Scientific Advisory Institutions

Scientific advisory institutions are the machinery by which a government obtains technical analysis it cannot generate internally, and by which that analysis is made to count. The category is broader than it looks. It includes legislative technology-assessment offices such as the UK's Parliamentary Office of Science and Technology and the defunded United States Office of Technology Assessment; chief scientific advisers, from Canada's Chief Science Advisor to the seven-member Group of Chief Scientific Advisors that the European Commission built after abolishing its single-adviser post; chartered academies working to commission, including the Council of Canadian Academies; emergency bodies convened under statute or under pressure, from SAGE to the World Health Organization's International Health Regulations emergency committees; regulatory expert committees such as the United States Food and Drug Administration's; and the intergovernmental assessment platforms, the IPCC and IPBES, whose distinguishing feature is that governments approve the summary line by line before it is published.

What unites them is not the science. It is the advice relationship: a body with expertise and no mandate producing a document for a body with a mandate and less expertise, under conditions where the second body may adopt it, ignore it, delay it, renegotiate its wording, dissolve the first body, or all five in sequence. The premise under test is the ordinary one — that independent expert analysis, formally structured, improves the decisions that follow, and that where advice fails the fix is more independence, more breadth and more transparency.

This brief keeps its predecessor's central finding and substantially changes the account of how advisory institutions fail. Outcome measurement still does not exist anywhere in this field, and the most authoritative statements of that come from the advisory establishment itself and from a Canadian government evaluation of a Canadian academy. What the earlier version of this brief did not have was the mechanism. Advisory bodies are very rarely destroyed by a decision-maker reading their advice and rejecting it. They are defunded, closed by administrative instrument, exhausted in procedural negotiation, deferred, or left running with no instrument that could tell anyone whether they work. Every well-documented case in the 2024–2026 record is one of those, and none of them is defiance.

The scope boundary with the sibling slot is worth stating precisely, because the two collide on a single phrase. Scientific governance models owns every decision science makes about itself with no external decision-maker in the loop: grant peer review, editorial decisions, funding lotteries, registered-reports policy, misconduct adjudication, research assessment exercises. Its unit of analysis is the allocation or gatekeeping rule. This brief owns every case where scientific knowledge is packaged for a decision-maker outside science, and its unit of analysis is the advice relationship — mandate, independence, the transmission of uncertainty, and what happens when the advice is inconvenient. The term “science advice” belongs here and the sibling brief does not use it; the terms peer review, replication rate and retraction belong there and this brief does not use them as load-bearing terms.

Three cases sit near that seam and are claimed here for stated reasons. Canada's Model Policy on Scientific Integrity looks like research-integrity policy but regulates a government scientist's right to speak to ministers and the public without pre-approval, which is the supply side of the advice relationship. Conflict-of-interest audits of regulatory and academy committees are ours rather than the sibling's because the output goes to a regulator or a legislature. And the 2025–2026 closures of United States federal advisory committees are ours because the finding they generate — that statutory bodies survive abolition and non-statutory ones do not — is a finding about advice relationships rather than about science's internal rules.

Established Start with the sentence the field's own leading body wrote about the field. A national research council consensus study on the use of science in policy concluded that “to date, there has not been much success in explaining the use of science in public policy”, and that “we lack systematic evidence as to whether these steps are having the results their sponsors hope for.” Its recommendation was to launch a research agenda rather than to report findings from one. That an institution assessing a field that includes its own product reached a negative finding makes it more credible, not less — and fourteen years later the finding has been independently reproduced by a government auditing a national academy it funds.

Established Canada supplies the cleanest version of that result, and it is an evaluation the Canadian government commissioned about a body it pays for. Innovation, Science and Economic Development Canada's evaluation of the Council of Canadian Academies covers 2005–06 to 2016–17: a $30 million upfront grant in 2005 for ten years, $3 million a year from 2015, total expenditure of $46.5 million averaging $3.9 million a year, 36 completed assessments, standard assessments averaging 27 months against an 18-to-24-month target, an average of 13 panellists and 11 reviewers per assessment whose volunteered time was valued at $407,400 each, and 86% of assessments cited in the academic literature with over 500 direct and 4,000 indirect citations across 43 countries. And then the sentence that matters: “the results and impacts of CCA assessments are difficult to measure and are not tracked by ISED.” The department's only instrument is an optional, voluntary sponsor feedback survey a year after release. Everything except the outcome is measured, and measured well.

Established The single largest fact about advisory institutions in 2026 is not about advice quality at all. It is that the world's largest advisory system has been cut by roughly a third in eighteen months. Following a February 2025 executive order directing agencies to eliminate advisory committees, more than 100 committees advising United States science agencies were terminated. The Department of Health and Human Services, including the National Institutes of Health, lost 77 of about 280 — over a quarter — against two terminations in the whole of 2024. The National Science Foundation lost 14 of 52, the Department of Energy six, and NASA 6 of 11. Activity fell with the count: HHS open meetings dropped from a ten-year average of 255 a year to 91 in 2025, and NASA committee meetings from 35 to 10. Seventeen members of the vaccine advisory committee were dismissed and the committee reconstituted; all 22 members of the National Science Board were dismissed in late 2025.

Established Buried in that record is the most transferable design finding in the subject. Statutory committees required by Congress could not easily be eliminated; non-statutory ones were closed by agency leadership. Read that against the canonical case from the other direction — the United States Office of Technology Assessment was never abolished, its authorising statute remains in force, and only its appropriation was zeroed — and the constraint specifies itself. A statute protects an advisory body from abolition and not from defunding. An appropriation keeps it alive and does not survive a change of majority. Nothing in the observed record protects both at once, and the bodies that have lasted longest are the ones whose agenda no single majority controls.

Established The best-documented pandemic case does not show a government ignoring science; it shows a government following it. A joint parliamentary inquiry found “a degree of groupthink” among advisers and officials alike, overreliance on models with insufficient data, and incorrect assumptions about public compliance — actual compliance exceeded what was modelled. Of 87 participants in the emergency advisory group, exactly one was from outside domestic institutions. The government chief scientific adviser told the inquiry there was “no significant disagreement between the Government and its scientific advisers on anything material”, while the health secretary said of an early judgement: “I bitterly regret that I did not overrule that scientific advice.” The political failure was excessive deference, not defiance.

Frontier And the remedy adopted in 2026 does not address that. The UK government's response of 25 March 2026 to the Covid-19 Inquiry's Module 2 report accepted every recommendation touching scientific advice: devolved chief scientific advisers invited to the emergency group from the outset of an emergency, a refreshed expert register during 2026 with open calls held out as a longer-term ambition, publication of technical advice “at the earliest opportunity” subject to commercial confidentiality, personal safety, national security and legal privilege, and reviewed terms of appointment by 2027 including compensation for institutions whose staff commit extended time. Every accepted recommendation addresses breadth, timing or compensation. None addresses deference. That may still be the right package; it is not responsive to the finding.

Established Where independence has been externally audited, the audits mostly find that the disclosure regime is not auditable. A peer-reviewed examination of a twenty-member national-academy committee on genetically engineered crops identified six members with financial conflicts — thirty per cent — comprising ten distinct conflicts, none disclosed, against the academy's statement that it had identified none. A national audit office looking at food-and-drug regulator advisory committees in March 2026 reached a structurally identical conclusion by a different route: across June 2018 to May 2025 it counted 15 recusals, one financial-conflict waiver and seven authorisations despite appearance issues, and objected not to those numbers but to the fact that conflict-of-interest guidance legally required more than 13 years earlier had never been finalised, that no public account exists of how financial conflicts are determined, and that there is no published guidance at all for guest speakers. Fifteen recusals in seven years is a low number. Nobody outside can tell whether it is low because there are few conflicts or low because few are found.

Frontier The forward-looking half of the conflict problem has been quantified once, and by journalists. An investigation matched 107 physician advisers to that regulator's committees over 2013–2016 against the federal payments database and reported that 40 later received benefits, 26 received over $100,000 from the maker of the product they had voted on or from competitors, seven received more than $1 million, and one received $1.9 million from a company after voting to approve its drug — none of it reported by the agency, because the disclosure regime looks only backwards. This brief could not open the investigation itself and takes these figures from a secondary summary of it, which is why they are flagged frontier and named rather than relied upon.

Established The transparency reform that every inquiry recommends is now well supported by experiment — and the experiments do not support the reason it is always sold. Two pre-registered UK randomised trials totalling 3,962 participants compared official persuasive text with balanced text disclosing risks, benefits, uncertainties and evidence quality. On COVID-19 vaccination (n = 2,928) there was no significant difference in rated trustworthiness of the information (5.28 against 5.39, p = 0.21, d = 0.08) or its producer (5.16 against 5.28, p = 0.14, d = 0.09). On nuclear power (n = 1,034) the balanced text scored significantly higher on both (4.70 against 4.92, d = 0.17; 4.35 against 4.63, d = 0.20; both p < 0.01). In both, the advantage was concentrated among participants with negative or neutral priors — and in both, the persuasive version was better at persuading, with the nuclear study finding significantly higher policy support in the persuasive condition despite its lower trustworthiness. Transparency is not a trust-maximising move. It is a trust-redistributing move, and an advisory body that adopts it expecting to become more popular will conclude within a year that it did not work.

Established Which is the stronger case for it, once the second and third results are added. A German experiment (N = 800) presented a public-health authority's statement under four conditions and then told participants it had been wrong: communicating uncertainty buffered the loss of trust, and explaining the uncertainty did no harm. A pair of pre-registered United States experiments (n = 600 and n = 1,001) found the effect of uncertainty on trust is a crossover, not a main effect: where the evidence agreed with a person's prior beliefs, adding uncertainty reduced trust in the information and the source; where it disagreed, it raised both. The interaction terms were large and consistent — β = −0.34 (95% CI −0.48 to −0.19) and −0.26 (−0.40 to −0.11) in the first study, −0.40 and −0.41 in the second — with a strong main effect of belief–evidence consistency (β = 0.58 and 0.50) and no evidence of polarisation in trust or belief. What publishing your reasoning reliably buys is reach into the sceptical part of the audience and insurance against the day the advice changes. Those are the two things the pandemic record shows were missing.

Four questions in this subject are genuinely open. Two of them have partial empirical answers that point in uncomfortable directions, and two do not have answers at all.

Frontier First: should advice speak with one voice or represent its own disagreement? The public-reception evidence exists and it is not encouraging for pluralism. Across three 2×2 factorial experiments (N = 320), participants read a scientific claim and a disagreeing source whose expertise and financial interests were varied. Expertise mattered robustly in all three: the original scientist was rated more trustworthy when contradicted by someone lacking relevant expertise (d = 0.39, 0.26 and 0.27, p < .001 throughout). The corollary is that expert disagreement measurably reduces trust in the contradicted claim. Publishing dissent therefore has a real and measured cost in immediate credibility, against an unmeasured benefit in accuracy and in resilience when advice changes, and no study in this file weighs the two against each other.

Established The same experiments contain a finding that inverts advisory design priorities. The disclosed vested interests of the disagreeing party had no significant effect on trust in the original scientist or on belief in the claim in two of the three samples; the effect appeared only among students. Scientific literacy increased sensitivity to expertise in all three studies but moderated the conflict effect only in that student sample. The authors' summary is that “expertise exerted a greater influence than conflict on judgements of trust” and that “vested interests of a disagreeing individual exert little to no influence over evaluations.” Advisory institutions spend nearly their whole procedural budget on conflict screening and almost none on making expertise gradients legible. Either the reallocation is overdue, or conflict screening exists to protect the institution's internal decision rather than the public's reception of it — in which case it is working and simply is not a communication instrument. Both readings are consistent with the evidence and this brief does not choose between them.

Established Second: how should deep uncertainty be carried into advice? The calibrated vocabulary the assessment bodies actually use is systematically misread, and the best available fix is small. A study of 556 respondents (from 841 invited, a 66% completion rate) randomised to three presentation conditions tested four IPCC probability terms against eight sentences drawn from IPCC reports. Interpretations regressed toward the middle in every term and in both directions: very unlikely, defined as below 10%, was read at a mean of 41%; unlikely, below 33%, at 44%; likely, above 66%, at 54%; very likely, above 90%, at 62%. A dual verbal-and-numerical format raised consistency with the guidelines from 20.76% to 30.12%, cut out-of-range responses from roughly 80% to 70%, and raised the median rank-order correlation with the intended ordering from 0.33 to 0.67. Even the best format tested left about seven readers in ten outside the intended range. A precision that exists only inside the institution is not precision.

Frontier Third: capture and the revolving door, where the forward-looking half of the problem is essentially unregulated. Every disclosure regime described in this brief is retrospective and self-certified: it asks what an adviser has received, not what an adviser might receive. The one systematic look at post-decision payments produced the largest figures in the file, and its own recommendation — non-compete terms for advisers and agency staff — has not been adopted by any body this brief could verify. The academy audit and the audit-office review land on the same structural point from opposite directions: one found undisclosed conflicts against an institutional denial, the other found a disclosure process that no outsider can inspect. Neither found a way to test whether the advice was actually distorted, which is the question everyone means to be asking.

Frontier Fourth, and least tractable: the accountability problem. An adviser carries influence without a mandate, and nobody has specified what follows. If advice is treated as authoritative, a decision-maker who defers has transferred responsibility to people who were not elected and cannot be removed — which is the pandemic case, on the chief adviser's own account. If advice is treated as one input among several, the formal apparatus is decoration and its independence guarantees protect nothing that matters. The UK's Principles of scientific advice to government, published 24 March 2010, tries to split the difference by requiring government to “publicly explain the reasons for policy decisions, particularly when the decision is not consistent with scientific advice”. That duty bites only in the rare case where a government openly departs from advice. It is silent on the common case, which is a government that adopts advice and is harmed by it.

Handwave What merely sounds open: whether the canonical defunded assessment office should be revived. The independent body commissioned to answer that question recommended against it, proposing to strengthen existing services instead. The statute still stands, so the barrier has been a single appropriation line for thirty years. Restoration bills exist and the most recent one this brief could verify dates from 2 February 2022 — it would amend the surviving statute to allow briefings without board review, let any member request an assessment, make board appointments bipartisan, add a rotator programme and require coordination with the existing research and audit services. Whether it has been reintroduced since is unverified here. The question is live as politics and closed as analysis.

Established The first bottleneck is existence, and this is new since the previous version of this brief. More than 100 advisory committees to United States science agencies were terminated following a February 2025 executive order directing agencies to eliminate them; one department lost 77 of about 280 against two the previous year, and its open meetings fell from a ten-year average of 255 to 91. A question about whether expert analysis improves decisions presupposes that the analysis is produced, by a body that exists next year. In the largest advisory system in the world, that presupposition is currently the contested one.

Established The second is that statutory protection and appropriated funding are separable, and each defends against a different attack. Non-statutory committees were closed by agency leadership in 2025 and statutory ones could not easily be; the canonical assessment office's statute survived its defunding intact. Nothing observed in this record protects a body against both abolition and defunding at once, and the design that has lasted longest works by removing the motive rather than adding armour: a work programme chosen by cross-party consensus has run since 1990 and had its operating contract unanimously renewed to 2028, while an agenda set by committee chairs did not survive a change of majority.

Established The third is that there is still no counterfactual anywhere in this literature. What gets measured is awareness, usage, satisfaction, output counts, citations and process compliance. A parliamentary science office's own evaluation reported that 96% of respondents had heard of it and over 80% consulted its briefings more than once a year, on 52 interviews achieved from 457 contacted — an eleven per cent response rate. A national academy's funder measured spending, throughput, volunteer value and citations across 43 countries and then recorded that impacts are “difficult to measure and are not tracked”. Every measured quantity is an input, an output or a perception. Not one is a policy outcome. The problem is not that evaluation is unfunded; Canada funded one and it found the gap anyway. The problem is that everything except the outcome is cheap to measure.

Frontier The fourth is procedural exhaustion at the approval stage. A summary scheduled for sixteen hours of negotiation that takes thirty-one, with lead authors gone before the vote, is a body spending its scarcest resource — the attention of the people who wrote the assessment — on the wording of the last document rather than on the analysis. The same pathology appears at the level of a whole institution: an intergovernmental panel meeting in March 2026 could not agree a delivery date for its next assessment at all, deferred the question, established an intersessional group and then abandoned it without consensus, and deferred its own mandated principles review into a dispute about the scope of the review.

Frontier The fifth is timing, which converts publication into a historical record rather than an input to scrutiny. The record of the meeting that ended a global health emergency on 4 September 2025 was published on 30 October 2025, nearly two months after the Director-General had acted on it. Pandemic advice in the best-documented national case was published roughly four months into the emergency. The UK's 2026 commitment is to publish “at the earliest opportunity”, which is a standard without a clock. A body that publishes its reasoning after the decision has produced an archive, not accountability.

Established The sixth is that rights and duties written into advisory policy are frequently unmonitored by construction. Canada's Model Policy on Scientific Integrity, first issued in 2018 and updated in February 2026, gives federal scientists the right to speak about their work “without approval or pre-approval” and defines interference as “any action that alters or suppresses the work or the impartiality of a scientist or researcher.” Its own status report found that of 25 required departments and agencies, 23 had a policy in force — and six had no process for investigating breaches, seven no training measures, seven had not told external collaborators the policy existed, and, in the report's own word, most significantly, nineteen had yet to implement a monitoring plan to assess whether it works. The right exists on paper in almost every department and is checked in almost none.

Established Nothing in this subject waits on a research result in the natural sciences. Every constraint identified here is institutional, procedural or financial, which is unusual on this map and is the correct reading of the evidence.

Established What it waits on, in order. An agenda-setting rule that does not make the body the property of whoever currently holds a majority — the office selected by cross-party consensus has run since 1990 and had its contract unanimously renewed to 2028, while the one commissioned by committee chairs did not survive a change of control. A form of protection that covers abolition and defunding simultaneously, which nothing in the observed record provides: statute stopped the 2025 closures and did not stop the 1995 defunding, and appropriation does the reverse. Conflict-of-interest disclosure that is externally auditable rather than self-certified, since the two independent audits available found undisclosed conflicts against an institutional denial in one case and an uninspectable process in the other. And a publication clock rather than a publication commitment, because “at the earliest opportunity” produced a two-month lag on the record of a decision that ended a global emergency.

Frontier One genuine methodological dependency, unusually for this category. An outcome measure. Until somebody constructs a defensible counterfactual — a matched comparison of decisions with and without formal assessment — the field cannot say whether any of this works, and its funders will keep buying citation counts because citation counts are what can be bought.

Established What depends on it. Every brief on this map that assumes competent technical policy assumes this machinery generically rather than by a typed edge. The strongest specific dependence runs to technocracy and democracy, which is the same accountability problem stated as a constitutional question rather than an institutional one, and to geoengineering governance, where an intergovernmental assessment body has already been asked to produce advice on solar radiation modification and the approval mechanism described here would apply to it.

Frontier The experiment nobody has run is still the obvious one: vary the advice, hold the decision-maker constant. Legislatures and executives make hundreds of technical decisions a year and commission formal assessments for a small, non-random subset. A matched comparison of decisions with and without a formal assessment, controlling for salience and contestedness, is difficult but not impossible, and it would be the first outcome evidence in the field. Thirty-eight years after the founding of the largest assessment body in the world, the outcome literature remains empty, and this brief's research did not find the exception.

Established The natural experiment now running is the largest ever conducted on this question, and nobody designed it. More than a hundred United States federal science advisory committees were terminated in 2025 while comparable committees in other jurisdictions were not, and while statutory committees within the same agencies survived. That is a difference-in-differences design with an assignment rule — statutory or not — that is plausibly independent of committee quality. The outcomes are observable: regulatory decisions, grant portfolios, agency guidance. The window for a before-and-after baseline is closing and this brief records that no such study is known to it.

Established The cheap experiment has been run once and never repeated. Externally auditing an academy committee's conflict-of-interest process found six undisclosed conflicts among twenty members. Doing it again, on a different committee, in a different country, would cost a fraction of one assessment's budget. Nine years later, this brief could find no replication. What has happened instead is a national audit office reaching the same structural conclusion about a regulator's committees by inspecting the process rather than the members — guidance legally required 13 years earlier and never finalised, no public account of how financial conflicts are determined, no guidance at all for guest speakers.

Frontier The prospective-conflict experiment is defined, unfunded and would take four years. Match advisers on a regulatory committee against a payments database for the four years after their votes rather than the years before, prospectively and by pre-registration rather than retrospectively and by journalism. The one retrospective pass over 107 advisers produced the largest numbers in this brief and the smallest amount of institutional response. A prospective version would establish whether the association survives controls for specialty and seniority, which the journalistic version cannot.

Established Three experiments on the communication side are already done and their results should change practice. Balanced, uncertainty-disclosing text is rated as trustworthy as persuasive text or more so, across 3,962 participants in two pre-registered trials. Disclosing uncertainty in advance protects trust when advice later proves wrong (N = 800). And the effect of uncertainty on trust is a crossover rather than a main effect, gaining among those whose priors disagree with the evidence and losing among those whose priors agree (n = 1,601), with no polarisation. The missing experiment is the institutional one: nobody has randomised an actual advisory body's output format and measured uptake by an actual decision-maker.

Frontier A natural experiment with a fixed reporting date is already scheduled. A national government committed in March 2026 to refresh its expert register during 2026, to explore open calls, and to revise advisers' terms of appointment by 2027. Those are dated, auditable commitments made in response to a published inquiry. Whether the register is refreshed, whether open calls appear, and whether the next emergency's advisory group is more institutionally diverse than the one that had a single participant from outside domestic institutions out of eighty-seven, are all checkable in 2027 and 2028 without anyone building anything.

The design choices that determine whether an advisory institution survives and whether its reasoning is inspectable are procedural, few in number, and well enough documented to tabulate. Five of them recur: who may commission the work, who approves the output, whether the reasoning is published, whether internal disagreement is recorded, and what protects the body's existence.

Established The approval row is where the differences bite hardest, and the intergovernmental case is the extreme. IPBES-11, held in Windhoek from 10 to 16 December 2024 with over 800 participants from more than 120 member states out of a membership of 147, scheduled 16 hours for approval of the Nexus Assessment summary and took over 31. Many lead authors had left before the text was approved. The recorded disputes were terminological: protection against conservation, whether armed conflict contributes to climate change, and the word polycrisis, which several member states disavowed. A Friends-of-the-Chair group cleared the last of it in thirty minutes. The assessment underneath — 165 experts from 57 countries and over 6,500 references — was not renegotiated; only its summary was. Participating authors write that the negotiations were “marked more by national interests than by science” and that “three years of work with its original scientific language was weakened by bureaucratically-driven wordsmithing.”

Established Who does that negotiating is measurable and it is not evenly distributed. A peer-reviewed comparison of IPCC approval sessions across the fifth assessment report (in person, 2013–14, 134 countries) and the sixth (virtual, 2021–22, 147 countries) found virtualisation raised average delegation size by about two, with much larger jumps for wealthy states — the United States to 25.3 delegates and Canada to 23, increases of roughly nineteen and twenty. Engagement, measured by mentions in the standard negotiation bulletin, was concentrated: India, Saudi Arabia and the United States each accounted for around a third of mentions in a working group, and the 30 largest delegations for 74 to 80% of all mentions. Several large delegations were largely silent. The mechanism of exclusion is banal: for two of the three working groups, 92% of Vanuatu's negotiating hours fell outside 09:00 to 17:00 local time, with Pacific delegations generally above half, and time zone significantly dampened the relationship between delegation size and engagement.

Established The publication row is where the WHO case stands out, and the improvement is dated and measured. A document analysis of 101 emergency committee statements from April 2009 to November 2023, covering 10 committees of which seven declared an emergency and three did not, found mean statement length rising from 1,020 words before COVID-19 to 2,992 after, with a range from 217 words in 2009 to 3,500 in 2023. COVID-19 statements were the first to carry ethics rules and conflict-of-interest declarations. The mpox statements introduced structured deliberative-session sections and were the first to lay out internal disagreement in public, including disagreement between the committee and the Director-General. At the fifth mpox meeting, held 4 September 2025 with 11 of 16 members and two advisers participating, the record states that the committee agreed the emergency was over “except for one Member, who expressed concerns regarding future access to vaccine.” One dissent out of eleven, published with its reason attached. No national chief-adviser system or academy described in this brief matches that standard.

The nearest neighbour is scientific governance models, and the two briefs split a single subject by who is in the room. That brief owns every decision science makes about itself with no external decision-maker present — how grants are ranked, what gets published, how misconduct is adjudicated. This one owns every case where knowledge is packaged for someone outside science who will act on it. The seam is the phrase “science advice,” which belongs here; the sibling reserves peer review, panel assessment and editorial decision. Read together the two make an argument neither makes alone: the same expert community is measured, on its internal decisions, as unable to rank the middle of a good field, and, on its external advice, as unable to demonstrate that its advice changed anything.

Technocracy and democracy is adjacent in the strong sense. This brief's fourth open question — what authority expertise has over a decision it did not make, and who is accountable when a minister defers — is that brief's whole subject, stated as constitutional theory rather than as institutional design. The pandemic record is the shared evidence base and the two briefs should be read as one argument.

Geoengineering governance is adjacent operationally rather than thematically: an intergovernmental scientific advice mechanism has already produced advice on solar radiation modification, and the line-by-line government approval procedure described in this brief's engineering section is the procedure that would apply to any assessment of deliberate climate intervention.

Within this map, AI-assisted governance shares the finding that an accountability apparatus can be well built and never observed to change an outcome, and megaproject governance is the useful contrast case: technical advice there meets a documented outturn record, so the counterfactual this field lacks is the one that field has. Outside the map: science and technology studies, public administration, and the sociology of expertise, which supply most of the critical literature and almost none of the quantitative kind.

Established Almost every evaluative document in this field is produced by the body being evaluated or by the body that funds it. A parliamentary science office's evaluation was conducted by an intern supervised by its own staff and is fifteen years old. An audit office's staffing and output figures are self-reported. A supranational executive's page describing its own advisory group lists 23 published opinions without publication dates and contains no account whatever of how the executive responds to advice, which is a notable absence in a body whose entire claim is transparency. The savings attributed to the closed assessment office come from an advocacy coalition campaigning for its restoration. The strongest evidence in this brief is against-interest: a national research council reporting that its own field cannot show its effects, and a national government's evaluation of a national academy stating that the academy's impacts are not tracked.

Established The institutional design finding that matters most is about what protects a body, and it has two halves that no observed institution holds together. In 2025 statutory advisory committees required by legislation could not easily be eliminated while non-statutory ones were closed by agency leadership; in 1995 an authorising statute survived intact while the appropriation was zeroed. Statute defends against abolition. Appropriation defends against closure and not against a change of majority. The only design that has demonstrably outlasted political turnover works by removing the motive rather than adding armour: a work programme selected by cross-party consensus has run since 1990 with its operating contract unanimously renewed to 2028, while an agenda commissioned by committee chairs did not survive a change of control. The causal claim is plausible, widely made, and still untested.

Established Who commissions determines whose questions get answered, and the Canadian case makes the mechanism visible. A national academy accepts requests from “foundations, non-governmental organizations, the private sector, or any level of government”, its panels are approved by a board on the advice of a scientific committee, and they deliberate in camera “to ensure the highest integrity and objectivity of the work.” Its funder's approval process for a new assessment runs about 12 months across twelve sequential steps and is described in the funder's own evaluation as “lengthy and unpredictable.” A year to authorise a question and twenty-seven months to answer it is a system that cannot respond to anything that has not been foreseen — which is why the same country runs a separate chief-adviser office with a departmental advisers network, a youth council and ad hoc expert panels alongside it.

Established In-camera deliberation is the fork in the road and both branches are occupied. A national academy protects candour by deliberating in private and publishing only the report. An emergency committee under a treaty instrument publishes structured accounts of its deliberation and names individual dissent with its reason. Neither model has been tested against the other on advice quality, uptake or public trust, and the choice between them is currently made by tradition rather than by evidence.

Frontier The institution that does not exist is an evaluator. No body anywhere in this record has the mandate, the budget or the access to run a matched comparison of decisions taken with and without formal scientific advice. Funders evaluate their own grantees; advisory bodies evaluate their own awareness; parliaments commission inquiries after failures. The evaluation function has no home, and the consequence is that the sector's oldest question — does any of this improve decisions — is not merely unanswered but unassigned. Everything else in this brief is downstream of that vacancy, and the closures of 2025 will be argued about for a decade precisely because no institution existed to establish what the closed committees were doing.

Established The ethical core of this subject is disclosure, and it is weaker than it looks in both directions. An advisory body's authority rests on the claim that its members have no stake in the answer, and that claim is generally self-certified. Where it has been externally tested by examining members, the tester found undisclosed financial relationships in thirty per cent of a twenty-member committee while the institution maintained it had identified none. Where it has been tested by examining the process, a national audit office found conflict-of-interest guidance legally required more than 13 years earlier and never finalised, no public account of how financial conflicts are determined, and no guidance at all for guest speakers — against 15 recusals, one waiver and seven appearance authorisations over seven years. Neither audit could test whether the advice was actually distorted, which is what everyone means to be asking.

Frontier The forward-looking half of the problem is unregulated everywhere this brief looked. Disclosure regimes ask what an adviser has received, not what an adviser might receive. The single systematic look at post-decision payments — 107 advisers, 26 receiving over $100,000 afterwards, one receiving $1.9 million from a company whose drug he had voted to approve — produced the largest figures in this brief and the least institutional response, and its own recommendation of non-compete terms has been adopted nowhere this brief could verify. This brief could not open the underlying investigation and takes those numbers from a secondary summary, which is stated here rather than buried.

Established Public money and public accountability meet in an unusual place in this field: the evaluation itself. A national government spent $46.5 million over eleven years on a national academy's assessments and can account for every dollar, every month of schedule, and every citation in 43 countries — and states in its own evaluation that the impacts are not tracked. A national scientific-integrity regime grants federal scientists a real right to speak and leaves 19 of 25 departments without a plan to monitor whether it is honoured. These are not scandals. They are the ordinary condition, and the ethical obligation they create is on the writer rather than the institution: do not describe a body as effective on the strength of measures its own funders say do not measure effectiveness.

Frontier The accountability question is genuinely unresolved and this brief does not resolve it. If advice is treated as authoritative, a decision-maker who defers has transferred responsibility to people who were not elected and cannot be removed — which is the pandemic case on the chief adviser's own account. If advice is treated as one input among several, the formal apparatus is decoration and its independence guarantees protect nothing consequential. A 2010 national code splits the difference by obliging government to “publicly explain the reasons for policy decisions, particularly when the decision is not consistent with scientific advice”, which addresses only the rare case. Neither the advisory literature nor the 2026 inquiry recommendations say which of the two is intended, and the recommendations adopted — more experts, wider disciplines, published advice, better terms of appointment — are consistent with either.

Established Two disclosures this brief takes on itself. A substantial share of what is known about advisory bodies comes from those bodies describing themselves, from their funders, or from advocacy coalitions campaigning for their restoration; those sources are marked in the reading list and are relied on for facts they would be embarrassed to get wrong rather than for framing. And the Covid-19 Inquiry's own Module 2 report could not be retrieved for this brief — every attempt returned an access error — so everything said here about it comes from the government's response to it, which is an interested party's account of criticism directed at itself. That is the single largest evidentiary gap in this brief and it is stated rather than smoothed.

Frontier The civilisational question this subject raises is whether technical advice can constrain a government that does not wish to be constrained, and the 2024–2026 record answers it more clearly than the earlier record did: it cannot, and the mechanism of non-constraint is administrative rather than argumentative. More than a hundred advisory committees in one country were closed by executive instrument in eighteen months. The canonical assessment office was defunded, not defeated in argument, for savings that rounded to nothing. A chief-adviser post was abolished by a change of president after a letter from nine campaigning organisations. In none of these cases did anyone have to engage with the content of the advice at all.

Established What advisory institutions do appear to do, and this is worth defending on its own terms, is produce a durable public record. Roughly 750 assessments still cited decades after the office producing them stopped existing. An academy's assessments cited across 43 countries with over 500 direct and 4,000 indirect citations. An intergovernmental assessment resting on 165 experts and 6,500 references whose underlying chapters no government renegotiated. Inquiry evidence that let a health secretary's regret be quoted against a chief adviser's account. An audit that put six undisclosed conflicts on the record and a national audit office that put an unfinalised thirteen-year-old guidance requirement on it. Frontier That is knowledge production and record-keeping rather than decision control — a real function, and not the one the model is sold as.

Established The general principle this case illustrates is about what institutions measure when the thing they exist for is hard to measure. A national academy system with excellent citation performance, disciplined timelines and a funder that tracks its spending to the dollar has no instrument that could detect whether its advice changed a decision, and says so. A scientific-integrity regime grants a real right in twenty-three of twenty-five departments and monitors it in six. A parliamentary office measures awareness and satisfaction. In each case the institution is not lazy or dishonest; it is measuring the quantities that are cheap, and the outcome is not one of them. Any field in this corpus whose central claim is that a process improves decisions should expect to find the same asymmetry, and should look for it before believing its own evaluation.

Speculative The long-run stake is whether a civilisation can institutionalise the habit of finding out before deciding. The observed record suggests the habit is real but fragile, survives best where no single majority controls the agenda, and produces its most reliable value as a public archive rather than as a constraint. That is a smaller claim than the one usually made for scientific advice and it is the one the evidence supports.

Established What already happened, because the timeline in this subject usually starts too late. The canonical legislative assessment office ran from 1972 to 1995 on an initial budget of about $5 million a year and roughly 200 staff, producing over seven hundred reports, and was defunded rather than abolished. A German office whose work programme is chosen by cross-party consensus has run since 1990. A Danish board lost its statutory basis in 2012 and continued as a private foundation. A single European chief-adviser post existed from 2012 and ended in October 2014; the seven-member group and academy consortium that replaced it were announced on 13 May 2015, with the first advisers named in January 2016 and the academy consortium launched in December 2016. Canada created its Chief Science Advisor post in 2017 and issued a model scientific-integrity policy in 2018.

Established 2024 to 2026: the years the machinery came under simultaneous strain. An intergovernmental biodiversity summary was negotiated over 31 hours in December 2024 against a planned sixteen. A February 2025 executive order triggered the termination of more than 100 United States federal science advisory committees. A pandemic inquiry's second report appeared on 20 November 2025 and the government's response, accepting every scientific-advice recommendation, on 25 March 2026. A national audit office reported on 24 March 2026 that a major regulator's conflict-of-interest guidance had been legally required for over thirteen years and never finalised. Canada updated its model integrity policy in February 2026 with nineteen of twenty-five departments still lacking a monitoring plan.

Frontier 2026 to 2028: three dated tests, none of which requires a breakthrough to resolve. A national expert register is to be refreshed during 2026 and advisers' terms of appointment revised by 2027. An intergovernmental climate panel deferred its next assessment's delivery date to a session in October 2026, having failed to agree it in March, and its trust fund is projected to be depleted by 2028 during what would be the cycle's most intensive phase. And the natural experiment created by the advisory-committee closures becomes measurable as regulatory and funding decisions taken without those committees work through.

Handwave Whether the defunded assessment office is refunded. Its statute has stood for thirty years and the barrier has been a single appropriation line for the whole of that period; the independent body commissioned to answer the question recommended against restoration; restoration bills have been introduced repeatedly, and the most recent this brief could verify is from 2 February 2022. Predictions in this category have been wrong in both directions for three decades and this brief supplies none.

Speculative Late 2020s to 2030s: either somebody constructs a defensible outcome measure and the field acquires an evidence base, or it continues to be evaluated on citations, output counts and satisfaction surveys. Nothing in the current record suggests the first. The plausible equilibrium is the one that already exists in the most durable case: consensus agenda-setting, an operator on a renewable multi-year contract, published reports, and no illusion that the body constrains anyone. Durability through modesty.

• Depends on Nothing on this map. No result produced by another brief here is on this topic's critical path; the constraints are institutional, procedural and financial, and are recorded below. That is an unusual sentence in a frontier-research corpus and it is the correct one here.
• Requires (not on this map) An agenda-setting rule that does not make the body the property of the current majority, which is the only design feature in the observed record that correlates with survival. Protection that covers abolition and defunding at once — statute blocked the 2025 committee closures and did not block the 1995 defunding, and appropriation does the reverse, so no observed body has both. Conflict-of-interest disclosure audited externally rather than self-certified, since one audit found six undisclosed conflicts in a twenty-member committee against the institution's denial and another found a determination process no outsider can inspect. A publication clock rather than a publication commitment, because a two-month lag turns a published record into an archive. And — a genuine methodological dependency, unusual for this category — an outcome measure, because on the advisory establishment's own account and on a national funder's evaluation of a national academy, nobody has one.
• Enables Technical analysis available to decision-makers who would otherwise rely on interested parties for it. No typed enabling edge is claimed: the briefs that assume competent technical policy assume it generically rather than through a named result.
• Adjacent Scientific governance models, the sibling slot, which owns everything science decides about itself with no external decision-maker in the loop; technocracy and democracy, which is this brief's accountability problem stated constitutionally; geoengineering governance, where the intergovernmental approval mechanism described here is already in use; and AI-assisted governance and future democracies, which share the gap between measured process and unmeasured influence.

“The failure mode is governments ignoring scientific advice.” Established It is not, in any well-documented case available to this brief. The pandemic record shows advice followed — “no significant disagreement between the Government and its scientific advisers on anything material” — and insufficiently precautionary. The drugs-adviser case shows an adviser removed rather than advice rejected. The European case shows a post abolished after pressure from non-governmental organisations, not from a minister. The 2025 case shows more than a hundred committees closed rather than overruled. The observable failure modes are defund, close, exhaust, defer, and never measure. A framing built around defiance is looking in the wrong place.

“Political interference with advice comes from politicians.” Established Not exclusively, and the exception is a documented success. In July 2014 nine environmental groups and non-governmental organisations, Greenpeace among them, wrote to the incoming president of a supranational executive urging that its chief scientific adviser post be eliminated, on the grounds that the incumbent's position on genetically modified crops was not representative of the scientific consensus. The post ended at the end of October 2014 with the outgoing president's term; the official account was that it “automatically came to an end” and no decision had been taken to reinstate it. It was not reinstated in that form. Advocacy organisations objecting to inconvenient advice is a mode of attack with a success on the record, and any account of advisory independence that models only ministerial pressure is describing half the cases.

“The Office of Technology Assessment was abolished.” Established It was defunded. Its authorising statute was never repealed and remains in force; the appropriation was zeroed, it closed its doors in September 1995 and ceased operations in February 1996 after a $3.6m closure appropriation, having produced roughly 750 assessments and papers on a final-year budget of $21.3m and a statutory cap of 143 staff. Reviving it needs an appropriations line, not a new act. Established And the independent study commissioned to answer the revival question recommended against it, proposing to strengthen existing services and create a congressional science adviser instead — a finding missing from essentially all restoration advocacy. One number should be quoted with care: a contemporaneous professional-society account gives “over 700 reports” against the scholarly source's roughly 750 assessments and papers. The two are compatible and this brief does not reconcile them.

“Intergovernmental assessment bodies politicise the science.” Established They politicise the summary, under time pressure, at the end of a multi-year process, and the distinction is load-bearing. A biodiversity summary took over 31 hours against a planned sixteen, with many lead authors gone before approval; the recorded fights were over protection against conservation, over whether armed conflict contributes to climate change, and over the word polycrisis. Participating authors write that negotiations were “marked more by national interests than by science” and that the scientific language was “weakened by bureaucratically-driven wordsmithing.” The assessment underneath — 165 experts, 57 countries, over 6,500 references — is not renegotiated. This matters because the obvious remedy, removing government approval, would also remove the reason governments treat the summary as binding on themselves.

“Consensus bodies speak with one voice because the experts agree.” Established Sometimes they publish the disagreement, and the record shows exactly when they started. Across 101 emergency-committee statements from April 2009 to November 2023, mean length rose from 1,020 words before COVID-19 to 2,992 after; conflict-of-interest declarations first appeared in the COVID-19 statements; and the mpox statements introduced explicit deliberative-session sections and were the first to record internal disagreement in public, including between the committee and the Director-General. The fifth mpox meeting records that the committee agreed the emergency was over “except for one Member, who expressed concerns regarding future access to vaccine.” One dissent out of eleven, published with its reason. That is a higher standard of disclosed reasoning than any national chief-adviser system or academy in this brief achieves.

“Publishing advice makes it available for scrutiny.” Established Only if it is published in time, and the record is poor. The statement of the meeting that ended a global health emergency on 4 September 2025 was published on 30 October 2025, nearly two months after the decision had been acted on. Pandemic advice in the best-documented national case was published roughly four months into the emergency, which is precisely when it could no longer inform scrutiny of the decisions already taken. The 2026 commitment is to publish “at the earliest opportunity” — a standard with no clock in it.

“Calibrated uncertainty language means readers get the message.” Established Readers do not. Against the standard definitions, very unlikely (below 10%) was read at a mean of 41%, unlikely (below 33%) at 44%, likely (above 66%) at 54%, and very likely (above 90%) at 62% — regression toward the middle in every term and in both directions, across 556 respondents. The best presentation format tested raised consistency with the guidelines only from 20.76% to 30.12%. A calibrated vocabulary misread by roughly seven readers in ten is a precision internal to the institution.

“Conflict-of-interest disclosure is what protects the credibility of advice with the public.” Established With the public it appears to do very little. Across three experiments (N = 320), the disclosed vested interests of a party contradicting a scientific claim had no significant effect on trust in the original scientist or belief in the claim in two of three samples, while the expertise of that party had a robust effect in all three (d = 0.39, 0.26, 0.27). Conflict screening may well protect the quality of the advice. The evidence says it is not what protects its reception, and institutions spend nearly all their procedural budget on it.

“Communicating uncertainty builds trust.” Established It depends who is reading, and the correction is precise. Two pre-registered experiments (n = 600 and n = 1,001) found a crossover interaction: where the evidence agreed with a reader's prior beliefs, adding uncertainty reduced trust in the information and the source; where it disagreed, it raised both (β = −0.34 and −0.26 in the first study, −0.40 and −0.41 in the second), with a strong main effect of belief–evidence consistency (β = 0.58 and 0.50) and no polarisation. Transparency is a trust-redistributing move, not a trust-maximising one, and that is a better reason to do it than the one usually given.

“The Danish technology board was closed.” Established It was not. Its statutory basis was abolished in 2012 after a proposal to remove its €1.5m annual funding drew around 111 consultation responses almost all opposed; it continued as a private foundation, merged with a participation consultancy in 2023, renamed itself in 2024, and downgraded from full to associate membership of the European assessment network in 2017. Institutional survival and institutional relevance came apart, which is a more interesting fate than closure and a more likely one.

“Advisory institutions are hard to evaluate because governments will not fund evaluation.” Established One government funded one and it found the gap anyway. The evaluation measured spending, throughput, timeliness against target, volunteer value and academic reach across 43 countries — and then stated that impacts are “difficult to measure and are not tracked”, with an optional voluntary sponsor survey as the only instrument. The constraint is not money. It is that inputs, outputs and perceptions are cheap to measure and outcomes are not.

“The circulating cost-of-abolition figures show what the office was worth.” Handwave Claims that particular reports saved hundreds of millions, or tens of billions, come from a coalition campaigning for restoration and carry no causal identification, no counterfactual and no independent audit. A body's supporters costing that body's outputs is an interested-party claim and should not be repeated without saying so.