Civic Technology

Civic technology is software built to connect people to government from the citizen's side of the interface: platforms for proposing and voting on policy, tools for reporting broken infrastructure, systems for making freedom-of-information requests, consultation software that clusters opinion to find consensus, crisis maps assembled by volunteers, and open data portals published in the expectation that somebody will use them. The claim, stated in one sentence and held with remarkable consistency for two decades, is that lowering the cost of citizen input changes what governments do.

The evidence for that claim is unusual in one respect and depressing in another. Unusual, because these platforms generate complete administrative records — every proposal, every signature, every report, every dataset view is logged, so this is one of the few areas of governance research where the denominator exists. Depressing, because the field instrumented its inputs exhaustively and its outputs almost not at all. A study of Decidim Barcelona reports 10,860 proposals and 18,192 comments broken down by sentiment, and no data on which proposals were accepted. A corpus of Taiwan's national platform runs to 20,923 proposals over nine years with no outcome variable. The peer-reviewed review of crowdsourced crisis mapping, after fifteen years and 51 deployments, lists “rigorous research on actual impact” as future work.

This brief takes the field's failures seriously and refuses two easy versions of them. It does not say that civic technology has been shown not to work, because that is a stronger claim than the record supports. It says something narrower and harder: that after roughly two decades and several hundred million dollars, the theory of change has never been evaluated on the thing it claims to do, and that where outcomes have been measured — in complaint routing, where a platform's own operator publishes the numbers — the effects are small, unevenly distributed, and pointed away from the people with the worst problems.

Where this brief stops. Future Civil Services owns the permanent bureaucracy and its workforce: headcount, pay, politicisation, and the in-house digital service units — 18F, the United States Digital Service, the Government Digital Service, the Canadian Digital Service — considered as employers and organisational forms. This brief owns technology built for and by civil society and citizens: platforms whose users are members of the public, whose builders are typically non-governmental, and whose asserted mechanism runs from citizen input to government action. The arbitration rule where they meet is who stands on the citizen side of the interface. The dissolution of 18F appears here only as a boundary fact, because it removed the government counterparty that civil-society technologists worked through; the workforce analysis is next door.

Two other boundaries are worth naming. Smart Cities owns municipal instrumentation, where the sensor faces the city rather than the citizen. Future Democracies owns deliberative instruments — citizens' assemblies, sortition, mini-publics — that reach the same participation question without software, and reach a strikingly similar verdict.

Established The most rigorous criticism of complaint-routing platforms comes from the operator of one, and it is more damning than any external critique. mySociety publishes a page instructing readers that FixMyStreet report data “should NOT be used to make comparisons at the national or local level,” and gives seven reasons: the platform competes with councils' own forms, social media, telephone and email, so it captures a fraction of reports; councils that adopt its paid product appear to have more problems than councils that do not; a high count may come from one active user, whose moving house moves the apparent problem rate; some councils refuse third-party reports entirely, creating data voids; resolution status depends on users returning to mark a problem fixed, which is unreliable; category definitions differ between councils; and reporting is demographically skewed in a way that contradicts real-world incidence. In February 2024 the organisation published a further statement recording that press outlets use its data to rank areas on potholes and fly-tipping, that when asked to supply such comparisons “we always say no,” and that because the platform is open source with publicly displayed reports it cannot always stop the story being written.

Established The redistribution hypothesis is confirmed, and confirmed against interest. mySociety's own analysis across English Lower Super Output Areas found that dog-fouling incidence peaks in the bottom two deprivation deciles while dog-fouling reports peak in the middle deciles. Rubbish reports concentrated at lower deprivation. Domain-level detail beat the composite index: the crime domain was the strongest predictor of rubbish reports, income deprivation tracked dog fouling better than multiple deprivation did, housing costs tracked abandoned vehicles — and, cutting the other way, drainage complaints concentrated in the most deprived areas against the general trend. A platform that lowers the cost of complaining lowers it for everyone but is taken up disproportionately by people who could already complain, so the queue it produces is a worse map of where problems are than the underlying distribution of problems. The platform does not merely fail to close the equity gap; it can generate a data product that misdirects attention away from the worst-affected areas while looking like evidence.

Established Where an outcome has been measured, the effect is real, small and heterogeneous. In a mySociety analysis of 663,591 reports, restricted to councils that mark fixes themselves, the 2019 baseline was 34% of reports fixed. Thirty-five per cent of reports carried a photograph, and the adjusted model put the fix rate with a photograph at 37.17% — an absolute gain of 3.2 percentage points (95% confidence interval 2.9 to 3.5) and a relative gain of 15% (13.5 to 16.6). Category effects ran both ways: highways enquiries +11.2 points and parks +9.7, against street lights −1.8 and rights of way −4.0. The authors flag their own selection problem — user-reported fixes were excluded because status updates come disproportionately from superusers, who also post more photographs, so the repair rates in that dataset are “not representative of the repair rates in FixMyStreet in general.” This is what a good result in this field looks like, and it is about an interface feature rather than about the platform.

Established Crisis mapping has a measured actionability rate and it is 0.35%. A peer-reviewed analysis of 51 mapping deployments between 2010 and 2016, with ten expert interviews split evenly between formal humanitarian organisations and volunteer technical communities, reports that after Typhoon Haiyan in 2013 volunteers gathered and processed nearly 230,000 tweets within two days, of which only 800 provided emergency responders with relevant, actionable data. The arithmetic is this brief's; the inputs are sourced. The authors introduce the figure with the warning that it “does not fall into the fallacy that some data is good and more is better.”

Established And the binding constraint on crisis mapping was never the data. In both flagship deployments — Haiti in 2010 and Nepal in 2015 — the crowdsourced datasets were, in the same review's words, “dismissed because of a lack of organisational infrastructure, and humanitarian politics.” The maps were made. Nobody on the response side had arranged to receive them. The barriers named are the absence of pre-crisis relationships, inconsistent cross-organisational collaboration, political limits on open data, and responders too overworked to process unstructured information during an acute emergency. The authors' own summary is the sentence to carry: this is “a human story not a technical one.”

Frontier The field's institutional layer died before its software did, and it did not die of running out of money. Code for America announced on 1 February 2023 that it was ending its relationship with roughly 60 volunteer brigades. Contractual agreements expired 30 June 2023 and the grace period for groups carrying the “Code for” name ran to 31 December 2023. The programme being closed carried more than $3 million a year and fourteen full-time employees plus contractors as of 2022; the organisation had received a $100 million donation in 2022 to launch a different programme. Seventeen cities regrouped as the Alliance for Civic Technologists. A brigade captain's assessment was that this was “likely the death knell for Code for New Orleans.” The organisation's stated reasons were a mission shift, post-pandemic resource constraints, declining volunteerism, and a judgement that it was “not best suited to serve as a central supporting organization for a network of volunteers.” The mortality mechanism in the best-documented closure in this field is portfolio reprioritisation by a well-capitalised intermediary, not the exhaustion of a grant.

Established Open data's gap between publication and use is measured at both ends and both measurements are bad. An index covering 115 countries and jurisdictions found only 7% of assessed datasets truly open, with nine in ten government datasets closed; by category, election results 11%, budgets 10%, company registries 5%, spending 3% and land ownership 1%. Openness is highest where the political cost of publishing is lowest. At the use end, an analysis of over 160,000 geospatial datasets across six national and international portals found heavy-tailed distributions in which most datasets are rarely viewed: median views of 15 on the United States portal, and mean downloads of 0.6 per dataset on the Humanitarian Data Exchange against 25 views. Publication was achieved; use was not.

Frontier Digital public infrastructure is the field's only success at scale, and it succeeded by inverting everything civic technology stands for. Aadhaar reached 1.377 billion unique identity numbers by mid-2023, is credited with lifting adult bank-account ownership from 35% to 78% between 2011 and 2021, and carried $10 billion in pandemic transfers to over 200 million women. Against that, field survey work in 32 villages of Jharkhand covering roughly 1,000 households found exclusion errors “as high as 20% in areas where biometric authentication was required for every sale” of subsidised food. Both numbers are real, and any account carrying one and not the other is advocacy. What makes this the counterexample rather than the maturity of the field is that the stack is state-built rather than civil-society-built, mandatory rather than voluntary, infrastructural rather than interface-level, and its participation mechanism is that you cannot eat without it.

Frontier The honest verdict is worse than “a record of failure,” because most of the record was never measured. Not one source consulted for this brief reports an outcome measure — a government decision changed, a budget reallocated, a policy reversed, a service improved against a comparison group — attributable to a civic technology platform at any scale. The evidence base consists overwhelmingly of input counts, attitudinal surveys and case narratives. That is not a demonstration that the theory of change is false. It is a demonstration that the burden of proof has never moved.

The genuine open questions in this subject divide into three, and only one of them is about software.

Frontier First and largest: does citizen input, delivered by any channel, change a decision that would otherwise have gone the other way? This is the field's founding claim and it is unresolved in the strong form. What is established is that input volume rises with a platform, that responsiveness feeds further participation, that an interface variable can shift a bureaucratic outcome by a few percentage points, and that the composition of input skews away from the most deprived. What is not established is counterfactual effect on a decision. The strongest causal result in the whole participation literature remains a pre-digital one — a panel across 3,651 comparable Brazilian areas from 1990 to 2004, where participatory budgeting raised the health-and-sanitation budget share by 2 to 3 percentage points and cut infant mortality by 1 to 2 per 1,000 — and it involves no technology at all.

Frontier Second: what actually kills these projects, and at what rate. There is no peer-reviewed survival analysis of civic technology projects. The nearest thing is a curated register tracking approximately 70 failures, whose maintainers state that it misses projects that never got off the ground. It has no denominator, so no mortality rate can be computed from it and this brief does not report one. What it does supply is a taxonomy of cause — market mismatch, where users do not engage intensely enough with generic civic platforms to sustain operations; absent user research, where “only a minority of civic tech organisations do any user research before choosing a tech tool”; collective-action complexity, where a platform requiring coordination loses to a hashtag; portfolio reprioritisation; and dependency obsolescence, with one project killed by the withdrawal of an identity service it relied on.

Frontier The register's most useful finding is that capital does not predict survival. Nine named dead platforms — Jumo, ChangeByUs, VoteIQ, Vote.com, Voter.com, Hotsoup.com, Speakout.com, Ruck.us and Votizen — “cumulatively… raised more than $20 million” and all shut down; Brigade launched with $9.3 million from named venture investors and remains barely operational. The one case quantified end to end is Citizinvestor, which hosted 76 projects, funded 37, and collected $282,737 — about 83% of pledged amounts — before its founders concluded it “didn't scale the way we wanted.”

Frontier Third: whether the mortality is a property of the field or of one business model inside it. The graveyard is populated overwhelmingly by venture-funded, generic, national-scale engagement platforms built on a social-network thesis. The tools that have run a decade or more are narrow and jurisdiction-bound: a street-reporting platform with hundreds of thousands of reports a year, a freedom-of-information channel that accounts for 15 to 21% of requests to the bodies it covers, participation software that Madrid's citizen-participation directorate was still presenting on in December 2025 roughly ten years after its development, and a national petition platform that has taken proposals under three administrations. If civic technology has a survival problem it may be specific to the species that tried to be a startup, and the field's own sustainability discourse — which is mostly about grant cycles — would then have been aimed at the wrong failure mode for a decade.

Frontier A fourth question, narrower and more tractable: does a platform-routed complaint fare differently from a telephoned one? No source consulted reports a counterfactual channel comparison. The 34% fix rate is a fact about reports that arrived through a website; whether the same pothole reported by telephone is fixed more or less often is unknown. This is the field's most obvious missing experiment and it does not appear to have been run.

Frontier Fifth: whether the diffusion counts mean anything. Installation registers list hundreds of deployments of participation software. Nobody publishes how many are live, how many have received a proposal in the last year, or how many were installed for one process and abandoned. An instance with no live process is counted identically to a flagship deployment. The single most valuable unbuilt dataset in this subject is a longitudinal census of participation-platform installations by activity, and it could be assembled from public endpoints by one competent researcher in a season.

Handwave What merely sounds open: whether better portals, better metadata and better discovery would unlock open data's value. This has been the field's standing remedy for fifteen years and the evidence contradicts it. Across six portals and more than 160,000 geospatial datasets, seventeen metadata quality metrics produced medians ranging from about 0.74 to below 0.5 — and the correlation between metadata quality and dataset use was weak. That points the causal arrow at demand rather than supply. Most of these datasets were not wanted. Publishing them was cheap, measurable and politically legible, which is why it happened.

Established The first bottleneck is that the receiving institution has no duty to act, and this binds before anything technical. Across every deployment in this brief, clearing a threshold obliges a government to respond, at most, and nowhere obliges it to decide. Taiwan's platform mandates a written ministerial reply within two months; Madrid's produced two threshold-clearing proposals in a decade and enacted neither; the co-creator of Taiwan's consultation process attributes its decline to recommendations lacking binding force, with the consequence that legislators did not take the process seriously. A pipeline whose terminus is a letter is a suggestion box with analytics.

Established The second is that the queue misrepresents need. The operator's own deprivation analysis shows reports peaking in middle deprivation deciles while incidence peaks in the bottom two. This binds harder than low participation, because a low-volume platform is merely ineffective while a skewed one is actively misleading to the authority receiving it — and the authority has no independent incidence data with which to correct it.

Established The third is the absence of an outcome measure anywhere in the field. Input is logged completely and output is not logged at all. This is not merely an evidentiary problem; it is a governance problem, because a funder cannot allocate against outcomes that nobody records, and so allocates against adoption metrics instead. The result is fifteen years of investment steered by installation counts, proposal counts and dataset counts — all of which are supply-side and all of which are cheap to move.

Frontier The fourth is the intermediary layer, which turns out to be the most fragile part of the stack. The volunteer network that supplied labour, local knowledge and legitimacy to American civic technology was closed by its own host in 2023 at a scale of more than $3 million a year and fourteen staff, in the same period that the sector's principal government counterparty was dissolved by executive action. Neither closure followed a published evaluation of the work. The field lost its principal non-governmental convener and its principal governmental counterparty within roughly twenty-five months, for unrelated reasons, neither of which was a failure of the work.

Frontier The fifth is fiscal hosting, which is a real dependency and is treated as an administrative detail. When the volunteer network was cut loose, groups had either to incorporate as charities or find a new fiscal sponsor; some moved to a host taking an 8% cut of funds raised, others to city-specific foundations. A volunteer technology group is not, in general, capable of running its own charitable entity, and the capacity to do so is unrelated to the capacity to build the tool. This brief could not verify what subsequently happened to the principal fiscal host used, and records that as a gap rather than completing the story.

Established The sixth binds only the digital public infrastructure case, and it binds on people rather than on projects. A single mandatory rail with a biometric gate produces exclusion at the rate of its authentication failure, and in the surveyed case that rate reached 20% where authentication was required for every transaction in a low-connectivity state. The prescription in the policy-ethics literature is flexible identity verification, which is precisely what a single-rail architecture is least able to supply, because the property that makes the rail cheap is that there is only one of it.

Established What is not a bottleneck, stated plainly because the field's own discourse says otherwise. Technical capability is not binding: this is ordinary web software and the deliberation methods predate it by decades. Cost is not binding: the platforms are cheap relative to the budgets they address. Data quality in crisis mapping is not the binding constraint, because a feed nobody has arranged to receive is not improved by cleaning it. And metadata quality is not the binding constraint on open data reuse, because the measured correlation between metadata quality and use is weak.

Established Nothing here waits on a research result. This is ordinary web software; the deliberation methods predate it by decades; the identity and payments primitives in the digital public infrastructure case are conventional cryptography and conventional database engineering. Whatever is holding this field up, it is not a discovery.

Established What civic technology waits on is institutional and it is a short list. A threshold calibrated to be reachable, since the measured clearance rates in this brief run from 2.1% down to roughly one in ten thousand. A binding duty to act on input that clears it, absent from every deployment in the record. Continuity across a change of administration, whose absence destroyed 182 approved proposals in one city. A formal receiving point inside the responding organisation, whose absence caused crowdsourced crisis data to be discarded in both Haiti and Nepal. And administrative capacity to deliver whatever the process decides, which no platform supplies and which several assume.

Frontier It also waits on a fiscal and organisational host, which is a dependency the field has consistently mispriced. Volunteer groups need charitable status, insurance, safeguarding and a bank account before they need a repository. When the national intermediary withdrew, that layer had to be reconstituted city by city, some of it through hosts taking a percentage of funds raised. The capability to run an entity is orthogonal to the capability to build a tool, and the field's funding models have generally assumed one implies the other.

Established What depends on it is narrower than the field claims and wider than the sceptics allow. Future Democracies depends on it for the delivery layer of any at-scale deliberative instrument, since a citizens' assembly that must reach millions cannot be convened in a room. Smart Cities shares its data-quality problem exactly: an instrument that measures reporting behaviour and is read as measuring incidence. AI-Assisted Governance inherits its evaluation deficit wholesale, and inherits the worse version of it, because a model trained on a skewed complaint queue reproduces the skew and launders it. And digital public infrastructure is now the substrate on which welfare delivery in several large states runs, so the exclusion rate is a dependency of the entitlement rather than of the software.

Established The obvious experiment is to randomise the threshold. These platforms are software and the signature requirement is a configuration value. Varying it across comparable municipalities and measuring proposals cleared, proposals enacted and subsequent participation would answer the field's central design question within a single budget cycle. It has never been done, and there is no technical or ethical barrier to doing it.

Established The second is a channel comparison, and it is overdue by about fifteen years. Route matched complaints through a web platform and through a council telephone line, and measure resolution. The platform-side number is already known — 34% fixed across 663,591 reports in a filtered dataset — and the comparison number does not exist anywhere in the literature consulted. Without it, every claim that civic technology improves service delivery is a claim about an absolute rate with no counterfactual.

Established The third has already been run and the field should stop ignoring it. Randomising the response rather than the platform is the design that produced the one clean causal result in complaint routing: users whose first report was fixed were 57% more likely to file a second, 24.1% against 13.6%, across a complete platform population of 399,364 reports from 154,957 unique users. That establishes that responsiveness drives participation, which runs opposite to the advocacy claim that participation drives responsiveness.

Established A negative result worth recording as an experiment in its own right. The highest-standing design in this subject is a field experiment across 200 zones with 23,856 reports over nine months and 679 physically measured waste piles, which found treatment effects of −4.23 m² (p = 0.112) and −7.78 m² (p = 0.303), no significant increase in cleanups, roughly 10% citizen response, and a programme eventually abandoned over cost and report reliability. It measured a physical outcome and found nothing. A field with one well-powered outcome study, whose result is null, should be describing itself very differently than it does.

Frontier The natural experiment already running in Taiwan is the most informative thing in the subject and it is a control pair. One national platform and one municipal platform, similar technology, overlapping era, same polity. The national one has taken proposals for nine years across administrations. The municipal one, which drew over 1,400 members and 40,000 visitors in its earliest form, is described in the research literature as obsolete, with the decline attributed in part to a 2016 episode in which “the city government offered only pre-selected options for public voting,” which “eroded public trust in the platform.” The variable that differs is not the software.

Frontier The installation census is the cheapest high-value study available. Take a published installations register, probe each instance for a live process and a proposal in the last twelve months, and report the survival curve. It requires no funding, no access negotiation and no new instrumentation, and it would replace the field's most-quoted diffusion statistic with a measured one.

Frontier And in digital public infrastructure, the decisive study is an authentication-failure audit published by the operator. The exclusion figure this brief carries comes from academic field survey work in 32 villages; the commentary carrying it states that no official exclusion data exists. An identity authority that publishes per-district authentication failure rates alongside its enrolment totals would settle a question currently argued between a projection and a survey.

Established The engineering that matters in this subject is unglamorous, and most of it is about the loop back to the citizen rather than the intake. Intake is a solved problem: a form, a map pin, a photograph, a signature count. What is repeatedly under-built is the disposition record — publishing what happened to each item, including rejections with reasons — and the funnel instrumentation that would let anyone say what fraction of input reached a decision. The documented citizen complaints in the record are dominated by silence rather than refusal.

Established The threshold is a configuration value, and it is the single most consequential engineering decision in a participation platform. The measured clearance rates across the deployments in this brief span four orders of magnitude, and they are set by an integer in a settings file rather than by anything about the software.

The two Taiwan rows are not a single series and this brief does not merge them. The 13,853 / 289 figures come from an academic case study covering an earlier period; the 20,923 corpus comes from a 2025 topic-modelling study covering nine years and reports no clearance count. Reconciling them would require data neither source publishes.

Established Report data is not incidence data, and building as though it were is the field's characteristic engineering error. The seven caveats the FixMyStreet operator publishes are each an engineering fact: competing intake channels mean the platform sees a fraction of reports; adoption of the paid council product inflates apparent problem rates; a single power user can dominate a locality; councils that refuse third-party reports produce data voids; fixed status depends on users returning, so stale reports are marked “unknown”; category taxonomies differ per council; and the demographic skew contradicts measured incidence. Any dashboard built on top of that data without reproducing all seven is a misinformation product.

Established Deliberation quality is measurable and the measurement is more encouraging than the outcome record. A study of Decidim Barcelona covering more than 40,000 participating citizens, 10,860 proposals and 18,192 comments — 16,217 first-level and 1,975 replies — classified alignment as 63.03% neutral (10,221), 32.05% positive (5,198) and 4.92% negative (798), and found that negative comments were more likely than neutral or positive ones to trigger complex discussion cascades. Counter-argument generated deliberation rather than suppressing it, which is the opposite of the expectation most platform designers build against. The same paper reports no acceptance or implementation data, which is the recurring shape of this literature.

Frontier Design for the change of administration, because that is when platforms lose their budget, their processes and their previously approved commitments. One city's transition discarded 182 previously approved participatory-budget proposals and halved the forward budget. Continuity is an engineering property — exportable records, durable commitment identifiers, published dispositions that survive a site rebuild — and almost nothing in this field is built for it.

Established Crisis-mapping architecture has the same shape and the same failure. The volunteer side scales: hundreds of mappers, 230,000 tweets triaged in two days. The receiving side does not exist. What the review literature identifies as missing is a formal contact point, pre-crisis relationships, and verification against authoritative datasets that volunteers cannot access. The integration surface is the deliverable, and it is the part nobody builds because it is not software.

Future Democracies is the nearest neighbour and the most useful one to read alongside this brief, because it reaches the participation question through citizens' assemblies, sortition and mini-publics — instruments with no software in them — and arrives at a similar verdict about binding force. The two briefs share the Brazilian participatory-budgeting result deliberately: it is the strongest causal evidence either can cite, and it is offline.

Future Civil Services meets this brief at exactly one place, the government's own digital service units, and the boundary is drawn at who stands on the citizen side of the interface. That brief owns 18F, the United States Digital Service, the Government Digital Service and the Canadian Digital Service as employers and organisational forms; this brief records only what their fate did to the civil-society side of the sector.

Smart Cities shares this brief's central measurement error in a different domain. A municipal sensor network that measures instrumented behaviour and is read as measuring the city has the same epistemic defect as a complaint platform that measures reporting behaviour and is read as measuring incidence, and both fields have responded by improving the instrument rather than by questioning the inference.

AI-Assisted Governance is downstream in a way that should worry both. Complaint queues, proposal corpora and open data portals are the training material for administrative machine learning, and every distributional defect documented here — reports peaking in middle deprivation deciles, inclusion impact at 1.9 out of 10, datasets published because publication was cheap — propagates into a model that will be harder to interrogate than the platform was.

Distributed Governance proposes to route around the institutions this brief identifies as the binding constraint. Read together, the two make an argument neither makes alone: that the constraint is a duty to act, and that a system with no duty to act is not improved by removing the body that could have been given one.

Technocracy and Democracy is adjacent through the exclusion literature. A mandatory identity rail that authenticates 80% of claimants in a low-connectivity district is a technocratic instrument making a distributive decision, and it is doing so without anyone having chosen the threshold at which the decision becomes unacceptable.

Established The defining institutional fact about this field is that its two convening institutions were removed within roughly twenty-five months of each other and neither removal followed an evaluation. On the civil-society side, a national intermediary ended its relationship with approximately 60 volunteer chapters on 1 February 2023, closing a programme carrying more than $3 million a year and fourteen full-time employees, with contractual agreements expiring 30 June 2023 and naming rights lapsing at the end of that year. On the government side, the federal digital consultancy that was the sector's principal counterparty was shut in the early hours of 1 March 2025 as “non-critical,” citing executive orders of 10 and 11 February 2025, with staff locked out immediately. One was a strategic reallocation; the other was an executive action. Neither was a finding about the work.

Established What replaced the intermediary is instructive about the sector's real capacity. Seventeen cities regrouped as a new alliance. Groups had to incorporate as charities or find fiscal sponsors; some moved to a host taking an 8% cut of funds raised, others to city foundations. Six to eight months of transition support were offered. That is what the institutional base of a two-decade-old field looks like when the subsidy is withdrawn: a few dozen volunteer groups shopping for a bank account.

Frontier The funder is the institution that actually governs this field, and it governs by adoption metrics. Because no outcome measure exists, allocation runs on installation counts, proposal counts, user counts and dataset counts — all supply-side, all cheap to move, and all producible without a government agreeing to anything. A $1 million grant supporting seven cities, a $9.3 million venture round for a platform that remains barely operational, and over $20 million cumulatively raised across nine platforms that all shut down are not anomalies in that system; they are what the system produces when the metric it optimises is uncorrelated with the outcome it wants.

Established The regulator, in the sense of a body that could impose a duty to act, does not exist anywhere in this record. Taiwan's national platform comes closest, obliging a ministry to respond in writing within two months of a proposal clearing 5,000 votes in a 60-day window. That is a duty to reply, not a duty to decide. Nowhere in the deployments surveyed does a citizen have a right that a cleared proposal be adjudicated, and the co-creator of one flagship consultation process attributes its decline precisely to recommendations lacking binding force and legislators consequently not taking the process seriously.

Frontier Where an institution does exist, it is a procurement institution, and it is now buying identity rather than participation. More than 50 countries requested multilateral technical support for digital public infrastructure following the model's promotion at the G20, and a United Nations-backed campaign targets 50 countries by 2028. The offer is an open-source identity platform, and the critique on record is that it creates “technological dependencies that can deprive governments of truly sovereign, locally-tailored solutions” with a “de facto lock-out of alternative solutions.” This is a far larger institutional programme than anything the participation wing of the field ever assembled, and it is being procured on the basis of GDP projections — $200 billion to $280 billion by 2030 across seventy low- and middle-income countries — produced by parties committed to the model.

Established The institution that has never existed is an evaluator. There is no independent body that assesses civic technology deployments against outcomes, no register of results, and no requirement that a funded project publish what happened. The consequence is visible in the source list of this brief: the most reliable evidence in the subject comes from a platform operator volunteering against-interest findings on its own blog, and from academics working on adjacent questions. A field with a curated graveyard and no evaluator has organised itself to remember its failures anecdotally and its successes promotionally, which is the exact inverse of what an evidence base requires.

Established Apply the interested-party rule harder here than anywhere else in this category, because in this subject nearly every substantive success claim originates with the platform, its foundation, or the deploying government. Case counts come from projects. Installation counts come from projects. Impact narratives come from foundations that funded the projects. A city's flagship self-assessment can carry an outstanding progress rating and sit on a multilateral partnership's website alongside genuinely independent assessments while being the city's own evaluation of itself. This brief marks interested parties in its reading list and, where an interested party's account of its own evaluation contains no numbers, says so rather than treating the absence as neutral.

Established The evidence-quality asymmetry in this field runs the opposite way to the usual one, and it deserves credit where it falls. The two organisations that publish results against their own interest — a street-reporting operator that documents seven reasons its data should not be used the way the sector uses it, and the same operator publishing a null result about its own freedom-of-information product, where the platform's 38.4% success rate sits within about a percentage point of the official 34.1% — are more reliable than any funder or index in the subject. A field that penalised that candour would end up with worse evidence, and the incentive structure currently does penalise it, because the honest number is the one that loses the next grant.

Established The representation question is structural rather than incidental, and it now has a second independent confirmation. Complaint volume peaks in middle deprivation deciles while incidence peaks in the bottom two. Open data's impact on social inclusion rates 1.9 out of 10 with 6% of governments showing relevant impact on marginalised groups. Two instruments, two domains, one distributional signature. Almost no deployment publishes the demographics of who cleared a threshold against who lives in the jurisdiction, and until they do, every equity claim in this field is unfalsifiable by design.

Frontier Public money and public accountability meet at the identity rail, and the accounting is not available. The exclusion figure this brief carries — up to 20% in areas where biometric authentication was required for every sale — comes from academic field survey work in 32 villages, and the commentary carrying it states that no official exclusion data exists. Separately, parliamentary scrutiny of a national audit produced an acknowledgement that active identity numbers may exceed the country's population, alongside a deduplication exercise in which roughly 15.5 million death registrations were received from 24 jurisdictions and about 11.7 million numbers deactivated. A system that mediates food entitlement for over a billion people publishes its enrolment totals and not its authentication failure rates, and that asymmetry is an ethical choice rather than a technical limitation.

Frontier Opportunity cost is the argument this field most needs to have with itself and never does. The Brazilian result was obtained by delegating binding budget authority. The measured effect of a photograph on a street report is 3.2 percentage points. Both are worth having; they are not in the same class, and they compete for the same reformist attention and the same municipal willingness to try something. Every hour spent improving a portal's metadata — a remedy the evidence says is weakly correlated with use — is an hour not spent arguing for a duty to act on what the portal produces.

Four questions this brief cannot answer and states as an obligation. Whether a platform-routed complaint fares better or worse than a telephoned one, which nobody has measured. What fraction of installed participation instances are live, which nobody has counted. Whether the exclusion rate found in one low-connectivity state generalises, which it must not be assumed to do. And whether any civic technology deployment anywhere has changed a government decision that would otherwise have gone the other way — a question this brief searched for and could not answer in either direction, which it records as a search result rather than as proof of absence.

Established The civilisational finding is that the strongest causal evidence in the entire participation literature involves no technology at all. A panel analysis across 3,651 comparable areas covering Brazil's municipalities from 1990 to 2004 found that participatory budgeting raised the health-and-sanitation budget share by 2 to 3 percentage points — 20 to 30% of that category's baseline share — and reduced infant mortality by 1 to 2 per 1,000, about 5 to 10% of the baseline rate. That is what a demonstrated participation effect looks like. It was produced by binding budget authority delegated to assemblies of people in rooms, and every element of the mechanism that mattered was institutional.

Established The general principle this case illustrates is about what gets measured when measurement is cheap. Civic technology instrumented the half of its own mechanism that costs nothing to instrument — proposals, signatures, reports, installations, dataset views — and left the half that would require negotiation with a government entirely unmeasured. Fifteen years of funding then flowed toward the metrics that existed. This is not a failure peculiar to this field; it is what happens whenever a supply-side metric is free and a demand-side outcome is expensive, and it should be expected in any domain where a technology intermediates between a citizen and an institution.

Frontier The second principle is more uncomfortable and comes from the field's own natural experiment. Taipei's platform did not decay slowly; the research literature dates its collapse to a single episode of a government offering only pre-selected options and thereby eroding trust. Participation platforms do not die from one bad decision unless participants had been treating them as consequential. The mechanism this suggests is not that better interfaces produce better governance, but that visible interfaces make bad faith legible — and that the cost of the legibility falls on the platform rather than on the government. If that is the real mechanism, civic technology has been selling the wrong product for two decades, and its actual function is closer to an accountability tripwire than to a participation channel.

Frontier At the largest scale, the case that matters is not participation at all; it is the identity rail. A single mandatory authentication layer intermediating access to food, cash transfers and banking for over a billion people is the most consequential thing anyone in this field has built, and it was built by a state rather than by civil society. Its benefits are enormous and its exclusion is measured in the same units as the benefit — households, not percentages of uptime. A civilisation that routes entitlement through one rail has made an irreversible bet that the rail's failure mode is tolerable, and it has generally made that bet without publishing the failure rate.

Established What already happened, because this timeline usually starts too late. Street-reporting and freedom-of-information platforms have been operating continuously since the mid-2000s. Participation software in the Consul lineage dates from roughly 2015; Taiwan's national platform opened on 10 September 2015; Decidim Barcelona's first large process generated 10,860 proposals shortly after. Crisis mapping's founding deployment was the Haiti earthquake of January 2010, and the independent evaluation of it was published in 2011. The field is not young, and the absence of outcome evidence is therefore not a function of immaturity.

Established 2016 to 2022: the municipal participation platforms peak and one of them dies. Taipei's iVoting ran from 13 April 2017 to 22 June 2022 and accumulated 454 proposals; the research literature dates its decline to a 2016 controversy over pre-selected voting options and describes it as obsolete. Madrid's platform accumulated more than 21,000 proposals from 2015 with two clearing its threshold, and a change of administration discarded 182 previously approved participatory-budget proposals and halved the forward budget.

Established 2023 to 2025: the institutional layer contracts. Code for America announced the end of its brigade relationship on 1 February 2023, with agreements expiring 30 June 2023 and naming rights lapsing 31 December 2023; seventeen cities regrouped under a new alliance. 18F was shut in the early hours of 1 March 2025 as “non-critical,” citing executive orders of 10 and 11 February 2025. Neither closure was preceded by a published evaluation.

Frontier 2023 onward: digital public infrastructure becomes the framing. More than 50 countries requested multilateral technical support for such systems following India's promotion of the model at the G20, and a campaign backed by a United Nations agency targets 50 countries by 2028. Whether the exclusion literature reaches those procurement decisions before the rails are laid is the live question, and nothing in the record suggests it will.

Handwave Any projection of when civic technology's theory of change will be tested. The experiments that would test it — a randomised threshold, a channel comparison, an installation census — are cheap, unblocked and unscheduled. They have been available for a decade and nobody has run them. A date attached to a study nobody has funded is not a forecast, and this brief declines to supply one.

Handwave Any projection of platform survival. There is no survival curve for this field, no denominator, and no register with a defensible sampling frame. Statements of the form “most civic tech projects fail within N years” are not supported by anything this brief could verify.

Speculative The plausible settlement is civic technology as reporting and transactional infrastructure rather than as a decision channel. That is where the evidence is genuinely reasonable — a photograph raising a fix rate by 3.2 percentage points against a 34% baseline across 663,591 reports is a real effect on a real outcome — and it is not where the field’s ambition has been pointed. As a decision channel the measured clearance and enactment figures sit near zero, and nothing in the record since 2014 suggests they move.

Speculative The durable contribution may turn out to be the administrative record itself. These systems make participation countable, and countability is the reason this brief can report a 0.35% actionability rate, a 1.9-out-of-10 social-inclusion score and reports peaking in the middle deprivation deciles while need peaks in the bottom two — rather than reporting anecdotes. A field that failed at its stated purpose while producing the first usable measurements of who actually participates has still contributed something, and it is not the thing it set out to contribute.

• Depends on Nothing on this map. This is ordinary web software and the deliberation methods predate it by decades; no brief in this corpus produces a result this one is waiting for. The constraints are institutional, evidentiary and fiscal, and they are recorded below.
• Requires (not on this map) A threshold calibrated to be reachable, since the measured clearance rates run from 2.1% to roughly one in ten thousand. A binding duty to act on input that clears it, absent everywhere in this record. A formal receiving point inside the responding organisation — the thing whose absence caused crowdsourced crisis data to be discarded after both the 2010 Haiti and 2015 Nepal earthquakes. Continuity across a change of administration, whose absence destroyed 182 approved proposals in one city. Administrative capacity to deliver what a process decides. Fiscal and charitable hosting for volunteer organisations, the layer that collapsed when the American intermediary withdrew in 2023. And outcome measurement with a counterfactual, which is the capability whose absence makes every other item on this list unarbitrable.
• Enables Nothing typed, and that is the finding rather than an omission. No brief in this corpus is waiting on civic technology to deliver a capability, because no capability it claims has been demonstrated against a comparison group. The relationships it has are shared problems and shared evidence deficits, not enabling edges.
• Adjacent Future Democracies, which reaches the same participation question through deliberative rather than digital instruments and reaches a similar verdict; Future Civil Services, which owns the state's own workforce and its in-house digital units and meets this brief at the 18F boundary; Smart Cities, which found the same pattern in municipal instrumentation and shares the reporting-versus-incidence error exactly; AI-Assisted Governance, the administrative counterpart, which inherits this field's evaluation deficit and its skewed training data; and Distributed Governance, which proposes to route around the institutions this brief finds to be the binding constraint.

“Civic tech projects have a well-documented pattern of grant-funded launch and abandonment.” Frontier The pattern is documented; the stated cause is not. The register that documents it lists roughly 70 failures with no denominator, so no mortality rate can be computed from it, and its maintainers say it undercounts. The causes it actually identifies are not grant exhaustion: market mismatch, absent user research, collective-action complexity, portfolio reprioritisation, dependency obsolescence. Nine named dead platforms had raised over $20 million between them; one launched with $9.3 million in venture capital. And the field's largest closure — a roughly 60-chapter volunteer network — happened at more than $3 million a year and fourteen staff in the year after a $100 million gift to the same organisation. Money was not the binding constraint in the best-documented case.

“Most civic tech projects fail within a few years.” Handwave There is no survival analysis of civic technology projects, no defensible sampling frame and no denominator anywhere this brief could reach. Any percentage attached to this claim has been invented. What can be said is that the failures which are documented cluster heavily in one species — venture-funded, generic, national-scale engagement platforms — while narrow jurisdiction-bound tools have run for a decade or more.

“Complaint platforms just redistribute who complains.” Established True, and the operator says so first: reports peak in middle deprivation deciles while incidence peaks in the bottom two. But the correction to the correction matters. Redistribution is not the same as no effect. Across 663,591 reports, 34% were marked fixed, and adding a photograph raised the probability of a fix by 15% relative, 3.2 percentage points absolute. The platform does route work and some of it gets done. What it does not do is route it in proportion to need.

“Ushahidi saved lives in Haiti.” Handwave The independent evaluation exists — Morrow, Mock, Papendieck and Kocmich, 2011 — and this brief could not obtain its full text; every host reachable was robots-disallowed, gated or returned an error, and the figure-level findings are recorded as a gap rather than reported at second hand. What is available is the organisation's own public summary of that evaluation, which describes the project as “one small part of a paradigm shift” and contains no numbers. What the peer-reviewed literature establishes across 51 deployments is that the Haiti data was “dismissed because of a lack of organisational infrastructure,” as was Nepal's in 2015, and that in the most-quantified case, 800 of nearly 230,000 processed messages were actionable. The capability is real; the impact claim rests on testimony.

“Crisis mapping failed because the crowdsourced data was unreliable.” Frontier Verification problems are documented from 2010 onward — unverified information bypassing editorial process, no established standards for quality control or verification, and the contemporaneous admission that the Haiti operation “didn't work perfectly.” But the mechanism the review literature identifies is organisational: no formal contact point, no pre-crisis relationships, responders too overworked to process unstructured information, and humanitarian politics. Improving the data quality of a feed nobody has arranged to receive changes nothing, and fifteen years of effort aimed at verification has been aimed at the wrong half of the problem.

“Open data has disappointed because the portals are hard to use.” Established It disappointed; the diagnosis is contradicted by measurement. Across six portals and more than 160,000 geospatial datasets, the correlation between metadata quality and dataset use was weak. Fifteen years of remedial work on discovery, metadata and portal design has been aimed at a supply-side explanation the evidence does not support. The demand-side reading is bleaker and better supported: most of these datasets were not wanted, and publishing them was cheap, measurable and politically legible.

“Nine out of ten government datasets are closed, but the important ones are open.” Established Exactly backwards. By category, the datasets most load-bearing for accountability are the least open: land ownership 1%, spending 3%, company registries 5%, budgets 10%, against 7% for the assessed corpus as a whole. Openness is highest where the political cost of publication is lowest, which is what one would predict and what almost nobody reports.

“Open data improves outcomes for the worst off.” Established On the only comparative measurement available, it does the opposite of what is claimed. Among the top ten performing countries, impact on entrepreneurship averaged 7.1 out of 10 and economic impact 4 out of 10, while impact on social inclusion averaged 1.9 out of 10, with 6% of governments showing relevant impact on marginalised groups. Where open data works it works for the constituency already best placed to use it — the same distributional signature as the complaint-platform deprivation finding, reached by an entirely different instrument.

“Decide Madrid and Consul are dead.” Handwave Madrid's citizen-participation directorate presented on its continued use of the platform at a community session on 11 December 2025, roughly a decade after the software's development. The finding about Madrid is about enactment, not about survival, and merging the two is an error in the sceptic's direction. A platform can run for ten years and change nothing, and saying so is a stronger criticism than saying it collapsed.

“Taiwan shows that digital democracy works.” Frontier Taiwan is not one case; it is two, and the difference between them is the finding. The national platform has taken 20,923 proposals between 10 September 2015 and 7 January 2025 under a live response duty. Taipei's municipal e-voting platform took 454 between 2017 and 2022 and is described in the research literature as having “become obsolete,” with the decline attributed in part to a 2016 episode in which “the city government offered only pre-selected options for public voting,” which “eroded public trust in the platform.” The variable that differs is not the technology.

“Digital public infrastructure is the mature version of civic technology.” Frontier It is the inversion of it. The identity numbers are real — 1.377 billion by mid-2023, bank accounts from 35% to 78% across a decade, $10 billion in pandemic transfers to over 200 million women — and so is the 20% exclusion figure from villages where biometric authentication was required for every sale. The stack achieved scale by being state-built rather than civil-society-built, mandatory rather than voluntary, and infrastructural rather than interface-level. Calling it the field's maturity requires not noticing what was traded away.

“MOSIP is open source, so adopting it does not create dependency.” Established The inference does not follow, and the critique on record is specific: that the platform creates “technological dependencies that can deprive governments of truly sovereign, locally-tailored solutions” despite its licence, with a risk of “de facto lock-out of alternative solutions.” The same analysis notes that the exemplar stack itself runs on cloud partnerships with United States hyperscalers, which is awkward for a sovereignty pitch. An open licence constrains a vendor's legal power and does nothing about the standard-setting power that comes from being the default.

“The evidence shows civic technology increases participation and trust.” Handwave The modal study supporting that claim is a cross-sectional attitudinal survey. The one read in full for this brief has 394 respondents across three districts of a single regency, uses ordinary least squares on five-point Likert scales, and reports digital transformation associated with trust at β = 0.571 and with participation at β = 0.581. Its own authors conclude that “technology alone does not guarantee democratic responsiveness” and that participation levels “remain modest.” Coefficients from designs like this are evidence about what respondents say, not about what governments do.

“18F's dissolution shows civic tech in government failed.” Established It shows no such thing. The unit was terminated in the early hours of 1 March 2025 as “non-critical,” citing executive orders of 10 and 11 February 2025, with staff losing email access immediately and no published evaluation of the work. The workforce analysis belongs to Future Civil Services. What belongs here is that the sector lost its principal government counterparty in the same twenty-five-month window in which it lost its principal civil-society convener, for unrelated reasons, neither of which was measured performance.