The InterNIC Contract and the Price of a Single Administrative Choke Point

Summary

• In April 1993, NSF launched InterNIC with Network Solutions handling central non-DDN domain, network-number and ASN registration, intake, processing and updates; AT&T handling directory and database access; and General Atomics/CERFnet handling information and referral; provisional agreement reproductions also set 3-, 5-, and 22-working-day requirements for Class C, B, and A assignments after a request was complete, but the surviving evidence does not measure compliance.
• Network Solutions’ documented request process used standard templates, automated parsing, error messages, requestor verification, a seven-day confirmation window, human final processing and visible ticket status, offering credible gains in scale, consistency and correction visibility.
• The surviving agreement copies are unauthenticated reproductions, the performance reports lack decisive number-request denominators, and the documented correction, inter-manager, procurement-party and registry-appeal routes supply no recovered series of use, duration, reversal or outcome.
• The record establishes concentrated central intake and processing dependency for defined non-DDN registration work, but not universal unavoidability, systematic exclusion, quantified applicant harm or proof that regional, military, local-registry and provider routes were unavailable to a specific affected population.

1 April 1993: a request enters the new channel

A non-military registration request approaching the Internet’s central service on 1 April 1993 encountered a deliberately structured path. The change was visible first in mundane instructions: which mailbox to use, which form to complete, where to telephone, and what would happen when the form failed a machine check.

The most detailed contemporary description is RFC 1400, published in March 1993. Its author, Scott Williamson, worked for Network Solutions, the incoming registration contractor. The document is therefore a first-party account of the intended transition and request process, not an independent examination of performance. Even so, it is unusually precise about the administrative sequence.

During the transition through 31 March, non-DDN users could submit existing DDN Network Information Center templates to either the old DDN NIC address or the new InterNIC hostmaster address. Email, fax and postal submissions received at either site were to be processed. DDN users remained on a distinct military route and were told to continue sending their registration requests to the DDN NIC. The transition divided service populations; it did not close the previous institution in one stroke.

From 1 April, a new non-DDN registration request was to be sent to Network Solutions’ automated registration mailbox using the new template. An obsolete template arriving there would be returned with parsing errors and the replacement form attached. The hostmaster mailbox would continue accepting the old format through 30 June. After that date, the old format was to be returned rather than processed.

The new form was designed for automated handling and stripped of DDN-specific references. Once submitted, a mail server parsed the template and performed a quick check of information that could be verified. RFC 1400 used a domain-name conflict as its example. The system then returned either a verification form or a rejection carrying error messages.

A verification was not itself the final result. The requestor had to inspect the machine’s interpretation. If the parsed information was correct, the requestor returned the form to a separate verification mailbox. If it was wrong, the requestor corrected it before replying. A rejected request had to be adjusted and resubmitted through the automated intake. Corrected verification data was checked again, and another verification form was issued. Only after Registration Services received an acceptable verification did the request move to final processing by InterNIC staff.

That sequence combined machine and human work. The parser managed intake, formatting checks, selected verification and a confirmation loop. Registration staff retained the final processing stage. RFC 1400 does not reveal the full internal decision path for a network-number application. It does not show how staff assessed a particular topology plan, how competing technical considerations were resolved, or what explanation accompanied a substantive reduction or refusal. Nor does its domain-ticket example establish the treatment of IP-number or ASN work.

The requestor had seven days to return the verification form. If no verification arrived within that period, the original request expired and had to be submitted again. The expiry rule identifies a procedural consequence, but the document supplies no incidence rate. It does not show how often requests expired, whether messages were lost, how often a parser mistake required another cycle, or whether resubmission produced material delay. Form conformance was a condition of progress; the available evidence does not turn that condition into a measured cost to applicants.

The sequence also offered conspicuous benefits. A confirmation message included a trouble-ticket number. A user could query the request’s status with the finger utility or by connecting interactively to the registration host. The example displayed whether a ticket was pending, how it had arrived, when it opened and which internal account owned it. For an expanding service, this replaced some of the uncertainty of an untracked mailbox with a visible administrative state.

Manual support remained available by telephone, email, fax and post. WHOIS service operated at both the DDN NIC and InterNIC sites during the transition. From 1 April, the DDN server was to contain DDN information, while the InterNIC server held records concerning IP addresses, domains, ASNs and associated points of contact. Root-zone distribution, domain registration, number assignment, contact records and WHOIS access shared parts of the same operational environment, but they were not the same function.

The walkthrough reveals the exact concentration that matters. At the central non-DDN service, Network Solutions operated the accepted electronic entrance, templates, parser, rejection and correction loop, staff-processing stage, ticket-status system and registration record system. That is evidence of consecutive administrative dependency. It is not yet evidence that every applicant in the world had to use this route, that a particular class lacked alternatives, or that the contractor exercised its position to cause injury.

The launch map—and the date when it stopped being complete

InterNIC was presented to users as a coordinated service, but NSF procured it through separate organisations. The 1 April 1993 announcement of the operational InterNIC services described a launch design divided among registration services, directory and database services, and information services. A unified public identity was meant to conceal unnecessary seams without erasing the underlying allocation of responsibility.

Network Solutions received the registration component. Its central non-DDN work included domain-name registration, domain-name-server registration, network-number assignment and ASN assignment. It accepted applications, operated the processing systems, maintained registration data and passed data onward for publication and access.

AT&T received directory and database services. That function concerned the storage, retrieval and discovery of records, documents, organisations, people, servers and other Internet resources. It was related to registration because completed data had to become available to users, but AT&T did not thereby become a second route for obtaining a network-number assignment.

General Atomics, operating through CERFnet, received the information and referral component. Its function was guidance: answering questions about using and connecting to the Internet, maintaining information resources and referring users to appropriate services or support organisations. A reference desk could help a requestor find the registration route; it could not substitute for the actor authorised to process an assignment.

Other institutions remained outside this contractor trio. NSF held the awards, supplied funding, monitored performance, approved specified actions, evaluated progress and coordinated disputes among the managers. USC’s Information Sciences Institute continued the IANA coordination function, including responsibility above or alongside registries and authority to delegate parts of number administration. The DDN NIC retained its route for military users. Regional and local arrangements were already beginning to change which registry served a particular population.

The launch structure can be stated compactly:

April 1993 function	Primary actor	Operational boundary
Central non-DDN domain, network-number and ASN registration	Network Solutions	Intake, templates, parsing, correction, staff processing, assignment, updates and status
Directory, database and publication access	AT&T	Storage, retrieval, public records and document access
Information and referral	General Atomics/CERFnet	Guidance, general support, resource discovery and referral
Award administration and coordination	NSF	Funding, monitoring, approvals, evaluation and inter-manager dispute resolution
IANA coordination and delegation	USC/ISI	Coordination of identifier authority and registry responsibility
Military registration	DDN NIC	Continued DDN-user service through a separate channel

This was the April 1993 map, not a stable description of the entire period to 1998. The archival reproduction of Amendment 4, dated 13 September 1995, says that a reference to General Atomics was being deleted to reflect the recent termination of its agreement. The reproduction also contains an apparent award-number inconsistency: it associates General Atomics with NCR-9218179, while the reproduced base text assigns NCR-9218179 to AT&T and NCR-9218749 to General Atomics. The accessible material does not justify silently correcting that inconsistency or determining from it the full legal history of the terminated award. It does establish that the original three-manager description was no longer complete by September 1995.

That change affects the article’s central question. Network Solutions did not govern an unchanged three-contractor system for five years. The registration dependency must be dated independently of the wider InterNIC organisation. Its systems remained important, but the surrounding information-service arrangement, regional hierarchy and federal oversight evolved during the award period.

The initial division also prevents a misleading merger of functions. A name is not an address. An ASN is not a domain. Reverse DNS links number administration to the naming system without collapsing the two. Directory publication is not assignment authority. Information referral is not an adjudication. Root-zone coordination cannot be used as a proxy for address-allocation experience. Each function had its own evidence, applicant population and possible failure modes.

Before reading the contract, a warning about the contract

The available agreement material is useful but not authenticated as a complete executed federal record.

The Cavebear private transcription is a private transcription. The Freespeech 31-page mirror-hosted scan is a mirror-hosted scan without an authenticated chain of custody. The ICANN archival reproduction is an archival reproduction rather than a canonical NSF copy. None is a canonical NSF copy. These sources allow clauses to be compared and located; they do not establish that the accessible set contains the complete executed agreement, all incorporated proposal material, every attachment, the full signature history, a verified page set or the complete chain of amendments.

Those limits are visible on the face of the surviving set. The reproduced agreement refers to Network Solutions’ September 1992 proposal, an October 1992 collaborative supplement and quality formulas in the proposal. Those materials are not fully available within the base reproduction. General conditions appear in the mirror-hosted scan, while the ICANN page primarily reproduces special conditions and later amendments. Amendment 4 incorporates another proposal and attachments by reference. An archival reader can reconstruct important terms, but not authenticate the entire operative package.

Accordingly, what follows is a reading of provisional reproductions. “The archival reproduction states” and “the mirror-hosted scan appears to provide” are evidentiary qualifications, not stylistic caution. A later court or government account may corroborate a component, date or calculation without proving that every reproduced page is complete and genuine.

The most obvious warning is a USD 1,000,000 conflict in the estimated amount. The cover reproduction states an estimated total of USD 4,219,339. The Article 8 reproduction states USD 5,219,339. The mirror-hosted scan appears to break the Article 8 figure into USD 4,854,061 of estimated cost plus a USD 365,278 fixed fee, which arithmetically equals USD 5,219,339.

The discrepancy between the cover total and the Article 8 total is exactly USD 1,000,000. Internal arithmetic cannot settle which headline amount represented the complete executed award. It only shows that the Article 8 components add to the Article 8 total. A 1998 federal district-court description independently repeated the USD 4,854,061 reimbursement component and USD 365,278 fixed fee. That corroborates the component arithmetic; it does not authenticate the full agreement copy, its attachments or its amendment history.

Several useful ratios can be recalculated from the Article 8 figures, provided they are identified as calculations rather than contractual rates. The USD 365,278 fixed fee is approximately 7.5252% of the reproduced USD 4,854,061 estimated cost and approximately 6.9985% of the reproduced USD 5,219,339 Article 8 total. The provisional initial allotment was USD 1,162,245 through 31 March 1994, approximately 22.2680% of the Article 8 total.

The cover reproduction and Article 8 reproduction should therefore remain side by side. Choosing one silently would create false precision. The defensible statement is that the sources conflict on the total while the Article 8 components are mathematically coherent and later judicially repeated.

What the provisional award tried to control

With those limitations carried forward, the archival reproduction describes a time-limited procurement instrument effective from 1 January 1993 through 30 September 1998. It divides the period into a three-month phase-in, five years of operational support beginning 1 April 1993, and a six-month no-additional-cost flexibility period.

The reproduced funding provisions describe cost reimbursement plus a fixed fee. That structure does not establish waste, indifference or weak performance. It means only that payment was not based exclusively on a verified output such as an accurately completed assignment within a measured end-to-end period. Service discipline had to come from requirements, reports, NSF supervision, annual funding decisions, performance review and the possibility of amendment or replacement.

Network Solutions appears in the reproduction as primarily responsible for the quality, timeliness and effective management of registration services. NSF retained support planning, oversight, monitoring and evaluation. The program officer could review progress and participate in technical or scheduling discussion, but the reproduction limited that officer’s ability to order work outside scope, increase estimated cost, extend the performance period or change express terms. Material changes required the grants and contracts channel and written action.

The funding mechanism gave NSF recurring leverage. The initial allotment covered performance through 31 March 1994, with further funds contemplated incrementally. The reproduced text required notice if expected costs over the following 60 days, added to costs already incurred, would exceed 85% of the amount allotted. Future support was tied to annual review, a proposed program plan, negotiation and available funds.

Monitoring was designed as a recurring process rather than an end-of-term inspection. The archival reproduction calls for weekly transition reviews during the phase-in and afterwards until NSF directed otherwise. Joint meetings among managers were contemplated at least quarterly during the first year. Quarterly reports were to cover technical status, accomplishments, problems, collaboration, changes in plans, pending approvals and expenditures. Annual submissions were to identify goals met, exceeded or missed and explain significant departures from the plan.

The reproduced final-report requirement contemplated an account of the work and problems detailed enough for a reasonably knowledgeable organisation to replicate the work. It also contemplated delivery of software and data if requested. That language expresses an ambition for continuity, but the existence of a delivery clause is not a test of whether another operator could assume a live service with its queues, correspondence, tacit policy knowledge and staff routines intact.

Amendment 4 added another visibility mechanism. Its archival reproduction required public availability, beginning in November 1995, of registration-service performance measures, turnaround times and performance against those measures. It also called for the performance-review report to be made available for public review. This indicates that public performance evidence had become an explicit concern. It does not supply the missing figures by itself.

The award also divided coordination from applicant review. The reproduced collaboration language allowed the NSF program official to resolve technical, managerial or scheduling disputes among InterNIC managers. That could address a disagreement between Network Solutions and AT&T over a handoff, or another managerial conflict within the project. It was not expressed as a hearing for an applicant challenging a number decision.

The principal-agent design was therefore more substantial than a passive subsidy. NSF retained money, reporting, approval, evaluation and coordination tools. Network Solutions still controlled the daily systems and staff through which central requests advanced. The important question is not whether formal oversight existed; it did. It is whether the surviving evidence shows how promptly, consistently or effectively those controls protected number applicants. On that point, the public record is thin.

The three-day clock that begins after completion

The provisional agreement reproductions include distinct turnaround requirements for classful network-number assignments: 3 working days for Class C, 5 working days for Class B, and 22 working days for Class A. These are reproduced requirements, not authenticated compliance results or observations of actual service.

Their denominator is crucial. The reproduced definition starts the clock at receipt of a completed request, including any specifically requested information about network topology and the use of previously assigned address space. It ends at assignment of a number.

That excludes several stages an applicant might experience. First receipt is outside the measured period if the submission is not considered complete. Parser rejection and correction are outside it. Time awaiting the requestor’s verification is outside it. Additional exchanges about topology, existing utilisation or projected need may precede completion. Publication delay after assignment is also outside the clock, as are submissions abandoned before reaching completion.

A contractor could therefore satisfy a three-working-day Class C requirement after completion while the applicant experienced a longer interval from first submission to usable record. That possibility is not evidence that long delays occurred. The converse matters too: an extended pre-completion exchange could reflect missing or inadequate evidence supplied by the requestor rather than contractor failure.

A serious evaluation would require at least four timestamps: first receipt, first parser or staff response, recognition of completion, and final assignment or refusal. It would also need the request type and size, any additional evidence requested, the requestor’s response time, pauses for coordination, and the time when the resulting registration data became available.

The reproduction refers to quality formulas in a proposal that is not present in the accessible agreement set. Without those definitions, the researcher cannot determine whether quality was sampled across names, numbers, ASNs and record updates; whether a mean, median, percentile or threshold was used; or how rejected, corrected, expired and abandoned submissions were treated.

The three-, five- and 22-day figures remain valuable. They show that the procurement tried to make number turnaround a specified obligation. They do not reveal how many applications entered each class, how many satisfied the standard, or whether the clock was applied consistently.

May 1994: scale in units that must not be mixed

The strongest contemporary evidence of workload comes from a contractor-submitted section of the Internet Monthly Report for May 1994. It records several different kinds of activity:

• 5,009 hostmaster emails
• 251 postal or fax applications
• 2,047 telephone calls
• 1,316 domains registered
• 565 inverse-address entries
• 12,645 Class C network-number units assigned
• 20 Class B network-number units assigned
• 57 ASNs assigned

None of these units can be added to produce a meaningful total called “requests.” An email may be an application, correction, status inquiry or policy question. A telephone call may concern any of several functions. An inverse-address entry is not a new allocation. A domain registration is not an ASN assignment.

Most importantly, the 12,645 Class C units do not represent 12,645 applicants. The report separately says that blocks of 256 Class C networks were assigned to multiple organisations and the DDN NIC. One application or allocation action could therefore account for many of the reported units. The figure demonstrates the scale of address units recorded in the month, not the number of unique applicants or adjudications.

The same report lists information-service usage for May without resolving the registration denominator. It gives 211,257 WHOIS client queries and 785,015 WHOIS server queries. It separately reports 48,859 Gopher connections, 24,748 WAIS connections, 8,779 FTP connections, and 1,387 Mailserv connections. Queries, connections, applications, assignments and telephone contacts remain distinct.

These figures make a strong efficiency case for common infrastructure. A purely manual office would have faced growing pressure from mail, calls, database updates, queries and multiple identifier types. Standard forms allowed repetitive checks to be mechanised. Parsing could keep malformed fields out of the staff queue. Error messages could tell a requestor what the system rejected. Verification could catch a machine’s misunderstanding before it became a public record. Ticket status could reduce uncertainty and avoid some support contacts.

The monthly report demonstrates activity at scale. It does not provide a completion rate for number requests, a parser-rejection rate, an abandonment rate, a turnaround distribution or an independent accuracy sample. It does not reveal whether one applicant generated many emails or whether one call concerned several records. It is operations evidence, not a fairness audit.

What the 1996 contractor account adds

A privately hosted transcription of the 1996 InterNIC Registration Services Annual Progress Report supplies later performance claims. Its provenance is weaker than an authenticated contractor report, and its statistics remain self-reported. Still, it is more informative than replacing the period with hypothetical questions alone.

The transcription says that, as of 31 December 1996, 95% of new registration requests were being processed within 24 hours. It provides no numerator, no request-type population and no measurement rule. “New registration requests” is not broken into domains, IP-number work, ASNs, contacts or other templates. The claim cannot be reported as a 95% IP-number completion rate.

A separate passage concerns domains specifically. It says that more than 85,000 domains were registered in November 1996 and that more than 90% of those requests were automatically processed by the parser within 24 hours. That domain-only statement cannot be blended with the 95% year-end claim. It is evidence of substantial domain automation in one month, not a measure of address-allocation turnaround or fairness.

The same transcription records less flattering general support evidence. During 1996, self-reported call-abandon rates ranged from under 8% to over 50%. Rates were typically 10% to 20% during the latter part of the year. Average customer waits were about four minutes, and the average time to return voicemail requests was not measured. The account also says that telephone-trunk problems identified in December caused many callers to receive busy signals.

These figures offer a bounded view of service strain. They do not provide the underlying number of calls, the distribution by month or day, the length of calls, or the share attributable to number applicants. A caller seeking an IP allocation cannot be separated from someone asking about a domain, billing or a contact record. The evidence supports a general finding that telephone support experienced substantial and variable pressure in 1996. It cannot be converted into a number-allocation injury rate.

The annual transcription also illustrates why activity totals are an unstable proxy for service quality. Automation could process a high share of domains rapidly while callers faced variable access to help. A one-day processing claim could coexist with seven-day billing-email backlogs or longer handling for selected manual templates. Different functions had different queues, and an aggregate success rate could conceal the population most dependent on staff judgment.

For number work, the necessary denominator is still absent. No reproducible completion rate can be calculated without the count of completed applications, unique applicants, rejected templates, corrected requests, expirations, abandonments, substantive refusals and class- or prefix-specific compliance results. No elapsed-time distribution connects first receipt, completion, assignment and publication. The sources also provide no verified outage duration or number-specific complaint rate.

The surviving evidence thus supports two propositions at once. Network Solutions built and operated automation capable of handling substantial reported volume. The public record does not permit a reliable estimate of how often number applicants experienced delay, correction, refusal or successful review.

Where concentration actually sat

The central chain can be analysed without calling every administrative requirement a harm.

At intake, Network Solutions controlled the published electronic address, accepted template and parser. A malformed or obsolete form could not advance through the automated path until corrected. Manual channels remained, but the same registration contractor ultimately operated the central non-DDN service.

At verification, the requestor had to confirm that the machine had interpreted the submission correctly. This was a protection against silent error. It also placed completion of the loop on the requestor within seven days. The incidence of expiry is unknown, so the rule demonstrates dependency and a possible resubmission consequence rather than a measured burden across a population.

Completeness was the next boundary. A number request could require topology and prior-use information. Such evidence was not bureaucratic ornament: address conservation and routing aggregation required registries to evaluate operational need. The governance issue is that the contractor’s completeness determination controlled when the reproduced turnaround clock began. Without timestamped request records, it is impossible to tell whether that discretion was exercised promptly or inconsistently.

Substantive processing followed. Network Solutions staff handled central non-DDN network-number and ASN work within the applicable delegation and technical framework. This was more consequential than directing a user to a document. Staff processing contributed to whether a requested identifier would be assigned, in what quantity and after what supporting information.

After assignment came registration and publication. The provisional reproduction defines availability as provision of registration data to the directory and database awardee. Network Solutions initiated the handoff; AT&T occupied the publication and access layer at launch. An assignment and its public record were connected but institutionally distinct.

That distinction creates a possible fault boundary. A delay could arise in processing, transmission, publication or correction. Yet no recovered reconciliation series shows how often records differed, how quickly handoff problems were resolved or whether users could identify the responsible manager. The agreement’s inter-manager coordination provisions acknowledge the need to govern such seams, but they do not measure failure.

Review formed the final stage. The early public administrative procedure was explicit about parser correction, hostmaster contact and ticket status. It was much less explicit about independent review of substantive number decisions. Silence in RFC 1400 cannot prove that staff reconsideration, IANA contact or informal escalation never occurred. It means the document does not provide an applicant-facing appeal procedure that can be evaluated.

The demonstrated finding is therefore concentrated central intake and processing dependency. Network Solutions occupied consecutive stages in the central non-DDN route: intake, parsing, correction, completeness, staff processing, assignment and registration update. The evidence does not identify a precise applicant class for which DDN service, RIPE NCC, APNIC, a local registry, provider-assigned space or parent-registry escalation was demonstrably unavailable. It also does not supply a documented operational injury caused by that absence of choice.

This is narrower than a universal choke point, but it remains institutionally significant. Concentrating consecutive stages can make automation coherent and records consistent. It can also make the quality of the central path depend heavily on one operator’s systems, staffing, definitions and handoffs. The historical record supports analysis of that dependency without presuming abuse.

Geography and applicant class changed the route

The central service never had a uniform relationship to every Internet applicant throughout 1993–1998.

Europe already offered the clearest counterexample. The RIPE NCC’s own history distinguishes its formal establishment in April 1992 from address distribution, which was not an initial activity but was added later in 1992. By the InterNIC transition, European number administration had begun moving through a regional structure.

RFC 1466, published in May 1993, described the Internet Registry as the default registry where no delegated registration authority existed. It argued for distributing registration because geographically closer registries could better serve local communities, languages and customs. It noted that RIPE NCC had already received a block of Class C space for European allocation.

RFC 1466 did not describe interchangeable commercial routes. It said that a network applicant could contact the central Internet Registry directly and might, depending on circumstances, be referred to a regional registry. The central registry was to be prepared to serve a subscriber when necessary. At the regional level, the document favoured a single recognised registry for each geographic area. Distribution reduced central dependency but could create a non-interchangeable regional dependency.

The Asia-Pacific path emerged on a different timetable. APNIC’s institutional history says its pilot was charted to begin in September 1993 and continued until June 1994. IANA publicly recognised APNIC in April 1994 by delegating the IPv4 ranges 202/8 and 203/8. An Asia-Pacific applicant’s available route therefore depended on the date, geography, maturity of the pilot and applicable delegation.

By RFC 2050 in November 1996, the published number-administration map was explicitly hierarchical: IANA, regional registries and local registries. It identified InterNIC with North America, RIPE NCC with Europe and APNIC with the Asia-Pacific region. Local registries operated beneath the regional level.

Provider relationships created another route. RFC 2050 advised many ISPs to obtain address space from an upstream provider in support of hierarchical routing and aggregation. Certain multihomed or exchange-connected providers could request directly from a regional registry. Applicants in areas without a designated regional registry could contact a regional registry, which might handle or refer the request.

Provider-assigned space was not equivalent to a competing central registry. It could carry renumbering implications if the customer changed connectivity provider. Direct registry space also did not guarantee global routability. Choice depended on operational circumstances, not merely preference.

The DDN route remained separate for its eligible population during the 1993 transition. A military user was told to continue with the DDN NIC. That route is evidence against a single service encompassing all applicants, but it was not a general alternative open to every non-military organisation.

By 1996, the most accurate description was not that Network Solutions served as the sole global number gate. InterNIC was one regional registry within a hierarchy that also included RIPE NCC, APNIC, local registries and provider-based assignments. Earlier, the central registry remained the default for areas without a delegate and retained important coordination functions.

Regionalisation narrowed the population exposed to the central request process. It did not create a market in which every applicant could switch registries after poor service. A European operator could not necessarily choose North America’s registry as a substitute; a provider customer might have to accept upstream space; a large or exceptional request might require higher-level coordination. Alternatives were date- and population-specific, and their non-interchangeability is central to understanding both resilience and dependency.

Four remedies that should not be confused

The record contains four different kinds of correction or review. Treating them as one would exaggerate applicant protection.

The first lane was operational correction. RFC 1400 documented parser error messages, corrected verification forms, resubmission, ticket status and contact with hostmaster staff. These mechanisms addressed malformed input, obsolete forms, parser interpretation and request status. Staff might also have reconsidered a submission informally. The document does not reveal the frequency, duration or outcome of those interactions.

The second lane concerned disputes among InterNIC managers. The archival agreement reproduction states that the NSF program official could resolve technical, managerial or scheduling disputes among the collaborators. This supported coordination within the procured service. It was not an applicant appeal from a number decision.

The third lane concerned disagreements between the awardee and NSF. The mirror-hosted general conditions appear to provide that disputes of fact unresolved informally would receive a written decision from an NSF grants officer, with the awardee able to request review within 30 days of receiving that decision. Because the complete incorporated general conditions and their provenance have not been authenticated within the accessible executed award set, this remains provisional contract evidence. In any event, it was a procurement-party process available to the awardee, not a remedy granted to a network applicant.

The fourth lane was the applicant-facing registry hierarchy stated in RFC 2050. An organisation dissatisfied with the assigning registry’s performance could appeal to the parent registry. The assigning registry was to make relevant documentation available. Further appeal could move to the parent’s parent and, after other avenues were exhausted, ultimately to IANA. Each registry was expected to document its appeal procedure.

RFC 2050 improved the published architecture of review. It separated the assigning registry from a higher-level reviewer and applied specifically to address decisions. It was a Best Current Practice written by representatives associated with InterNIC, APNIC, RIPE NCC and IANA, not an audit of how appeals operated in practice.

None of the surviving material supplies a number-specific series of appeals. There is no recovered count of challenges, average duration, reversal rate, outcome distribution or evidence that similarly situated applicants received consistent review. The absence of such a series does not show that appeals were unused or ineffective; it prevents a measured conclusion either way.

The remedy map therefore changes over time and by actor. In 1993, applicants could correct forms and contact staff, while NSF managed its contractors. The provisional general conditions supplied a contractual review lane for the awardee. By November 1996, the published number-registry system articulated an applicant-facing route through parent registries to IANA. Each addressed a different dispute.

Performance evidence and continuity at the end of the chain

A strong claim of systematic number-applicant harm would require several datasets that have not been recovered together.

Request counts would have to identify completed number applications and unique applicants rather than address units, messages or calls. Turnaround records would need end-to-end timestamps, not only the interval after completion. Correction data would need rejected templates, corrected forms, expired requests and abandoned submissions. Decision data would need substantive refusals, reductions and class- or prefix-specific outcomes.

Reliability would require verified service interruptions and their durations. Accountability would require complaint counts, escalation rates, reversals and resolution times. Portability would require evidence showing how quickly another operator could assume the service, including software, data, pending queues, correspondence, policy history and staff knowledge.

The reproduced agreement contemplated reports capable of addressing some of these questions. Amendment 4 required public performance measures. The 1996 transcription refers to performance and processing data, while the surviving public presentation is incomplete for number-specific analysis. The existence of an intended reporting system is not the same as possession of its full series.

GAO’s 2016 legal opinion provides bounded retrospective evidence about the later transition. It says that, when administration moved toward the Department of Commerce in 1998, an added provision required Network Solutions to deliver copies of software, data and documentation generated under the agreement through October. GAO reported delivery of database files on ten 8-millimetre cartridges, a binder listing the databases and several additional binders containing data and software documentation.

That evidence demonstrates a transition obligation and physical delivery. It does not establish that the materials were complete enough to operate the service, that they captured active queue state or historical correspondence, or that a successor performed a tested takeover from them. Operational replaceability remains an inference to be examined, not a demonstrated failure.

GAO’s inquiry was primarily about government property and the later IANA transition. Its account is heavily concerned with domain-name and root-zone questions and supplies no number-request denominator. It cannot fill the missing record of address applications, delays or appeals.

Later domain fees, trademark litigation, registrar competition and disputes over root-zone property are similarly bounded. They may illuminate the history of the same company or award, but they do not prove that number applicants were refused, delayed or denied a remedy. Function-specific evidence remains necessary.

The missing metrics are concentrated at the points where the article’s thesis would otherwise become causal. There is no reproducible number-request completion rate, verified outage duration, number-specific complaint rate, escalation rate, reversal rate or switching time. The evidence shows an administrative dependency and incomplete observability, not an estimated welfare loss.

Counterfactual: one manager for all three services

Consider an analyst-built alternative in which NSF awarded registration, directory and database access, and information services to one integrated provider. This is a counterfactual for testing the procurement’s structure, not a claim that such a model was historically feasible or preferable in 1993.

Integration could reduce coordination costs. Registration data could move from intake to publication within one organisation. A single operations team could trace a fault across parsing, staff processing, database availability, public query and general support. Users would not have to identify whether Network Solutions or AT&T owned a problem at the handoff.

Automation investment might also become more coherent. One identity and ticketing system could connect requestors, organisations, domains, numbers, ASNs, contacts, publication records and support history. Changes to data formats would not require agreement across awards. General information staff could see the state of a registration without a separate referral.

The risk is broader failure containment. The same provider would receive requests, process assignments, publish records, answer questions and report on its own performance. A system outage or failed organisational transition could affect intake, authoritative processing, publication and guidance at once.

Monitoring would also become less modular. Separate organisations created boundaries where NSF could compare accounts of delivery and test whether data passed from registration to publication. Separation did not make either manager independent of federal funding or guarantee truthful reporting, but a fully integrated provider would eliminate even that organisational contrast.

Replacement would involve a larger bundle of systems and expertise. Switching the registration service already required software, data, queue and policy continuity. Replacing an integrated InterNIC would add directory infrastructure, public access tools, archives, information services and a broader support operation.

The counterfactual strengthens the efficiency case for NSF’s division. Separate awards limited the breadth of any one provider’s formal role and allowed specialised operations. It also clarifies why concentration remained inside the registration component: the division separated unlike services rather than creating substitutes for the same authoritative act.

Counterfactual: several replaceable registration providers

A second analyst-built alternative would allow several registration providers to serve the same applicant population while sharing an authoritative ledger and a central conflict-resolution function.

Each provider might accept a standard request, check formatting, gather supporting information and submit an approved transaction to a central service. Applicants could move to another provider when support was inaccessible. A provider outage would not necessarily halt every intake channel. NSF could compare turnaround, correction rates and applicant experience across providers.

The authoritative core would still need strict serialisation. Two providers could not assign the same number block. A shared system would have to manage available space, pending reservations, completed assignments, exceptions and reversals without conflicting states. Delayed replication could be operationally dangerous.

Consistency would pose a second challenge. Providers might interpret topology, utilisation and aggregation requirements differently. Applicants could seek the most permissive decision. Common policy, standard evidence, transaction logs and a binding reviewer would be necessary. Distribution of intake would therefore move rather than eliminate central authority.

Automation investment could fragment. Multiple providers might duplicate parsers, status systems and support infrastructure, or depend on NSF to build a common platform. Interoperability, security, liability, audit formats, reservation rules and migration procedures would need to be defined before operational experience had made the requirements obvious.

Monitoring could improve if providers generated comparable records, but only if the principal controlled the measurement rules. One provider might start its clock at first receipt while another started at completion. One might count resubmissions as new requests. Without standard denominators, apparent competition could produce incomparable performance claims.

Applicant review would also require institutional design. A customer-service choice is not a substitute for appeal when the central policy service rejects an assignment. A provider switch might help with explanation or form preparation, while the authoritative decision remained unchanged. The shared reviewer would need documentation, deadlines and authority over every provider.

Failure containment would be better at the intake layer but uncertain at the ledger. Several front doors could survive one provider’s outage. A failed central reservation or uniqueness service could stop them all. The architecture would distribute some operational risk while preserving a non-competitive core.

Historical feasibility cannot be assumed. Early 1990s routing constraints, scarce address space, incomplete policy standardisation and the cost of real-time coordination may have favoured a clear owner for each decision domain. Regional delegation addressed scale through geographic hierarchy rather than competing providers sharing one pool.

This alternative nevertheless exposes useful design questions. Could intake be replaceable without fragmenting the ledger? Could applicants switch support providers while decisions remained consistent? Could comparable logs make performance review more credible? Could the central reviewer be separated from frontline processing? The surviving InterNIC evidence does not test those possibilities, so no welfare, delay or switching-cost estimate follows.

The proportional finding

NSF’s 1993 InterNIC design combined specialisation with a common public identity. Network Solutions handled central non-DDN registration; AT&T handled directory and database access; General Atomics/CERFnet handled information and referral. NSF retained funding, monitoring, approvals, evaluation and inter-manager coordination. USC/ISI continued IANA coordination, and the DDN NIC retained the military route.

The registration system offered useful administration. Structured forms, machine parsing, explicit errors, requestor confirmation, ticket status, common infrastructure and human final processing were plausible responses to rapidly increasing volume. The May 1994 operational totals and later domain-automation claim show substantial activity, though not independent proof of number-service quality.

Within the divided institution, Network Solutions controlled consecutive stages of the central registration path. That concentration made a request dependent on one contractor’s accepted channel, parser, correction loop, completeness determination, staff processing and registration update. It is properly described as concentrated central intake and processing dependency.

The surrounding system changed. RIPE NCC distributed addresses in Europe, APNIC emerged in the Asia-Pacific, local registries and providers served additional populations, DDN users retained another channel, and RFC 2050 placed InterNIC within a regional hierarchy with a parent-registry appeal route. These alternatives were specific to time, geography and applicant class rather than interchangeable choices.

The evidence of supervision and remedy is incomplete but not absent. NSF had contractual tools; applicants had correction and staff contact; the provisional general conditions described an awardee–NSF review; and RFC 2050 later articulated applicant appeal through parent registries to IANA. What is missing is the outcome series needed to judge those mechanisms in number cases.

The title’s “price” is therefore an investigative question rather than a quantified loss. Central administration delivered scale and consistency while placing multiple stages of a defined chain at one contractor. The surviving record does not prove that this dependency was universally unavoidable, systematically abused, or responsible for measurable harm to number applicants.