Summary
- A registration-system design published in 1993 distinguished malformed submissions, correction and resubmission, verification, seven-day expiry, pending tickets, and final staff action. These were different administrative states, not interchangeable forms of substantive refusal.
- Public reports expose successful assignments, registry growth, staffing, queues, and budgets. They do not provide a comparable annual population of unique number-resource requests and final dispositions. The 16,871 network numbers assigned in Europe during 1993 were resources, not applications, applicants, approvals, or decisions.
- The public material examined here contains zero complete, comparable chains connecting an initial number request to a final substantive refusal and then to reconsideration or reversal. That finding applies to this bounded body of evidence; it does not establish that no such case occurred or that no relevant private record survives.
- Confidential engineering plans gave registries a legitimate reason to protect applicant files. That privacy interest limits the case for publishing raw submissions, but it would still permit de-identified reporting of request types, corrections, expiries, withdrawals, partial grants, substitutions, refusals, and review outcomes.
- Without a common request population, historical rejection rates, consistency estimates, disparate-treatment findings, workload effects, and remedy-effect estimates remain unmeasurable. The evidentiary gap proves neither fair treatment nor unfair treatment, and it reveals no motive for the absence of comparable public totals.
The first rejection might not have been a decision
In March 1993, a proposed registration system placed a machine-readable form at the beginning of an administrative sequence. An incoming message could fail automated checks and return to its sender with an error explanation. The sender could correct the message and submit it again. A submission that passed that stage could trigger a verification request. If the recipient did not return the verification within seven days, it would expire. A trouble-ticket number could reveal that an item remained pending before staff completed the final action.
That sequence appears in RFC 1400, Transition and Modernization of the Internet Registration Service, which scheduled non-DDN registration to move to InterNIC on 1 April 1993. Its account is important because it shows how many distinct events can hide beneath the ordinary word rejection. A malformed message returned by a parser was different from an application rejected on its merits. A request sent back for correction was different from one abandoned by its sender. An unanswered verification that expired after seven days was different from a verified ticket still awaiting staff attention. None of those states was equivalent to a final judgment that the requested number resource was unwarranted.
Even a completed judgment need not have produced a simple approval or refusal. The address-management guidance published in May 1993 allowed an applicant judged not to warrant a Class B number to receive a Class C block instead. That result delivered a resource while withholding the resource originally requested. It could reasonably be described as a substitution, a reduction, a partial grant, or an adverse decision on the original demand. Calling it simply approved would erase the reduction. Calling it simply rejected would erase what the applicant received.
The same difficulty follows a request beyond its first disposition. An applicant might correct a defective message, allow verification to expire, withdraw while waiting, accept a smaller resource, seek reconsideration, or return later under changed facts. An administrator might affirm an earlier result, modify it, provide a substitute, or reverse it. A later assignment could represent a changed decision, but it could also arise from a new request, a new rule, different supporting information, or another administering registry.
A historical rejection count therefore has to begin with classification, not arithmetic. It must distinguish a communication failure from an adverse merits judgment and an adverse judgment from a later remedy. It must also preserve the connection between the thing requested and the thing granted.
RFC 1400 establishes that the proposed machinery contemplated error messages, correction, verification, expiry, ticket status, and final staff action. The worked ticket illustration concerned domain registration, and the document did not release a population of Internet number requests. It consequently shows what records the proposed system could have generated, not what complete number-request records were actually created, retained, or made public.
That boundary is decisive. Designed capability is evidence about institutional architecture. It is not evidence that every contemplated field existed in operation, remained connected to the same request, survived later changes, or entered a public statistical series.
Before counting refusals, define the request
The most intuitive denominator would be all applications. Yet even that phrase is unstable unless the record explains when an attempted contact became an application and how repeated submissions were joined.
One organization might send an incomplete message, receive an error, submit a correction, obtain a ticket, provide further information, and accept a smaller resource. Depending on the administrative record, the sequence might generate two or more messages, one parser rejection, one corrected submission, one verified ticket, several staff exchanges, one quantity reduction, and one assignment. If the organization returned months later, a second ticket might refer to the same underlying need or to a genuinely new demand.
Each unit answers a different question. An applicant is the person or organization seeking a resource. An application is the substantive demand presented for consideration. A message is a communication and may duplicate, correct, or supplement another message. A ticket is a tracked work item whose scope depends on the rules of the system. A network number is an assigned resource. A block can represent multiple network numbers. An assignment or allocation is an outcome rather than a requester. A registry is an administering institution, not a count of applicants.
Confusion between those units can create a plausible-looking rate that describes nothing coherent. Dividing assigned network numbers by applications mixes resources with requests. Counting every corrected message as a new application enlarges the denominator for applicants who complied with administrative instructions. Treating an expired verification as a merits refusal attributes a substantive judgment where the record shows only failure to complete a procedural step. Counting substitutions as full approvals conceals the difference between what was sought and what was supplied.
A sound refusal rate would use final substantive refusals as its numerator and comparable, substantively decidable applications as its denominator. Both sides would have to cover the same period, resource type, administrative rules, and finality standard. Pending requests could not be mixed with completed decisions. Withdrawals and verification expiries would need separate categories. Partial grants and substitutions would have to retain their relation to the original request.
An approval rate demands the same discipline. It cannot be inferred from successful resource output unless each output can be connected to a distinct finally decided application. If a single approved request produced several network numbers, an output count would overstate successful applications. If one application received only part of the quantity requested, a binary approval label would understate the adverse part of the decision. If a resource assigned in one year arose from an application submitted in another, annual output and annual intake would describe different cohorts.
The denominator must therefore begin at a consistently observable intake point and follow each underlying request through closure. Corrections should remain transitions within that history unless a stated rule treats them as new applications. Duplicate tickets should be reconciled. Renewed demands should be separated from reconsiderations. The final record should identify whether the request was fully granted, partially granted, satisfied through substitution, refused, withdrawn, expired, closed as a duplicate, or still pending.
Only then does a count become a measure of treatment rather than a count of administrative traces.
The lifecycle visible through the surviving record
The public sources support a cautious map of request states. They do not establish that every registry used identical terminology or implemented every stage in the same way.
| Request state | What the bounded public evidence establishes | Comparable public count | Evident meaning |
|---|---|---|---|
| Attempted contact | Archival holdings include correspondence and operational material | No complete total located | A contact may or may not have become a valid number request |
| Parser or format rejection | The 1993 design allowed an erroneous message to be returned with an explanation | No number-request total located | Failure of message form, not necessarily a merits decision |
| Correction and resubmission | A sender could correct the error and submit again | No comparable total located | Continuation of an earlier demand unless rules defined otherwise |
| Verification pending | Verification preceded final staff action in the proposed design | No population located | The submission had not yet reached final disposition |
| Seven-day expiry | A verification not returned within seven days could expire | No count located | Procedural closure with no demonstrated judgment on merit |
| Ticket pending | A trouble-ticket number could identify pending work | No complete number-request denominator located | Tracked workload, not a final outcome |
| Final staff action | Staff action followed the earlier automated and verification stages | No complete decision population located | Could include more than a binary grant or refusal |
| Reduction or substitution | The 1993 guidance allowed a Class C block where a Class B request was judged unwarranted | No frequency or comparator population located | Resource supplied, but the original demand was not fully granted |
| Withdrawal | Complete accounting would require it as a separate disposition | No uniform count located | Applicant departure should not be presumed to be refusal |
| Final substantive refusal | Policy criteria made an adverse merits judgment possible | No complete comparable series located | The requested resource was denied after substantive consideration |
| Reconsideration or reversal | Linked decisions are required to measure remedy | No comparable linked series located | A later result matters only if connected to the earlier outcome |
The absences in this table concern publicly established comparable totals. They should not be translated into a claim that the information never existed.
A record may have been created and kept confidential. It may have survived only as correspondence rather than as a structured status. It may have been retained temporarily and later discarded under a rule that has not been examined here. It may remain in a private or closed collection. One registry may have preserved it while another did not. A field contemplated in a public design may never have been implemented. A surviving folder may contain isolated cases but not a complete population.
Those possibilities belong to different evidentiary categories. Non-public information is not the same as unrecorded information. A confidential file is not the same as a destroyed file. A file not found in a bounded inquiry is not proof of universal absence. A description of possible machinery is not proof of implementation. An archive inventory is not proof that every transaction named by a broad folder heading survives.
The public material can establish particular procedures, institutional claims, output totals, workload narratives, and surviving collections. It cannot bridge the missing joins by implication.
Three windows onto different administrative realities
The historical record between 1983 and 1999 is not a single damaged ledger. It consists of materials created for different purposes by different institutions. The earlier SRI/DDN-NIC holdings, the InterNIC design published in 1993, and RIPE’s regional reports expose different slices of administration. None supplies the complete population required for a rejection-rate calculation.
What survived from SRI/DDN-NIC
An independent archival guide describes the SRI ARC/NIC collection as 351 linear feet distributed across 281 boxes. It identifies NIC monthly progress reports, contract deliverables, naming-and-addressing records dated from 1972 through 1989, and miscellaneous email and correspondence. The guide states that monthly reports contain reference-activity statistics and describes extensive back-and-forth correspondence.
For the earlier part of the 1983–1999 period, this is substantial counterevidence to any claim that administration left no documentary trace. Correspondence survived. Monthly operational reporting survived. Naming-and-addressing material survived. The physical scale of the collection indicates a broad institutional record, although physical extent says nothing by itself about the proportion devoted to number requests.
The Guide to the SRI ARC/NIC Records remains a finding aid rather than a case census. A heading such as naming and addressing could encompass policy, forms, technical discussions, correspondence, reports, or individual transactions. Monthly reference statistics might count inquiries, services, messages, or completed actions rather than unique applications. Contract deliverables might describe responsibilities without listing adverse dispositions. Miscellaneous correspondence may contain detailed cases, but the guide alone cannot establish which cases exist or whether successful and unsuccessful requests were preserved on the same basis.
The collection therefore demonstrates survival, not completeness. Inspection could reveal a refusal letter, a correction exchange, or a changed decision. Such a discovery would prove that the particular event was recorded. It would not establish its frequency. Several examples would still lack a statistical denominator unless researchers could locate the corresponding population of requests and determine whether the preserved examples were representative.
The institutional description published in August 1990 helps place those records in context. RFC 1174 stated that USC/ISI performed the IANA function and SRI’s DDN-NIC performed the Internet Registry function described in that document. It characterized IANA’s delegation authority as discretionary and Class A and Class B identifiers as increasingly scarce. It recommended retained central functions, delegated registries, and centralized updating of aggregate registration information.
Those statements define roles and policy claims. They do not show how often discretionary authority produced an adverse decision, how an applicant participated in that decision, or whether an outcome could be challenged through an enforceable remedy. Performing the Internet Registry function is operational responsibility; it does not by itself reveal legal membership, board control, applicant consent, or the distribution of decision power within the institutions involved.
Central updating of assigned identifiers records technical uniqueness and administrative output, but it does not establish route operation or a complete history of unsuccessful demand.
The archival evidence and the policy evidence must consequently remain separate. The guide shows that correspondence and reporting survived. RFC 1174 describes institutional functions and recommended authority. Neither supplies a comparable request-to-disposition population.
What InterNIC was designed to reveal
RFC 1400 offers a different kind of evidence. It describes intended machinery rather than an inspected collection of number-request cases. Its principal value lies in the distinctions it draws between error, correction, verification, expiry, pending status, and final action.
If implemented with stable identifiers and retained fields, that design could have supported a detailed administrative history. A parser error could have been linked to a corrected submission. Verification timestamps could have shown whether a sender replied within seven days. A ticket could have recorded how long an item remained pending. A final status could have distinguished completion from continuing work.
The conditional matters. The document does not provide an InterNIC ticket export or a population of Internet number applications. It does not state how many messages failed parsing, how many senders corrected them, how many verifications expired, or how many verified tickets proceeded to approval, substitution, withdrawal, or refusal. It does not establish that every contemplated field was implemented, consistently retained, or publicly accessible throughout the period.
A ticket number itself would not solve the measurement problem. Tickets can have different scopes. One may contain a single correction while another contains a lengthy exchange. A reopened ticket may represent continuing work or a new demand. Separate tickets may concern the same underlying resource need. Pending status reveals that work remained open, but says nothing about the final decision or its reason.
The proposed design nevertheless rebuts the idea that early registration administration could recognize only successful assignments. It contemplated intermediate states and feedback to the sender. That is meaningful institutional counterevidence. It supports a claim about designed traceability, not a claim that a complete lifecycle population was produced and preserved.
What RIPE chose to report about its operations
The RIPE NCC Annual Report for 1993 states that 83 local registries operated in the distributed European system and that European Internet registries assigned 16,871 network numbers during the calendar year. These figures reveal scale and successful output during the first full calendar year of regional operation.
Their units are exact. Eighty-three is a count of local registries, not applicants, allocations, or requests. Sixteen thousand eight hundred and seventy-one is a count of network numbers assigned, not unique organizations, completed applications, approval decisions, or blocks. One request might have produced more than one network number; a block might be represented differently from an individual network number; a partial grant might still contribute to the assignment total.
The RIPE NCC Annual Report 1995 exposes another aspect of institutional activity. It records growth from 141 local Internet registries at the beginning of 1995 to 308 at year end, an increase of 167 registries from the stated starting point. It describes an increasing IP-request queue during an understaffed first half, hostmaster hiring, and elimination of the queue. At year end, registration services had six staff members, one part-time.
The report’s account is valuable evidence of workload pressure and the institution’s stated response. Yet the queue has no published size in the material examined here. There is no distribution of waiting times and no breakdown showing how queued items ended. An item could leave the queue through approval, correction, transfer, withdrawal, refusal, or another form of closure. Queue elimination records an operational result, not an approval rate.
The year-end staffing figure also has a limited temporal meaning. It follows the reported hiring and cannot be projected backward as the staffing level throughout the first half. Six staff members, one part-time, describe the registration-services team at a particular point. Without a compatible request count, the figure cannot yield requests per staff member. Without decisions joined to their dates, it reveals nothing about the consistency or substance of outcomes.
RIPE reporting also included budget disclosure. That strengthens the countercase against depicting the institution as wholly opaque: assignments, registry scale, staffing, queue conditions, and budgets all appeared in institutional reporting. Budget transparency, however, adds no population of unique applications or final dispositions. It can illuminate institutional resources without showing how many applicants were corrected, delayed, partially granted, substituted, refused, or reconsidered.
The examined reports thus make outputs and workload visible while leaving the full disposition population unavailable. This is an observed reporting asymmetry, not evidence of why aggregate request outcomes were absent. Annual reporting may reasonably focus on institutional activity, successful assignments, staffing, and finances. The sources do not establish whether comparable disposition totals were never compiled, retained privately, omitted for practical reasons, or excluded for another reason.
A true numerator can still answer the wrong question
The 16,871 network numbers assigned in Europe during 1993 are the most tempting basis for a calculation because they offer a large, exact figure. Their precision does not turn them into approved applications.
To see the denominator problem, assume artificially that every assigned network number represented exactly one successful request and that every request in the comparison population reached a final decision during 1993. Neither assumption is established by the public report. Holding the reported output constant while changing only the imaginary request population produces sharply different results:
| Illustrative request denominator | 1993 network-number output treated artificially as successful requests | Illustrative approval rate |
|---|---|---|
| 18,000 requests | 16,871 successful requests | 93.7% |
| 25,000 requests | 16,871 successful requests | 67.5% |
| 50,000 requests | 16,871 successful requests | 33.7% |
The first and third results differ by 60.0 percentage points even though the observed output remains unchanged. None of the three denominators is a historical estimate. The exercise demonstrates only the dependence of a rate on a population that the report does not supply.
The artificial assumptions also conceal several additional mismatches. A single successful request may have generated multiple network numbers. A request may have been partially granted. A block may not map one-to-one to the unit reported as a network number. An assignment completed in 1993 may have originated in an earlier year. A request received in 1993 may have remained pending into the next year. Withdrawals, expiries, duplicate closures, and substitutions may sit outside both a simple success numerator and a simple refusal numerator.
Even if the total number of requests were discovered, an approval rate would remain ambiguous until the outcome categories were reconciled. Should a Class B request answered with a Class C block count as approved, refused, or partially granted? Should an expired verification enter the decision denominator? Should a withdrawn request count against the registry? Should a corrected resubmission count once or twice? Should unresolved tickets remain outside the cohort until finality, or remain visible as pending?
Different answers could all be legitimate for different research questions. The critical requirement is consistency: the numerator and denominator must describe the same unit, cohort, stage, and definition of finality.
The same warning applies to every other prominent figure. Eighty-three local registries describe institutional participation in 1993. The change from 141 to 308 local registries describes institutional expansion during 1995. Six registration-services staff at year end describe staffing. A 351-linear-foot collection across 281 boxes describes archival extent. None is a request population.
These figures remain historically useful. They show regional scale, organizational growth, staffing capacity, operational pressure, and documentary survival. Their proper use strengthens the analysis. Relabelling them as applicants or decisions would weaken it.
The boundary cases carry the governance question
The most revealing cases are those that disappear under a binary approval-refusal classification.
A quantity reduction or resource substitution is the clearest example. The May 1993 address-management guidance stated that a Class B applicant would ordinarily have to document more than 32 subnets and more than 4,096 hosts within a 24-month engineering plan. It also allowed an applicant judged not to warrant a Class B number to receive a Class C block.
That rule identifies a possible institutional response but gives no frequency, applicant list, or comparative case series. It establishes neither how often substitution occurred nor whether different registries applied the criteria consistently. It does, however, show why the original request and the final grant must remain visible together. From the standpoint of resource delivery, the applicant received something. From the standpoint of the demand submitted, the applicant did not receive the requested Class B number.
Expiry sits at another boundary. Under the proposed 1993 arrangement, a verification not returned within seven days expired. The rule identifies a clear administrative closure point, but the reason for non-response remains unknown. The sender may have abandoned the request, missed the verification, encountered a delivery failure, misunderstood the requirement, or pursued another route. The source does not establish how often any explanation applied.
Classifying every expiry as refusal would convert non-completion into a merits judgment. Omitting all expiries would hide attrition before substantive review. Separate reporting preserves both facts: the request did not reach a completed merits disposition, and the public record does not explain why.
Correction can also represent either assistance or burden. A parser’s error response may help a sender submit usable information. Repeated correction demands may also create delay or contribute to abandonment. Without counts connecting errors to resubmission and final outcome, neither interpretation can be measured. The design of a correction path shows procedural capacity, not its completion rate or effect on applicants.
Withdrawal poses a similar problem because it may remove a request without producing an assigned resource or a refusal. A requester may voluntarily abandon an obsolete need, react to delay, accept guidance that the request is premature, or decide not to disclose further information. Unless withdrawal is separately recorded, an analyst cannot distinguish applicant choice from unresolved or adverse administrative action.
Reversal is the most demanding state because it requires at least two linked decisions. A later assignment counts as a reversal only when the record connects it to an earlier adverse result and shows that reconsideration altered that result. A later grant following new facts, a changed request, a new rule, or another registry’s action is a different event.
These boundary cases are not marginal. They determine whether the history describes technical intake, administrative attrition, substantive scarcity decisions, or effective redress. A ledger that records only assigned resources makes all four harder to distinguish.
Confidentiality explains a limit, not the missing denominator
The strongest argument against publishing individual request files comes from the same address-management guidance that made adverse quantity decisions possible.
For a Class B request, the ordinary threshold required an engineering plan projecting more than 32 subnets and more than 4,096 hosts over 24 months. Those plans were to be held in strict confidence. The Internet Registry could also require accounts of regional assignments and audit engineering plans.
The privacy rationale is substantial. An engineering plan could reveal anticipated growth, topology, dependencies, operational design, or commercially sensitive intentions. A refused applicant might face greater exposure than a successful applicant because explaining the adverse result could require publication of the very plan that failed to justify the requested resource. Public access to raw dossiers could discourage candour or reveal details unnecessary for institutional accountability.
This countercase rules out a simplistic demand that every application and every reason should have been published. Confidentiality is compatible with legitimate administration, and the absence of public applicant files should not be treated as evidence of concealment or inconsistent treatment.
It does not follow, however, that privacy required the absence of aggregate dispositions. A registry could protect applicant identities and engineering plans while reporting how many unique requests entered each stage. It could separate parser failures, corrected submissions, expired verifications, withdrawals, full grants, partial grants, substitutions, substantive refusals, and pending cases. Broad reason categories could preserve the institutional basis for decisions without exposing the underlying plans. Rare categories could be combined or released after a delay where identification risk was high.
The sources establish confidentiality as a rational limit on raw-file disclosure. They do not establish it as the historical cause of the missing aggregate request ledger. Aggregate totals may have been absent for another reason, retained internally, compiled under incompatible units, or located in records not inspected for this inquiry.
This distinction protects both sides of the argument. It takes applicant privacy seriously without allowing confidentiality to become a universal explanation for missing institutional measures. It also recognizes that transparency existed in several forms: feedback and tickets were contemplated; RIPE disclosed assignments, staffing, queues, registry growth, and budgets; and extensive correspondence and operational reports survive in archives.
The historical problem is therefore not complete darkness. It is a mismatch between what the visible records can count and what a treatment analysis needs to compare.
Consistency and remedy require joined histories
A rejection rate is only the first missing measure. Claims about consistent treatment require more than counts of approvals and refusals because different outcomes can be justified by different facts.
Two applicants may seek different resource types or quantities. They may submit under different rule versions or at different points in a period of institutional expansion. Their engineering evidence may differ. One request may arrive complete while another requires correction. A result labelled approved may mean full satisfaction in one case and a reduced substitute in another.
A meaningful comparison therefore needs the administering registry, request type, resource and quantity sought, resource and quantity supplied, receipt and decision dates, applicable rule version, correction history, final disposition, standardized reason, and any later reconsideration. Confidential evidence need not be published, but the record should show whether it was considered under a common category.
The bounded public material supplies fragments of that picture rather than a comparable population. It provides policy criteria, possible administrative states, successful output totals, registry counts, staffing and queue narratives, budget disclosure, and archival holdings. Those fragments support precise questions. They do not yield population-level answers.
Written criteria alone establish that an institution articulated a standard. They do not show uniform application. Different outcomes alone do not show inconsistency because the underlying requests may differ. The evidence consequently permits neither a finding of consistent treatment nor a finding of inconsistent treatment.
A disparate-treatment analysis faces an additional obstacle. It would require defined comparison groups and relevant applicant characteristics connected to equivalent requests and outcomes. The public material examined here lacks both a complete applicant population and consistent comparator fields. No defensible estimate can be made of whether one class of applicant faced a different rate of correction, delay, reduction, refusal, or reversal.
Workload effects are equally dependent on joined histories. RIPE’s account of an increasing queue during an understaffed first half of 1995, followed by hiring and queue elimination, supplies a credible institutional narrative. Several effects are possible. Pressure may have lengthened waiting times while leaving substantive decisions unchanged. It may have increased correction cycles, altered classifications, or had no measurable effect on final outcomes. The report does not distinguish among them.
A before-and-after analysis would require requests dated to the periods before, during, and after the queue pressure, along with request types, correction events, waiting times, dispositions, and contemporaneous staffing. Registry participation also changed sharply during the year, rising from 141 to 308 local registries. That expansion could have altered intake distribution and support demands. It cannot be treated as a fixed background condition.
Remedy analysis requires the most explicit linkage. It must begin with a defined adverse outcome, identify reconsideration, preserve the grounds, and connect that review to a final result. A correction after a parser error is a procedural opportunity, but its effectiveness depends on how many affected senders resubmitted and reached final action. A substituted resource may resolve part of a demand without reversing the original judgment. A later grant may reflect new evidence rather than review.
The public corpus assembled for this inquiry contains zero complete, comparable chains running from an initial number request to a final substantive refusal and then to an identified review result. That is a finding about this evidence set. It does not establish that applicants never sought reconsideration, administrators never changed outcomes, or relevant records do not survive elsewhere.
Without joined decisions, sequence cannot stand in for remedy. Later success is merely later success until the record shows what earlier decision was reconsidered and why the result changed.
The minimum account that would make measurement possible
A useful public account would begin with an internal identifier connecting every correction, verification event, ticket, decision, and review to the same underlying request. The identifier need not reveal the applicant publicly. Its purpose would be to prevent messages and resubmissions from being mistaken for additional applications.
For each request, the record would preserve the reporting registry, resource type, quantity sought, quantity supplied, rule version, receipt date, verification or expiry date, correction events, ticket dates, final disposition, broad reason category, finality date, reconsideration status, and review outcome. It would also indicate whether supporting material was confidential and whether the administrative record remained retained.
That is one compact specification, not a demand to publish raw case files. Annual public totals could be grouped by registry, request type, rule version, and disposition. Full grants, partial grants, substitutions, refusals, withdrawals, expiries, duplicate closures, pending cases, and unresolved items should reconcile to a defined intake cohort. Corrections and messages could appear separately as workload measures rather than being counted as additional final decisions.
The separation would permit several analyses that the surviving reports cannot support. Researchers could calculate the share of finally decided requests ending in full grant, partial grant, substitution, or refusal. They could measure how many submissions expired before merits review. They could compare waiting-time distributions without treating delay as a decision. They could observe whether correction cycles differed across request types or administering registries. Linked review records could reveal how often an adverse result was affirmed, modified, substituted, or reversed.
Such totals would still fall short of proving fairness. Broad categories can conceal differences in complexity. Administrators may apply reason codes inconsistently. Aggregation can hide disparities within groups. A serious evaluation would still require clear definitions, sampling, and controlled examination of confidential evidence.
But the public account would supply the indispensable common population. It would allow legitimate privacy restrictions to coexist with measurement of institutional treatment.
What archival work could still change
The surviving archive guide identifies promising material for the earlier period: monthly progress reports, contract deliverables, naming-and-addressing records through 1989, and extensive correspondence. Those holdings deserve inspection precisely because the finding aid cannot answer the denominator question.
Monthly reports might contain counts of inquiries, messages, completed actions, or number requests. Researchers would need to establish the unit, determine whether number-resource activity was separated from other services, and test whether adverse and successful outcomes were reported consistently. Correspondence might preserve individual refusals or corrections, but examples should be indexed against the surrounding holdings rather than selected only because they are contentious or unusually complete.
Contract deliverables might illuminate reporting duties, operational definitions, or record-retention expectations. Any stated duty would still need to be compared with the surviving material. An obligation to produce a report does not prove that every expected field survives; a preserved report does not prove comprehensive case coverage.
For the InterNIC period, the decisive materials would be ticket exports, field definitions, status codes, retention rules, deletion evidence, and documentation of what was public or private. Any export would need reconciliation for duplicate, reopened, or migrated tickets. The historical definitions of each status would matter more than modern interpretations of its label.
For RIPE’s regional administration, annual totals of unique requests by resource type and final disposition would transform the analysis. Consistent queue snapshots, reason categories, rule versions, decision dates, and linked reconsiderations would permit the reported institutional narratives to be tested against request outcomes.
Evidence of absence requires more than an unsuccessful search. A defensible retention conclusion would require a defined search scope, inspection of likely repositories, knowledge of contemporary record practices, and, where relevant, retention schedules or deletion records. Even then, the result might apply only to a particular period, office, or administrative system.
The reverse is also true. Finding a refusal proves that at least one refusal was recorded. Finding a reversed decision proves that at least one result changed. Neither discovery supplies a rate until it can be situated within a comparable population.
Archival discoveries should change one proposition at a time. They should not be asked to supply a universal conclusion that their scope cannot bear.
The missing denominator is the finding
The surviving record supports a firm but bounded judgment.
Early Internet number administration was not wholly recordless. Policy documents described institutional roles and possible adverse resource judgments. The proposed 1993 machinery distinguished errors, corrections, verification, expiry, tickets, and final action. RIPE reported successful assignments, expanding registry participation, staffing, queue conditions, and budgets. Extensive SRI ARC/NIC correspondence and operational reports survive in a large archival collection.
Each source illuminates a different part of the institutional landscape. The archive proves survival of correspondence and reporting, not completeness of request files. RFC 1400 shows what its proposed machinery could have recorded, not what complete number-request history was implemented or retained. RIPE’s figures document institutional activity and output, not the population of applications from which final outcomes emerged.
The examined public evidence therefore cannot produce a defensible historical rejection rate. It cannot show whether comparable applicants received consistent treatment, whether identifiable groups experienced different outcomes, whether workload altered decisions, or how often reconsideration changed an adverse result. It also cannot establish the opposite conclusions. Published rules and successful outputs do not prove consistency or fairness any more than a missing disposition ledger proves inconsistency or unfairness.
Confidential engineering plans justify restraint in publishing raw applicant material. They do not erase the possibility of de-identified institutional accounting, and the sources do not establish confidentiality as the reason comparable aggregate dispositions are missing.
The central historical asymmetry is narrower and more consequential: successful resources can be counted, participating registries can be counted, staff can be counted, budgets can be disclosed, queues can be described, and archive boxes can be measured, while the common population connecting attempted request, correction, expiry, withdrawal, reduction, refusal, and review remains unavailable.
Until inspected records produce that population, the most responsible conclusion is also the most useful one. The public record can show the states administrators contemplated, the outputs institutions reported, and the materials that survived. It cannot show how often comparable number requests ended adversely or whether a later action provided an effective remedy.
That is not a verdict on motive. It is the boundary between a documented institutional history and a rate the evidence cannot yet sustain.