Best Practice Forums and the Problem of Institutional Self-Citation

Summary

• Best Practice Forums were introduced in 2014 to collect experience and produce tangible IGF outputs. The Multistakeholder Advisory Group selects their topics and supports intersessional work; openness and collective authorship broaden knowledge but do not create independent corroboration.
• Institutional self-citation occurs when an IGF statement is used to validate a later IGF claim without a new evidentiary step. Such citations can establish chronology, mandate and the existence of a view. They cannot by themselves establish prevalence, causation, effectiveness or representative support.
• The IGF's own history contains useful safeguards: published calls, named submissions, draft review, questions about adverse effects, explicit limitations and records of agreement or dissension. These practices are uneven and need to become a common publication standard.
• Every consequential claim should carry a source class, conflict disclosure, independence assessment, contrary-case search and confidence judgment. Every material objection should receive a disposition: accepted, partly accepted, answered, unresolved or outside scope.
• The 2026 integration of the BPF model into Policy Networks creates an opportunity to preserve participatory inquiry while replacing institutional memory loops with claim-level evidence and durable dissent records.

A report is an outcome of participation, not proof that participation found the truth

Best Practice Forums occupy an unusual place in Internet governance. They are not negotiating rooms, standards committees or peer-reviewed journals. They are temporary communities asked to gather practice, compare experience and produce a document useful beyond the people who joined the discussion. The IGF's description of BPFs is careful on one point: the objective is to collect existing and emerging practices, not to develop new policy. The outputs are meant to inform policy discussions, standards development, business decisions and public understanding.

That is a demanding knowledge function. A reader may use a BPF report to decide which incident-response arrangement to fund, how to frame a cybersecurity norm, whether a connectivity intervention travels across regions, or what a technical community should do during conflict. The report may not bind anyone, but it can influence people who do. Its evidentiary burden therefore depends on the consequence of the claim, not on the non-binding status of the publication.

Multistakeholder authorship helps. A network operator can expose an implementation constraint that a lawyer missed. A rights group can identify people excluded by a technical remedy. A government official can explain legal authority. A company can disclose how a measure works at scale. An academic can distinguish a plausible mechanism from a demonstrated effect. No single profession has the whole picture.

Yet variety of affiliation does not automatically produce independence. Five contributors can belong to different stakeholder categories while relying on the same original assertion. A public meeting can hear many speakers who cite one another. A draft can receive broad praise without being tested by anyone who bears the cost of the recommended practice. The number of entities measures access and activity; it does not measure whether a factual claim survived a serious attempt to disprove it.

The basic distinction is simple. Participation tells the reader who had an opportunity to contribute and what views entered the room. Verification tells the reader why a proposition should be believed. A credible BPF needs both, and it should never use the first as shorthand for the second.

The 2014 innovation answered a real weakness in the annual forum

The BPF model was not created in search of a bureaucratic product. It answered a recurring criticism that rich IGF discussions were difficult to carry into later decisions. The 2012 Commission on Science and Technology for Development working group called for more tangible outputs while preserving the IGF as a non-binding forum. It also recommended that outcome documents map both converging and diverging opinions. That pairing matters: greater influence was supposed to come with greater fidelity to disagreement, not with a smoother institutional voice.

The IGF 2014 account records that earlier best-practice sessions had not been documented sufficiently because resources were lacking. In 2014, the MAG chose five topics, assigned lead experts, created mailing lists and began intersessional work. The first subjects included multistakeholder participation, unwanted communications, computer security incident response teams, local content and online child protection. Drafts were opened for comment, and the final documents reflected what the IGF described as the rough-consensus views of lead experts and contributors.

This was an institutional improvement. Knowledge gathered over months could be inspected, cited and challenged after the meeting. A written report makes assumptions visible in a way that a sequence of speeches does not. It allows later communities to compare practice across years. It can also reduce the advantage of people able to attend in person, because evidence and comments can be submitted before and after the annual event.

The same improvement created a new risk. Once an institution has reports, summaries, messages, meeting records and retrospective accounts, it possesses an expanding library about itself. Those documents are easy to find, carry official presentation and use familiar language. A later author can build an apparently dense reference list while rarely leaving the institution's own record. The resulting narrative may be accurate about what the IGF said and still weak about what happened outside it.

The problem is not quotation from the archive. Any serious history of the BPFs must use the archive. The problem begins when the archive crosses an unstated boundary: a record that a claim was made becomes evidence that the claim is true; a report that a practice was shared becomes evidence that it worked; a statement that participation was open becomes evidence that relevant interests were represented.

The BPF innovation should therefore be completed, not reversed. Tangible output needs tangible provenance.

The MAG is a steward of topics, not an independent verifier of every conclusion

The MAG's formal role explains why evidentiary responsibility cannot be left to institutional endorsement. Its terms of reference make programme design its central task. Members advise on themes, sub-themes and issues, select workshops, organise main sessions and support intersessional work. Members serve in a personal capacity but are expected to maintain links with stakeholder communities. Decisions seek rough consensus, with each member having an equal say.

For the BPFs, topic choice is consequential. Selecting cybersecurity norms rather than procurement concentration, or access in conflict rather than surveillance during reconstruction, does not determine the conclusion. It does determine where scarce coordination, staff time and attention go. The 2020 review of earlier BPFs observed that the number conducted in a year depended partly on the Secretariat's capacity to support them. In 2025, the IGF progress report stated directly that resources permitted four intersessional streams despite demand for more, and the MAG prioritised continuing areas.

This is ordinary agenda power. Every publication programme has it. The legitimacy question is whether the selection criteria, resource constraint and rejected alternatives are visible enough for readers to distinguish importance from availability. A topic can be important and still have been chosen from a narrow field. Continuation can reflect demonstrated value, a strong organising group, institutional familiarity or simply the feasibility of producing something within the year.

MAG involvement gives a BPF an authorised place in the IGF programme. It does not mean MAG members have replicated the evidence in the final report. Nor should they be expected to become subject experts for every claim. Their stewardship is strongest when it protects openness, balance, methodological candour and completion. Treating selection or endorsement as substantive certification would ask the MAG to provide assurance its design does not promise.

That boundary protects the MAG as well as the reader. A report can say that its topic was selected through the recognised IGF mechanism while identifying the authors and reviewers responsible for factual judgments. If an error is found, correction need not become a referendum on the entire programme. Authority to convene and responsibility for evidence remain connected but distinct.

Institutional self-citation has three forms, and only one is harmless

Self-citation is often discussed as though it were vanity: an organisation cites its own publications to inflate importance. The more serious governance problem is epistemic. An institution can unknowingly reuse the same assertion through documents with different titles until repetition looks like corroboration.

The first form is archival self-citation. A report cites an earlier IGF page to establish when a forum began, who selected a topic, what consultation occurred or what the institution formally said. This is appropriate. The organisation is usually the best source for its own documented act, provided the claim is limited to that act.

The second is derivative self-citation. A session report feeds a yearly summary; the summary feeds an IGF message; the message is cited in a BPF proposal; the final BPF report cites the proposal and message. Four documents appear, but there may be one underlying discussion. The chain can legitimately show how the issue moved through the institution. It cannot be counted as four independent confirmations of the substantive proposition.

The third is validating self-citation. An IGF publication points to earlier IGF outputs as evidence that an intervention is effective, a view is widely shared, an institution is legitimate or a problem is global. This is where the circle closes. The proposition needs observation outside the institution, a reproducible comparison, an independently collected record or at least a clearly identified external witness with relevant knowledge.

The distinction should appear in the report. Citations can be tagged in ordinary prose or a compact evidence table: institutional record, contributor testimony, operator measurement, legal authority, independent research, affected-community account, or contested estimate. Readers do not need a philosophical essay beside every footnote. They do need to know whether a reference establishes an official position or tests the world described by that position.

Without that label, polished institutional material receives an evidentiary premium. It is formatted, stable and easy to quote. Contrary experience may arrive as a short email, a transcript intervention or an account in another language. Good editorial practice should not allow presentation quality to decide epistemic weight.

The 2025 cybersecurity report shows how an institutional genealogy can become a support loop

The 2025 BPF on securing Internet access during conflict provides a useful, recent example because its report is unusually explicit about its origin. The final publication traces the topic from an IGF 2024 main session to the Riyadh IGF Messages, then to a BPF proposal selected by the MAG. The BPF developed a draft problem statement, sought comments, discussed it at meetings and the annual session, and revised its framing.

This sequence shows responsiveness. Feedback added the perspective of people directly affected by lost access, requested clearer treatment of human rights, broadened the temporal scope beyond active crisis and challenged the meaning of "core Internet resources." Those are substantive changes, not decorative consultation.

But the topic genealogy is not independent proof of the problem statement. The 2024 main session and the IGF Messages are connected outputs of the same annual forum. Their repetition can show that entities considered the issue urgent and that it persisted across IGF formats. It cannot establish, without another evidentiary step, the frequency of particular disruptions, which institutions caused them, which responses succeeded or whether a proposed neutral role is feasible in every legal and conflict setting.

The report partly recognises this boundary. Its section on crisis examples states that the examples are illustrations, not full case studies or detailed situation analyses. Its annex lists meetings, a public call and three written contributors to the draft problem statement. It points to legal instruments and external organisations. These disclosures let a careful reader see the difference between consultation and case verification.

What remains missing is a claim-level account. Which assertion rests on a entity's direct experience? Which was corroborated by network measurements, humanitarian records, court material or multiple independent observers? Which government, operator or affected community disputed the framing? Which examples were considered and excluded? A reader can reconstruct some of this from notes and references, but reconstruction should not be the price of understanding a high-consequence recommendation.

The lesson is not that the report is unreliable. It is that a transparent institutional genealogy should be followed by an external evidence genealogy. The first explains why the IGF asked the question. The second explains why the answer deserves confidence.

Official material is indispensable evidence of institutions and limited evidence of outcomes

A disciplined BPF would state what each source class can and cannot prove. Official material is strongest for mandate, procedure, formal position and recorded action. A government order can establish that an order was issued. A company transparency report can establish what the company reports under its stated method. An IGF meeting summary can establish that a view was raised and perhaps how the rapporteur characterised the discussion.

None is useless because it is self-authored. Institutions possess records unavailable elsewhere. The mistake is to extend the inference. A ministry's statement that a shutdown was necessary does not prove necessity. A platform's count of removed accounts does not prove social benefit. A technical body's deployment claim does not prove that small operators faced the same conditions. An IGF summary describing broad agreement does not establish the preferences of people who did not participate.

The required corroboration depends on the claim. Legal authority should be checked against the operative instrument and, where disputed, judicial or expert interpretation. Technical effect should be checked with measurements, implementation records and counterexamples. Social effect should include affected-community evidence and attention to selection. Prevalence needs a defined denominator and collection method. Causation needs a plausible comparison, not a sequence of testimonials.

This discipline also prevents false balance. Independent evidence does not mean finding one critic for every statement. A poorly supported objection remains poorly supported. Nor does external mean neutral. An academic paper can have a narrow sample; a civil-society report can select cases strategically; a commercial measurement service sees only its own network. Independence concerns the relationship to the claim and institution, not moral superiority.

The report should ask three questions of each important source. Did the source directly observe the fact? Does the source have an interest in the conclusion? Can another reader inspect the method or underlying record? Those questions produce more useful assurance than stakeholder labels alone.

The BPFs' own review identified organisational limits but was mainly an insider review

In 2020, the MAG commissioned a review of BPF experience from 2014 to 2019. The resulting BPF on BPFs report documented 26 forums and proposed definitions, selection metrics and practical improvements. It was candid about unclear purpose, short timelines, uneven dissemination and the difficulty of assessing impact only months after publication. It also recognised that lessons from unsuccessful approaches can be as valuable as a collection of successes.

This was useful institutional learning. It led to a modalities document endorsed by the MAG, clarified that BPFs normally collect practice rather than make new policy, and encouraged more systematic assessment of proposals and returning forums. The report did not merely celebrate output volume.

Its evidence boundary is equally important. The review says it reached people directly involved in organising and leading the BPFs: MAG facilitators, co-facilitators, lead experts, key contributors, Secretariat staff and consultants. Those people were well placed to describe coordination burdens, deadlines and institutional opportunities. They were less able, by design, to represent people who considered a BPF irrelevant, withdrew after disagreement, could not participate, or used a report and found it misleading.

An insider review can answer "How did we experience the instrument?" It cannot alone answer "Whose knowledge did the instrument exclude?" or "Did the report improve decisions outside the IGF?" The distinction becomes especially sharp when the entity under review is itself multistakeholder inclusion. Organisers may accurately report that a call was open; nonparticipants are needed to explain why the call did not reach them or why the expected return did not justify the effort.

The report considered citations, references and reuse as possible impact measures but noted that the interval before continuation decisions was too short. That caution was correct. Citation is delayed, and institutional documents frequently cite one another. A better later evaluation would separate independent uptake from reuse within the IGF family, and substantive adoption from ceremonial mention.

The BPF on BPFs should therefore be read as a strong operational self-assessment, not an independent evaluation of legitimacy or effect. Stating that limit does not diminish its recommendations. It tells later reviewers what additional evidence to collect.

Open contribution is a route to evidence, not a substitute for a sampling frame

The BPF model relies heavily on public calls, mailing lists, online meetings and annual sessions. These methods are appropriate for discovery. They can uncover cases an appointed research team would never find. They let people with practical experience identify terms, jurisdictions and failures that formal literature overlooks.

They also produce a self-selected set. People respond because they know the call exists, work in a language used by the forum, have time, see professional value, trust the convenor and possess material they can disclose. Organisations with communications staff can turn experience into polished submissions more easily than an affected individual or small operator. An open door does not equal a representative room.

The best participation research on the IGF reinforces this caution. Nadia Tjahja, Trisha Meyer and Jamal Shahin mapped on-site participation from 2006 to 2019, identifying 18,968 unique entities and 7,326 organisations; 10,000 people attended only once. The authors also found that evaluating the multistakeholder model from existing participation records is difficult. Their work does not prove that BPF contributors are unrepresentative. It shows why a list of names and broad stakeholder categories cannot answer that question by itself.

Research on the East Africa IGF similarly offers a bounded warning rather than a verdict on the global forum. A 2016 case study found capacity building and policy transfer alongside reinforcement of entities with stronger institutional endowments. The setting, period and regional body differ from a global BPF. The transferable mechanism is resource advantage: sustained participation, travel, professional time and network access can shape framing even when formal access is open.

A BPF should consequently publish a participation profile relevant to the question, not only the IGF's standard sectors. A report on civilian connectivity should show the presence of affected local groups, network operators, humanitarian responders, governments implicated in restrictions and organisations that challenge neutrality claims. A report on artificial intelligence should distinguish deployers, workers, model providers, public buyers and people subject to automated decisions.

No profile will prove representativeness. It will reveal missing knowledge and constrain the language of the conclusion.

Contribution counts need denominators, provenance and treatment records

Some BPFs publish useful detail about their calls. The 2019 local-content forum reported 36 contributions and opened a draft for comment. The 2019 cybersecurity forum listed submissions from companies, government bodies, civil society, technical organisations and individuals, then separately listed feedback on the draft. Its questionnaire expressly asked about implementation challenges, adverse effects and tensions between commitments and outcomes. These choices make the evidentiary route more inspectable.

Counts still need interpretation. Thirty-six contributions may be broad for a specialised question or narrow for a global claim. Seven organisational submissions can contain original data, repeated advocacy or both. A consultation can receive no opposing submission because there is no opposition, because the opposing party did not see value in participating, or because the question made its position difficult to express.

The report should therefore publish a compact contribution account. How many submissions were received? From which countries and relevant roles? Which were based on direct observation? Which supplied supporting material? How many were solicited through targeted outreach rather than volunteered? Which could not be published? How many distinct original cases were represented after duplicates and derivative accounts were grouped?

Treatment matters as much as receipt. A submission can be listed and ignored. Each material contribution should be traceable to a section, an editorial response or an explanation that it fell outside scope. This need not expose confidential sources or burden a short report with correspondence. An anonymised identifier and disposition are enough where protection is necessary.

Such a record would also improve future participation. Contributors could see that a careful objection changed a definition, narrowed a claim or remained unresolved. People are more likely to invest scarce time when the institution demonstrates that input has a path to effect.

The objective is not a vote count. Ten unsupported claims do not outweigh one reproducible counterexample to a universal proposition. The record lets the reader evaluate quality without mistaking silence for agreement.

Some BPFs already model the candour a common standard would require

The archive contains practices worth preserving. The 2017 Gender and Access report included a limitations section. It acknowledged that an online survey could not fully reach women and gender-diverse people without Internet access and that an English-only instrument created a language barrier. Those admissions directly constrain the population to which findings can be generalised.

The 2019 cybersecurity call asked contributors whether agreements produced adverse effects or tensions with their intended results. That is a valuable design choice because a best-practice exercise otherwise attracts success stories. The question creates permission to report harm and implementation failure, not only aspiration. Publication of named contributions and separate draft feedback also helps readers distinguish initial evidence from later review.

The 2016 stocktaking synthesis preserved criticism that some sessions needed to become more fact-based. It recorded calls for balanced management of intersessional groups to avoid capture, tracking of stakeholder participation, and reports that note agreement and dissension. These observations came from the IGF community, so they do not independently prove the prevalence of each problem. They do demonstrate that the institution has long recognised the risks.

The 2025 conflict-and-crisis BPF showed another good practice when it published the evolution of its problem statement and acknowledged that directly affected people were initially missing from the framing. This is more informative than a final text that appears to have emerged fully formed. It reveals correction.

The weakness is unevenness. A reader should not need to hope that a particular coordinating team chooses to publish limitations, ask for negative cases or preserve dissent. Those elements should travel with the BPF method regardless of topic. Flexibility is useful for deciding whether evidence arrives through a survey, case comparison, technical measurement or legal analysis. It is not a reason to omit basic assurance.

A common standard can be short. State the question, population and scope. Classify material sources. Describe contributors and missing perspectives. Publish limitations. Record material objections. Identify who made final editorial judgments. Preserve versions and corrections. Topic-specific methods can then remain genuinely flexible.

Dissent is not whatever remains in a comment archive

Multistakeholder reports often use inclusive prose: entities noted, stakeholders emphasised, the community recognised. These formulations are convenient when many people contributed. They can also erase the distribution and strength of disagreement. A reader cannot tell whether "some" means one unopposed technical clarification or a material objection from every affected local organisation present.

The 2012 improvements report asked IGF outcomes to map converging and diverging opinions. That obligation should reach BPF publications. A dissent record is not a list of every editorial preference. It concerns objections that, if accepted, would alter a conclusion, recommendation, scope, confidence level or description of affected people.

Each such objection needs a disposition. Accepted means the text changed accordingly. Partly accepted means a defined element changed. Answered means the editors rejected the objection and state why. Unresolved means evidence or agreement was limited public evidence. Outside scope means the objection matters but the forum did not examine it. Duplicate means the substance is addressed under another entry.

Names should be published with consent. The substance should survive even when the person needs anonymity. In sensitive areas, identifying a local operator, dissident or humanitarian source can create risk. The record can describe the role and evidence without revealing identity.

The final report should also distinguish dissent about fact, interpretation, value and remit. A entity may agree that a shutdown occurred but dispute its measured extent. Another may accept the effect but reject the legal conclusion. A third may argue that the BPF should not recommend institutional neutrality. Compressing these objections into "views differed" wastes the information they provide.

Dissent improves rather than weakens a non-binding forum. Decision makers can identify where further evidence is required and avoid citing the report as a consensus it never claimed. Future forums can test the unresolved point. Minority reasoning may become the most valuable section when technology or geopolitical conditions change.

A claim register would expose circular support without turning the report into an audit manual

The most practical reform is a claim register for consequential propositions. It can sit behind the narrative as an annex and remain readable. Each row should contain the claim, claim type, geographic and temporal scope, source class, independence assessment, contrary evidence, known conflicts, confidence and editorial owner.

Claim type matters. A descriptive claim states what occurred. A prevalence claim says how common it is. A causal claim links an action to an outcome. An effectiveness claim says a practice works. A normative claim says it should be preferred. An institutional claim says a body is authorised, representative or accountable. Each needs different support.

The independence assessment should follow the information, not the URL. If five IGF documents derive from one session, they form one institutional evidence family. If an independent paper relies entirely on those five documents, it does not add observation merely because it has a university affiliation. Conversely, a company measurement can provide independent evidence of an IGF proposition if it collected the data separately, while its commercial interest is disclosed and the method is inspectable.

Contrary evidence should include failed practice, non-adoption and alternative explanation. If a report says a multistakeholder incident arrangement improved response, it should ask where similar arrangements did not improve response, whether better funding explained the difference, and what happened where a single public authority acted instead. A best-practice label is a hypothesis about transfer, not a prize for a good story.

Confidence should be tied to scope. A practice can be well supported in a defined operator class and uncertain globally. This is better than flattening every conclusion into cautious but vague prose. Readers can use bounded evidence.

The narrative remains an article, not a spreadsheet. The register is the scaffolding that prevents rhetoric from hiding unsupported transitions.

Negative-case search must be designed before the success stories arrive

Calls for "best practices" predictably attract organisations proud of their work. The label selects on claimed success. If the forum waits until drafting to seek failures, it will have too little time and fewer relationships with disappointed entities.

Every BPF should begin with a negative-case plan. For each proposed practice, identify what failure would look like, which groups are likely to observe it, and where evidence might exist. Contact implementers that discontinued the measure, jurisdictions that chose another model, users who could not access the benefit, and oversight bodies that reviewed complaints. Ask successful contributors to disclose failures, cost, prerequisites and groups not served.

The plan should also distinguish absence of evidence from evidence of absence. A small operator may lack measurement capacity. A government may not publish an adverse review. A community may experience harm without a recognised reporting channel. These gaps lower confidence; they do not automatically prove failure.

Case selection needs a denominator. If the forum presents six successful computer-response teams, how many candidates were considered? Were the six chosen because documentation was available, because they were recommended by organisers or because they represented variation in income, scale and legal setting? A transparent selection note can be only a paragraph and still prevent showcase cases from being mistaken for a success rate.

Where direct comparison is impossible, the report can use mechanism testing. State why the practice is expected to work, then look for each required condition. A coordination forum may depend on legal permission to share incident data, stable staffing, trusted contacts and a protected communications channel. If a case lacks one condition yet succeeds, the theory needs revision. If it has all conditions and fails, the report should not preserve the recommendation by changing the definition of success.

Best-practice research becomes valuable when reality is allowed to embarrass the category.

Conflict disclosure should follow claims, editorial power and institutional benefit

Stakeholder affiliation is not a sufficient conflict declaration. A entity listed as civil society may receive project funding from a company affected by the report. An academic may advise a government. A technical expert may have designed the practice under review. A MAG facilitator may have advocated continuation of the BPF. None of these relationships disqualifies the contribution. They change how independence is assessed.

Reports should disclose interests relevant to the claim: funding, employment, authorship of a cited initiative, litigation, procurement, regulatory responsibility and organisational advocacy. The disclosure should cover coordinating editors and external reviewers, not only people quoted in the report.

Institutional benefit also matters. A BPF report may conclude that more intersessional work, funding or IGF coordination is needed. That conclusion can be correct. Because the producing institution benefits from greater relevance and resources, it needs support outside its own strategic documents. A recommendation for the IGF to do more should specify the observed gap, alternatives considered and measurable result expected.

This is where institutional self-citation and conflict meet. An IGF report cites an IGF review to show that an IGF format has impact, then recommends expansion of that format. The chain is not fraudulent. It is incomplete. Independent user evidence, documented adoption outside the IGF and critical evaluation would make the recommendation stronger.

Conflict policy should protect contribution rather than shame affiliation. Internet governance expertise often comes from people inside the relevant institutions. Excluding them would discard essential knowledge. Disclosure allows readers to distinguish observation from advocacy and gives editors a reason to seek balancing evidence.

The key question is not "Was the author interested?" It is "Could the claim be checked without relying on the author's interest?"

External review should test the strongest claims, not certify the institution

An independent review panel for every BPF would be expensive and could reproduce elite access in another form. A lighter model is possible. Before publication, the coordinating team should identify the five to ten claims most likely to influence decisions or cause harm if wrong. Reviewers who did not participate in drafting then attempt to verify those claims and find contrary evidence.

Reviewers should be selected for method and subject knowledge, with conflicts disclosed. At least one should come from a population affected by the practice, not only from a recognised global organisation. Where language or local context is material, review cannot be performed entirely from Geneva, New York or the annual host city.

The review result should be public: confirmed within scope, narrowed, contested, unsupported or not reviewable. Editors retain responsibility for the final text and may disagree with a reviewer, but the reason should be visible. This is not peer review in the journal sense, and it should not be advertised as such. It is an adversarial check on a limited set of claims.

The review should also trace citation families. A reference list of fifty items may collapse into three original observations. Tools can help identify duplicate and derivative sources, but judgment remains necessary. Two reports may use different data; two apparently independent articles may quote the same press release.

Publication should not wait for impossible certainty. A BPF can report that evidence is mixed, local or incomplete. The danger lies in giving provisional knowledge a collective institutional voice without preserving its limits.

Impact should be measured by changed decisions and corrected claims, not institutional citation volume

The BPF on BPFs considered references, citations and reuse as possible signs of impact. These metrics are easy to count and potentially misleading. IGF summaries cite BPF reports; later BPFs cite earlier ones; partner organisations mention the IGF in support statements. The institution can generate a large citation footprint without showing that anyone changed a decision because the evidence was useful.

A better impact review begins with intended users. Which regulator, operator, standards group, company, community organisation or international body was expected to use the report? Did it cite a specific finding, alter a practice, fund a measure, reject an option or request more evidence? What happened next? Adoption is not automatically success; an adopted recommendation can fail.

The review should also count corrections. Did external users identify factual errors? Were claims narrowed after new evidence? Did a dissenting case later become central? An institution that publishes visible corrections may appear less flawless and be more reliable.

Independent citation remains useful when classified. A report quoted by scholarship, judicial material, regulatory analysis or operator guidance has travelled beyond its own family. The context matters: a citation can criticise rather than endorse. Reuse of text is different from use of evidence.

The time horizon should match the claim. A report released in November cannot demonstrate global policy impact by the next selection meeting. Continuation decisions can instead assess process quality and whether the next year asks a genuinely new question. Longer reviews can assess uptake and outcome.

This approach reduces pressure to overstate immediate success. A BPF can be renewed because the evidence remains incomplete, not because its first report has already transformed policy.

The move into Policy Networks is a chance to carry forward inquiry and retire the circle

The IGF states that standalone BPFs operated through 2025 and that, beginning in 2026, the BPF model was integrated into Policy Networks to simplify intersessional structures and strengthen coordination. Administrative consolidation does not solve the evidence problem. It may even increase derivative citation if longer-running networks build larger internal archives and familiar contributor groups.

It also creates an opportunity. A network continuing across years can maintain a claim register, revisit unresolved objections and plan negative-case research before publication deadlines. It can separate an annual update from a cumulative evidence review. It can invite independent reviewers early and track whether recommendations worked.

The MAG should make evidence quality part of selection and renewal. Proposals should identify the question, affected populations, likely evidence classes, conflicts, contrary-case plan and method for recording dissent. Returning networks should show which earlier claims were confirmed, corrected or abandoned. Citation counts should be separated into IGF-family and independent use.

Resource allocation must support this standard. A small coordinating team cannot conduct global verification, translation, outreach and editing merely because participation is voluntary. The MAG should choose fewer claims when resources are constrained rather than preserve breadth through unverified synthesis. Limitations are more legitimate than invisible shortcuts.

The resulting publication need not sound academic. Clear narrative, case detail and practical recommendations remain the point. Evidence discipline improves readability because it tells the reader what is known, by whom, where and with what uncertainty.

The strongest multistakeholder report is one that shows where the stakeholders did not settle the matter

Institutional self-citation is not unique to the IGF. Governments, companies, universities and civil-society organisations all build narratives from their own records. The risk is acute for Best Practice Forums because collective authorship can be mistaken for external validation and because their outputs are expressly designed to travel into other decision venues.

The remedy is neither suspicion of every IGF citation nor a demand for unattainable neutrality. It is a visible division of labour among sources. IGF records establish what the institution convened, heard, selected and published. Contributors provide experience whose basis and interest should be described. Independent research and external records test prevalence, mechanism and effect. Affected people identify consequences that institutional categories may miss. Dissent shows where synthesis stopped.

This standard honours the original purpose of the BPFs. They were meant to collect practice and make the IGF's knowledge useful between annual meetings. A report that states its limits, records failure and preserves disagreement is more useful than one that converts an open conversation into a seamless voice.

The MAG's role is not to certify universal truth. It is to steward a credible space in which claims can be assembled, challenged and corrected. That requires topic transparency, adequate support and publication rules that do not reward the most institutionally fluent source.

The decisive test is what happens when a report encounters a serious objection. If the objection disappears into meeting history while the final text cites the institution's earlier confidence, multistakeholderism has become a citation loop. If the objection changes the scope, lowers confidence, prompts a search for contrary cases or remains visible as unresolved, participation has produced knowledge rather than merely agreement.

Best practice is not what a respected forum says often enough. It is what survives contact with evidence from beyond the forum, including evidence supplied by people who do not share its conclusion.