When Jon Postel Said Yes: Discretion Inside the Early IANA Function

Summary

• RFC 790 records AMPRNET at network 44 and identifies Postel as the assignment contact, but no contemporaneous file preserves who requested the Class A block, who selected its size, the decision reasons or an appeal path.
• The record separates Postel’s technical and administrative roles from USC/ISI’s institutional home, DARPA’s funding and later contract, the IAB’s policy role, SRI’s registry work and the regional registries’ delegated operations.
• Repeated registers and operator compliance show operational reliance, while missing early contract terms, incomplete decision records and limited review routes leave an independently reviewable mandate unproven.

The best-preserved trace of Jon Postel saying yes to an Internet number request is not a decision file. It is a registry entry with the reasons missing.

In September 1981, RFC 790, Assigned Numbers, listed “AMPRNET,” the Amateur Radio Experiment Net, at network 44. The reference code beside it, [HM], identified Hank Magnuski. The document explained that a Class A address had a seven-bit network number and a 24-bit local field, giving network 44 a numerical space of 16,777,216 possible local-address values. It also directed anyone needing a link, socket, port, protocol or network number to contact Postel at the University of Southern California’s Information Sciences Institute. “The assignment of numbers is also handled by Jon,” it said.

Those are contemporaneous facts. They establish an outcome, a responsible contact and the architectural size of the assignment. They do not reveal who initiated the request, whether Magnuski asked specifically for a Class A network, who else discussed it, what forecasts were supplied, whether a sponsor participated, or why network 44 was selected rather than a smaller unit.

The request itself enters the public record through later accounts. The current custodian, Amateur Radio Digital Communications, says Magnuski requested address space in 1981 for licensed radio amateurs worldwide and was assigned 44/8. Brian Kantor, who later administered AMPRNET, wrote in a 2017 first-person recollection that Magnuski obtained the network through a telephone call to Postel. Kantor’s account also confused USC/ISI with the separate SRI Network Information Center, a warning against using memory to settle institutional chronology.

These are interested retrospective accounts from later administrators and the current custodian. They corroborate the community’s later account that Magnuski made a request and that a telephone call featured in it. They do not independently verify the procedure, participants or rationale in 1981. ARDC subsequently assumed formal control of the block and sold part of it in 2019, giving the organisation both custodial knowledge and a present interest in the allocation’s history.

Between the later account of a request and the contemporaneous registry entry lies an inference: some administrative action converted the proposal into a globally unique network number. RFC 790 placed Postel at the public entry point for assignments, making him the accountable registry contact. It remains unknown whether he alone selected the address class, approved the use, set any conditions or consulted others. “Postel said yes” is therefore not a quotation from the reported call and not proof of a solitary decision. It describes the recorded result under a system that told applicants to contact him.

The result mattered because other participants treated the register as a common reference. The entry made a unique number available to AMPRNET for implementation. The same number could not be assigned compatibly to an unrelated network. Software authors and administrators could use the published assignment without negotiating separately with the amateur-radio community. Practical reach arose through coordinated reliance on the register.

The allocation proved durable. Network 44 remained associated with amateur packet-radio experimentation through successive administrators and enormous changes in the value and management of IPv4 space. Durability is evidence that the registry outcome was operationally real. It is not evidence that the original decision was supported by a utilisation forecast, tested against comparable applications or open to independent review.

That distinction is the institutional problem. A register can show that a decision was made and that others relied on it while revealing almost nothing about how judgment was exercised. The stable output may be the strongest evidence of achievement and, simultaneously, the reason the missing procedure is easy to overlook.

Which office was speaking?

Postel’s roles have often been compressed into a single image: one engineer at one desk controlling the Internet’s names and numbers. The documents show a denser arrangement in which his responsibilities overlapped without becoming identical.

In 1972, Postel was at UCLA’s Network Measurement Center. He worked on host-side protocols, co-authored technical documents, edited the RFC series and helped coordinate numerical identifiers. The name “Internet Assigned Numbers Authority” was not yet the documented title of a standing organisation. The immediate setting was an ARPA research network whose hosts needed compatible socket numbers and other shared values.

The Network Information Center was separate. It operated at SRI and performed publication, directory and information services for the network. A recommendation that the NIC maintain a list did not make Postel the NIC, and work later performed at USC/ISI did not turn USC into SRI. The distinction matters because the person deciding or proposing an identifier, the institution publishing information, and the organisation holding a government contract could be different actors.

Postel did not move directly from the 1972 UCLA socket work into a continuously documented USC office. A 1992 IAB biographical compilation, RFC 1336, says that before ISI he worked at MITRE, spent several months at Keydata and then worked at SRI International. The same biography says he joined ISI in March 1976. A much later Government Accountability Office reconstruction, relying in part on information supplied by USC’s counsel, says the work that became the IANA functions moved with him in 1977. The safest conclusion is that Postel was at ISI by 1977; the surviving public accounts disagree over whether his appointment began in 1976 and when the assignment work formally moved.

USC/ISI became the institutional home of the coordination function. USC employed Postel and colleagues, housed systems and performed defence-funded work. Yet the precise terms of the early arrangements cannot be assumed. The GAO reported in 2016 that it could not obtain the 1970s-to-1990s DARPA contracts under which the IANA functions were considered to have been developed and performed. That missing record prevents a confident statement that an early contract prescribed how Postel should decide an individual network request or gave DARPA a particular case-level review right.

Postel’s RFC editorship was another role. In a recorded interview from 18 February 1988, he recalled that maintaining the RFC series “sort of fell to me” after Steve Crocker volunteered UCLA to do the work. That is evidence about how Postel remembered the editorship’s origin. It does not establish an unpaid “volunteer office,” and assigning an RFC number was not the same administrative act as assigning an IP network.

The advisory structure developed separately. RFC 1120 records that Vint Cerf established the Internet Configuration Control Board in 1979 while serving as DARPA programme manager. In late 1983, Barry Leiner reorganised it as the Internet Activities Board. The board coordinated Internet design, engineering and management, but an IAB recommendation was not interchangeable with a USC/ISI registry entry or a DARPA contracting decision.

The IANA label appears clearly in the RFC record by December 1988. RFC 1083 listed Joyce K. Reynolds as the Internet Assigned Numbers Authority contact. It separately listed Postel as the IAB’s Deputy Internet Architect and as RFC Editor. The contact address and telephone number were shared at USC/ISI, but the functions were presented in distinct blocks. Protocol standards were managed for the IAB by IANA; RFC submissions went to Postel as editor; comments about the protocol list went to him in his IAB capacity.

That page is strong evidence against the claim that every IANA action was literally one man acting alone. Reynolds had a documented operational role. The IAB supplied policy and standards direction. USC/ISI supplied an organisational setting. SRI’s NIC performed separate registry and information functions at relevant dates. Applicants supplied technical plans and made the resulting identifiers useful. Network operators and software implementers supplied the reliance that gave an assignment practical force.

Postel also worked in the Domain Name System, including root-zone and top-level-domain coordination. A root-zone provisioning request could have global operational consequences, but it was not an IP-address allocation. Protocol authorship, RFC editorship, number coordination, IAB participation and DNS-root administration must be separated before any claim about discretion can be tested.

A number for discard, with conflicts already in use

The 1981 AMPRNET entry shows an outcome without its reasons. The socket-number documents of 1972 provide the reverse: a visible method with an incomplete record of what happened after conflict appeared.

On 26 March, Vint Cerf and Postel published RFC 322, Well Known Socket Numbers. The Network Measurement Group wanted a standard socket number for a process-discard service. Before choosing one, the authors asked each host’s technical liaison to report, by note or telephone, the functions and socket numbers already in use. They planned to publish a catalogue and recommended that it be maintained at the NIC.

The request exposed the administrative problem before proposing a ruler. A network-wide standard would be useful only if it did not unknowingly conflict with existing public services. The centre lacked the necessary information. Host liaisons possessed local facts, while the measurement group had a proposed common use. Consultation was not ceremonial; it was a way to discover constraints that no central table yet contained.

Two months later, Postel published RFC 349, Proposed Standard Socket Numbers. Its status was explicit: “I propose.” He suggested a central “czar” to hand out official numbers for standard protocols and to publish numbers used for host-specific services. The proposal divided the space into four bands: network-wide standard functions, host-specific functions, future use and experiments. It also proposed socket 1 for Telnet, 3 for file transfer, 5 for remote job entry, 7 for echo and 9 for discard.

The ranges constituted a rule rather than an isolated preference. They separated common services from local and experimental uses and preserved uncommitted space. The individual assignments were intelligible in relation to recognised services and protocol work. Even so, the proposal did not specify a general test for deciding when a function was sufficiently standard, which existing use should prevail in a conflict or what evidence could overturn the coordinator’s view.

By December, the status had changed. RFC 433, Socket Number List, co-authored by Postel and Nancy Neigus of BBN, said the socket-number coordinator had established assignments for public functions. The four bands remained, and the proposed numbers for Telnet, file transfer, remote job entry, echo and discard appeared as established assignments. Other services had been added, several with links to specifications or named technical contacts.

This was a documented yes, but not one that can be reduced to Postel’s unexplained preference. The trail begins with a request for reports from host liaisons, passes through a public proposal and ends in a list co-authored with Neigus. The published record shows multiple information sources and technical precedents. It does not preserve the incoming reports, any meeting minutes or a reasoned comparison of every competing use.

The conflicts were real. RFC 433 said several hosts were performing useful public services on sockets that conflicted with the new scheme and expressed the hope that the problem could be resolved with minimal disruption. Its host-by-host table showed the incompatibilities. At SRI-ARC, socket 5 served Echo and socket 7 CPYNET, while the common assignments gave 5 to remote job entry and 7 to Echo. UCSB used 5 for a non-standard remote-job service. NASA Ames listed what it called “Sorry Remote Job Entry” on 5, Echo on 7 and Discard on 9.

The RFC preserves the collision between standardisation and installed practice; it does not document its resolution. It contains no order directing a host to move, no follow-up account showing which service changed, and no finding that the transition succeeded. Any claim that Postel confronted the hosts and settled these conflicts would go beyond the surviving evidence.

What it does show is a correction channel. Anyone who knew of an omitted function or an entry that should be corrected or deleted was invited to contact Postel or Neigus. That made the publication corrigible and placed two named people at the receiving end of objections. It was not an independent appeal. The people associated with producing the list also received requests to change it, and the RFC did not promise a written disposition.

For an applicant, the difference between a local and an official socket was consequential. A host-specific service could continue to exist, but an official common assignment supplied a network-wide expectation. Implementers could design clients around the standard value. Conversely, a local service occupying that value faced pressure from the common convention even without a coercive order. The register changed the coordination environment in which software choices were made.

The episode therefore supports a narrower and more defensible account of Postel’s discretion. He was publicly identified as the coordinator; his proposal supplied the organising scheme; the final document said assignments had been established. Yet host liaisons, Cerf, Neigus, protocol authors and local operators contributed information or implementation. The decision was person-centred without being information-poor or wholly solitary.

Its operational achievement is visible in the recurring published register, the linkage of identifiers to functions and specifications, the identification of responsible contacts, and the invitation to correct errors. The documents were designed to reduce incompatible use of shared identifiers. They do not provide a measured collision rate, response times, refusal statistics or evidence that every conflict was resolved. Administrative performance should be credited where it is documented, not inflated to fill the gaps.

The blank space behind network 44

Nine years later, RFC 790 made Postel’s role more explicit while leaving the substance of the AMPRNET choice less visible.

The architecture imposed hard limits. An IP address contained 32 bits. Under the classful scheme, Class A left 24 bits for the local part, Class B left 16 and Class C left eight. An assignment had to be globally unique, and the chosen class determined the size of the local numerical space. The table recorded which network numbers were assigned and which remained open.

What RFC 790 lacked was an allocation test. It did not require an applicant to submit host counts, a deployment schedule, a utilisation target or evidence that a smaller class would be insufficient. It did not describe conditions for reclamation or an appeal from a refusal. Those omissions do not prove that Postel or his colleagues asked no questions. They show that the public document does not tell applicants or later readers what questions governed the choice.

AMPRNET fitted the experimental character of many entries in the table. Packet radio was part of the environment in which internetworking techniques were being developed. A globally unique network number could allow geographically dispersed amateurs to design systems without choosing identifiers that conflicted with another Internet network. The designation “Amateur Radio Experiment Net” and the Magnuski contact support the conclusion that the registered purpose was technical experimentation.

They do not establish how the address class was chosen. Perhaps Magnuski requested a Class A network because the project was conceived as worldwide. Perhaps Postel or another participant selected it because the architecture offered a convenient unit and large portions of the table were still unassigned. Perhaps the choice followed an informal precedent applied to experimental networks. No contemporaneous decision record located here distinguishes among those possibilities.

The unassigned range in RFC 790 is relevant evidence, but it must be used carefully. The table shows that many Class A numbers remained unassigned after 44. That condition could have reduced the perceived opportunity cost of a large allocation. It does not quantify request volume, prove that addresses had no value, or show that administrators expected all unused space to remain plentiful. Nor does it establish that a Class A assignment was routine for every credible experiment.

Later statements by ARDC that large blocks were easy to obtain because demand was limited are institutional recollections, not measurements of the 1981 queue. The registry inventory supports the narrower observation that the recorded free pool was extensive. It cannot supply a denominator of applications, a rejection rate or the coordinator’s subjective valuation.

Modern scarcity must not be projected backwards. The later sale of part of 44/8 demonstrates that the assignment eventually acquired a financial dimension unimaginable from the registry entry alone. It does not show that Postel transferred a known modern asset or that Magnuski represented such a future value in 1981. An administrator should be judged against reasonably knowable conditions, not hindsight converted into motive.

A policy framework became more visible after the assignment. RFC 820, published in January 1983, said that number assignments handled by Postel were subject to an agreement between DARPA’s Information Processing Techniques Office and the Defense Data Network Program Management Office. Its appendix summarised a September 1982 meeting and recommended divisions among research, defence and commercial uses. AMPRNET appeared in the research category.

That document cannot supply the missing 1981 rationale. The meeting post-dated the AMPRNET entry. Its allocations were repeatedly described as recommended policy, and the appendix concluded that the policy had not yet been fully implemented; Postel was still acting as coordinator for all assignments. It is evidence that sponsor and programme institutions were developing prospective constraints, not proof that one of those rules governed network 44.

The contract record adds material support but not the missing decision rule. The GAO’s later reconstruction supports that USC performed IANA-related work under defence-funded arrangements. Because the early contract texts were unavailable, it does not establish which staffing obligations, approval rights or substantive criteria applied to the 1981 request. The fact of funding rebuts the idea that the registry was merely a private possession. It does not answer who authorised the particular size of 44/8.

The applicant consequence is clearer. A registered Class A number gave AMPRNET a durable, globally unique basis for experimentation. The wider consequence was exclusion: the number could not be assigned compatibly to a different network. The public correction or review path is not clear. RFC 790 told applicants to contact Postel for assignments and current information, but it did not name a separate reviewer or specify how to challenge his office.

Network 44 therefore remains a useful form of evidence for examining informal authority: a result too concrete to dismiss and a rationale too incomplete to reconstruct. The shared register carried the identifier forward for decades. The archive does not show whether the same judgment would have been applied to a similar outsider, what happened to unsuccessful applicants, or how an error could have been reversed before reliance made reversal costly.

When one yes had to become many

The move toward regional registries supplies an implementation trace rather than a complete decision file. It also shows why the later phase cannot be narrated as Postel personally granting every request.

In August 1990, RFC 1174 presented the official recommendation of the Internet Activities Board to the Federal Networking Council. It described USC/ISI as performing the central IANA function and SRI International’s DDN-NIC as performing the Internet Registry function for network and autonomous-system numbers. It said IANA had discretionary authority to delegate parts of its responsibility and that rapid growth, internationalisation and increasing scarcity made further distribution timely.

The proposed method retained central IANA and Internet Registry functions. The Internet Registry would allocate blocks to organisations approved by the Coordinating Committee for Intercontinental Research Networking and delegate further assignment authority. It would remain the default where no delegate existed, maintain aggregate databases and redistribute copies. Candidate registries were to meet with IANA and the Internet Registry to review procedures and produce documentation.

This was an institutional recommendation, not an individual recognition decision. The sender was the IAB chairman; the recipient was the FNC chairman; the contemplated actors included IANA, SRI’s registry, CCIRN and candidate organisations. The text made discretion explicit but placed it in a chain.

By October 1992, RFC 1366 supplied selection criteria. It reported general support from the Federal Engineering Planning Group on behalf of the FNC, the co-chairs of the International Engineering Planning Group and Réseaux IP Européens. A regional registry should be unbiased, widely recognised by providers and subscribers, legitimised by networking authorities in its area, established beyond the registry function, adequately resourced, committed to common allocation guidelines and willing to coordinate sub-allocation strategy with the central registry.

These criteria left judgment open. “Widely recognised,” “legitimacy” and “appropriate resources” were evaluative terms, not mechanical tests. A candidate could satisfy the words in different ways. The documents did not prescribe a vote threshold, comparative tender or independent appeal from rejection. But they gave candidates and reviewers a public framework that the 1981 AMPRNET record lacked.

An implementation outcome appears in RFC 1467, Status of CIDR Deployment in the Internet. Published in August 1993, the informational status report said that, under a milestone dated 31 October 1992, regional-registry selection criteria had been put in place and IANA was accepting requests from prospective registries. It reported that the RIPE Network Coordination Centre requested regional-registry status and was given the range 194.0.0.0 through 195.255.255.255 to administer for the European Internet community. RIPE NCC had independently received 193.0.0.0 through 193.255.255.255 earlier and could manage that space under the same guidelines.

This is evidence of a request, a resource transfer and an operational consequence. It is not a signed recognition decision. RFC 1467 was written at the Corporation for National Research Initiatives as a report on CIDR deployment; it was not RIPE NCC’s application, an IANA decision letter or a transcript of deliberation. It does not name the person who evaluated the request, reproduce the submission, identify alternative candidates or provide a reasoned disposition.

The report nevertheless moves the record beyond policy recommendation. It says network service providers received blocks from RIPE NCC or from the Internet Registry acting for North America and the Pacific Rim. It describes the registration function moving closer to end users and states that the process had operated as hoped without major problems. That performance judgment came from an institution-produced status assessment drawing on agencies, operators, vendors and technical groups. It did not include applicant satisfaction data, error rates or rejected registry requests.

Postel’s leadership at USC/ISI makes his involvement in the central function plausible, but the evidence supports an IANA action inside a multi-institutional programme, not a solitary telephone yes. The applicant was RIPE NCC, not a downstream network seeking addresses. The delegated resource was a block for regional administration. The practical effect was to move first-instance assignment work closer to European applicants while the central system retained authority over the aggregate space.

The published regional-registry criteria concerned service capacity, acceptance by network authorities, coordination, neutrality and adherence to central guidelines. A weak delegate could misallocate blocks, lose registration data or apply inconsistent standards across a continent. The choice therefore concerned the identity of a continuing administrator, not merely the size of one applicant’s assignment.

RIPE NCC’s own institutional history says European operators began coordinating through RIPE in 1989, decided in 1990 to fund a staffed coordination centre and formally established RIPE NCC in April 1992. It says address distribution was added to the centre’s work later in 1992. These details help explain how RIPE NCC could present itself as regionally grounded and operationally prepared. Because the account is produced by the recipient institution, it is evidence of its history and self-understanding, not an independent audit of IANA’s action.

The correction path remained incomplete at the delegation level. RFC 1366 allowed network subscribers to contact the central Internet Registry directly, which could serve applicants where necessary. That provided a fallback for service delivery. It did not specify how a rejected regional-registry candidate could appeal, how a regional delegate could be removed, or what record would be transferred if it failed.

By May 1993, RFC 1466 restated the regional criteria and said the distributed registry was empowered by IANA and the Internet Registry. By November 1996, RFC 2050 described a hierarchy of IANA, regional registries and local registries. InterNIC served North America, RIPE NCC served Europe and APNIC served the Asia-Pacific region. Regional registries were established under IANA authority and required regional Internet-community consensus.

RFC 2050 also made individual address decisions more reviewable. It called for documented network plans, set utilisation guidelines and distinguished conservation, routability and registration. It permitted a registry to request published material verifying that an organisation was what it claimed to be; it did not require such organisational evidence in every case. Applicants dissatisfied with an assigning registry had a right to appeal to its parent, with relevant documentation made available. After other avenues were exhausted, an appeal could reach IANA for a final decision.

The document proves that an appeal policy existed. It does not prove that an applicant successfully used it, how often appeals occurred or whether IANA’s final review was independent of the policy applied below. Nor does the written hierarchy itself demonstrate that service continued smoothly after any named coordinator departed. It makes organisational continuity more plausible by distributing records and roles, but that is an institutional inference rather than a measured outcome in the cited material.

The RIPE NCC episode changes the portrait without yielding a personal decision file. At the start of the period, an official socket could emerge from consultation followed by a named coordinator’s published list. By 1981, a network assignment could be visible mainly as an entry beside an applicant’s initials. By the regional-registry milestone reported for 1992, a candidate organisation was operating within public qualifications, multiple policy bodies and an explicit division of responsibility. Discretion remained, but it was no longer contained at one desk.

What reputation could do, and what the record cannot prove

Why was a person-centred arrangement tolerated long enough to become infrastructure?

The most defensible answer begins with observable service. Assigned-number RFCs appeared repeatedly. They identified responsible contacts, distinguished assigned from unassigned values and connected many entries to technical documents. RFC 433 exposed conflicting socket uses rather than hiding them. RFC 790 told applicants where to seek an assignment. RFC 1083 separated IANA, IAB, RFC Editor and NIC contacts. These are documented administrative outputs.

A named coordinator reduced uncertainty about where a request should go. That arrangement may also have reduced coordination steps, but the public RFC inventory does not establish response times. It may have worked particularly well when applicants and administrators shared technical networks and professional norms, but the archive does not provide a complete applicant population against which to measure access. The plausible explanation should not be promoted into a finding merely because it fits the surviving success stories.

Postel’s expertise nevertheless mattered. He worked on the protocols whose identifiers he catalogued. RFC editorship exposed him to proposals and specifications. His advisory work connected him to technical debates. This combination could help a coordinator recognise duplication, understand why a parameter was needed and identify the people responsible for a protocol. Expertise was a governance input because assignment was not clerical copying: RFC 2050 later acknowledged that conservation, routability and registration could conflict and required judgment in individual cases.

Expertise did not settle mandate. A protocol author could know what an identifier meant without being authorised by every affected operator to allocate it. An applicant could accept Postel’s help without representing future entrants. An IAB member could influence policy without becoming the contracting principal. A DARPA-funded institution could deliver reliable work without its procurement relationship amounting to international consent.

Postel’s 1988 interview helps explain the culture, within limits. He remembered early host-protocol work as being performed largely by graduate students who expected established professionals to arrive and replace their designs. They did not, and the host-side protocols created by the working group endured. He also described the RFC series as an informal conversation that became more formal as its audience expanded.

His observation about missing reasons is especially relevant, although it concerned technical discussion rather than IANA appeals. Postel said ideas rejected or set aside often returned because no document preserved the full earlier argument. The analogy is strong: a community can remember that a question was settled while losing the reasoning needed by a newcomer or successor. It would be wrong, however, to treat that interview as evidence that number applicants were denied appeals or that assignment refusals followed the same pattern.

Reputation evidence comes later still. Vint Cerf’s October 1998 memorial, RFC 2468, remembered Postel as a mediator, careful decision-maker and steadfast provider of service. It linked his identity closely to IANA and described the loyalty he inspired among colleagues. The document is compelling evidence of how an influential peer understood Postel immediately after his death.

It is not an administrative audit. A memorial does not supply a denominator of requests, test consistency across applicants or uncover cases in which personal familiarity affected access. Its language is evidence that trust existed, not independent proof that every decision deserved it.

The strongest benign explanation for the arrangement is therefore bounded. The work produced visible and repeatedly updated registers. Applicants had named contacts. Technical context was concentrated in people able to interpret requests. Much of the recorded space in 1981 remained unassigned. Peer consultation and sponsor institutions existed even when their role in individual decisions was not published. Under those conditions, a compact arrangement could appear proportionate to the problem.

The weakness is equally bounded. The surviving record overrepresents published, durable and celebrated outcomes. Routine refusals, abandoned requests, informal corrections and unrecorded dissatisfaction were less likely to become RFCs. No conclusion about universal fairness can be drawn from their absence. Good service can earn reliance while leaving consistency, review and equal access unmeasured.

When reliance outran the desk

By the early 1990s, the documents stopped treating scale as an abstract future concern.

RFC 1174 cited rapid escalation in the number of networks and the Internet’s internationalisation. RFC 1366 said demand had grown significantly over two years and required a more systematic allocation process. RFC 1467 supplied a more concrete operational picture: in 1993 the NSFNET/ANSNET policy-routing database contained more than 13,000 networks and was growing by about eight per cent a month, although not all entries represented active networks and the database did not cover the whole Internet.

These figures do not measure IANA request volume, but they do document a rapidly expanding operating environment. More networks produced more registration work, larger routing tables and greater consequences from poorly aggregated assignments. The problem was no longer only whether two experimenters might select the same number. An allocation could affect the global routing system’s ability to carry many other networks efficiently.

RFC 2050 made the trade-offs explicit. Conservation sought fair distribution and resistance to stockpiling. Routability favoured hierarchical allocations that routers could aggregate. Registration required a public record for uniqueness and troubleshooting. The document acknowledged that these goals could conflict with each other and with the interests of an applicant or provider.

The applicant now faced a documented evidentiary burden. Registries examined network topology, subnetting, routing plans, previous assignments and projected utilisation. A registry could request deployment plans and organisational verification. Direct allocation did not guarantee that providers would route the resulting prefix. Saying yes to address space and saying yes to global reachability had become distinct decisions made by different actors.

Review became more valuable because an allocation could impose costs beyond the applicant. A generous block consumed finite space. A poorly aggregated set of prefixes enlarged routing tables. A denial or forced provider-based assignment could impose renumbering and dependency costs on a network. The administrator made neither the applicant’s investment nor every operator’s routing investment, yet its judgment influenced both.

The material authority around USC/ISI also became more specific in the surviving contract record. The GAO obtained information about Task 4 of DARPA’s final Tera-node Network Technology contract with USC, in force from July 1995 to July 1999. Task 4 required network-infrastructure activities that included acting as the Internet Assigned Numbers Authority. USC was to supply the personnel, materials and facilities needed for the work.

That contract demonstrates that, by 1995, performance of the IANA functions was an institutional deliverable rather than a personal pastime. It identifies a government purchaser, a university contractor and a defined period of performance. It cannot be projected backwards to specify the rules for AMPRNET in 1981, and a federal procurement contract did not by itself confer political authorisation from every international network using the resulting registries.

The difference between operational authority and broadly accepted mandate became unusually visible in the DNS root in January 1998. This was not an IP-number assignment, and it should not be used to imply that earlier number allocations were covert exercises of political control. It was a separate function in which Postel’s accumulated reputation met distributed infrastructure.

A contemporary Washington Post report said Postel asked operators of six of the twelve secondary root servers to obtain root-zone information from an ISI server rather than the established Network Solutions source. Postel stated that the change would make no alteration to the data and that the servers would return to the previous arrangement when the test ended. The report found no apparent disruption to users.

The operators’ compliance revealed the force of personal reliance. Gerry Sneeringer of the University of Maryland explained: “If Jon asks us to point somewhere else, we’ll do it. He is the authority here.” A University of Tokyo operator also changed his server after receiving Postel’s message.

The limits are just as important. The reported scope was six operators, not the entire root-server system. The data were said to be unchanged. No apparent user outage was reported. Federal officials instructed Postel to restore the prior arrangement, and the operators reverted.

A later technical history, RSSAC023v2 from the Root Server System Advisory Committee, describes Postel asking Jim Koda and Paul Vixie to create an ISI primary server as a test, inviting several operators to use it, and requesting a return to the old primary a few days later. That institution-produced retrospective supports the test account. The timing, during a dispute about future governance, led contemporaneous officials to suspect a broader purpose. The surviving evidence does not conclusively resolve motive.

The episode demonstrates neither a seizure of the Internet nor a harmless routine change beyond debate. It demonstrates that a request from a trusted coordinator could alter distributed infrastructure before the institutions around it agreed on the request’s meaning. Personal credibility supplied operational capacity; government authority supplied a capacity to reverse the action. The disagreement exposed that those sources of authority were not identical.

For number administration, the lesson is indirect but important. A common register works because others rely on it. The same reliance that makes a technically correct assignment effective can magnify an ambiguous instruction. As reach and value grow, governance must distinguish the administrator’s ability to produce compliance from the administrator’s authority to decide the underlying policy.

If one more person had signed

A historically plausible alternative to the early arrangement was not a modern regulator with hundreds of staff. It was a small review group for decisions above a defined threshold.

The 1972 socket record already contained the necessary participants. Cerf and Postel solicited host reports. Neigus co-authored the final list. Host liaisons supplied local information. Protocol authors provided specifications. A rule could have required two coordinators to approve a new network-wide socket, a large network assignment or a decision that displaced an existing use.

The group might have recorded the request, the governing text, the known conflicts and a short reason. Routine entries could still have been delegated to one coordinator. Reconsideration could have gone to members who did not make the first decision. A network such as AMPRNET could have left behind a page explaining whether the applicant requested a Class A network, what alternatives were considered and why the chosen unit fitted the experiment.

The costs can only be stated as ex ante risks. Waiting for another reviewer might have delayed an answer, but no response-time series shows how much. A panel drawn from the same professional circle might have reproduced the same assumptions. A requirement to publish reasons could have exposed defence, commercial or security-sensitive plans unless confidential submissions were separated from public findings. Formal thresholds might also have encouraged argument over whether a case was “consequential” enough for review.

A small group could become a gatekeeping club. Applicants outside the established research community might find it harder, not easier, to persuade several insiders. Consensus rules could produce stalemate while software developers selected unofficial values. A second signature would reduce dependence on one memory without automatically broadening representation.

The benefit would have been a different kind of evidence. RFC 433 shows that consultation and co-authorship were feasible. What it lacks is a disposition of the conflicts it recorded. A panel minute or short decision could have shown which use prevailed, whose operations changed and what principle would govern the next comparable case. That would improve reviewability even if the substantive answer remained identical.

The question is not whether a panel would certainly have been faster, fairer or wiser. None of those outcomes can be demonstrated retrospectively. Its distinctive contribution would have been plural responsibility and a preserved reason before reliance hardened an entry into infrastructure.

For network 44, such a record would now guard against two opposite errors. It would stop critics from assuming that Postel casually bestowed a future fortune. It would also stop admirers from treating the allocation’s durability as proof that the original sizing decision was fully reasoned. Recorded reasons protect legitimate discretion from mythology as much as they expose weak judgment.

Delegation was not theoretical

Regional delegation was more than a counterfactual. During the 1990s, it became operating policy.

RFC 1174 supplied the institutional outline. RFCs 1366 and 1466 supplied qualifications and address-management rules. RFC 1467’s report of RIPE NCC’s request and address-space delegation supplied implementation evidence. RFC 2050 later described allocation standards, documentation, auditing and appeal through a hierarchy. This was not a clean replacement of judgment by rules. It redistributed where judgment occurred and made some of its constraints visible.

Regional processing offered a plausible response to language, time-zone and local-network knowledge. Staff closer to applicants could understand regional topology and provider arrangements. A central registry could allocate aggregate blocks, preserve global uniqueness and remain available where no regional service existed. Parent registries could review lower-level decisions without processing every ordinary application themselves.

The design introduced new risks. Regional criteria could diverge. A registry could become dependent on incumbent providers or local political interests. Applicants in different regions could receive different service. A central appeal body might lack local context, while a regional body might resist central correction. Delegation also required reliable databases, defined service areas and a practical way to move records if a delegate failed.

The 1992 criteria addressed some of these risks through regional legitimacy, neutrality, resources and coordination commitments. They did not specify a mature removal mechanism. RFC 2050 created a right to appeal address decisions, but IANA remained the final authority after other avenues were exhausted. Distribution reduced first-instance dependence on Postel without eliminating discretion at the centre.

Replaceability was an intended institutional advantage, not an automatically proven result. A staffed registry, documented policy and replicated data make it easier in principle to survive the departure of an individual. Continuity still depends on access to records, legal authority, technical systems and the cooperation of operators. If these remain tied to one institution or personality, delegation can merely move the dependency.

The RIPE NCC implementation episode reveals a more realistic distribution of authority than the image of Postel granting permission to Europe. European operators organised RIPE and created a coordination centre. The IAB and federal networking bodies promoted distribution. IANA and the Internet Registry retained central authority. Published criteria framed the arrangement. The delegate then administered address space for regional applicants. No one actor alone supplied the mandate or the operating capacity.

This architecture also clarified the difference between participation and authorisation. Regional engineers could contribute expertise and establish local acceptance. Central coordinators could maintain global uniqueness. Sponsors could fund systems. Applicants could submit plans. None of those roles, considered alone, answered every question about who was entitled to set policy. The hierarchy worked by combining them and specifying at least some routes for review.

Delegation did not make reputation irrelevant. Organisations still depended on trusted staff, competent managers and credible technical communities. It changed reputation’s institutional position. Personal confidence no longer had to carry the entire weight of intake, allocation, recordkeeping and correction. It could operate within published criteria, multiple organisations and an appeal chain.

What the surviving record can bear

Placed beside one another, the socket register, network 44 and the RIPE NCC implementation report do not support a single verdict about personal rule. They show a change in what the administrative record could explain.

RFC 433 documents solicitation, a public proposal, an established list, named co-authorship, visible conflicts and a correction mailbox. It does not show whether the incompatible host practices were resolved or whether affected operators obtained independent review. Network 44 entered a shared register under a system that directed applicants to Postel, but its class-selection rationale, other participants and appeal path remain unknown. RFC 1467 later reported that a prospective regional registry requested status, received defined address space and participated in a distributed assignment process under published criteria; it did not preserve a signed recognition decision or a personal rationale from Postel.

The funding record has the same bounded quality. US defence support and USC/ISI’s institutional role are documented. By 1995, TNT Task 4 specifically required performance of the IANA functions and required USC to provide the necessary personnel, materials and facilities. The unavailable earlier contracts prevent that evidence from being projected onto 1972 or 1981 as proof of exact duties, supervision rights or case-level DARPA intervention.

Where Postel’s judgment is directly visible—as in the socket-number proposal—its effect depended on publication, technical uptake and the choices of other operators. Network 44 shows the same administrative system producing a durable outcome without revealing whose judgment selected its size. Across the record, expertise, institutional support, published policy and operational reliance explain the reach of the function without supplying a complete causal chain for every assignment.

Postel helped deliver a coordination service whose registers became indispensable references. The achievement is visible; universal consistency, independent mandate and easy replaceability are not. The later turn toward documented criteria, divided responsibilities and appeal routes did not abolish discretion. It made discretion less dependent on what one trusted contact remembered, decided and left unexplained.