Summary
- RIR processes rightly avoid courtroom formality, yet their existing requirement for reasoned objections does not tell entities how to distinguish observation, dataset, model, legal interpretation, institutional estimate and policy value.
- Evidence should be classified by the claim it supports. Operational testimony can establish that a mechanism occurs, while prevalence needs a defined population; registry statistics can describe recorded cases, while causal and welfare claims need additional analysis.
- Every consequential empirical claim should expose a minimum test: proposition, source, period, population, definitions, method, uncertainty, relevant interests and a route for correction or rebuttal.
- Rebuttal is not a demand that volunteers reproduce privileged systems. The claimant bears responsibility for making an assertion testable; staff may protect confidential records while publishing aggregate method, assumptions and limitations.
- Chairs should maintain an evidence ledger linked to the issue map, distinguish factual disputes from value choices and explain why evidence was accepted, discounted or left uncertain. A numerical score must never replace reasoned judgment.
- Better rules would strengthen lived operational evidence, not exclude it. They would prevent institutional format, repeated anecdote or spurious precision from receiving authority that the underlying claim cannot support.
Consensus has a reason rule but not an evidence rule
RIPE's PDP says that suggestions and objections must be justified with supporting arguments. LACNIC describes consensus in terms of meaningful opinions and the absence of irrefutable technical objections, stressing nature and quality rather than number. APNIC asks chairs to distinguish minor and major objections and determine whether concerns were resolved or given due consideration. RFC 7282 similarly directs attention away from vote totals and toward issues.
These are important safeguards. They prevent a bare "no" from becoming an automatic veto and a crowd of endorsements from becoming proof. They require chairs to examine why entities take a position.
But an argument can be reasoned and still rest on weak evidence. An operator says a proposed rule will cause widespread route instability based on two incidents. Staff predicts low cost using an undisclosed system assumption. An author cites a registry count that excludes failed applications. Counsel identifies legal risk without stating whether it is prohibition, uncertainty or prudence. Each contribution has a reason. The public still needs a way to test its reach.
Current norms leave much to experienced chairs. That flexibility is valuable, but it creates inconsistency and favours familiar forms. A polished impact assessment can appear stronger than scattered field reports even when its model is narrower. A veteran's confident anecdote may receive more attention than a newcomer's careful but uncertain dataset.
An evidence rule need not dictate outcomes. It should make claims legible: what is being asserted, which observation supports it, how far the inference extends and what would count against it. Consensus remains judgment about policy and trade-offs. The evidence constitution ensures that judgment is made over claims whose foundations are visible.
Evidence is not one substance
Policy discussions often use "evidence" as though it were a single commodity that some entities possess and others lack. In reality, number policy relies on several distinct forms.
Operational testimony reports direct experience: an allocation request, routing failure, transfer delay or customer burden. Administrative data records cases processed by a registry. Technical measurement observes routes, certificates, database entries or traffic under defined methods. Documentary evidence includes policy text, contracts, minutes and implementation notices. Comparative evidence examines another region or institution. Models estimate future depletion, demand, cost or behaviour. Legal analysis interprets authority and exposure. Value claims state what is fair, proportionate or mission-aligned.
These forms answer different questions. One documented incident can prove that an event is possible. It cannot establish prevalence without a denominator. A registry total can describe records under its definitions. It may not capture discouraged or abandoned requests. A model can illuminate consequences under assumptions; it does not observe the future. Legal advice can identify constraints but cannot decide whether a lawful trade-off is desirable.
Confusion produces two opposite errors. Entities dismiss valuable testimony because it is "only anecdotal," even when the claim is existence or mechanism. Or they use a vivid case to claim that a practice is region-wide. Institutions publish precise counts that appear objective even though category definitions changed.
Evidence classification is therefore not hierarchy alone. It is fit between form and proposition. The strongest evidence is the material capable of supporting the particular inference being made, with limitations disclosed. A policy forum should ask not "is this evidence?" but "evidence of what, over which population, and with what uncertainty?"
The proposition must come before the citation
A citation cannot be evaluated until the claim is stated precisely. "Transfers are being abused" might mean false documentation occurs, acquired space is routed differently from stated need, intermediaries evade a rule, or outcomes conflict with a policy value. Each requires different evidence.
Authors should express consequential claims as testable propositions. Identify the actor, behaviour, period, population and expected mechanism. "Several recent transfers show a problem" becomes "among completed transfers reviewed under the stated criteria during this period, a defined pattern appeared, and the proposed restriction is expected to reduce it through this mechanism."
Precision does not require false certainty. A proposition can be exploratory: available cases suggest a risk, but prevalence is unknown. That statement supports investigation or a reversible safeguard more readily than a permanent broad restriction.
Chairs can help entities refine claims without becoming researchers for one side. An issue map should separate existence, frequency, cause, consequence and remedy. A speaker may have direct evidence for the first and a hypothesis for the others. Recording the distinction preserves contribution while preventing inference inflation.
Staff assessments need the same discipline. "Implementation will be complex" should identify functions, dependencies, expected effort and uncertainty. "Policy will accelerate exhaustion" should state the pool, eligible population, uptake assumption and horizon. Institutional letterhead does not remove the need for a proposition.
Once the proposition is clear, disagreement becomes more productive. Entities can challenge a definition, offer a counterexample, supply another denominator or accept the fact while rejecting the remedy. Without precision, they argue around an adjective.
Anecdote can prove existence but not prevalence
Operational communities depend on stories because many important events are not captured in public datasets. A small provider may describe how a documentation rule delayed service in a market where corporate records are slow. A national registry may explain how local allocation patterns differ. An engineer may recount a routing consequence that a proposal author did not anticipate.
This testimony is evidence. Its strength depends on the claim. A documented case can refute "this never happens" and reveal a mechanism. It can identify variables a model omitted. It can show that a burden is real for at least one organisational form.
The danger is extrapolation. A entity says "we see this all the time" without defining time, cases or observation. Repetition by several colleagues may reflect one shared event. Meeting culture rewards vivid narratives, and a chair may reasonably treat them as urgency.
A light protocol can preserve value. The speaker states whether the experience is firsthand, the approximate period, affected type of organisation, frequency relative to their own work and any reason it may not generalise. Confidential customer details remain protected. If the claim is consequential, a follow-up note can document it.
Opponents should not demand impossible disclosure as a tactic. Commercial sensitivity and security matter. They can ask what inference the testimony supports and whether a narrower policy could address it. The chair can record prevalence as uncertain while recognising the mechanism.
An evidence rule that excludes anecdote would entrench institutions with databases. A rule that leaves extrapolation unchecked privileges confident storytellers. Fit-for-claim treatment protects both operational voice and analytical integrity.
Registry data describes the registry's field of vision
RIRs hold uniquely important administrative data. They know requests received, resources registered, processing outcomes, transfer records and system changes. RIPE's PDP expressly assigns the secretariat a role providing facts and statistics; other RIRs publish impact assessments and policy experience material.
These data can be the best available source. They are not a complete image of the policy environment. A registry observes people who enter its process. It may not observe organisations deterred by criteria, networks using substitutes, informal markets or errors never reported. Definitions reflect policy and system design.
A count therefore needs a data dictionary. What is a request, account, transfer, rejection or incident? Were duplicates removed? What period is covered? Did system or policy changes affect comparability? Is the unit an organisation, ticket or resource block? Which records are excluded for quality or confidentiality?
Administrative data can also contain feedback from the rule under debate. If current policy discourages applications, low request volume may be cited as evidence that demand is low. The institution's own barrier helped produce the observation. This does not make the count false; it limits the inference.
Staff should publish aggregate method and uncertainty. Where raw records cannot be released, reproducible queries may also be sensitive, but category logic and validation can often be described. An independent reviewer can examine protected material for high-stakes claims.
The public should treat registry data as evidence from a privileged vantage, not as neutral reality. Its strength is custody and operational continuity. Its boundary is the institution's field of vision.
Models are arguments with arithmetic
Scarcity and implementation debates frequently rely on projections. How quickly will a free pool be exhausted? How many accounts are eligible? How much engineering time will a rule require? What behaviour will change after a restriction?
Models are essential because policy concerns the future. Their numbers can create unwarranted authority. A precise date or cost appears factual even when it depends on uncertain uptake, demand, staffing and behaviour.
Every material model should disclose inputs, assumptions, structure, sensitivity and validation. If all eligible organisations are assumed to apply, show that as a maximum scenario rather than expected demand. If historical uptake informs the central case, explain why it remains relevant. Provide ranges where uncertainty is real.
Scenario comparison is often more useful than one forecast. A low, central and high case reveals which variable drives the decision. Entities can then debate that variable. A model whose conclusion remains across plausible assumptions is stronger than one that turns on a narrow estimate.
Cost models need marginal analysis. Separate the minimum implementation required by policy from an ideal system redesign. Identify recurring and one-time cost, transferred member burden and opportunity cost. Benefits should receive comparable treatment even when harder to quantify.
Models should be updated when evidence changes and compared with outcomes after implementation. Persistent over- or under-estimation becomes evidence about the method. The institution should preserve old forecasts rather than replacing them quietly.
Arithmetic helps discipline intuition. It does not convert assumptions into observations. A chair should describe a model as supporting a scenario under stated conditions, not proving what will happen.
Legal evidence needs a public category
Legal analysis presents a special challenge. Advice may be privileged, jurisdiction-specific and uncertain. Entities may lack access to comparable counsel. Yet a reference to legal risk can stop a proposal.
The public record should distinguish at least four categories: clear prohibition, material exposure under uncertain law, drafting ambiguity and prudential preference. Each has different consequences. A prohibited act requires redesign. Exposure may require mitigation or board judgment. Ambiguity calls for clearer text. Prudence is a policy argument, not a legal command.
Counsel should identify the proposition, applicable jurisdiction, factual assumptions and range of plausible interpretations to the extent possible. A non-privileged summary can preserve confidentiality while making the constraint testable. Where even the legal basis cannot be disclosed, the accountable board should own the decision and state why secrecy is necessary.
Entities can rebut legal claims through text, public authority, comparative institutional practice or independent advice. They should not be expected to disprove a hidden memorandum. The burden remains with the actor invoking the constraint to provide enough public reason.
Legal conclusions also change over time. New judgments, statutes or guidance may alter risk. Policy records should date the assessment and avoid treating it as permanent precedent.
An evidence protocol does not ask chairs to adjudicate law. It tells them what kind of institutional claim is before the community and which body must decide. Without a category, "legal concern" becomes an unanswerable adjective with veto power.
Source quality is not speaker status
Open policy communities rightly allow anyone to contribute. In practice, speaker reputation affects how claims are received. Experienced entities are trusted because they have often been correct. Staff is trusted because it administers the system. A person from a large network may be assumed to possess broad data.
Reputation can help allocate attention, but it is not evidence of the proposition. A veteran can extrapolate too far. A newcomer can supply a decisive measurement. Institutional status can guarantee access to records while not guaranteeing a fair interpretation.
Chairs should separate source competence from source authority. Ask whether the speaker is positioned to observe the claim, whether method is disclosed and whether relevant interests are stated. An operator reporting their own incident has direct competence. Claiming region-wide effect requires more.
Affiliation disclosure helps readers understand context, but it should not become automatic discount. A transfer broker may possess valuable market evidence while having an interest in policy. A registry team may understand implementation while preferring existing systems. Interests inform scrutiny; they do not erase facts.
Anonymous or protected evidence requires care. It may be necessary for commercial or security reasons. The intermediary presenting it should state how identity and credibility were verified, what details are withheld and which inference remains safe. Chairs should not invoke unseen community concern without such handling.
The norm should be symmetrical: no claim becomes self-proving because of title, employer or tenure. At the same time, entities do not start from artificial equality of expertise. Competence is demonstrated by relationship to the evidence and openness about limits.
Rebuttal is a designed opportunity
Policy rhetoric often says that anyone may challenge a claim. The practical ability to rebut depends on disclosure, time and access. A volunteer cannot reproduce a registry query without definitions. An operator cannot answer a new cost model during a five-minute microphone interval. A entity cannot contest legal risk when its category is hidden.
The claimant should make the assertion testable. Publish method, assumptions and limitations proportionate to consequence. The challenger should identify the disputed proposition and offer contrary evidence, an alternative explanation or a request for missing support. Chairs should ensure a public response.
Rebuttal does not require proving the opposite. Showing that data cannot support the claimed prevalence may be enough. Providing one valid counterexample can defeat a universal claim. Demonstrating sensitivity to an uncertain assumption can reduce confidence even without another model.
Time must be adequate. Material staff analysis released near closure should pause the decision. Revised evidence should be versioned. Entities should be able to ask technical questions without those questions being counted automatically as opposition.
Confidential evidence may require an independent reviewer. The reviewer reports whether the protected material supports the public claim and identifies limitations. Selection, competence and conflicts should be transparent. This is not perfect equality, but it avoids requiring blind trust.
A rebuttal record should show what changed. Was a count corrected, a claim narrowed, an assumption retained or an objection left unresolved? The chair's final reasoning can then use the current evidentiary state rather than the first institutional statement.
Open doors do not create adversarial testing by themselves. The process must provide enough information and time for challenge to be meaningful.
Absence of evidence has several meanings
Policy debates regularly encounter missing data. No documented cases appear, no reliable denominator exists, or a registry cannot measure the behaviour in question. Entities may treat this as proof that a problem is absent or as proof that it is hidden.
Both moves are too quick. Absence may reflect genuine rarity, poor reporting, a definition that excludes cases, deterrence, confidentiality or lack of instrumentation. The relevant question is whether the system would likely observe the event if it occurred.
An evidence ledger should record the observation mechanism. Are applicants required to report? Do staff codes capture the issue consistently? Can public routing data reveal it? Did anyone search the relevant period? A negative finding is stronger when detection is reliable.
Policy response should match uncertainty. A severe but poorly observed risk may justify monitoring, a pilot or reversible safeguard rather than a permanent restriction. A claim of widespread harm with no detection mechanism may justify data collection before action. Urgency can alter the balance but should be explicit.
The burden of proof is not fixed in the abstract. Proposals changing rights or imposing substantial cost should explain why uncertainty warrants intervention. Existing policy also has consequences and should not receive immunity merely because it is the baseline. Evidence of inaction risk belongs beside evidence of action risk.
Chairs should avoid the phrase "there is no evidence" without specifying search and scope. A precise statement might say that no documented case was found in defined registry records, but external or deterred cases remain unmeasured. That sentence supports a better policy choice than false certainty.
Counterexamples deserve explicit treatment
A counterexample can be powerful because it tests the boundary of a rule or claim. If an author says every organisation can meet a documentation requirement, one valid organisational form that cannot may defeat the universal proposition. It does not necessarily defeat the policy; the text may add an exception or adopt a narrower claim.
Chair summaries often absorb counterexamples into "concerns" without showing whether the underlying generalisation changed. An evidence ledger should link each counterexample to the proposition, verify relevant facts and record disposition.
Not every exception should control. A rare edge case may be handled operationally. Entities should ask frequency, severity, detectability and remedy. The author can explain why the benefit outweighs residual burden. What matters is that the counterexample is not dismissed because it is singular when singularity is logically sufficient.
Comparative RIR practice can serve as a counterexample to claims of impossibility. If another registry implements a similar mechanism, local staff should explain relevant differences. The comparison does not prove equal cost or legality. It narrows the claim from impossible to contingent on system, structure or jurisdiction.
Adverse examples should be sought as well as welcomed. Authors and staff can include cases where the proposal fails, groups bearing cost and conditions under which the model reverses. This strengthens trust and improves design.
The culture of consensus sometimes rewards compromise language that hides falsification. A proposal adds vague discretion without resolving the case. Better evidence practice states whether the exception is covered and who decides. The counterexample then improves the rule instead of disappearing into reassurance.
Evidence quality and policy values must remain separate
Empirical evidence can establish consequences. It cannot decide every normative question. A reliable model may show that a restriction extends pool life by a certain range. The community must still decide whether the distributional cost is justified. Accurate data may show that an abuse is rare; severity may still warrant action.
Conflating evidence with values creates technocratic closure. Entities argue over numbers because the real disagreement about fairness, mission or risk tolerance remains unstated. The side controlling data appears to control the decision.
Chair issue maps should separate factual propositions from value choices. First ask what likely happens, to whom and with what uncertainty. Then ask which outcome the policy should prefer and why. Evidence informs both but does not erase responsibility for the latter.
Value claims also require reasons. "Fair" should identify the principle: equal access, need, prior reliance, regional benefit or protection of the commons. Different principles can fit the same facts. The final record should state which trade-off the community accepted.
This separation protects science from strategic use. A dataset is not discredited because it supports an unpopular policy, and a policy is not legitimate solely because it cites a dataset. Authors cannot outsource moral choice to a projection. Staff cannot turn administrative efficiency into a factual necessity.
Rough consensus is suited to mixed judgment because it examines objections rather than applying a formula. Evidence rules improve the factual field; they should not replace the public act of choosing among defensible values.
Chairs need an evidence ledger, not a scorecard
An evidence ledger can be a simple public table attached to the proposal's issue map. For every consequential claim, it records the proposition, type of evidence, source, method link, period, population, uncertainty, relevant interest, challenge and current status.
Status terms might include supported within scope, contested, corrected, superseded, limited public evidence for claimed inference and unresolved. These are descriptive, not verdicts on the policy. Chairs explain material classifications and invite correction.
The ledger reduces repetition. A entity can endorse an existing claim or add evidence without restating the entire argument. Authors can see which assumptions need work. Staff can correct misunderstandings before they become allegations of bad faith.
It also preserves version history. If a model changes, the old result remains linked and marked superseded. If revised text removes the issue, the ledger records that disposition. Later reviewers can understand why the proposal advanced.
Numerical evidence scores should be avoided. A single rating creates false comparability and invites gaming. A confidential case may be highly credible but not reproducible publicly. A large dataset may be precise but irrelevant to causation. Narrative classification is more honest.
The chair is not expected to validate every technical method personally. They can request expert review, identify limits and judge whether uncertainty is acceptable for the decision. The ledger makes that judgment visible.
Most importantly, it prevents institutional memory from retaining only conclusions. Future experience can revisit the claims. If predicted harm occurs, the record shows who anticipated it and on what basis. If it does not, the model can improve. Evidence becomes a learning asset rather than ammunition consumed in one debate.
Confidentiality can coexist with testability
Number policy touches commercially sensitive network plans, customer relationships, abuse cases and security practice. Demanding full public release would exclude important evidence and may violate duties. Confidentiality must not become an all-purpose authority shield.
The first technique is aggregation. Publish counts, ranges and category definitions while suppressing identifying detail. The second is redaction with an explanation of what kind of information was removed and why. The third is synthetic examples that preserve mechanism without claiming to be observed cases.
For high-stakes disputes, a trusted reviewer can inspect protected material under clear terms. The reviewer reports scope, method, whether the public proposition is supported and any limitation caused by unavailable records. The selection process should avoid dependence on the claimant alone.
Entities can also submit confidential evidence to chairs under a published handling rule. Chairs should not rely decisively on it without a public summary sufficient for response. The source's identity may remain protected while the claim and confidence are visible.
Data minimisation matters. Reviewers should receive only what is needed. Retention, access and destruction should be defined. An evidence rule must not create a new store of sensitive operational material merely to display rigour.
Sometimes testability cannot be achieved. A board may need to act on protected legal or security evidence. It should state that it, not the open forum, made the judgment, describe the category and provide a review or expiry where possible. The public should not be told that consensus resolved a claim it could not examine.
Confidentiality changes the form of accountability. It does not eliminate the need for it.
The standard of proof should follow consequence
Not every policy claim requires the same support. Correcting a cross-reference needs less evidence than restricting transfers, changing eligibility or authorising revocation. A proportional standard asks how consequential, irreversible and uncertain the decision is.
Low-impact editorial changes may rely on documentary comparison and public review. Reversible pilots can proceed on plausible evidence if monitoring and expiry are strong. Durable restrictions affecting member rights need broader data, mechanism analysis, alternatives and distributional assessment. Emergency action may accept lower initial certainty but requires narrow scope and prompt review.
The baseline deserves scrutiny too. Requiring strong evidence only from change protects existing rules regardless of harm. A proposal can show credible evidence that current policy causes a problem; opponents defending the status quo should address it. The burden is allocated by claims, not by institutional comfort.
Serious allegations require particular care. Claims of fraud, capture or widespread abuse can damage people and institutions. Define conduct, verify cases and avoid inferring motive from outcomes. Policy can address vulnerability without declaring unproven wrongdoing.
Chairs should announce the expected evidence depth early. Authors then know whether a survey, administrative analysis or comparative review is needed. Staff can plan assessment. Entities can challenge proportionality.
There should be no formula such as "three sources" or "statistical significance" for every case. Number-policy populations may be small, and rare events may matter. The standard is sufficient support for the consequence under acknowledged uncertainty.
Proportionality allows the process to remain usable while demanding rigour where power is greatest.
Post-implementation evidence closes the argument loop
Policy adoption should not end evidence review. Forecasts become testable after implementation. Institutions can compare expected and actual demand, cost, delay, incidents and distributional effects.
The original evidence ledger provides the baseline. Which claims drove adoption? What uncertainty remained? Which safeguards or thresholds were supposed to matter? A later report can examine those propositions rather than offering generic success language.
Data collection should be designed before implementation. Otherwise the institution may discover that it cannot measure the objective. Privacy and burden need consideration. Sampling may be enough.
Negative findings should be published. If a policy did not produce the expected benefit, amendment is not institutional failure; refusing to learn is. If unpredicted harm appears, affected operators need a route to trigger review. If staff estimates were accurate, the record strengthens future trust.
Post-implementation comparison also calibrates evidence sources. Were operational anecdotes early signals of a wider issue? Did a model overstate uptake? Did registry data miss deterred applicants? The answers improve future standards.
Sunset clauses and scheduled review can make this learning consequential. A temporary rule continues only if evidence supports it. Permanent policy can still have review triggers based on thresholds or incidents.
An evidence constitution is incomplete if claims vanish after winning. Policy communities should retain responsibility for the consequences they predicted and the uncertainties they accepted.
What a minimum evidence protocol would require
Every material empirical assertion should identify its proposition, evidence type, source, period, population or case boundary, definitions, method, uncertainty and relevant interests. It should distinguish observation from inference and state what would challenge the claim.
Authors include an evidence table with the proposal. Staff assessments follow the same categories and disclose model assumptions. Entities can submit structured operational testimony without needing academic form. Chairs maintain the public ledger and issue map.
Material new evidence receives a response period. Corrections are versioned. Confidential material uses aggregation, redaction or independent review. Legal assertions identify category and responsible decision-maker. Comparative evidence states relevant institutional differences.
Chair findings explain which claims were accepted within scope, which remained contested and which value choices controlled despite uncertainty. They do not claim that unresolved data became fact through consensus. Formal decisions link the ledger.
After implementation, the institution reports against key predictions. Members and boards review whether evidence practices were followed and fund necessary research independence. Training helps chairs classify claims without becoming gatekeepers to specialised language.
The protocol should begin as guidance and be evaluated. Overly burdensome fields can be simplified. Examples can show how one incident, a registry count and a model each receive appropriate treatment. Accessibility must remain central.
The purpose is not to professionalise ordinary entities out of the room. It is to prevent hidden standards from favouring institutional documents and confident insiders. A visible minimum gives everyone the same questions to answer.
Rigour without a courtroom
Number-policy communities should not become courts. They do not need formal admissibility hearings, cross-examination theatre or expert credentials as a condition of speech. Such machinery would slow decisions and advantage wealthy organisations.
They do need a shared ethic of testability. Claims that justify power should expose their basis. Institutional evidence should remain challengeable. Operational experience should be scoped rather than dismissed. Uncertainty should be recorded. Rebuttal should receive information and time.
This ethic strengthens rough consensus. Chairs can focus on whether material issues were adequately examined rather than whether the loudest side produced more citations. Entities can disagree about values after narrowing factual disputes. Future reviewers can see which assumptions proved right.
The absence of explicit rules does not create neutrality. It creates informal evidence hierarchies. Staff format, seniority, fluency, repetition and precision become proxies for reliability. People who understand those customs prosper; others struggle to make valid knowledge legible.
A public protocol replaces hidden hierarchy with accountable judgment. It will not remove disagreement or error. It will make the reasons for accepting one claim over another available for correction.
The central principle is modest: evidence must be strong enough for the inference and visible enough for fair challenge. Policy values then determine what to do with it. That separation preserves both expertise and community authority.
RIRs administer scarce and consequential resources through processes built on trust. Trust is not the absence of verification. It is confidence that claims can be examined without regard to who controls the archive, the microphone or the system. Number policy already asks for reasons. Its next constitutional step is to govern the evidence those reasons carry.
Evidence assistance should be a common resource
A visible protocol will expose an inequality the protocol cannot solve alone: some entities have analysts, counsel and institutional data, while others have a relevant experience and no time to turn it into a formal submission. If evidentiary discipline simply raises the price of being heard, it will narrow the community.
RIRs should provide neutral evidence assistance. A small support function can help a entity state the proposition, remove confidential details, identify the correct denominator and separate direct observation from inference. Assistance should be available to supporters and opponents under published rules. The resulting submission remains the entity's, and the support given is disclosed at an appropriate level.
The institution can also publish reusable materials: data dictionaries, query notes, glossary entries, examples of scoped testimony and explanations of common statistical errors. Public aggregate datasets should use stable definitions and note breaks in series. When staff already holds a relevant analysis, entities should not need personal relationships to discover it.
Independent research grants may be justified for proposals with major distributional effect. Selection should focus on the question and method, not the preferred outcome. Results, including inconclusive ones, belong in the public record. Funding must not buy a special vote.
Chair training is equally important. Chairs need to recognise universal claims, missing denominators, causal leaps and undisclosed value choices. They also need to avoid humiliating entities whose evidence is incomplete. A clarifying question can narrow a claim without dismissing the speaker.
Evidence assistance turns rigour into infrastructure. It recognises that an open door is limited public evidence when only institutions can afford the language of proof. The community should make valid knowledge easier to express, then apply the same testability standard to everyone.