Technocracy and Democracy

Technocracy in its usable sense is not rule by scientists. It is delegation: an elected principal hands a decision to an unelected body because it expects the body to decide better, and then finds it cannot easily take the decision back. Independent central banks, fiscal councils, regulatory agencies, constitutional courts and — twice in Europe in one month of 2011 — whole governments of non-partisan ministers are all instances of the same move. So, in a weaker form, is every occasion on which a minister who retains the formal power to decide declines to depart from technical advice.

This brief owns the tension. Three questions run through it and the literature persistently conflates them. Does delegation produce better outcomes? Do the governed accept it? And, least often asked and most consequential, on what warrant? — that is, what is actually known about the accuracy of expert judgement in the conditions under which authority is handed to it. The answers do not point the same way. On the first, the evidence is inconsistent in a specific and diagnosable manner, and it is asymmetric: much better evidence exists that destroying a technocratic institution does harm than that building one did good. On the second, the survey record contradicts almost everything written about a legitimacy crisis, and the one behavioural measure contradicts the survey record. On the third, expert judgement turns out to be well calibrated at exactly the tasks delegation does not require of it and poorly calibrated at the tasks it does.

The scope boundaries with three sibling slots are worth stating precisely, because this brief borrows evidence from all three and owns none of their subject matter. Scientific advisory institutions owns the machinery of advice: who sits on committees, how they are appointed and conflicted, how assessments are approved, what happened to more than a hundred advisory committees in 2025 and 2026. Its unit of analysis is the advice relationship. This brief takes that record only where it bears on the authority question — what warrant technical analysis has over a decision it did not make, and who is accountable when a minister defers. Future democracies owns democratic innovation: citizens' assemblies, sortition recruitment, participatory budgeting, electoral-system reform, and whether recommendations travel. This brief takes sortition only as a proposed answer to technocracy, and asks whether the answer has a mandate problem of its own. Institutional design owns whether institutions can be engineered at all; this brief takes its enforcement null and its central-bank results as evidence about the specific class of institution that removes a decision from electoral reach.

One further borrowing is declared at the outset because it crosses a seam the corpus otherwise keeps closed. The best quantitative evidence about how accurate expert judgement actually is comes from studies of science's own internal machinery, which belongs to scientific governance models as facts about allocation and gatekeeping rules. This brief cites four of them, and only four, as facts about expert judgement under uncertainty, because the warrant question cannot be answered without them and no comparable measurement exists in the political-delegation literature. The borrowing is marked wherever it occurs. What this brief does not do is settle the normative dispute: whether a democracy may legitimately place a decision beyond electoral reach is a live argument conducted mostly by philosophers and mostly without reference to the empirical record, and both sides of it are stated below at their strongest.

Established Start with what “independence” actually means, because the word conceals at least three designs. The cleanest documented delegation event in an advanced economy proceeded in stages: an announcement letter in May 1997, a target set by the finance ministry in June, Royal Assent in April 1998, commencement that June. The statute gives the central bank a duty to maintain price stability — but the Treasury defines price stability, the government sets the numerical target, and the Treasury holds reserve powers to direct the bank in “extreme economic circumstances”. That is instrument independence with a government-set goal and a live statutory override. It has never been goal independence, and most of the comparative literature treats the three designs as one.

Frontier Now the outcome question, and the honest headline is that five serious results disagree. The strongest identification attempt in the earlier literature — a reform-dummy design over 52 countries, deliberately avoiding subjective indices — finds no effect in low-constraint or high-constraint political systems and a marginally significant effect only in the middle. A second finds the level relationship absent in both periods examined and its rescuing first-difference result disappearing once initial inflation is controlled for. A third finds an effect only in non-advanced economies and none in advanced ones; a fourth, which the third names, finds the reverse. And the fifth, the newest and most careful about identification, is close to a null: a doubly robust estimator applied to a directed acyclic graph of inflation transmission across 60 countries from 1998 to 2010 with 17 measured variables returns +0.01 percentage points with a 95% confidence interval from −1.48 to +1.50, alternative specifications reaching −0.61 points and the high-income subgroup showing +0.48 — independence associated with higher inflation. Its authors conclude that there is “only a weak causal link from independence to inflation, if at all” and that “a strong inflation-boosting impact from introducing central bank independence cannot be ruled out.” All five use overlapping country panels. They cannot all be right, and no sixth paper has adjudicated.

Established Then the single most important new result in this brief, which runs the other way and is better identified than anything above. An IMF working paper published in 2026 exploits 132 central bank governor transitions across 28 central banks — eleven advanced economies and seventeen emerging markets, together about 70% of world GDP — over 2000–2024, classifying 50 as politically motivated and 82 as routine using three independent reviewers working from news sources. Politically motivated transitions are 48% of emerging-market transitions against 13% in advanced economies. On a local projection difference-in-differences, such a transition is followed by short rates falling about 2 to 3 percentage points in the first year, realised inflation rising 2 to 4 points, one-year inflation expectations rising 2 to 3 points, and GDP growth accelerating 1 to 2 points in the short term. Long-term inflation expectations rise about 1 point at two years, and only for governors classified as unorthodox; long-term growth expectations do not move at all, which the authors read as private agents not believing the accommodation affects potential output.

Frontier The asymmetry is the finding, and it changes the shape of the argument for delegation. We have poor evidence that constructing this institution lowered inflation and considerably better evidence that dismantling it raises inflation. Two readings are available and the literature does not distinguish them: a genuine asymmetry in the world, in which credibility accumulated over decades can be destroyed in a news cycle, or a measurement asymmetry, because destruction has a date and a treatment group and construction does not. Either way, the strongest defensible case for technocratic delegation on the current record is not a performance case but a conservation case — not that handing the decision to experts made things better, but that visibly taking it back makes things worse. Those are different arguments, and the second is much weaker as a justification for creating new delegations.

Established The measurement instruments are part of the problem, on the record inside the literature. Two of the standard independence indices, applied to the same 17 countries, translated the legislation identically in exactly one country on one of nine criteria; their average interpretation spread on four national legal systems is close to 50%; they disagree in nearly 60% of countries on whether the bank may buy government debt in the primary market. A later check finds the two correlating at 0.67 over 1992–2012, falling to 0.59 in the crisis window — two measures of the same thing sharing under half their variance exactly when the thing mattered most. This is not peculiar to central banking, and the institutional section generalises it. When the technocratic instrument is a measure, the choice of measure can determine the answer, and the choice is made by experts without public involvement.

Established And the most direct refutation of the naive design premise is a regression with sixteen observations. Regressing measured de facto independence of financial, telecom and competition regulators on the standard index of formal independence gives a slope of 0.37 with a standard error of 0.43, and R² = 0.05. Some regulators are more independent in practice than on paper; others the reverse. The authors' conclusion is that formal independence is neither necessary nor sufficient for independence in practice. The sample is small and cross-sectional and that must be said — but the designed property explains five per cent of the variance in the property it was designed to produce.

Established Now the warrant question, and here the brief borrows. The best measurements of how accurate expert judgement is under uncertainty come from studies of how science allocates its own money, which is another brief's subject and is used here only as evidence about judgement. Across 102,740 funded grants from 1980 to 2008, reviewer percentile score explained r² = 0.0078 of subsequent publications and citations, at AUC 0.54 against a chance baseline of 0.50. The natural inference — that expert panels are arbitrary — is wrong, and the correction matters more than the finding: near-zero agreement estimates are largely an artefact of range restriction, and across the full range of applications single-rater reliability is 0.34 (95% CI 0.31–0.37), rising to 0.61 for three-rater panels. Expert judgement reliably separates good from bad and cannot separate good from good. A randomised trial of 607 reviewers assessing documents seeded with nine major and five minor deliberate methodological errors makes the boundary concrete: they found a mean of 2.58 to 3.05 of the nine major errors, with training producing only slight improvement. Technical authority is well founded when the question is whether something falls inside or outside a competence boundary, and weakly founded when the question is which of several defensible options to choose — which is what delegation actually asks.

Established The same pattern appears in the one place where expert forecasts were elicited before an institutional result was known. Before the analysis of an eleven-year randomised institution-building programme across 236 Sierra Leonean villages, 126 experts forecast its effects. On physical infrastructure the mean forecast of 0.218 SD almost exactly matched the realised 0.204 SD. On institutions, the thing the programme existed to build, they predicted 0.095 SD against a realised 0.062 SD that was not significant after adjustment — and the national policymakers closest to the programme predicted around 0.25 SD, roughly four times the truth. Expert judgement was well calibrated about the concrete and badly calibrated about the institutional, and proximity made it worse rather than better. Institutional questions are exactly the ones technocratic delegation is justified by.

Established Against that stands the strongest single result in favour of aggregated expert judgement, and it should be given its full weight. Asked to predict which published experimental findings would replicate, prediction markets priced at 75.2% and surveys at 71.1% against an observed 61.1% across 18 experiments, with individual beliefs correlating at r = 0.52 with outcomes. The community's aggregate judgement carried real information the formal machinery — which had published all of those papers — did not act on. The case for expertise is strongest where it is aggregated, elicited on a resolvable question, and not asked to produce a ranking. That is a much narrower warrant than delegation assumes, and it is a real one.

Established On acceptance, the survey record contradicts the standard story consistently and at scale. Across 31 European democracies, three European Values Survey waves and 84 country-wave observations, an average of 56% think technocrats rather than politicians should make decisions — about 57% in 1999–2001 and about 58% in 2017–2021. Twenty-two years, no measured decline. Support is higher where technocratic appointments have a history and rises with perceived executive corruption; economic conditions do not move it meaningfully. A nine-country study of 9,449 respondents recovers three distinct dimensions — expertise, elitism, anti-politics — and concludes that “there is no populism without expertise”: respondents with populist attitudes want outsider experts, not fewer experts.

Established And the acceptance is stage-specific in a way that maps almost exactly onto the normative argument. A survey of 7,357 respondents across seven democracies finds publics preferring independent experts for policy design (+9 percentage points) and implementation (+9 points), with no significant difference for the decision itself; environmental policy is the one area where experts are preferred for the decision, at about seven points. Set that beside the sharpest philosophical statement of the objection to expert rule, which distinguishes shortcuts that enhance citizen judgement from shortcuts that substitute for it, and explicitly permits informational shortcuts including expert recommendations where citizens retain the capacity to evaluate on the merits. The public's revealed line and the philosopher's argued line are in the same place. That convergence is the most interesting single fact in this brief, and it is not evidence that either is correct.

The genuinely open questions divide into three that are empirical and answerable, one that is empirical and currently unanswerable, and one that is normative and may not be the kind of question evidence settles. Keeping them apart is most of the work, because the literature routinely answers an easy one and presents it as an answer to a hard one.

Frontier Open and answerable, one: whether the build-versus-break asymmetry is real or an artefact of study design. Everything turns on this. If credibility genuinely takes decades to accumulate and can be destroyed in a news cycle, delegation is a ratchet worth protecting even where its construction cannot be shown to have helped. If the asymmetry is instead a measurement artefact, the honest summary is that we do not know whether this class of institution works in either direction. The IMF study's own caveats point at the problem rather than resolving it: governments may undermine independence in response to macroeconomic risks that independently drive inflation, anticipation effects bias estimates toward zero, and the exchange-rate results show pre-trends suggesting endogeneity. No study in this record distinguishes the two accounts, and one could be designed.

Established Open and answerable, two: where delegation works, it appears to push the distortion somewhere else. In the reform-dummy study, the same medium-constraint countries where independence reduced inflation show government expenditure rising with it — about a percentage point of GDP immediately and roughly five percentage points in the long run, with no significant effect on the budget balance. A falsification test using a five-year lead of the reform dummy returns nothing, which is what it should. The authors hedge that the result is not always robust. If it holds, delegating one instrument does not remove a government's incentive; it relocates its expression — which would mean the measured success of a technocratic institution is partly a measure of where the analyst chose to look. Frontier The magnitudes usually quoted are also the flattering half of a sensitivity check: the same paper reports that using median rather than mean inflation shrinks the implied economic magnitudes by about 60%, four points instead of twenty-two.

Frontier Fiscal councils have been evaluated once seriously, and the evaluators state the identification failure themselves. The study covers EU member states only, because comparable data exist nowhere else. Forecast-accuracy effects are statistically insignificant in almost every specification — the coefficient on GDP-growth forecast error is −0.338 with a standard error of 0.286 on 330 observations — and the one solid result, on rule compliance, is procedural and vanishes once forecast error is added as a regressor. Verbatim: “We could not identify any suitable instrument.” The same table contains a result that ought to be better known: tighter fiscal rules go with more optimistic budget forecasts, the rule-strength index entering the primary-balance forecast-error regression at +0.301 (s.e. 0.105). The rule created the incentive to game the input to the rule. And where an identification strategy has been applied — the geographic diffusion of rules across 142 countries, 1985–2015, as an instrument — the positive correlation between rules and balances “disappears when endogeneity is correctly addressed”, surviving only for well-designed rules. That work was done at the institution that recommends the rules, which raises its weight rather than lowering it.

Established Open and answerable, three, and now answered for one large class of delegation: independent courts do not measurably make constitutional rights real. Matching de jure constitutional text against de facto outcomes across nine rights over 1946–2010, with specifications running from 188 to 5,315 country-year observations, the interaction between constitutionalising a right and having an independent court to enforce it is predominantly non-significant, and the authors state without hedging that they do not find countries with independent courts better at upholding their constitutional commitments. What does predict better outcomes is whether the right is organisational — unionisation, political parties, religion — because those create constituencies with means of mobilisation beyond litigation; their conclusion is that “rights enforcement ultimately falls on citizens themselves.” This is the enforcement null, and it is the most direct available test of whether a counter-majoritarian expert body can hold a line the majority wants to cross. Judicial review has nonetheless been adopted by very nearly every constitution written in the past half-century — one of the largest institutional transplants in history, and one whose marginal effect nobody has detected.

Frontier Open and currently unanswerable: whether stated acceptance of expert rule and behavioural acceptance are the same thing. Against twenty-two years of flat majority support sits a study of 371 general elections across 20 Western European countries, 1945–2022, finding that the share of ministries held by non-partisan technocrats significantly decreases turnout — about 3.4 percentage points from a fully partisan to a fully technocratic cabinet, absent in the early post-war decades and significant from the 1980s on. Either stated preference is real and lower turnout is its rational price — the decision has moved, so the election matters less — or stated preference is cheap talk and not voting is what people actually do when experts govern. No source in this record measures attitude and behaviour on the same people. The attitude data are cross-sectional and post-treatment; the turnout study has no attitude measure. Both readings are supported, by different instruments measuring different things, and this brief declines to pick one.

Established One measured divergence between expert and public judgement about politics itself is worth recording, because its direction is not the assumed one. Whether democracy is measurably declining is a live dispute over expert-coded indices, and it belongs to future democracies, which owns democracy measurement and its critiques. What belongs here is a by-product of the exchange: across nineteen waves of expert surveys matched to parallel public surveys, experts rated the state of United States democracy about 10 points higher than the general public did, on a hundred-point scale, across 2017–2023, and among 544 surveyed political scientists those who had been invited to code for the index under challenge rated democracy higher, not lower, in 12 of 13 countries. Where expert and public assessment of a political system diverge, the experts were the less alarmed party. That is an awkward datum for the account in which detached expertise catastrophises what ordinary people experience as tolerable, and an equally awkward one for the account in which experts are the early-warning system.

Frontier Open, normative, and possibly not settleable by evidence: whether a democracy may legitimately place a decision beyond electoral reach. The objection at its strongest is not that expert bodies decide badly. It is that a citizen who did not participate has no reason to expect the body's decisions to track her interests and values — politically blind deference, illustrated by a voter who chooses among representatives by tossing a coin. The standard proposed is that “all those subject to these regulations can see themselves as their authors.” The defence at its strongest replies that a randomly selected body produces inclusive equality because “only chance distinguishes us,” that public justification supplies a form of deliberative accountability even without electoral sanction, and that participants return to ordinary life. Both positions are peer-reviewed, both are internally coherent, and neither has been refuted. What this brief adds is that the empirical record bears on the dispute and appears nowhere in it — the stage-specific survey result, the enforcement null, the calibration evidence, and the only quantitative measurement located of whether procedural legitimacy transfers to people who were not in the room. A vignette study with n = 4,225 finds a legitimacy gain to non-participants of Glass's Δ = 0.079, small and concentrated among low-trust citizens; a companion with n = 1,309 finds a fairness gain of about +0.7 against a referendum's +1.47, which disappears entirely when the recommendations are ignored. That conditional generalises past mini-publics: procedural legitimacy is contingent on the output being acted on, and no delegation design in this brief can guarantee that.

Established The first bottleneck is that delegation is revocable by ordinary means, and is revoked at a measurable rate. A century-scale coding of central bank legislation across 155 countries and 370 reforms records 91 — roughly one in four — as reducing independence. That is the base rate, not an aberration, and the 2026 transition study puts a second number beside it: 50 of 132 governor transitions across 28 central banks between 2000 and 2024 were classified as politically motivated, rising to 48% of transitions in emerging markets.

Established The case record is unambiguous about the mechanism. One country deleted the governor's fixed term by presidential decree in 2018, then dismissed governors in 2019, 2020 and 2021; by February 2024 it had had five governors in five years, and average inflation had gone from about 14% in 2017–18 to 63% in 2022–23. Another dismissed its governor in January 2010 for refusing to divert reserves; a court challenge followed, the governor resigned anyway, and inflation exceeded 50% by the decade's end. A third rewrote its central bank law three times in eighteen months, prompting a formal European opinion whose core objection was that this “is incompatible with the principle of legal certainty”. And the supranational remedy was an opinion, which did not stop the thing it objected to: one of that 2011 opinion's three objections was to a draft law merging the central bank with the financial supervisor, and the merger happened, the supervisory authority being terminated on 1 October 2013 and its powers taken over by the central bank in modified form. The objection was registered; the institutional change proceeded.

Established Nor does this require an authoritarian turn. A country in the 93rd percentile of the World Bank's government-effectiveness measure amended its central bank's mandate three times in five years — adding maximum sustainable employment in 2018, requiring assessment of effects on house-price policy in 2021, and reverting to inflation-only in 2023.

Established The second bottleneck is that a purely procedural safeguard binds only a minister who chooses to be bound. In September 2022 a fiscal statement described by a parliamentary library as roughly £45bn of tax cuts was made with no forecast requested from the fiscal watchdog, which consequently could not say whether the fiscal rules would be met. The legislative fix defines a “fiscally significant” measure as one costing at least 1% of GDP — about £30bn — and was called “broadly sensible but largely performative” by an institute economist and “a forecast lock” rather than a fiscal lock by a member of the committee scrutinising it, with an open question about announcing changes just below the threshold.

Established The third is that expert authority is not usually contested; it is bypassed administratively, and nobody has to engage with the content. Following a February 2025 executive order directing agencies to eliminate advisory committees, more than 100 committees advising United States science agencies were terminated; one department lost 77 of about 280 against two terminations in the whole of 2024, and its open meetings fell from a ten-year average of 255 a year to 91. A European chief scientific adviser post was abolished in 2014 after nine environmental groups wrote to an incoming president urging its elimination on the grounds that the incumbent's position on genetically modified crops was unrepresentative. Neither case involved anyone reading the advice and rejecting it, and one of them was executed by civil society rather than by a minister. The sibling brief on advisory institutions owns this record; what belongs here is the implication for authority, which is that the mechanism protecting a delegated body must survive administrative attention, not argument.

Frontier The fourth is that the terms in which delegated expertise communicates are not received as sent. Tested against the calibrated probability vocabulary a major assessment body uses, 556 respondents read “very unlikely,” defined as below 10%, at a mean of 41%; “unlikely,” below 33%, at 44%; “likely,” above 66%, at 54%; and “very likely,” above 90%, at 62% — regression to the middle in every term, in both directions. The best presentation format tested raised consistency with the published guidelines only from 20.76% to 30.12%. A vocabulary misread by roughly seven readers in ten is a precision that exists only inside the institution, and an authority claim that rests on calibrated uncertainty rests on something the public is not receiving.

Established The fifth is that outcome measurement does not exist for any of it. No source in this record measures whether a delegated decision was better than the decision an elected body would have taken. The fiscal-council evaluation could not identify a suitable instrument and says so. The independence–inflation literature disagrees with itself across five results. The one design built to produce an implementation count for citizen recommendations has produced satisfaction statistics instead. What binds is not that delegation is known to fail; it is that no instrument exists that could detect either result, which means the argument is conducted on priors and stays conducted on priors.

Established Nothing here waits on a research result. That sentence is unusual in this corpus and it is the correct one. Every question in this brief is about institutions, statutes, elections and public judgement, and none of them is gated by a discovery.

Established What delegation waits on is institutional, and the record specifies four conditions. Goal-setting retained by an elected principal, which is what the flagship design actually has and what its imitators often lack. An override that exists but is costly to use, because a reserve power conditioned on extreme circumstances is a different object from a procedural safeguard a minister may simply decline to trigger, and one fiscal statement of roughly £45bn made without requesting a forecast is what the second kind is worth. Delegation confined to the stages publics will actually grant — design and implementation, each preferred by about nine points, with no majority for delegating the decision itself. And durability across a change of government, which one central-bank reform in four fails.

Established It also waits on something the literature does not name as a dependency: a measurement instrument. Every argument in this brief is conducted without an outcome measure, and the two sides recruit different instruments — surveys against turnout series, formal indices against de facto surveys, cross-sections against event studies. Until one field publishes a series that both sides accept, the dispute is not resolvable by evidence and will be settled politically.

Established What depends on it is the more consequential direction. Every proposal anywhere in this corpus that turns on a government binding itself — a long-horizon technology programme, a carbon price, an emergency reserve, a spending rule — is implicitly a delegation proposal and inherits this brief's base rate. Scientific advisory institutions depends on the authority question this brief owns, because an advisory body's whole value proposition is that its output has standing it did not receive from an election. Future democracies depends on it inversely: the mini-public is proposed largely as an answer to expert rule, and if the legitimacy objection to technocracy also applies to sortition, as the philosophical literature argues it does, then the answer and the problem share a defect.

Established The most valuable experiment available is not an experiment: it is an adjudication. Five peer-reviewed or institutional results on the independence–inflation question use overlapping country panels and reach mutually inconsistent conclusions, and the newest of them — a doubly robust estimator returning +0.01 percentage points with a confidence interval from −1.48 to +1.50 — is closest to a null. A pre-registered re-analysis on a common sample, with the index choice, the sample split, the estimator and the treatment of initial inflation varied systematically, would settle which of the five is an artefact of specification. The data are public. Nobody has done it.

Frontier The build-versus-break asymmetry has a designable test, and the design already exists in half-form. The 2026 governor-transition study built its treatment variable by having three independent reviewers classify 132 transitions from news sources as politically motivated or routine. The same construction applied in the other direction — dating and classifying grants of independence with the same reviewer protocol, then running the same local projection difference-in-differences — would show whether the asymmetry is in the world or in the estimator. If construction events dated as sharply as destruction events still yield nothing, the asymmetry is real and the conservation argument for delegation is the correct one. If they yield an effect of comparable magnitude, four decades of null-ish cross-sectional results were a dating problem.

Frontier Second, and pointed: measure attitude and behaviour on the same people. The survey literature and the turnout literature reach opposite conclusions about public acceptance, and no source in this record measures both on one population. A panel that asked about expert rule and tracked participation would separate the two hypotheses, and nothing else in the current toolkit can.

Frontier Third: publish the de facto independence measure annually. The formal index is updated continuously; the practical one exists as a one-off expert survey with sixteen observations and an R² of 0.05 against the formal index. Repeating it at scale and on a schedule would convert the single most damaging finding in this brief from a curiosity into a monitorable series.

Established Fourth, a natural experiment is already running and its readout is dated. The 2025–26 termination of more than a hundred United States advisory committees was a large, abrupt, differentially applied reduction in institutionalised expert input, with agency-level counts published and a control group — the statutory bodies that survived — defined by an unrelated legal characteristic. Whether agency decisions measurably changed in the affected domains is answerable within a few years by anyone willing to define the outcome in advance.

Established Fifth, and the cheapest: elicit forecasts before results. The Sierra Leone study is the template. Before the analysis was run, 126 experts forecast the effects; they were accurate on infrastructure (0.218 predicted against 0.204 realised) and wrong on institutions by roughly half, with the most contextually knowledgeable group wrong by a factor of four. Attaching a pre-registered forecast elicitation to any major delegation reform — asking the officials who designed it, the academics who recommended it and a lay sample what it will do — costs almost nothing and produces the calibration series this field entirely lacks.

Established Negative results worth recording as experiments in their own right, because each closes a line of argument. The enforcement null closes the claim that an independent court is what makes a constitutional right effective: the right-by-court interaction is predominantly non-significant across nine rights and sixty-four years. The eleven-year follow-up on the Sierra Leone programme closes the “institutions just need time” defence: institution-building effects were +0.028 SD at four years and +0.062 SD at eleven, not significant after adjustment, while the physical infrastructure persisted at two-thirds strength. And the fiscal-council evaluation closes the pretence that the effectiveness question has been answered, by printing its own null coefficients and its own failure to find an instrument. A field with three well-documented nulls is in better condition than a field with none, and this one publishes them.

Established The design details that matter are mundane and are usually omitted from the summary. In the flagship central-bank statute: a committee of the governor, deputy governors, internal appointees made after consultation with the finance minister, and external members appointed by the finance minister; a finance-ministry observer who may attend and speak but not vote; a governor's casting vote; a duty to meet at least monthly, publish decisions as soon as practicable, and publish minutes recording individual votes within six weeks. The numerical target sits outside the Act — which is why it could be changed from a retail-price measure at 2.5% to a consumer-price measure at 2% in December 2003 without touching the statute. Publishing individual votes is the single most legitimacy-relevant clause in the design, because it converts an institutional opinion into a set of named, disagreeing human judgements, and the evidence below suggests that is what publics actually read.

Established The fiscal-council class is not one institution, and its resourcing differs by orders of magnitude. On a 2019–21 snapshot, the largest national budget office ran to 264 full-time staff; several councils had fewer than five; one is a six-member council with no full-time secretariat at all; another had a single member of staff. Any claim that “fiscal councils work” is a claim about a class whose members differ in scale by a factor of several hundred. Two entries in the standard table also carry founding dates for a predecessor organisation rather than for the fiscal-council function — one body created in 2013 under EU rules is listed against a 1970 debt committee — so the “began” column should not be read as the age of the design.

Frontier Where an independent forecaster grades itself, the interesting comparison is not against private forecasters. Across all 27 fiscal outlooks published between 2010 and 2023, the body reports itself of similar accuracy to external forecasters — slightly better three years out, slightly worse at one, two and four — with systematic over-estimation of growth and under-estimation of borrowing. The one comparison that favours delegation is against the ministry it replaced: more accurate and less biased than the previous official forecasts over the preceding twenty years. That is the single clearest pro-delegation quantitative finding in this brief, it is the institution grading itself, and the two periods are not comparable macroeconomic environments.

Established The instrument that most directly encodes the technocracy–democracy trade is not an economic one, and it is short. A 2010 United Kingdom instrument on scientific advice to government sets four principles — clear roles and responsibilities, independence, transparency and openness, and applying the principles — committing government to leave advisers “free from political interference,” to let them “communicate publicly their advice to government” even where it is inconsistent with policy, to publish advice unless there are over-riding reasons such as national security, and — the load-bearing clause — to “publicly explain the reasons for policy decisions, particularly when the decision is not consistent with scientific advice.” That is the trade written down: the expert may not decide, and the decider must give reasons for departing. The clause permitting dismissal of advisers was retained over objections from two campaigning organisations, so the same instrument that created a duty to justify departures left intact the power that made the duty necessary. It was produced by a review commissioned after a government adviser was dismissed for publishing a harm ranking that contradicted policy, the minister writing that he had “lost confidence in your ability to advise me.”

Established The protection architecture has exactly two components and they defend against different attacks. The 2025 United States record specifies the constraint with unusual clarity: statutory advisory committees required by Congress could not easily be eliminated, while non-statutory ones were closed by agency leadership. Read that against the canonical case from the other direction, where a legislative technology-assessment office was never abolished, its authorising statute remains in force, and only its appropriation was zeroed. A statute protects a delegated 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. Applied to decisional rather than advisory delegation, the same logic explains why a central-bank statute is a weaker guarantee than it looks: the statute survives, and the governor does not.

Frontier One design does exist that attempts to close the accountability loop in the other direction, and its output is instructive. Two Brussels parliaments wrote into their regulations in 2019 a committee of 45 randomly selected citizens sitting with 15 members of parliament, in which government and parliament must report within six to nine months what they did about each recommendation and the citizens reconvene to assess the follow-up. Five have run, and reported satisfaction is high. What has not been published is an implementation count, from the one design in existence built specifically to produce one. That absence is treated as content below rather than as an omission.

The nearest neighbour is scientific advisory institutions, and the two briefs split one subject by whether the expert body decides. That brief owns the machinery — appointment, conflict screening, assessment approval, publication practice, and the administrative destruction of more than a hundred committees in eighteen months. This brief owns the authority: what warrant technical analysis has over a decision it did not make, and who is accountable when a minister defers. The shared evidence base is the pandemic record, and read together the two make an argument neither makes alone. That brief shows that advisory institutions almost never fail by being overruled. This one shows that when they do have authority, the measured failure is deference rather than defiance.

Future democracies is adjacent in the strong sense, because sortition is proposed precisely as the democratic answer to expert rule. The relationship is not complementary but awkward: the objection that a randomly selected body cannot claim a mandate is structurally the same objection as the one made to technocracy, and the philosophical literature that presses it hardest presses it against both. That brief owns recruitment, transmission and implementation; this one takes only the mandate problem, and takes it because the answer to technocracy inherits the question.

Institutional design is the same problem one level up and supplies this brief's most transferable negative result, the enforcement null. Scientific governance models is adjacent by borrowing only: four of its findings about the accuracy of expert judgement are used here as evidence about warrant, and its subject — the rules by which science allocates and gatekeeps its own work — is not touched.

Outside the map: monetary and fiscal economics, where the quantitative content lives and where the identification problems are best understood; the politics of regulation; comparative public opinion, which supplies the half of this subject the economics literature ignores; and the philosophy of democratic legitimacy, which is where the argument that matters is actually being conducted, largely without reference to any of the numbers above.

Established The accountability loop can be observed closing, and it closes on nothing. A parliamentary committee reported in November 2023 on making an independent central bank work better, raising mission creep, appointment patterns and intellectual diversity. In February 2024 the finance minister declined to publish the indemnity deed, declined a memorandum of understanding on debt management, opposed publishing draft remit letters, and declined an independent review of appointments; the governor's position was that a memorandum would add nothing already public. Five recommendations about the accountability of a delegated body; at least four declined by the delegating minister and the delegated body between them. The formal scrutiny apparatus exists, functions, publishes and is answered in the negative, and nothing in the design makes the negative costly.

Established The protection architecture is the institutional finding with the widest application, and it is a specification rather than a complaint. Statutory United States advisory committees required by Congress could not easily be eliminated in 2025; non-statutory ones were closed by agency leadership. The canonical legislative assessment office was never abolished — its authorising statute is still in force — and only its appropriation was zeroed. A statute defends against abolition and not against defunding; an appropriation defends against defunding and not against a change of majority; no observed body has both at once. Applied to decisional delegation, this explains why statutory central-bank independence has not prevented ninety-one reductions in ninety-plus years of coded reforms: the statute is not the thing under attack. The governor is.

Established The institution that does not exist is an evaluator. Nobody owns the question of whether a delegated decision was better than the alternative. The international financial institutions evaluate the bodies they recommend. The fiscal watchdogs evaluate themselves under statute. The academies evaluate assessments they commissioned. The parliamentary committees evaluate process. No standing body anywhere in this record has the mandate, the data access and the independence to answer the counterfactual question, and the one attempt to answer a neighbouring version of it — whether fiscal councils improve forecasts and compliance — was performed by the institution that recommends fiscal councils and returned nulls it published anyway. The absence is structural: an evaluator with that mandate would have to be delegated, and would inherit every problem in this brief.

Frontier The institution that does exist and is under-recognised is the ratings and indices industry, which supplies the technocratic authority nobody voted for and nobody audits. Independence indices determine what the comparative literature concludes. State-capacity indices determine whether published findings hold or reverse: seven established measures correlate pairwise between 0.70 and 0.94 and load 86.91% of common variance onto one component, and yet 45 countries diverge by more than 0.40 standardised units between two of them — most at the intermediate levels where policy-relevant countries sit — and three published findings about democracy and state capacity reverse or vanish depending on which index is substituted. And the one governance index that acquired real institutional power — over 70 countries forming reform committees oriented to it, consultancies paid to move rankings, 111 countries swinging more than 40 places in fifteen years — had its scores altered under internal pressure for four named countries and was cancelled in September 2021 after an independent investigation found no written procedures for methodology change and unwritten rules that permitted manipulation. This is technocratic authority in its purest form: unelected, consequential, self-certifying, and in the one case where it was audited, corrupted at the top.

Established Who benefits from the current arrangement is worth stating plainly, because it explains the stability. Elected principals benefit from delegation twice: the decision is taken and the blame is elsewhere. Delegated bodies benefit from an accountability apparatus that measures process and cannot measure outcome. International financial institutions benefit from a design they recommend, evaluate and staff. Publics report majority support for exactly the stages — design and implementation — where delegation is least contested and least consequential, and no majority for the decision stage where the actual authority sits. Nobody in the arrangement has an interest in the counterfactual being measured, which is a more economical explanation for the missing evaluator than any account based on difficulty.

Frontier What would change the institutional picture, stated as a testable condition. One jurisdiction publishing a pre-registered outcome evaluation of a delegation reform — with the forecast elicited before the result, from the officials who designed it and from a lay sample — would create the first entry in a series this field has never had. It costs almost nothing, requires no new institution, and would tell us within a decade whether the argument at the centre of this brief is empirical at all. That nobody has done it, in a field this old and this consequential, is the institutional finding.

The evidence base on this topic is unusually crowded with interested parties, and the layering is the point. Established The international financial institution that recommends fiscal rules and councils in surveillance is also the body that compiles the datasets used to evaluate them, the author of the leading quantitative evaluation of them, and the author of the 2026 governor-transition study that supplies this brief's strongest pro-independence result — an institution with a standing commitment to central bank independence measuring the consequences of undermining it. Separately, the organisation that wrote the principles for independent fiscal institutions evaluated those institutions and cited the first body's explicitly tentative result as established, while thanking three of the evaluated councils for contributions and interviews. That is one interested party citing another interested party's hedged finding as settled, with the subjects helping to write it. None of this makes the results wrong — the governor-transition study prints its own selection, anticipation and pre-trend caveats — but on the one question where the evidence is strong, the strong evidence was produced by a party with a position.

Established Two sources in this record are nonetheless unusually honest, and the brief leans on them for exactly that reason. The fiscal-council paper prints its own null coefficients, states that no valid instrument could be found, and warns that reducing forecast errors could encourage sophisticated creative accounting its analysis could not capture. The independent forecaster, required by statute to grade itself, writes: “We are no better equipped to see into the future than other forecasters.” Self-evaluation that publishes its own nulls is worth more than self-evaluation that does not, and it is why both are cited here for findings against their own interest and not for findings in favour.

Frontier One caution about a source class, retained because it still applies. The commercial and normative literature on this subject is large and mostly argument rather than evidence; a widely cited theoretical letter arguing that the distributional consequences of monetary policy undermine the legitimacy of independent agencies contains no empirical content and should never be cited as evidence of a backlash. The survey record is the evidence, and it does not show one.

Established Public money and public accountability meet here in a way this brief has to name rather than resolve. Delegation moves decisions with large distributional consequences — interest rates, fiscal rules, regulatory standards, rights adjudication — out of the arena where the people bearing those consequences can vote on them, justified on the ground that the decisions are technical. On the evidence assembled here the technical warrant is narrower than the justification requires: expert judgement discriminates reliably between competence and incompetence and unreliably among defensible options, and is well calibrated about concrete outcomes and badly calibrated about institutional ones. Delegation is being justified by a competence claim the measurement literature only partly supports, and the part it supports is not the part doing the work.

Established There is a second-order obligation about how expertise presents itself, and the evidence has a clear answer that is not the comfortable one. Publishing uncertainty and reasoning does not on average make an expert body more trusted; it redistributes trust toward the sceptical and away from the already-persuaded, with no net polarisation, and it insures against the day the advice changes. Institutions that adopt transparency for the popularity it will not deliver are setting themselves up to abandon it, and the honest case — reach and insurance — needs to be made instead. The corollary is worse: the persuasive framing outperformed the balanced one at producing policy support, so any delegated body measuring its success by whether its recommendation was adopted has a standing incentive to be less transparent, and the measurement that would resist that incentive — did the decision turn out well — is the one nobody in this field has.

Four questions this brief cannot answer and states as obligations rather than smoothing over. Frontier Whether the build-versus-break asymmetry is a property of the world or of study design, which determines whether conserving existing delegations is a principled position or a rationalisation. Whether stated support for expert rule and the observed turnout decline are compatible, which no study measures on one population. Whether the philosophical objection to blind deference is answered by any actual institutional design, including the ones its critics propose. And whether any delegated decision anywhere has been shown to be better than the decision the elected body would otherwise have taken — to which the answer, on everything consulted here, is that no such demonstration exists, in either direction, for any class of delegation examined. A brief asserting that technocratic delegation works, or that it fails, is going past the evidence.

Established The civilisational stake is not whether experts are right. It is whether a polity can put anything beyond the reach of its own next election, and the observed answer is that it cannot, at a measurable rate. Of 370 coded central-bank reforms across 155 countries, 91 reduced independence. Of 132 governor transitions across 28 central banks, 50 were politically motivated. More than 100 advisory committees in one country were closed by executive instrument in eighteen months. Nothing in this record protects an institution against both abolition and defunding at once. A civilisation that needs to bind itself over horizons longer than an electoral cycle currently has no instrument that reliably does so, and every long-horizon proposal in this corpus that assumes one is assuming a capability the record denies.

Established The two technocratic governments of 2011 supply the electoral arithmetic, and it is the part nobody disputes. The two parties sustaining one of them lost a combined 23.6 percentage points at the first subsequent election; the technocrat's own list, standing on his own record, took 8.30%, while a party contesting its first general election took 25.56%. In the other country the governing party that installed the technocrat fell from 43.92% to 12.28% in thirty-two months and to 4.68% within thirty-nine — a loss of 89% of its vote — and its junior coalition partner defected three months before the election, voting against the second bailout, and was still wiped out of parliament entirely. Defecting did not save it. Frontier The causal claim must not be made: these parties were also incumbents through a sovereign debt crisis and an external adjustment programme, with no counterfactual in which they governed the same crisis themselves. The defensible statement is that supporting a technocratic government did not insulate its sponsors, not that it destroyed them.

Frontier The general principle this case illustrates is about the difference between authority and legitimacy, and it is one the corpus meets repeatedly. A delegated body can have complete formal authority, a well-drafted statute, published minutes and named votes, and still have no answer to a citizen who asks why she should regard its decisions as hers. The two currently proposed answers are outputs and procedure: it decides better, or it was selected fairly. This brief finds the first unproven on the best available evidence and the second contested at the level of principle, with the leading philosophical objection applying equally to the leading proposed alternative. Neither answer is currently in a state to carry the weight placed on it, and the institutions carry on regardless — which is itself the most robust finding here, because it means legitimacy in practice is supplied by habituation rather than by either argument.

Speculative The long-run stake, stated as the smallest claim the evidence supports. If the build-versus-break asymmetry is real, then delegated institutions are a ratchet that civilisations can lose and cannot easily rebuild, and the correct posture toward existing ones is conservation regardless of whether their construction can be shown to have helped. If it is a measurement artefact, then a great deal of constitutional architecture has been transplanted worldwide on a warrant nobody has ever verified, and the enforcement null suggests that is exactly what happened with judicial review. Those two futures require opposite policies and are currently indistinguishable on the published evidence. Distinguishing them is a cheap piece of empirical work that nobody has done.

Established What already happened, because this timeline usually starts in 1997 and should start earlier. The delegation of monetary policy is the oldest large case and its modern statutory form dates from the late 1990s; the coded record of central-bank legislation runs a century and contains 370 reforms across 155 countries, of which 91 reduced independence. Judicial review, the other mass delegation of decisions to unelected experts, spread through post-war constitutions from 1947 to 1949, again after decolonisation, again after 1989, and is now near-universal in constitutions written since about 1975 — adopted, on the enforcement evidence, without anyone having measured its marginal effect. The two technocratic governments of November 2011 are the closest thing to a controlled comparison this subject has. The written statement of the trade — the duty to explain publicly a decision that departs from expert advice — dates from March 2010 and followed a dismissal.

Established 2024 to 2026: the destruction evidence arrives and the construction evidence does not. A 2026 study of 132 governor transitions establishes the harm from politically motivated removals at 2 to 4 percentage points of inflation; a doubly robust estimate of the reverse direction returns essentially zero with a wide interval. More than 100 United States federal science advisory committees are terminated by executive instrument in eighteen months. A national fiscal framework acquires a statutory trigger for independent forecasting that its own scrutineers call largely performative.

Frontier Late 2020s: several more reversals, and the question is only where. On a base rate of one reform in four reducing independence, and with roughly two-fifths of emerging-market governor transitions politically motivated, further reversals are near-certain over any ten-year window. What is not forecastable is whether any of them will be in an advanced economy with a long delegation history, which is the case that would actually test the conservation argument.

Frontier Late 2020s: whether anyone runs the pre-registered re-analysis that would settle the independence–inflation dispute. This is the cheapest variable on this list and the least likely to move, because no career is advanced by adjudicating between five existing papers.

Speculative Roughly a generation out: the plausible split is stage-specific and is close to the current arrangement. Delegation survives comfortably at the design and implementation stages, where publics grant it by about nine points, and stays contested at the decision stage, where they do not — which would mean the present settlement is stable rather than transitional. The tail measure is the one to watch: monthly trust in one major delegated body ran flat through the worst inflation of its currency's existence while the share reporting no trust at all rose from about 9% to 12%. A zero-trust share that keeps growing while the mean stays flat is what an unmeasured problem looks like.

Handwave Beyond that, no date. Whether the turnout series continues to fall where technocratic cabinets are used, whether the philosophical dispute about blind deference is ever resolved in either direction, and whether any polity adopts a delegation design that survives both abolition and defunding are not forecastable from anything in the current record. The oldest institution in this brief is younger than the average lifespan of a written constitution, and no design decision should be made on a guess about any of the three.

Handwave Anything beyond that horizon. The oldest institution discussed in this brief is younger than the mean lifespan of a written constitution, which the endurance literature puts at roughly nineteen years. A body of doctrine about delegating authority to unelected experts, resting on institutions that have not yet outlived the average constitution, does not support century-scale forecasting, and this brief declines to supply any.

• Depends on Nothing on this map. Delegation waits on no result produced by another brief; the constraints are constitutional, statutory, electoral and normative, and every one of them is recorded above. The absence of an outcome measure is a missing instrument rather than a missing finding, which is why no dependency edge is claimed for it.
• Requires (not on this map) Instrument independence with the goal retained by an elected principal, which the model design has and the imitators that failed usually did not. An override that exists but is costly to use. Delegation confined to the stages publics will actually grant. Durability across a change of government, which one coded central-bank reform in four fails. Statutory protection and a secured appropriation together, which the observed record shows no body has ever had at once — a statute defends against abolition and not defunding, an appropriation defends against defunding and not a change of majority. And a calibration instrument for expert judgement, of the kind produced once, by accident, when a research team elicited 126 expert forecasts before analysing an eleven-year randomised result and found them accurate about infrastructure and wrong about institutions. All six are institutional or methodological capabilities rather than discoveries.
• Enables In principle, credible commitment for any brief whose bottleneck is a government's inability to bind itself. No typed enabling edge is claimed — on this evidence the commitment is revoked in roughly one reform out of four, and an edge asserting durability would assert exactly what the record denies.
• Adjacent Scientific advisory institutions, where the same authority question is asked of bodies that advise rather than decide; Future democracies, which supplies the leading proposed alternative and its own mandate problem; Institutional design, one level up and carrying the enforcement null; Scientific governance models, borrowed from for the warrant question only. Outside the map: monetary and fiscal economics, comparative public opinion, and the philosophy of democratic legitimacy, where the normative argument is actually being conducted.

“Independent central banks lowered inflation.” Established Not a settled finding. Five peer-reviewed or institutional results on overlapping data disagree about where the effect exists — middle-constraint political systems only, non-advanced economies only, advanced economies only, nowhere once initial inflation is controlled for, or essentially nowhere on the most careful identification available, which returns +0.01 percentage points with a 95% interval from −1.48 to +1.50 and whose authors write that an inflation-boosting effect cannot be ruled out. And the famous large magnitudes are the mean-inflation version: the same paper reports that using the median shrinks them by about 60%, from twenty-two points to four.

“Well, the evidence for central bank independence is overwhelming anyway.” Established It is overwhelming in one direction only. What is well identified is the harm from removing independence: 132 governor transitions across 28 central banks and 70% of world GDP, with politically motivated removals followed by inflation up 2 to 4 points. Citing that as evidence for the value of building the institution runs the argument backwards, and the difference matters because a conservation argument does not justify creating new delegations.

“An independent constitutional court is what makes constitutional rights real.” Established The interaction between constitutionalising a right and having an independent court to enforce it is predominantly non-significant across nine rights and 1946–2010. What predicts better outcomes is whether the right is organisational — unions, parties, religious bodies — because those create constituencies that can defend it outside a courtroom. Judicial review has nonetheless been adopted by nearly every constitution written in the past fifty years: one of the largest institutional transplants in history, and its marginal effect has never been detected.

“Technocratic institutional forms spread worldwide because they work.” Established The better-supported explanation is isomorphic mimicry, defined in the standard diagnosis as “the tendency to introduce reforms that enhance an entity's external legitimacy and support, even when they do not demonstrably improve performance,” producing a capability trap in which governments adopt reforms to secure financing and never improve. Adoption is evidence about incentives to adopt, not about effects. The same authors state that their proposed remedy has no outcome evidence, and this brief preserves that: the diagnosis is well evidenced and the cure is not.

“Technocratic government restored market confidence.” Established Does not survive the monthly series. The largest fall in sovereign spreads happens between June and September 2012, bracketing a central-bank remark of 26 July and a bond-buying announcement of 6 September, and Italian yields in mid-2012 — seven months into the technocratic government — were higher than in October 2011 under the government it replaced. The confidence event is dated to the central bank, not to the technocrats, and over the whole episode GDP, unemployment and debt all moved the wrong way; the figures are in the civilisational section below.

“Publics have had enough of experts.” Established Contradicted by the survey record. 56% across 31 democracies say technocrats rather than politicians should decide, flat across twenty-two years and three waves; a nine-country study of 9,449 respondents finds no populism without expertise, with populist respondents wanting outsider experts rather than fewer. Frontier The honest qualification is the turnout study — 3.4 percentage points lower turnout from a fully partisan to a fully technocratic cabinet across 371 elections — and a zero-trust tail that grew from about 9% to 12% while the mean held flat. Neither is visible in the survey means, and this brief refuses to resolve the two against each other.

“The failure mode is elected governments ignoring expert advice.” Established It is not, in any well-documented case in this record. The pandemic inquiry found “a degree of groupthink,” exactly one of 87 emergency-advisory participants from outside domestic institutions, a chief scientific adviser stating there was “no significant disagreement between the Government and its scientific advisers on anything material,” and a health secretary saying “I bitterly regret that I did not overrule that scientific advice.” The measured political failure was excessive deference. Elsewhere the failure mode is administrative — committees closed by executive instrument, posts abolished on a change of president, appropriations zeroed, with nobody engaging the content. A framing that asks whether expert authority is being defied is asking about the rarest case in the file.

“Political interference with expertise comes from politicians.” Established Not exclusively. Nine environmental organisations lobbied in July 2014 for the abolition of a European chief scientific adviser post on the grounds that the incumbent's position on genetically modified crops was not representative of the scientific consensus; the post ended in October 2014 and was never restored in that form. Advocacy organisations objecting to inconvenient expertise is a documented and successful mode of attack, and any account of expert independence that models only ministerial pressure is describing half the record.

“Conflict-of-interest disclosure is what protects the credibility of expert authority with the public.” Established With the public, it appears to do almost nothing. Across three factorial experiments (N = 320) presenting a scientific claim and a disagreeing source, the disclosed vested interests of the disagreeing party had no significant effect on trust or belief in two of three samples, while that party's expertise had a robust effect in all three (d = 0.39, 0.26, 0.27). Institutions spend their procedural budget on conflict screening and almost none on making expertise gradients legible; the audience is doing the opposite. Two readings are available and this brief does not choose between them: either institutions should reallocate, or conflict screening protects the internal decision rather than its public reception and simply is not a legitimacy instrument.

“Publishing uncertainty and reasoning makes expert bodies more trusted.” Established Not on average, and the correction is the most useful thing in this subject. Two pre-registered United Kingdom trials totalling 3,962 participants found balanced, uncertainty-disclosing text rated as trustworthy as persuasive text or slightly more so (d between 0.08 and 0.20) — and in both, the persuasive version was better at persuading. Two United States experiments (n = 600 and n = 1,001) then found the effect is a crossover: where evidence agreed with prior beliefs, adding uncertainty reduced trust in the information and the source; where it disagreed, it raised both, with interaction terms of β = −0.34 and −0.26 in the first study and −0.40 and −0.41 in the second, and no evidence of polarisation. A German experiment (N = 800) adds that disclosing uncertainty in advance buffers the loss of trust when the advice later turns out to be wrong. Transparency is not a trust-maximising move. It is a trust-redistributing move — reach into the sceptical part of the audience, plus insurance against being wrong.

“Sortition solves the legitimacy problem that technocracy has.” Established It has a version of the same problem and one of its own. The philosophical objection to empowered mini-publics is that a citizen who did not deliberate has no reason to expect the body's decisions to track her interests — politically blind deference, the same structure as deference to an expert board. And the lottery is not a lottery over the public. Roughly 7% of invitations are answered, and conventionally recruited assembly members reported 96% turnout in German federal elections against a national rate near 77%. A stratified lottery corrects for age, gender, education and region and does not correct for the disposition to answer a letter from the government, the trait most closely associated with political participation. The strongest reply is peer-reviewed and unrefuted, and its author concedes that informed participants risk ceasing to be representative.

“Making technocratic performance measurable improves it.” Established The best-documented case says the opposite. Over 70 countries formed regulatory reform committees oriented specifically to one governance index's indicators; at least nine of the twenty most rank-volatile countries in one region received over US$100 million in donor-funded business-environment projects between 2005 and 2019; over fifteen years 111 countries swung more than 40 places out of 190. An independent investigation reporting in September 2021 found scores altered under internal pressure for four named countries, no written procedures for methodology change, unwritten rules permitting manipulation, and a device-deletion policy that destroyed evidence. The index was cancelled. When you make an institutional target legible and scored, you get the score rather than the institution.

“Delegated bodies deliver because they are insulated and monitored.” Established Insulation and monitoring are not the same instrument and the evidence separates them. In hand-coded engineering assessments of 4,700 public projects across 63 federal organisations, plus a companion study of 3,628 projects across 31, the autonomy index is robustly positively correlated with project initiation, full completion and average completion rate, while the incentives-and-monitoring index is robustly negatively correlated with all three. The summary source gives directions and no coefficients, so no magnitude is attached here. If it generalises, the part of technocratic delegation that works is the insulation rather than the accountability apparatus bolted to it.

Two factual corrections that circulate widely. Established The technocratic prime minister's list did not collapse in 2018 — it took 8.30% in 2013, contested 2018 inside a list that polled 1.3%, and was struck off the register in 2019. And the flagship central bank's target has not been a retail-price measure at 2.5% since December 2003; it is a consumer-price measure at 2%, and quoting the older figure is two decades out of date.