of a finding from elsewhere, offered as corroboration and flagged as such.

The other two measurement briefs are the closest neighbours on this map: Human Flourishing, where the failure is cardinality and the institution is a finance ministry, and Human Development Metrics, where the failure is weighting and the pathology is rank-chasing. Here the failure is inflation and counterfeiting, and the pathology has a market price. Also within this map: Intelligence Measurement, for what happens when a score becomes portable between institutions; Digital Citizenship, which owns the identity layer this brief takes only the uniqueness question from; Institutional Design, where the pooling equilibrium is the general problem; Distributed Governance and AI Governance, for Sybil resistance and for scoring systems applied to conduct; Future Legal Systems, for enforcement attached to a court finding rather than to a score; and Civic Technology, for what a public-sector reputation layer would sit on.

Outside it: mechanism design and the economics of information, which supplies both the memory theorem and the tradeable-name result; platform economics, which supplies the field experiments; consumer credit regulation, which supplies the fairness record; and scientometrics, which supplies the citation currency and its pathologies. Content moderation, platform liability, biometric privacy law and state surveillance capacity belong elsewhere; this brief takes from them only what bears on whether a reputation currency exists.

Established The most useful institutional pattern in this brief is that the strongest evidence against platform reputation came from inside platforms. The finding that eBay's headline metric is nearly uninformative was produced with eBay's cooperation by a team including a researcher at its own research labs; the inflation result required a marketplace's private feedback data; the Airbnb result required listing-level data. All three run against the host's interest and are weighted up accordingly. Frontier The corollary is uncomfortable: the research frontier is inside firms, and what is knowable about reputation systems is whatever those firms choose to permit.

Established The China record is the clearest case in this brief of an institution making a design choice that most commentary assumes it did not make. Eight companies were authorised in 2015 to experiment with individual credit reporting and all eight were denied personal credit-investigation licences, on conflict-of-interest grounds and doubts about the indicators used, then folded into a shared database. The private scores closest to a transferable personal reputation asset were refused, by the state usually described as having built one. Established The corporate system remains sector-specific — tax on an A-B-M-C-D scale, customs on certified, general-credit and discredited categories, publicity periods of three months to three years — and the unifying law has been a draft since November 2022.

Established Regulation is the binding constraint on the deployed identity layer, and it bit fast. A supreme court deletion order in one country, a GDPR breach finding in another, and a suspension of operations in a third, against 10 million verified humans. Frontier Any reputation infrastructure that begins with biometric enrolment inherits that record, and it is a constraint on the substrate rather than on the reputation idea itself.

Frontier Two institutional requirements are recorded below because they are cheap and undone. A transfer registry making provenance visible: reputation is already transferable in fact, and the only question is whether the transfers are recorded. And a platform publishing its internal quality score beside the public one: the platforms already compute both, the difference is the size of their own inflation, and none publishes it. Neither requires a discovery.

Established The sharpest ethical finding here is that accuracy and fairness moved in opposite directions on the same data. A machine-learning model raised mortgage acceptance from 89.8 to 92.0 per cent and slightly narrowed cross-group acceptance dispersion, while raising cross-group rate disparity by 23 per cent and doubling within-group rate dispersion for Black borrowers. Established Removing race from the inputs barely degraded performance, because the model reconstructs it from correlates — which means the standard remedy is not a remedy. A better reputation instrument admits more people and prices them further apart.

Established The exclusion problem is measured and it is large. 26 million US adults with no credit record and 19 million unscorable, with invisibility at 28.9 per cent in low-income neighbourhoods against 3.6 in upper-income ones and clear differences by race and ethnicity that emerge young and persist. Frontier Any proposal to widen reputation into a general medium of exchange has to answer what happens to the fifth of the population with no reputational record, and the honest answer in the record is that behavioural exhaust would score them — which is the surveillance answer.

Established On China, accuracy is itself the ethical obligation, and the corrections cut in both directions. Against the myth: no unified citizen score by explicit policy, individual blacklisting at 0.15–0.3 per cent a year, the sharpest sanctions attached to an existing court judgment, private scoring licences refused. Established Against complacency: approximately 11 million ticket purchases denied, sanctions reaching private schooling and entertainment spending, and blacklist publicity periods up to three years. Frontier A targeted enforcement mechanism attached to a court finding is a serious civil-liberties object in its own right, and it is a different object from the one Western commentary describes. Getting it wrong makes the real thing harder to argue about.

Frontier And the fake-review market is an ethical object with an unusual shape. Its victims are buyers who cannot see the counterfeit, its beneficiaries are sellers of predominantly young low-quality products, and its long-run outcome is that the buyers of reputation end up worse off than before the campaign — ratings below the starting point, one-star share doubling from about 8 to over 15 per cent, sales falling. Speculative Self-limiting after a profitable interval is not the same as deterred, and the interval is the median 53 days in which the platform has not yet voided the units.

Established The terminal position is that the framing has already been answered by events, and the answer is not the one it expects. “Reputation as a medium of exchange” is usually offered as a design proposal awaiting implementation. It is not. Wherever reputation acquires exchange value and transfer is unobservable, a market in reputation appears immediately — 23 Facebook groups averaging 16,000 members, 1,500 products, a known unit price, a measured effect and a measured decay. The proposal has been running as an experiment for years, in the direction nobody wanted.

Established Every real system in the record does one of three things, and none of them is a reputation economy. It forbids transfer, as the soulbound-token design does on paper and has never done in deployment. It polices transfer after the fact at scale, as a platform voiding 43 per cent of counterfeit units at a median 53-day lag does. Or it abolishes the public currency, hiding quality inside the ranking algorithm, which is now standard practice and was the explicit recommendation of the researchers who measured the score's emptiness. Frontier Non-transferability, enforcement, abolition. Those are the three observed responses, and each is a retreat from the framing rather than an implementation of it.

Speculative What survives is a change of question, and it is the one the theorem actually poses. Kocherlakota's result is that money is a primitive form of memory — not that reputation should replace money, but that money exists because records are incomplete. Frontier The interesting civilizational question is therefore not whether to build a reputation currency but what happens to money as records become more complete, and the measured answer so far is: records got vastly more complete, money did not recede, and the reputational layer that grew alongside it inflated to near-uselessness and acquired a counterfeiting market. That is a result about the theorem's assumptions, and it deserves more attention than it gets.

Speculative The unexplored middle remains the most consequential open position, and it is worth restating as the brief's last claim. The negative result depends on transfer being unobservable. Nobody has built transferable-but-provenance-visible reputation; no fetched model analyses it; and the technology best suited to it chose non-transferability instead. Handwave Whether a name with a visible chain of custody behaves like a used capital good with a service history, or collapses into the same pooling by a route the model does not contain, is a genuinely open question — and it is open because nobody tried, not because anybody proved it shut.

These horizons track platform policy changes, credit-regulation cycles and one pending law rather than technology:

• 10 yr: Frontier Expect the concealment trend to continue: platforms move quality signals into ranking and away from displayed scores, because it works and the alternative has not been found. Frontier Expect alternative-data credit scoring to expand, because the measured performance gain for the previously unscorable is large, and expect the fairness objection to arrive through regulation rather than through model design, because removing protected characteristics does not remove their influence. Speculative China's Draft Social Credit Construction Law, pending since November 2022, either passes or does not inside this window, and the enterprise credit evaluation plan is the nearer-term movement. Speculative A provenance-visible transfer experiment is technically available today and there is no sign anyone intends to run one.
• 25 yr: Speculative The plausible split is that reputation remains a price modifier in every general market and becomes a genuine access mechanism only inside bounded systems with real exit costs — professional licensing, credit, academic standing — which is what the oldest position in section 3 predicts. Speculative If any general-purpose reputation currency is attempted at scale, the two things to watch are its inflation curve and its counterfeit lag, because those are the two parameters every measured system converged on. Handwave Forecasting which is asserting a design decision rather than extrapolating a measurement.
• 50 yr: Speculative If records become complete enough for the memory theorem's preconditions to be approached, the interesting outcome is not a reputation economy but a change in what money is doing, and there is no measured evidence about that at all. Speculative The alternative and better-supported path is that reputational data becomes ubiquitous, is used internally by allocators, and is never visible to the people it describes — which is the current trajectory extended. Handwave Both are extrapolations from a theorem and a platform-policy trend.
• 100 / 250+ yr: Handwave Beyond useful forecasting. The only long-horizon datum is that bounded merchant communities have run reputation-based exchange without state enforcement for many centuries, and that every attempt to unbound one has been recent. Handwave A historical pattern with no measured mechanism is a story rather than a base rate.

• Depends on Nothing on this map. No brief here produces a result this one waits on; the missing pieces are one formal analysis nobody has published and two registries nobody has built. No typed depends-on edge is claimed.
• Requires (not on this map) One missing formal result and two registries. A model of transferable reputation under observable transfer is the specific gap a theorist would recognise: the governing negative result — names with good histories are traded in every equilibrium, and there is no equilibrium in which only good types buy them — is derived under the assumption that clients cannot observe a change of ownership, and no work fetched for this brief analyses the case where every transfer is recorded and the name carries its provenance. That is not a restatement of the brief's title; it is a named assumption in a 1999 American Economic Review paper that nobody has relaxed. A transfer registry that makes reputation provenance visible is the corresponding institutional object, and the point is that reputation is already transferable in fact, at a commission averaging $6.24 plus a refund of price, payment fees and sales tax, so the only open question is whether the transfers are recorded; a public ledger supplies exactly the observability the theorem's escape clause needs, and the leading ledger-based proposal chose non-transferability instead of testing it. And a platform that publishes its internal quality score beside the public one: platforms already compute both, the gap between them is the size of their own inflation — on the one marketplace where the comparison was published, 15 per cent of private feedback was negative against 4 per cent of public ratings at three stars or below, with 28.4 per cent of the privately damning giving four stars or more, and an effective score with a median of 0.67 against a public median of 1.00 — and no platform publishes it voluntarily.
• Enables In principle, any decentralised allocation mechanism on this map that proposes to substitute standing for collateral — uncollateralised lending against credentials, Sybil-resistant governance, community-weighted voting — rests on reputation being non-purchasable. No typed enabling edge is claimed, and the reason is a finding: under unobservable transfer, good names are traded in every equilibrium and no equilibrium exists in which only good types buy them, so the enabling condition has been proved absent rather than merely unmeasured.
• Adjacent Mechanism design and the economics of information, which supplies both the memory theorem and the tradeable-name result; platform economics, which supplies the field experiments and the inflation record; consumer credit regulation, which supplies the coverage and fairness measurements; scientometrics, which supplies the citation currency; and within this map Human Flourishing, Human Development Metrics, Scientific Governance Models and Digital Citizenship.

Established “China operates a national citizen reputation score.” It does not, by explicit policy, and correcting this is worth the space because the error is routine in Western commentary and it distorts every argument built on it. Chinese policy documents state there will not be a unified score rating individual behaviour; an all-encompassing scoring system was not part of the original plan; the documents focus on information collection, consolidation and aggregation rather than scoring, and envisage many different ratings for different uses. The Unified Social Credit Code so often reported as a score is an identification number. Established What exists is a compliance-file and blacklist architecture aimed mainly at companies — regulatory focus splitting 73.3 per cent companies, 13.3 government entities, 10.3 individuals, 3.3 social organisations, with annual blacklisting at 1–2 per cent of companies and 0.15–0.3 per cent of citizens. The specific misreportings, named by two independent China specialists: that all surveillance feeds one centralised database; that every recordable action carries a point value; that everyone receives a score; that enrolment is mandatory; and the causal link routinely drawn from the court blacklist to a commercial loyalty programme.

Established “Sesame Credit is the social credit system.” It is not, and it never became a licensed credit bureau. Eight companies including Sesame Credit were authorised by the central bank in 2015 to experiment with individual credit reporting, and all eight were subsequently denied personal credit-investigation licences, on conflict-of-interest grounds and doubts about the creditworthiness indicators used, then made equal shareholders in a shared database of non-bank online lending information. Established A China law specialist's characterisation is “a Loyalty Rewards Program with fantastic swagger rather than a real credit rating”: voluntary enrolment, benefits limited to deposit waivers and affiliate discounts, no negative consequence for a low score. Frontier The corrected picture is the more interesting one: the state usually credited with building a reputation currency refused licences to the private schemes that came closest to one.

Established “Platform ratings tell you how good a seller is.” They tell you much less than their scale implies. A percent-positive with a mean of 99.3 and a median of 100 against an effective measure with a median of 0.67; 0.39 per cent of transactions receiving negative feedback while more than 1 per cent opened a dispute and over 3.3 per cent showed silent dissatisfaction; a perfect-rating share moving from 33 to 85 per cent in nine years with 67.7 per cent of the rise attributed to inflation; 91.6 per cent of listings on another platform at 4.5 stars or above. Established Ratings above the ninetieth percentile of a compressed scale are not fine-grained information about quality; they are a pooling equilibrium.

Established “Fake reviews are a marginal enforcement problem.” They are a functioning market with a known price and a measurable, temporarily profitable return: $6.24 average commission plus a refund of price, 2.9 per cent payment fees and about 6.6 per cent sales tax; +0.11 stars, doubled weekly reviews, 0.260 off log sales rank and about 20 search positions. Established And the platform voids about 43 per cent of the units at a median 53-day lag, which is an enforcement rate that would be considered a failure in any currency. Frontier The self-limiting long run — ratings ending below where they started, one-star share doubling, sales falling — is real and it is not deterrence, because it arrives after the profitable interval.

Established “Citation is a corrupted currency.” Overstated, and this brief resists rounding it up. 95.1 per cent of top-2 per cent standing survives removing self-citations entirely; the failure is confined to a thin tail, and only fifteen people in the whole database fall below the top decile once self-citation is excluded. Frontier The counterfeiting is real, measurable and small — which is a better security record than any platform rating system in this brief. The database's own authors caution that comparisons are meaningful only within a subdiscipline.

Established “Altmetrics are an alternative reputation currency.” Not on the evidence. A meta-analysis of 914 correlations from 111 studies across 13 indicators found Mendeley, usage metrics, ResearchGate, Twitter and peer ratings most strongly associated with citations, while Facebook, Wikipedia, Google+, blogs, news and Reddit were only weakly associated, with no consistent moderator. Speculative Channels showing no association are measuring a different construct, and aggregating heterogeneous channels into an “attention” score pools sources with different meanings — the same defect as pooling five-star ratings, arriving by a different route.

Established “Blockchain reputation systems have been shown to work.” Their own authors say otherwise: the leading proposal describes itself as “sketches” requiring “extensive red teaming”, calls its central formula “a first pass for experimentation”, and states that no deployed system demonstrates the mechanisms at scale. Established The largest deployed identity layer has been ordered to delete a minor's records by a supreme court, found in breach of GDPR, and suspended in a third jurisdiction. Speculative “Proof-of-personhood solves the reputation problem” fails on a simpler point: uniqueness was never the failure mode in any measured system in this brief. Inflation, counterfeiting and exclusion were.

Frontier “Alternative-data credit scoring is straightforwardly inclusive.” It reaches the previously unscorable at AUC 72.2 and cut default rates about 50 per cent for low-score customers in the one adopting retailer, which is a real gain. Established And a better instrument widened cross-group rate disparity by 23 per cent and doubled within-group rate dispersion for Black borrowers. Established Removing race from the model barely degrades performance, because the model rebuilds it from correlates — so the obvious fix is not one. Both findings are real; the site should say both.

Speculative And the framing itself: “reputation as a medium of exchange” describes something that already exists and is not what was wanted. Reputation is priced at 8.1 per cent, rewarded, accumulated, and — where transfer is unobservable — bought and sold at a known unit price. Established In no measured system is it what a buyer surrenders to obtain a good, which is what a medium of exchange means. Handwave The strongest dissolving objection is that this could never have been otherwise, because reputation is a belief and beliefs cannot be exchanged, only caused — so a reputation economy is always a system that exchanges records, and records behave like documents. Nobody has proposed a way to test that against the memory theorem, and it is carried here at the weakest flag precisely because it explains every result in this brief and predicts nothing new.