Summary

  • The NSFNET Acceptable Use Policy governed traffic carried by a subsidised backbone. Contemporary connection guidance treated provider selection, policy eligibility, address registration, circuit installation, routing configuration and end-to-end testing as separate decisions.
  • The policy acquired practical force through institutional layers. NSF retained interpretation authority, regional networks incorporated acceptable-use provisions into customer relationships, and commercial traffic depended on additional provider and interconnection arrangements.
  • MCI Mail, the disputed Dialog service and the prospective ANS-CIX interconnection illuminate different parts of this system. Together they document permission, classification and negotiation, but they support only a bounded inference about the usefulness of routable identifiers.
  • The AUP allocated eligibility for a valuable transport subsidy. It did not allocate, revoke or reassign Internet Protocol numbers, and its significance diminished as commercial backbones and exchanges offered more credible alternatives to the NSFNET path.

A technically possible service still faced a purpose test

In November 1990, the commercial Internet was already visible inside the research Internet. MCI Mail, SprintMail and CompuServe had electronic-mail connections. UUNET marketed TCP/IP service. Performance Systems International operated NYSERNet and sold Internet-connected services on both coasts. Fee-based databases and information services could reach academic users.

The qualification in RFC 1192, Commercialization of the Internet: Summary Report, was institutional rather than technical. Whenever those services used the NSFNET backbone, their use still had to support the authorised purpose of NSFNET. A packet could be correctly addressed, accepted by a regional network and technically capable of crossing a router while its proposed use remained outside the ordinary terms of subsidised carriage.

The historical question can be expressed as a causal sequence:

AUP text → interpretation authority → regional-network admission or carriage terms → route availability or an alternative-network requirement → effective value of a globally unique identifier.

The sequence contains several kinds of decision. Policy language defined a formal boundary. NSF designated an office to answer interpretation questions. Regional and mid-level networks managed customer relationships and their own facilities. Backbone and commercial providers determined what carriage arrangements were available. Network operators configured routes. Registration institutions maintained globally unique numbers and contact records.

Contemporary connection manuals recognised those separations. RFC 1359, Connecting to the Internet, advised an institution to compare providers, ask what each service cost, establish what the quoted price included, inspect the provider’s acceptable-use policy and determine whether the service could reach every required part of the Internet. It then treated registration for an IP network address as another task. Installation still required a circuit, hardware, routing configuration, activation of the router and end-to-end testing.

Registration supplied a unique identifier. A provider and its interconnected networks supplied carriage. The identifier could exist before a circuit came up, remain registered while a route was unavailable and become more useful when interconnection improved.

RFC 1174 described the administrative boundary in August 1990. It observed that numbers were assigned to private networks that sought global uniqueness without intending to connect to federally supported systems. The document recommended abandoning a single registry field for “connected” status and recording network access and transit policies separately. A registry could collect and publish policy information, while network administrators decided what traffic to carry.

The AUP therefore addressed neither the technical question “Is this number globally unique?” nor the administrative question “How large an address block is justified?” Its narrower question was economically important: “May this use traverse NSFNET backbone services on the programme’s ordinary subsidised terms?”

That mattered because NSFNET had become a valuable interregional path. Its regional aggregation joined universities, laboratories, databases, instruments and other networks into a service whose usefulness exceeded the sum of its circuits. A user could possess a valid address and a local connection while lacking equivalent reach to distant networks. The title’s allocation metaphor begins after registration, where transport eligibility might change what a registered number enabled its holder to do.

Why a public backbone needed a boundary

NSFNET was a public programme intended to expand research and education networking, connect federally supported computing resources and stimulate advanced network development. It was not an ordinary commercial carrier financed entirely by customer payments.

RFC 1192 estimated annual backbone costs around 1990 at approximately $10 million. NSF supplied less than $3 million, the Michigan Strategic Fund contributed $1 million, and IBM and MCI supplied much of the remaining value through their work with Merit. At the regional level, NSF support varied while state funds, membership charges, connection fees and institutional contributions covered other costs. Campuses financed their local networks, equipment and staff separately.

Authorised regional and federal networks received the backbone as a service without a direct usage charge. That arrangement created a common interregional path instead of requiring every university or laboratory to negotiate a national collection of bilateral transit agreements. Access to remote computers, databases, instruments, colleagues and information services increased the value of belonging to a regional network.

The benefits extended beyond institutions located near large telecommunications markets. RFC 1192 noted that remote institutions could face long and expensive leased lines to a point of presence, while a regional network covering a large, sparsely populated territory carried costs that a compact metropolitan network might avoid. Public support could preserve service to institutions that a provider pursuing only the lowest cost per customer might not prioritise.

This financial design also created a legitimate boundary problem. If ordinary revenue-generating traffic used the common backbone without contributing an attributable share, public funds and private contributions intended for research infrastructure could subsidise commercial operations. Competing carriers financing their own facilities could confront a service whose most valuable national component was offered without a direct usage charge to authorised networks.

Capacity supplied another reason for caution. RFC 1192 described monthly traffic growth of 15 to 20 percent or more and recalled the congestion that preceded the 1988 T1 upgrade. Those aggregate figures identify the engineering environment rather than responsibility for congestion. They explain why programme officials treated unrestricted demand as a fiscal and operational risk.

An acceptable-use policy could therefore protect a defensible public purpose: reserve capacity for research, prevent uncompensated private appropriation and sustain the programme while commercial infrastructure developed. Commercial activity was never uniformly prohibited. Corporate research arms, fee-based information services, standards work and communications incidental to authorised research could qualify. The difficult question was how to distinguish those activities from ordinary sales, advertising and private business.

The wording changed as officials confronted that question. The three principal formulations survive with different degrees of documentary authority:

Period Documentary status Purpose and permitted use Interpreter and scope
1988 to mid-1990 Contemporaneous summary in RFC 1192; the complete original draft has not been identified here Backbone use had to support scientific research and other scholarly activity. Commercial services supporting the programme purpose could receive case-by-case consideration. RFC 1192 attributes review to NSF but does not reproduce a complete procedure, remedy system or enforcement code.
June 1990 interim policy Purpose language confirmed contemporaneously by RFC 1192; detailed text survives in a later April 1993 archival posting rather than an original dated NSF release The stated purpose was research and education in and among US academic institutions through access to unique resources and collaborative work. The detailed copy listed qualifying uses without claiming to exhaust them. The archived text gave the NSFNET Project Office authority to determine whether particular uses were consistent and stated that it governed backbone use.
February 1992 policy Directly reproduced in RFC 1359 and included in the March 1992 congressional hearing record Open research and education were covered, including open scholarly work by research arms of for-profit firms. Professional development, standards activity, grant administration and relevant product announcements could qualify. Ordinary for-profit activity, advertising and extensive private business did not qualify unless an exception applied. The NSF Division of Networking and Communications Research and Infrastructure resolved interpretation questions. The text expressly covered NSFNET backbone services and expected connecting networks to adopt their own policies.

The ledger prevents the detailed February 1992 clauses from being projected backwards onto the 1988 draft. RFC 1192 supports the earlier research-and-scholarship purpose and case-by-case treatment of some commercial services, but it does not reproduce a complete 1988 instrument. The June 1990 wording is better preserved, although the detailed copy available here was posted in 1993. The February 1992 text has the strongest direct provenance.

That last formulation also demonstrates why legal identity alone was an inadequate classifier. A university could conduct extensive private business outside the policy. A commercial research laboratory could engage in open scholarly communication within it. A database company might sell a service used for authorised academic research while offering the same service to ordinary business customers. A network equipment company could collaborate with university engineers and conduct unrelated sales from the same organisational environment.

Routers could identify networks and destinations more readily than they could infer human purpose. Treating a whole corporate network as commercial risked suppressing legitimate industrial research. Classifying it broadly as research-oriented risked admitting ordinary business traffic. Administration therefore relied on institutional declarations, case-by-case interpretation, customer classification, contractual promises and retrospective attention to reported misuse.

That compromise avoided routine inspection of message content. It also created ambiguity. An organisation could narrow a service, segregate traffic, seek informal guidance or purchase another path before a formal dispute arose. Nonqualifying traffic might pass unnoticed. Corporate users might abandon permissible collaboration out of caution. The surviving records cannot separate informal resolution, limited enforcement, incomplete preservation and pre-emptive self-restraint quantitatively.

How the federal purpose entered customer relationships

The path from a federal purpose to a customer’s connection was institutionally layered. NSF stated the programme purpose and reserved interpretation authority. Merit remained responsible under its cooperative arrangement with NSF. IBM and MCI participated in the backbone partnership. Advanced Network & Services later supplied common network facilities. Regional and mid-level networks admitted members, arranged transport, selected upstream providers and handled customer relationships. Commercial carriers supplied alternative paths. Route operators determined which announcements and traffic their networks accepted.

The February 1992 AUP acknowledged this dispersion by limiting itself to NSFNET backbone services and expecting connecting networks to formulate their own policies. A regional network could permit uses on its facilities that would not qualify for the federal backbone. When traffic needed NSFNET or another destination network, the policy of that transit or destination network re-entered the path.

The March 12, 1992 House hearing on the management of NSFNET records how one participant described the contractual layer. Michael M. Roberts, vice president for networking at EDUCOM, appeared on behalf of EDUCOM’s Networking and Telecommunications Task Force. The hearing materials described EDUCOM as a consortium of 630 universities and colleges with interests in information technology and said Roberts was responsible for its networking and telecommunications programmes.

When chairman Rick Boucher asked how nonconforming corporate traffic connected through ANS CO+RE would be charged, Roberts answered that each individual FARNET network had contractual agreements with the universities or corporations that signed up and that the relevant provisions were incorporated into enforceable agreements. He added that a party dealing directly with ANS had a fee-based arrangement. His answer is contemporary testimony about how regional and direct-provider relationships were supposed to carry policy and payment obligations. The hearing record does not reproduce the regional contracts to which he referred or name a customer against whom one of those provisions was invoked.

A more specific customer-level account appeared the following year. In an April 29, 1993 archival posting, Karen Starr described an Idaho State Library project that installed Internet nodes at the state library and two public libraries. According to Starr, the libraries joined Westnet, a regional network, and all of them had to sign an “Acceptable Use Agreement” among the paperwork required for installation.

Starr reproduced language under which a signer acknowledged having read and understood NSF’s interim backbone policy, promised a “best effort” to comply and accepted that abuse might lead to disconnection from the Internet. Her account connects named institutions, a regional network, the interim NSFNET policy and a stated consequence.

Its provenance is important. The material is Starr’s dated participant account and reproduction, not an image of an executed Westnet form or a signed customer file. It does not provide signature dates, terms of service, circuit capacity, fees or a definition of abuse. The reproduced phrase “My company” sits awkwardly with Starr’s description of public libraries, suggesting that the text may have been a standard form. It also leaves unanswered who would determine that abuse occurred, whether notice or an opportunity to cure preceded disconnection, whether the consequence concerned Westnet service or every possible Internet path, and whether any of the libraries was ever threatened with disconnection.

The attached interim policy placed interpretation authority in the NSFNET Project Office. That allocation of authority did not necessarily give NSF sole control over a customer remedy. Westnet handled the reported customer relationship, and the reproduced text does not divide the final decision among Westnet, NSF and an upstream operator. Starr’s account supports the presence of a customer-facing compliance promise and a possible disconnection consequence in Westnet installation paperwork. It supplies no instance of the consequence being used.

The same archival thread contains a CICNet policy approved by its board in December 1991. CICNet welcomed public and private organisations, recognised purposes extending beyond academic research and required outgoing traffic to comply with the policy of any transit or destination network. Its board retained authority over its own policy and service. This was a design for regional policy propagation: a broader local mission could coexist with narrower rules beyond the regional network. The document is a published policy rather than a member admission file, executed agreement or service-termination record.

The three records describe different institutional positions. Roberts spoke from EDUCOM about contractual arrangements used by FARNET networks and direct fee relationships with ANS. Starr described paperwork encountered by three Idaho libraries joining Westnet. CICNet published a rule for traffic crossing into other networks. Read together, they show how the federal purpose could enter regional terms without converting every regional network into a passive arm of NSF.

The commercial-carriage arrangement added another layer. The controlling modification to the NSF-Merit cooperative arrangement has not been located in an inspectable form here, nor has an executed ANS CO+RE agreement with a named regional network. Later analyses describe connectivity, gateway attachment and cooperative-agreement options, but their descriptions cannot substitute for signed terms when the questions concern covered traffic, prices, obligations and remedies.

The NSF Office of Inspector General’s March 1993 report supplies the strongest official assessment of the commercial arrangement. It reported that common facilities used for NSFNET service could carry commercial traffic under conditions protecting the public programme. NSF required that commercial traffic not diminish the awarded service and that after-tax profits from the commercial subsidiary enter an infrastructure pool supporting national and regional networking. The report regarded the infrastructure pool as consistent with the award conditions while criticising NSF’s failure to announce the commercial-access decision adequately.

Contemporary and later accounts also describe a principle that commercial traffic should recover an appropriate share of network cost. The available official assessment documents shared facilities, protection of the awarded service and an infrastructure-pool arrangement. It leaves the precise contractual calculation, enforcement responsibility and relationship among average cost, attributable cost and commercial profit incompletely visible.

Permission, classification and interconnection in the record

The most prominent episodes from this period occupy different parts of the institutional chain. MCI Mail concerned permission for application interoperability. Dialog became the subject of a contested commercial-service account. ANS and CIX announced a provisional interconnection plan. None can be read as though it supplies the documents missing from the others.

MCI Mail and the uncertain approval record

Retrospective accounts by Robert Kahn, Vint Cerf and the Corporation for National Research Initiatives state that CNRI sought federal permission in 1988 to interconnect the commercial MCI Mail system with the Internet as an electronic-mail experiment. They report that permission was granted and that the gateway was completed under Cerf’s direction in the summer of 1989. RFC 1192 contemporaneously confirmed by November 1990 that MCI Mail and other commercial mail systems had Internet connections.

The name of the 1988 approving body remains unresolved. The 1999 CNRI account identifies the Federal Networking Council, while the 1992 hearing chronology states that the FNC was created during 1990. RFC 1192 identifies the predecessor Federal Research Internet Coordinating Committee in a May 1989 programme plan, but it does not contain the MCI request or permission instrument.

The available record therefore supports an interagency federal approval described retrospectively by Cerf and institutional histories. It does not securely identify the 1988 decision body or reproduce the terms of approval.

The significance of the episode lies in permission-dependent interoperability. Technical feasibility alone did not settle whether a commercial mail service could bridge into the federally supported community. Approval preceded an application-layer connection that allowed mail exchange. MCI Mail users gained messaging interoperability; the record contains no corresponding change in the assignment, filtering or restoration of an IP network number.

Nor can one successful experiment sustain a broad claim about the distribution of interpretive burdens. MCI and Cerf possessed institutional knowledge and federal relationships, but the record supplies no comparison with unsuccessful applicants, review times, professional expenses or abandoned proposals.

Dialog and incompatible accounts of commercial reach

Dialog Information Services became the customer most frequently invoked in the 1991-1992 commercial-interconnection controversy. The accounts are adversarial and incomplete.

In his written submission to the House hearing, PSI president William Schrader alleged that Dialog bought commercial service from ANS, encountered restricted reach because few regional networks had signed the applicable agreements, could reach fewer than five percent of Internet users and then converted to ordinary ANS service under a research-and-education classification. Schrader was a commercial-network participant and an ANS competitor. His testimony directly records PSI’s allegation; it is not an ANS routing record or Dialog customer file.

Mitchell Kapor offered a related account for CIX. He said Dialog initially entered as an ANS commercial customer and changed classification after finding insufficient regional participation. Kapor and Schrader spoke from the same side of the interconnection controversy, so their accounts should not be treated as independent operational confirmation.

NSF director Stephen Wolff emphasised another side of the policy. Commercial information providers, including Dialog, could distribute fee-based information to research and education users. Douglas Van Houweling similarly argued that such a provider could work through a regional network. In that interpretation, charging for access to a database did not automatically make the traffic unacceptable when the service supported authorised research.

Brett Perlman’s later study, Pricing the Internet, reported reach of about 25 percent and attributed the limitation to regional networks that had not signed. That account ultimately relied on period trade reporting. Its denominator is not reconciled with Schrader’s “fewer than five percent” figure, and neither percentage can safely be converted into a measured share of the global Internet.

No Dialog contract, ANS customer notice, applicable filter list, dated routing table, regional refusal letter or correspondence confirming the reported reclassification appears in the sources used here. Dialog remains a named contemporary dispute about whether commercial classification and regional participation constrained reach. The attachment, denominator and operational result cannot be reconstructed neutrally from the adversarial accounts alone.

The prospective ANS-CIX arrangement

On June 9, 1992, EFFector Online published an ANS CO+RE and CIX announcement. Its language was prospective: the organisations announced that they “will interconnect” for a provisional period to increase connectivity among clients and members while continuing work on technical and economic terms for a permanent arrangement.

The announcement identified the intended parties and prerequisites. ANSNet was described as interconnecting with 17 networks carrying commercial and research traffic; CIX represented seven commercial networks. A mid-level network would need to sign an ANS agreement and join CIX to exchange commercial traffic with CIX members through ANSNet. ANS was not becoming a CIX member. ANS and CIX said they would forego cross-payments during the provisional period.

Those details document a negotiated plan and its intended scope. The announcement gives no activation time, router interface, advertised prefix, named participating mid-level network, traffic measurement or later implementation certificate. Its prospective wording cannot be converted into an observed route.

The absence of cross-payments between ANS and CIX likewise says little about participants’ total costs. A mid-level network still required whatever ANS agreement applied, CIX membership, circuits, equipment, staff and potentially other transit. The announcement supports neither a claim of free interconnection nor an assumption that all costs had been settled.

The June item is consequently evidence of intended provisional interconnection. It is not a record that the plan became operational on the announcement date or that it resolved the earlier Dialog dispute.

What the cases amount to

The episodes assemble a coherent institutional picture without completing the causal history implied by the title. MCI Mail shows that a commercial application could require federal permission before connecting to the research Internet. The Dialog controversy preserves competing claims that customer classification and regional agreements limited reach. ANS and CIX documented their intention to create a provisional provider-level accommodation. Roberts’s testimony, Starr’s Westnet account and the CICNet policy show several ways acceptable-use obligations could enter regional relationships.

The inspected material contains no prefix-level case joining a governing AUP version, an identified interpreter, executed carriage terms, an observed routing change and a completed remedy. It likewise supplies no complete causal chain for a named customer and no measured incidence across applicants or connected organisations. That gap keeps the claim about identifier usefulness at the level of a bounded historical inference. The documentary finding is the policy and contracting architecture; the possible identifier effect depends on how that architecture interacted with actual routes.

Enforcement, remedies and the distance between an address and reach

The February 1992 AUP named the NSF division responsible for interpretation, but interpretation was only one form of control. Monitoring and enforcement were less centralised.

At the House hearing, NorthWestNet executive Eric Hood described the system as self-policing and warned that corporate participants could become so cautious that useful collaboration was lost. NSF officials Nico Habermann and Stephen Wolff said enforcement was retrospective. They did not describe routine inspection of private-message content. Wolff said NSF acted when abuses came to its attention. Both officials accepted the chairman’s description of the arrangement as an honour system.

This approach reduced surveillance and allowed mixed research and commercial communities to communicate without continual content inspection. It also made enforcement dependent on notice, institutional judgement and how customers classified their activities. Some nonqualifying uses could escape attention; some users could narrow legitimate work out of caution. The testimony supports both possibilities without indicating their frequency.

A purpose-based rule was especially difficult to express through early routing systems. A corporate network might carry open research, technical-standards work, proprietary development, administration and sales at the same time. A route associated with the entire network could not reliably identify the intention behind each packet. A network-wide commercial classification might suppress qualifying collaboration, while a broad research classification might admit ordinary business traffic.

CAIDA’s retrospective study, The Evolution of Internet Exchanges, describes regional networks trying to use NSFNET for AUP-conforming traffic and CIX for other traffic. NSFNET had reached T3 capacity, approximately 45 Mbps, while the CIX interconnect remained T1, approximately 1.5 Mbps. Available routing technology could not classify the purpose of individual traffic and direct it over the appropriate path. Operators preferred one route, declined another or maintained hand-crafted routing approximations.

This history identifies a mechanism by which institutional classification could become routing policy. The documented differences concerned paths, capacities and routing practices rather than the disposition of address applications.

Remedies were distributed across the same institutional structure. A user could request an interpretation from the designated NSF division. That was a decision by the responsible office, not review by a separate adjudicator. A customer could negotiate with a regional network under its service relationship. Starr’s reproduction described possible disconnection but provided no notice or reconsideration procedure. A regional board could revise or reconsider its own policy. A customer might change classification, as CIX witnesses alleged Dialog did. A network could purchase commercial carriage, arrange a bilateral exchange, join CIX or select another provider. ANS and CIX could negotiate interconnection. A dissatisfied user could narrow a service or leave the market.

Those responses had different meanings. Reclassification changed the terms under which traffic was represented. Alternative transit avoided the NSFNET restriction through another commercial relationship. Provider interconnection addressed exchange between networks. Market exit could leave no formal decision or route record.

No published provision in the inspected AUPs, RFC 1359, the House record, the archived regional policies or the ANS-CIX announcement created an independent body empowered to review an NSF interpretation and order carriage. Informal reconsideration may have occurred, but the public material does not define it as a formal right.

The outcomes also differed in severity. The AUP directly addressed restriction of NSFNET backbone carriage. Traffic outside the authorised purposes could be excluded from that subsidised service, admitted after interpretation or moved to a commercial arrangement.

Failure to obtain any Internet connection would have been more severe. The named material here does not document such an outcome. Commercial mail gateways, regional commercial memberships, PSI, UUNET, CERFnet, ANS CO+RE, CIX and bilateral connections demonstrate that alternatives existed, although availability and quality varied by place and date.

Another possible outcome was access through an operationally different path. Early commercial exchange could offer lower capacity, incomplete counterpart reach, inconvenient geography or additional engineering demands. Those differences are documented at the architectural level. Their customer-specific consequences depended on contracts, circuits and routes.

Denial of a number resource belonged to a separate administrative domain. No inspected record shows NSF invoking acceptable use to refuse, reduce, revoke or reassign an IP network number. RFC 1366 discussed address management through host projections, subnet plans, conservation, delegation and aggregation. RFC 1359 separated address registration from provider policy and routing. The AUP entered after those identifier questions, when a network sought carriage over NSFNET services.

The routing inference follows from the separation. If two provider groups do not exchange a route, a registered destination behind one group is less useful to a customer behind the other. If interconnection later becomes active, the identifier can remain unchanged while its reachable counterpart set grows. The functional effect occurs in carriage and reach after registration.

Prices, alternatives and distributional uncertainty

The House hearing contains several monetary figures, but they describe different services, payers and billing periods.

Schrader’s March 12, 1992 submission marketed PSI service ranging from unlimited electronic mail for an organisation at $25 per month to a high-performance, full-service supercomputer connection costing more than $50,000 per year. The first was a recurring monthly price for organisation-wide mail. It was not described as full IP transit and included no stated circuit capacity. The second was an annual service description at the opposite end of PSI’s range. The record does not specify its line speed, local-loop charge, installation fee, equipment, CIX cost, staffing or exact destination reach.

Wolff told the same hearing that some schools could not afford the $20,000 to $30,000 required to connect to a regional network. The payer in his example was a campus, with the regional network as its immediate provider and the backbone component subsidised above it. He said NSF responded to proposals by supporting some institutions for two years. He did not identify a standard capacity or specify whether each amount represented installation, annual service, circuits or a combined package.

The ANS-CIX announcement concerned cross-payments between two network organisations during a provisional period. It did not state the costs of CIX membership, a regional network’s ANS agreement, leased circuits, routers, operations staff or downstream transit.

These amounts cannot be compared as prices for one product. Mail service, a high-performance connection, a campus regional attachment and an inter-provider settlement occupy different layers. None provides a like-for-like tariff for equivalent capacity, geographic coverage, reliability and destination reach.

RFC 1359 explains what a valid comparison would have required. A customer needed to identify circuit installation and monthly charges, onsite and offsite routers, hardware, software, maintenance, membership fees, support, training and staffing. The customer also needed to ask whether the provider could reach every required part of the Internet. A quoted connection price could conceal substantial differences in what the service included.

The historical material documents differences in service composition, capacity, topology and contractual obligations. It does not support a universal commercial price premium or show that subsidised NSFNET access was always less expensive after local and regional costs were included. Negotiation and engineering could add burdens, but their monetary incidence is unmeasured.

The same caution applies to firm size. A purpose-based policy plausibly made institutional fit valuable. Universities and nonprofit research organisations could describe much of their traffic in the language for which NSFNET was built. Corporate research arms could qualify while engaged in open scholarly work. Commercial information services could serve authorised research users. General sales, advertising and private business required a different classification or path.

An organisation with mixed traffic might need advice, internal segregation, provider negotiation and uncertainty management. A firm with more staff and capital might have been better equipped to absorb those tasks than a small provider. The sources contain no representative applicant sample, review-time series, professional bills, abandonment rate or comparison of similarly situated large and small firms. The distributional mechanism is plausible, while its scale and direction across actual applicants remain unmeasured.

Starr’s Westnet account indicates that even public libraries encountered a formal compliance promise and possible disconnection consequence in installation paperwork. It does not report substantial review expense or unequal treatment. The MCI Mail experiment records successful navigation by a prominent applicant without supplying a comparison group. PSI’s criticism identifies a competitive concern from an interested participant rather than market-wide incidence.

A balanced account must also consider that the AUP may have aided commercial network development. By declining to provide unrestricted subsidised carriage, NSFNET created demand for private backbones and exchanges. Commercial providers gained customers who needed paths outside the federal purpose rule. The policy could protect public research infrastructure while creating transition burdens for users whose activities did not fit its mission.

The central institutional issue is not whether research organisations benefited; that was the programme’s purpose. It is whether a mission boundary attached to a highly valuable route was sufficiently transparent, reviewable and avoidable for users outside the programme’s default constituency.

Two alternative institutional histories

Counterfactuals clarify the mechanism by asking which features of the historical cases would change under different institutions. They do not describe events that occurred.

A subsidised backbone with neutral carriage prices

Suppose the same NSFNET facilities, capacity and research mission had used a neutral published tariff instead of purpose-based carriage eligibility. Research and education users would receive grants, vouchers or explicit credits funded separately. Providers could buy the same backbone service at a stated price for the same capacity and service level. Research and commercial traffic could share the path without requiring a judgement about the purpose of each communication.

This arrangement would need more than a nominal price. The tariff would have to distinguish setup charges, recurring transit, local circuits, equipment, support and usage-related elements. Cost accounting would need to prevent research subsidies from leaking into commercial service without favouring the incumbent operator. Comparable providers would require equivalent interconnection opportunities. Capacity allocation and congestion management would operate under published technical rules. Subsidies would have to reach research institutions without requiring them to pre-finance unaffordable service.

Decision rights would change as well. NSF could determine the research-subsidy budget and eligibility for grants, while the carrier would no longer classify packet purposes. A tariff administrator would publish service classes and cost methods. Interconnection decisions would require objective criteria, and disputes about discriminatory treatment would need review by an institution separate from the provider being challenged.

The public record under such a regime would include the tariff, cost-allocation method, service commitments, interconnection criteria, subsidy rules, application dates, reasons for refusal and a review procedure. Customer-specific information could remain confidential while governing terms and aggregate outcomes remained visible.

MCI Mail would then present a different institutional problem. Its operator could purchase carriage under published terms instead of requesting a purpose-based exception for a commercial mail system. Technical, security and interoperability approval might still be required, but commercial identity would not determine ordinary backbone eligibility.

The Westnet libraries would still face security, operational and payment conditions. Their connection paperwork would not need to ask whether all backbone use matched an NSF purpose. Disconnection could instead follow specified events such as non-payment, network harm or breach of published technical conditions, with notice and review terms stated in advance.

Dialog is harder to predict because its contract and routes are missing. If the reported limitation arose from special commercial agreements and regional participation requirements, uniform carriage and interconnection terms might have reduced the distinction. If reach was limited by regional networks independently declining exchange, inadequate circuits or other provider choices, neutral backbone pricing alone would not have supplied the missing connections.

CIX would still have reasons to exist. Competing networks require exchange points, route diversity and bargaining alternatives even without an AUP. Its capacity, location and membership structure would continue to matter. Neutral NSFNET pricing would remove one policy-based reason for a separate path, rather than eliminate the general need for commercial interconnection.

Several burdens would also remain. Remote campuses would still need local circuits. Institutions would still purchase routers and employ staff. Congestion and capacity planning would persist. Moving the subsidy from a common backbone to grants or credits might increase regional fees, burden institutions with weak grant support or create a large administrative programme. RFC 1192 considered those risks when discussing whether subsidies should move down the distribution chain.

Historical evidence capable of distinguishing this alternative would include comparable published tariffs, invoices separating setup and recurring elements, research credits recorded independently, equivalent offers to competing providers, application and activation dates, routing tables, regional membership charges, performance records and reasoned interconnection decisions. An address issued on time but routed only after a commercial contract would still indicate a carriage constraint.

The predicted benefit is narrower than universal equality. Neutral pricing would replace a purpose judgement with a visible financial and technical rule. Its effect on access would depend on the tariff, subsidy design, competition and review arrangements.

The same AUP with abundant commercial alternatives

Now retain the February 1992 AUP on NSFNET but imagine several commercial backbones comparable to NSFNET in geographic coverage, capacity, reliability and interconnection. A commercial user could switch providers without replacing its internal network, waiting for a prohibitive circuit build or losing access to the research community. Prices would be comparable only after matching capacity, installation, recurring transit, equipment, membership, support and reachable destinations.

The reach assumption is demanding. An alternative would need to connect to the destinations the user required, rather than merely advertise Internet service. It would exchange traffic with major regional and research networks or provide neutral connections to them. Performance would include latency, capacity, reliability and operational support. Switching burdens would include renumbering risk, router changes, new circuits, contract termination and the interval between ordering service and obtaining stable end-to-end connectivity.

Decision rights in this market would remain distributed. NSF could reserve its subsidised service for research and education. Customers would choose among commercial providers. Providers would negotiate interconnection, but no single provider would control access to most important destinations. Independent exchanges and bilateral agreements would offer multiple paths, giving customers a credible exit when one carrier’s terms became unacceptable.

Relevant evidence would include provider tariffs, network maps, points of presence, interconnection agreements, route measurements, outage and performance data, installation intervals, switching charges and customer outcomes. A list of provider names would not demonstrate abundance. The question is whether a customer in a particular location at a particular date could purchase a genuinely comparable service.

Under those assumptions, NSF’s AUP would continue to allocate a subsidy but would have little leverage over general commercial reach. A company whose traffic did not qualify could purchase another path. Its identifier could remain broadly useful because equivalent networks would carry routes to the same counterparties. The AUP might change which users received subsidised service while having little bearing on general connectivity.

The early 1990s did not clearly satisfy those conditions. Commercial providers existed, and CIX offered exchange outside NSF’s purpose restriction on that path. CAIDA’s history nevertheless records differences in capacity and routing. Regional participation varied. Some routes were geographically inefficient, and early routing tools handled policy separation poorly. These are signs of non-equivalence in technical dimensions, though they do not provide a complete national comparison of price and quality.

The observable discriminator is what happens as alternatives become comparable. If commercial users can obtain equivalent reach, performance and switching terms while the AUP remains in force, the AUP’s possible influence on identifier usability should recede. If important destinations remain dependent on NSFNET despite nominal alternatives, the alternatives are not comparable in the dimensions that matter.

Address-issuance timing provides a further control. Prompt issuance followed by delayed useful routing points towards infrastructure, contracting or interconnection. Delay under address-management criteria would indicate another mechanism. The reviewed material contains no case in which the AUP caused the latter.

An effect that diminishes when comparable transport becomes abundant is best understood as a product of infrastructure scarcity and subsidy leverage. It does not become an identifier-assignment rule merely because routes determine what an identifier can reach.

From a mission-bound backbone to a commercial mesh

The practical significance of the AUP changed with the network around it.

During NSFNET’s early expansion, the common backbone aggregated regional networks and research resources into a path that was difficult to replace. The 1988 policy connected that path to scientific research and scholarship. The MCI Mail experiment showed that a commercial application could enter the community after an interagency permission process, although the original approval instrument remains unavailable.

The June 1990 policy stated the mission more specifically around research and education among US academic institutions. RFC 1192 simultaneously documented commercial providers, corporate regional membership and debate over moving subsidies away from the backbone. Policy and market structure were evolving together.

ANS and ANS CO+RE changed the implementation problem. Research traffic and commercial service could use common facilities under separate institutional and financial arrangements. This approach could preserve the research programme while permitting broader use. It also required decisions about cost allocation, regional agreements and interconnection with competing commercial networks.

The March 1992 hearing exposed disagreement over those decisions. PSI and CIX representatives argued that the arrangement favoured ANS. NSF, Merit and regional-network witnesses emphasised private investment, research service and regional or commercial alternatives. The OIG later accepted important elements of the public-purpose rationale while criticising inadequate public notice and recommending better documentation and oversight.

The February 1992 AUP was the clearest preserved statement of scope and interpretation authority. Starr’s account indicates that language tied to the interim policy still appeared in Westnet installation paperwork in 1993. Successive federal texts, regional policies and commercial agreements meant that the rule confronting a user depended on date, provider, customer and path.

The June 1992 ANS-CIX announcement recorded an intended accommodation between network groups. Its prospective language and missing activation record prevent the announcement date from becoming a routing date. It remains valuable because it shows that broader exchange among ANS clients, mid-level networks and CIX members required negotiation and institutional prerequisites.

On October 23, 1992, Public Law 102-476 authorised NSF to support research and education access to computer networks used substantially for other purposes when those additional uses increased the networks’ capability to support research and education. The amendment gave NSF greater room to work with mixed-use infrastructure. It neither converted the AUP into an address policy nor recorded the immediate removal of every carriage restriction.

By 1995, commercial backbones, exchange points and purchased research connectivity had displaced the dedicated NSFNET backbone service. That transition was not an automatic triumph of private infrastructure. Public funding had developed protocols, trained operators, financed regional networks and aggregated demand. The Merit, IBM and MCI partnership itself combined public and private resources. Research support continued after the backbone was retired.

Infrastructure power also remained. Commercial carriers could set prices, decline interconnection, prefer routes and provide uneven coverage. The transition redistributed the decisive questions. A customer increasingly asked which provider would sell transit, where that provider interconnected, what service level it offered and whether another provider supplied a credible exit.

The address system continued through this change. NSFNET’s retirement did not withdraw customer IP numbers. Identifiers survived while the dominant subsidised path and its mission rule disappeared. Address totals cannot reveal how many holders possessed limited routes, how many services changed classification or how many applicants purchased alternatives. Routing-table growth cannot disclose the purpose of traffic or the private costs of negotiation.

The historical record is strongest on institutional design. A federal policy defined eligibility for subsidised backbone use. NSF retained formal interpretation authority. Regional networks managed their own policies and incorporated acceptable-use provisions into customer relationships. Commercial carriage depended on separate financial and interconnection arrangements. Early routing technology applied policy more readily to whole networks than to the purpose of individual communications. Alternative paths differed in capacity, topology and participation.

That design supports a restrained functional inference. When the subsidised backbone supplied unusually valuable reach, a purpose restriction created a risk that a correctly registered network would reach fewer counterparties on ordinary terms. An organisation might request an interpretation, change classification, negotiate a commercial agreement or use another path. These responses could change the usefulness of connectivity while leaving the identifier unchanged.

The central evidentiary limit remains. No surviving prefix-level case in the material used here completes the path from an applicable AUP through interpretation and executed terms to observed reach and a documented remedy. The sources also do not measure how often organisations encountered such a sequence. They cannot quantify review costs, abandonment, small-provider disadvantage, a national commercial price premium or the frequency with which informal guidance resolved uncertainty.

The word “allocation” therefore survives as a constrained metaphor. The AUP allocated eligibility for a scarce, aggregated transport benefit. Its possible consequences for Internet numbers arose through provider relationships, interconnection and reach after registration.

This boundary preserves both sides of the history. NSF had a defensible reason to protect a publicly supported research service. Commercial participation could be permitted and could help finance infrastructure. Yet a mission test attached to an unusually valuable route could alter the options available to users outside the programme’s ordinary constituency.

The address ledger recorded uniqueness. Contracts and routes determined usefulness. The NSFNET AUP mattered in the distance between them.

Sources