Some numbers here are wrong

Places where the v0 model produces a number that is wrong about the world rather than informative about it. These are not bugs in the code: the pipeline is doing what it was told. They are failures of measurement, and anyone building on this needs to know them before quoting a score.

Evidence for each is a diagnostic in data/out/diagnostics.json or a figure from the published output. The in-session panel runs in data/delphi are one panelist each, carry no dataset version and were scored against older frames, so a gap between one of them and a current score is not a finding (see A9).

Every entry is on dataset 9.0.0, the 125-country frame of D153, except where it names another run. A figure that came from a counterfactual rescore or a count the published files do not carry keeps the run it comes from: the scored-row counterfactual in A13, the question-format split in A15 and the B-READY 2025 package count in A12, which were made on the 53 countries of the 8.x frame. The B-READY and CPIA API counts in A6 and A12 were taken against the 125 when this file was brought to 9.0.0. A3 also quotes the 16-country run that last scored the WGI rows, and A4 the ten-country one. A12 quotes one figure D131 measured on V-Dem v15. A13's regime groups and its electoral democracy correlation read the 2025 rows of the V-Dem v16 country-year file for all 125. No entry quotes a panel figure (D139). Where a figure predates the current frame, the scale it was measured on is not the scale in use.

A1: Experimentation is not measured, it is inferred from patents

Severity: medium.

Five of ten indicators are observed for most countries: resident patents, trademarks and industrial designs per head (D126), new public software repositories per head (D145), and, for the 16 countries GEM covered first, early-stage entrepreneurial activity and fear of failure. Venture deals, firms through regulatory sandboxes and university spinouts are still gaps (since D152 the first two count deals and firms, where they used to read money and regimes). Business R&D share is retired: its only working source reads the make-up of a spending stock, and state enterprises count as business (D142). Mean confidence is 0.318 on dataset 9.0.0, the lowest of the nine, just under Shared Purpose (0.319), and under the 0.40 objective. Of the 125 countries, 85 are scored on the three filing rows and the repository row, 15 add the two GEM rows, and 25 rest on fewer. Three are scored on the filing rows alone: China, Cuba and Belarus, which the repository row's access gate holds out. Seven sit on two rows at the coverage floor, at confidence 0.110 to 0.175: Belgium, the Gambia, Guinea, Côte d'Ivoire, Mali, Myanmar and the Republic of the Congo. GEM's later reports are held out because, on the 53 countries of dataset 7.2.0, the 14 that GEM skips are mostly lower-income and the gain would sit where income already is (D125); those reports were not read for the 72 countries D153 added. See D21.

Resident patents and resident trademarks per head measure formalised, completed, defensible invention, which is close to the opposite of the many-cheap- experiments construct the dimension is supposed to capture. The repository row is the one cheap attempt the dimension observes, and it sees one platform: public GitHub work, located by IP address, so VPN users are misplaced and Singapore reads high partly as a regional hub. It tracks income (r 0.716 on the normalized row), and Experimentation's correlation with log GDP per capita is 0.728 (n 123).

On dataset 9.0.0 the Netherlands scores 56.0, 12th of 125. That is not a finding about Dutch innovation: its trademark and design rows are missing, venture capital is a gap, and the score rests on resident patents, the two GEM rows and new repositories. Belgium has the same two rows missing and no GEM rows, and scores 34.3 on patents and repositories alone. Uruguay scores 29.2, 37th, having legalised and regulated a national cannabis market and run a fintech sandbox. Argentina scores 21.3, 48th, having produced more technology firms of scale per head than anywhere else in the region.

The filing rows count filings at a national office, so a country whose residents file at a regional office reads thin or stale. The Netherlands and Belgium file trademarks and designs at the Benelux office. Six members of OAPI, the regional office for francophone Africa, are in the frame, and the national series they do have are old. A value more than 15 years old does not count (D159), so on dataset 9.2.0 Mali's patent row (one filing in 1981), the Republic of the Congo's (1988) and Burkina Faso's patent and design rows (2010) are set aside, and each of the three keeps one scored row and publishes no Experimentation score. Côte d'Ivoire's patent row is from 2012 and still counts, at the recency floor. At the other end China scores 100, first of 125, on the three filing rows alone: each sits at the top of its frame, the filings were subsidised (D126), and the repository row that would temper them is held by the access gate.

No panel figure is quoted here, and no panel run is on dataset 9.0.0 (D139, D153).

Fix. A venture deal count is still missing: first rounds per million people, counted by deal and not by amount (D152). The only inspectable aggregate, the OECD SME and Entrepreneurship Financing scoreboard, carries amounts rather than deals for 6 of the 16 original countries and omits Brazil, India, South Africa and Singapore. GEM has not surveyed 14 of the 53 original countries since at least 2019, so its coverage cannot grow evenly from the published reports.

A2: Per-capita normalisation flattens India

Severity: medium.

India scores 12.2 on Experimentation on dataset 9.0.0, 67th of 125. It comes from dividing absolute counts by 1.4 billion people. The arithmetic is correct and the result is not informative: it says India files few patents per head, which was never the question. Counted whole, India is third of 125 on scientific articles, sixth of 121 on resident patents and first of 122 on new public repositories. Anticipation reads 34.5, 73rd of 124, on articles per head and statistical performance; the per-head stocks of researchers and research spending are conditions beside it and do not enter it (D122). No panel figure is quoted here, for the reason A9 gives.

Fix. Per-capita is right for most indicators and wrong where capability is concentrated in institutions rather than spread across a population. Consider a per-capita and absolute-capacity pair for research and experimentation indicators, reported separately.

A3: Trust's low income correlation comes from which countries the survey reaches

Severity: high. Trust's 0.354 against log GDP per capita mixes two bases of evidence and understates how closely each tracks income; Coordination still tracks income through border time.

The figures in this entry are from dataset 9.0.0 unless they name the 16-country run. Coordination publishes a score for 124 of 125 countries, from border time, budget execution and V-Dem's expert-coded civil-society index. Trust publishes for 124 of 125, from five rows: contract enforcement days, frozen at 2019; bribery incidence (D123); government compliance with the courts (V-Dem, D131); and two Joint EVS/WVS items, A165 social trust and G007_34_B trust in people met for the first time (D140), which cover the same 72. Trust correlates with log GDP per capita at 0.354 (Spearman 0.398, n 123) and Coordination at 0.533 (Spearman 0.594, n 123), so both are usable as research baselines but not as clean claims of wealth-free capability. D23 retired the WGI perception composites and D44 retired homicide.

Both dimensions once leaned on the Worldwide Governance Indicators, which are expert and firm perception composites that track income closely by construction. On the 16-country run, with those indicators in place, Coordination correlated with log GDP per capita at 0.90 and Trust at 0.88.

Drop every indicator correlating with log GDP per capita at |r| ≥ 0.7 and each dimension kept exactly one indicator. The table is from that same run, and the indicators in it are retired, so it cannot be recomputed on the current frame:

The pattern to read is the class column. Every indicator that failed the old test was class P and every indicator that survived was class C. On 9.0.0 Trust's rows correlate with log GDP per capita at 0.604 for the A165 social measure (n 71), 0.533 for bribery incidence (n 120), 0.449 for court compliance (n 123), 0.318 for trust in strangers (n 71) and 0.160 for contract enforcement days (n 123). The A165 measure is a watch item rather than a verdict. Court compliance carries the most income: without it Trust's correlation falls from 0.354 to 0.194 (n 121). The two survey rows pull the other way: dropping trust in strangers would raise the correlation to 0.473, and dropping A165 to 0.437, although A165 alone tracks income more closely than any other Trust row.

The 0.354 is a mix, not a separation. The countries without the survey rows are poorer: a median income per head of 10,800 dollars across the 52 of them with an income figure, against 25,095 across the 71 with both rows. Inside each group Trust tracks income more closely than across the frame: at 0.742 among the 71 scored on five rows, and at 0.418 among the 52 scored on the institutional rows alone. Scored on the institutional rows alone in every country, Trust would correlate at 0.579 (n 123). The frame-wide figure is lower than all three because the two groups are scored on different rows, so the fall from 0.606 on the 53 countries of dataset 8.3.0 is a change of evidence and not a sign that Trust has come apart from income. Do not quote 0.354 as Trust's distance from wealth.

Coordination's rows sit at 0.544 for border time, 0.281 for V-Dem civil-society strength and 0.210 for budget execution; without border time Coordination's correlation falls to 0.339 (n 113). All three remain a partial operational proxy rather than a direct test of cross-agency delivery.

Fix. These dimensions need observable, behavioural indicators that are not WGI and not frozen at 2019. V-Dem and budget execution are useful additions for Coordination, but they do not show whether agencies delivered the same objective. Trust's social family is two perception items from one survey release and no behaviour. Trust needs a recent court-throughput or case clearance series, and broader institutional-performance evidence. See also A9, which is the same problem seen from an executive-led state, and D20, where documented cross-agency delivery is being collected as evidence.

Overturned by. Behavioural indicators that cover the country set and show that each dimension remains distinct from income after its indicators are combined. Until then, treat both as thin operational proxies and treat the Trust score as provisional.

A4: The four WGI series are one measurement wearing four names

Severity: closed. None of the four are scored. See D23.

Government effectiveness, regulatory quality, rule of law and control of corruption correlate with each other between 0.93 and 0.98 on the ten-country run that last scored them. They are spread across Coordination, Trust and Shared Purpose, so a single underlying perception measure is being counted three times in three different dimensions.

None of the four are scored. The measurement-class analysis in D23 argued for keeping none of them: they were the mechanism turning three dimensions into restatements of income per head. The rows stay in the registry as retired, with the reason on each.

A5: Voice and accountability is answering a different question in Shared Purpose

Severity: closed. The indicator is retired. See D23.

GOV_WGI_VA.EST measures the democratic channel for participation. Shared Purpose asks whether people can see themselves in a common project. When the row was last scored, before D23, Singapore read 20.9 while being one of the most effective collective actors in the set.

The spec is explicit that political uniformity is not a capability, so the fix is not simply to raise Singapore.

Voice and accountability is retired. On dataset 9.0.0 Shared Purpose rests on three rows, tax revenue, income inequality and EVS/WVS civic participation (D128, read through A15), at mean confidence 0.319, under the 0.40 objective, and it publishes a score for 123 of 125 countries. Its correlation with log GDP per capita is 0.259 (n 122), the lowest of the nine dimensions. Singapore scores 32.3, 117th of 123, on the two rows it has. Volunteering, national belonging and polarization are gaps. National belonging has no usable series: its only cross-national item, national pride, is not belonging and fails the regime test A13 describes, so it is not wired, and the gap counts against confidence (D151). No panel figure is quoted here, for the reason A9 gives.

V-Dem's polarization item brings the Singapore question back from the other side: a regime with no organised opposition can read as calm, as the United Arab Emirates does. It is published beside Shared Purpose as a behavioural check and not scored. See A13 and D121.

Voter turnout, from the same pinned V-Dem file, is published beside Shared Purpose as a behavioural check and not scored, because it reads this same democratic channel as well as compulsory voting and managed mobilisation. See D129.

A6: Doing Business indicators are frozen at 2019

Severity: low, correctly handled.

Five indicators come from the discontinued Doing Business programme: time and procedures to start a business, border compliance time, contract enforcement time, electricity connection speed. All are stuck at 2019 and the World Bank has archived the codes.

The recency term already marks them down, so this is visible rather than hidden. It is listed because these five have to migrate to B-READY, which is in the API as IC.BRE.* inside World Development Indicators. Its 2024 round, the only one in the API, covers 37 of the 125 countries on the business entry and the dispute resolution scores, short of the half-frame gate. The migration waits on coverage and not on the publisher. See A12.

A7: Learning understates Korea and Japan

Severity: medium.

Learning is scored on three rows: the Human Capital Index, whose last full round is 2020; the share of firms offering formal training; and research citation impact (D124). Enrolment, public education spending and the vocational share of secondary enrolment are conditions beside the score (D122, D141). None of the three scored rows observes what adults can do, so countries with exceptional measured outcomes do not get full credit for them.

On dataset 9.0.0 Korea scores 52.1, 35th of 124, and Japan 52.4, 34th, although their Human Capital Index rows are third and second of 121. The other two rows pull them down: Korea reads 17.4 on firm training and 53 on citation impact, and Japan 40 and 30, against 52.5 and 100 for Singapore, which leads the dimension at 84.2. No panel figure is quoted here, for the reason A9 gives.

Fix. A learning-outcomes series (PISA or PIAAC) would resolve most of this. It is a gap because coverage across the country set is uneven, not because the data does not exist.

A8: Every correlation here is a hint, not a result

Severity: structural.

Every correlation in diagnostics.json is computed on the 125 countries loaded, and most on fewer: 123 have an income figure, and 123 have all nine dimensions scored. Two dimension pairs sat at 0.94 on the 16-country run and read as near-duplicates. On dataset 9.0.0 no dimension pair passes the redundancy threshold, and the highest is Anticipation with Learning at 0.75 (n 124). Pairs separating is not the same as the nine being separate: one shared factor carries 0.505 of the variance over the 123 countries with all nine scored, well above the 0.175 chance gives at that size, and that factor correlates 0.815 with log GDP per capita (n 122). The one-factor test (D137) is the stronger reading. The 95% interval D153 gives for that r runs from 0.745 to 0.867, about six hundredths either side: enough to say the factor tracks income strongly, and too wide to place it on one side of the 0.8 edge of the strong band.

The redundancy and wealth-proxy findings are strong enough to act on because they also have a mechanical explanation, not because the coefficient is large.

The 125 are nearly every country of a million people that can be measured, so a larger set will not tighten these figures much: the frame expansion memo puts plus or minus 0.03 on r at about 350 countries, which do not exist.

A correlation here is also a statement about which countries are in the frame. From 8.3.0 to 9.0.0 Trust's correlation with log GDP per capita fell from 0.606 to 0.354, below the 95% interval of 0.40 to 0.76 that 51 countries gave it, while Agency's fell from 0.579 to 0.466 and Building's rose from 0.434 to 0.573, inside theirs. A3 shows that Trust's fall is a change in which rows the added countries have. Never quote one of these correlations without its n and its dataset version, and quote the interval where one is given.

A9: Coordination reads Singapore and other executive-led states low

Severity: low. The size failure is absent from the 9.0.0 scores; one row moves states that coordinate through the executive down by about a third.

The figures in this entry are from dataset 9.0.0.

Coordination publishes a score for 124 of 125 countries. Cuba sits below the coverage floor on one row. Three rows feed it: border time to export (Doing Business, 124 countries, frozen at 2018 or 2019), budget execution fidelity (GF.XPD.BUDG.ZS, 113 countries, 2016 to 2024) and V-Dem civil-society strength (v2x_cspart, 125 countries, 2025). 113 countries are scored on all three and 11 on two. Confidence runs from 0.100 to 0.413, mean 0.371, and the score correlates with log GDP per capita at 0.533 (n 123).

The entry was opened for a different failure, and the rows in use do not produce it. On the perception layer D23 retired, Uruguay scored 18.8 and Costa Rica 33.7, because the Worldwide Governance Indicators and the Logistics Performance Index read a small country with a small port as a weak one. On the current rows Uruguay scores 81.5 and Costa Rica 90.8 against a frame median of 78.6, the 12 least populous countries in the frame have a median of 86.3, and the score's correlation with log population is -0.17 (n 124).

What is left falls on a narrower set of states. With equal row weights, civil-society strength is a third of the score, and it measures whether society organises independently of the state. A state that coordinates through its own agencies reads on that row as one that does not coordinate:

Singapore is the case the entry was written for, a small state with whole-of- government coordination as its organising principle, and it sits 80th of 124, below the frame median. Every other country in the table sits lower still, at 84th or below, while scoring above 80 on the two operational rows. The row does not read autocracy alone: El Salvador reads 36.4 on it against 90.3 on its other two rows, and Mexico 45.7 against 88.6. Part of the reading is defensible, since D83 admits the row for the coordination a state does with society. The part that is wrong is reading its absence as an absence of coordination among institutions.

The panel gap is not evidence here. The in-session run that data/delphi/latest.json points at has one panelist, so isPanel is false; its provenance is in_session, so isEvidential passes; and it carries no dataset version, so the scorer attaches it to no cell and 9.0.0 publishes no delphiScore at all. Its Coordination cells were anchored on the retired perception scores, so their distance from a current score measures the change of ruler and not the countries. D139 keeps the run off any comparison with the indicators.

Fix. Cross-agency delivery records, a declared gap, would measure the construct directly and would let civil-society strength sit in a family of its own beside them, the way D57 splits Trust. A reviewed panel on the 9.0.0 frame, from the gateway or in session under D154, would make the panel comparison readable again. Until one exists, no panel figure belongs in this entry.

A10: The frame is every measurable country of a million people, and that is not the world

Severity: structural.

Every country in the benchmark sets the endpoints of every indicator scale and is measured against the result. See D47. So 0 and 100 mean "weakest and strongest of these 125". D153 sets who the 125 are by one rule: every World Bank economy of at least one million people that reaches MIN_INDICATORS_FOR_SCORE observed rows in all nine dimensions. A score is a position in this set and carries no claim about a country outside it.

The rule selects on measurability, so what it leaves out is not a random sample. Of the 160 economies of a million people or more, 35 are outside: 22 that clear eight dimensions, among them seven francophone African members of OAPI, the regional office that files their patents and trademarks, with no Experimentation rows, and the five Gulf monarchies outside the frame, with no Gini or tax ratio for Shared Purpose; and 13 that clear seven or fewer, such as Afghanistan, Yemen, North Korea and Hong Kong. The 57 economies under a million are out by the size line, 15 of which would clear all nine, and Taiwan is out because the World Bank API does not carry it. Each of the 35 is missing because a publisher does not reach it in at least one dimension, so the frame leans toward the countries the international datasets cover. Counted in the frame expansion memo, it holds 24 of the 38 economies in the world's poorest income quartile and 33 of the 39 in the richest, and 54% of it is a democracy by V-Dem's Regimes of the World, against 44% of the 158 it classifies.

Two consequences follow.

Scores are only comparable inside one dataset version. Adding a country moves the endpoints it touches and restates every number. That is done as an announced rebase with a major version bump, and 8.3.0 numbers do not compare with 9.0.0 ones. Anything quoting a score has to quote the version with it.

Clamping happens only in history. No observed cell clamps: on dataset 9.0.0, 0 of 3,721, because a current value cannot fall outside a frame its own country helped build. The outOfFrame flag fires only where a historical value sits outside the current frame, which is 17 of 607 momentum baskets. A trend carrying a clamped basket member is part distance-to-the-clamp rather than movement in the country, and every surface that prints a trend prints that count.

A11: Building measures industrial output, and reads as delivery capacity

Severity: medium.

Building asks whether a country can build and deliver. Its four measured indicators are manufacturing value added, high-technology export share, electricity connection speed and economic complexity. Output per worker sits beside them as a condition (D122). All four describe industrial output. Nothing in the measured set can see a national programme that was specified, funded and delivered.

On dataset 9.0.0 Brazil scores 38.3, 43rd of 125, at confidence 0.568, its second best evidenced dimension after Adaptability. The score is a correct statement about Brazilian industrial output and it is read as a statement about Brazilian delivery capacity, which is a different construct. In the same decade Brazil built and ran Pix, which settled 7.98 billion transactions in July 2026, and GOV.BR, which reports 175 million active accounts.

The two indicators that would carry the delivery construct, large_project_delivery and firm_scale_up_rate, are both gaps.

Fix. Two parts, one done and one open. Documented deliveries are recorded in data/evidence/records.json against the gap they bear on, outside the score, so the cases are written down with sources and limits instead of being argued in prose. 451 records cover the 53 countries of the 8.x frame and are filed against 20 indicators, 16 of them declared gaps; the 72 countries D153 added have none. Brazil's 29 run from Casa da Moeda in 1694 to the federal cash grant after the 2024 Rio Grande do Sul floods. See D20. The open part is a comparable delivery series across the country set, without which the gap cannot be promoted to an indicator.

Watch for. The same reading error in reverse. A country with strong industrial output and a poor record of finishing public programmes scores well here, and the benchmark currently has no way to say so.

A12: Coordination and Trust are scored on thin evidence

Severity: high. Trust is partly measured and Coordination remains narrow.

The figures in this entry are from dataset 9.0.0 unless they name another run.

Coordination remains a narrow operational proxy: budget alignment, border processing and civil-society judgements do not show whether several institutions delivered a shared national objective (A9). Trust prints for 124 of 125, but country-level confidence runs from 0.072 to 0.498, and the spread is the survey: the 72 countries with both EVS/WVS rows sit at 0.366 to 0.498, and the 52 scored without them at 0.164 to 0.295. Cuba is below the floor on both. Shared Purpose has a third row in 66 countries; the 57 on two rows sit at 0.079 to 0.316 and print, drawn dashed with a marked axis and a confidence band that says do not quote it alone. Cuba and Haiti publish no Shared Purpose score. That is a mitigation and not a fix.

Trust still has a narrow family balance. D57 splits the dimension into a social family, which asks whether people rely on strangers, and an institutional family, which asks whether they rely on courts, government and the civil service. The social family has both its rows observed, A165 and trust in strangers (D140), in the same 72 countries, so it is two perception items from one survey release and no behaviour. The institutional family has three of five: contract enforcement days, bribery incidence and court compliance. 52 of the 124 scored countries rest on the institutional family alone. Among them are the ten whose survey values are held because the release prints separate EVS and WVS rows (D64): Germany, the Netherlands, the United Kingdom, Armenia, Czechia, Romania, Russia, Serbia, Slovakia and Ukraine. Court case clearance and institutional trust remain gaps. familyBalance publishes this coverage.

The two social items disagree where "most people" means one's own circle. China reads 63.5% on A165, fourth of 72 after Denmark, Norway and Finland, and 13.4% on trusting a stranger, 60th. Japan (33.7%, 16th, against 10.4%, 65th), Singapore (34.4%, 14th, against 18.0%, 51st), South Korea (32.9%, 18th, against 17.5%, 53rd) and Malaysia (19.6%, 33rd, against 14.9%, 56th) split the same way, and so do Belarus (40.0%, 12th, against 20.1%, 47th) and Tajikistan (20.6%, 32nd, against 10.3%, 66th). The second row pulls their Trust scores down: Japan to 64.4, 51st of 124, Singapore to 70.4, South Korea to 67.5, Malaysia to 58.2 and China to 55.3, 68th. Ethiopia goes the other way, 11.9% on A165 and 47.8% on strangers, eighth of 72, and is the value to check first (D140); Myanmar (15.1% against 41.3%, 11th) splits the same way. Read a high A165 alone as a statement about a respondent's own network, and read China's Trust score through its court compliance (4.5 of 100) as much as through either survey row.

Fix. Land court throughput and case clearance, cross-agency delivery records, institutional trust, and behavioural measures of corruption experience. The V-Dem row is a partial Coordination repair, not a replacement for delivery records. The generative panel can interpret the dimensions while those data are missing, but its values stay beside the indicator score and never become observations. The ten countries D64 holds have no open route to a survey value: pooling their rows needs the registered microdata download, which D154 excludes.

What the World Bank can and cannot supply. GF.XPD.BUDG.ZS, primary government expenditure as a proportion of the original approved budget, covers 113 of 125 with a latest year between 2016 and 2024, 23 of them at 2024, and its scored row correlates with log GDP per capita at 0.210. The value is two-sided: both underspending and overspending can indicate weak execution, so the registry converts it to absolute distance from 100 before scoring. The CPIA cluster rates only countries that borrow from the World Bank's concessional window, and its public administration rating covers 38 of the 125 from 2018 on. Of the two Enterprise Survey corruption series, IC.FRM.CORR.ZS asks a firm what it believes firms similar to itself pay, so it records belief and is ineligible, while IC.FRM.BRIB.ZS asks whether the responding firm was itself asked for a bribe across six public transactions. The second covers 121 of 125, 106 of them at 2023 or later, and D123 scores it in Trust's institutional family. Its correlation with log GDP per capita is 0.533. It carries the reticence risk D123 records: China reads 0.14% and Korea 0.02% while Vietnam, surveyed by the same programme, reads 31%. Trust still needs court data and a broader social comparison. The remaining shortlist is OECD Government at a Glance and a harmonized court or audit source.

B-READY is what the frozen rows become. IC.BRE.* replaces Doing Business inside World Development Indicators. The API's 2024 dispute-resolution rows cover 37 of the 125. The official 2025 downloadable package reached 25 of the 53 countries of the 8.x frame, two below the half-frame gate there, and has not been counted against the 125. It remains a candidate rather than a scored replacement. Its dispute-resolution and operational- efficiency fields are promising, but the package mixes expert and firm-survey inputs and does not publish the court clearance numerator and denominator the Trust gate requires. Revisit the next release instead of forcing a partial series into the frame.

Government compliance with the courts (court_compliance, V-Dem v2jucomp, D131) covers 125 of 125. It is an expert code of a public act, and it reads regime: D131 measured r 0.88 with V-Dem's electoral democracy index on the v15 release. The row reads v16, and the pinned file does not carry the index, so that figure is not recomputed. Two autocracies read high: Singapore, 14th of 125, plausibly because its courts seldom rule against the state, and Jordan, a closed autocracy, 22nd. Nicaragua, Azerbaijan, Venezuela, China and Tajikistan sit in the bottom five, with Myanmar, Belarus, Cuba, Laos and the United Arab Emirates next above them.

V-Dem civil-society strength is an adapter-backed Coordination row (D83), but its expert coding keeps confidence low and does not answer cross-agency delivery. Voter turnout, volunteering and civic participation are absent from the catalogue under any database id, and the interpersonal and institutional trust items were never World Bank series. The Joint EVS/WVS adapter supplies both social rows; Trust still needs an adapter for court clearance and Coordination needs one for delivery beyond the V-Dem and budget rows. The shortlist for court clearance is OECD Government at a Glance. ingest: 'manual' remains available for sources with no usable API.

The one indicator added to fix contamination was the contamination: D42 showed that homicide_rate raised Trust's wealth correlation by 0.288, so D44 retired it. Generative estimates are therefore a useful research layer, but they cannot repair a missing observation series.

This artefact remains open while the dimensions rely on narrow proxies. For Trust, D57 sets the stricter condition: one harmonised social measure and one comparable institutional-performance measure, because two indicators from the same family would clear the floor without answering the dimension. The current release meets that structure with two social and three institutional rows, but court throughput, a behavioural social row and the wealth and redundancy review remain open. The budget series opens the Coordination door; it does not close its measurement problem.

A13: Polarization reads calm where there are no camps to polarize

Severity: low. The series is published as a behavioural check and is not scored (D121). Across the 125 closed regimes do not read calm as a group; the failure is single autocracies among the calmest countries in the frame.

The political_polarization check is V-Dem's v2cacamps_osp: whether supporters of opposing political camps meet in a hostile rather than a friendly manner, on a 0 to 4 scale. The question counts hostility and not disagreement, which is what the registry gap asks for. It cannot tell a society where camps meet in friendship from one where no opposition camp is allowed to exist.

The 2025 values in V-Dem v16 do not sort cleanly by regime. Grouped by V-Dem's Regimes of the World classification (v2x_regime), the 26 liberal democracies in the frame average 1.64, the 41 electoral democracies 2.60, the 41 electoral autocracies 2.86 and the 17 closed autocracies 2.68. Across the 125 the closed group reads calmer than the electoral autocracies and more hostile than both groups of democracies, so the calm end of the scale is not where closed regimes sit as a group. The closed group is also the widest, from 1.05 for Laos to 3.99 for Myanmar. The calm reading correlates with the electoral democracy index (v2x_polyarchy) at 0.37 (n = 125).

The failure is carried by single countries and not by a group. Of the 16 calmest of the 125, five are autocracies: Kyrgyzstan (0.96, 10th), Laos and Tajikistan (1.05, 12th), the United Arab Emirates (1.07, 14th) and Uzbekistan (1.09, 16th), beside Ireland at 0.41, Norway at 0.62, Japan at 0.86 and Denmark at 0.93. Jordan, a closed autocracy, reads 1.43, and Cuba 1.75, near the liberal democracy mean. Rwanda reads 2.67 and Singapore 1.83, both electoral autocracies and neither calm. The same release codes Rwanda 0.96 and Singapore 1.27 for 2024, so a single year moved Rwanda by 1.7 points with no change in its regime class. A latest-year expert code of this item can swing that far, and a reader comparing two countries on it is partly comparing coding years.

Scored as a third Shared Purpose row against the 8.1.0 frame, it would raise the United Arab Emirates by 11.1 points, Singapore by 7.7 and China by 4.8, and let Haiti publish a Shared Purpose score for the first time, at 35.9, on one World Bank row and this one. Rwanda would fall 4.2 and Vietnam 3.3. The largest gains go to calm democracies, Panama by 18.0, Paraguay by 13.3 and Japan by 11.8, which is the reading the item is meant to give. The United Arab Emirates gain is A5 inverted: A5 retired a perception composite that penalised political uniformity, and this item rewards it. Political uniformity is not a capability, and low measured hostility under repression is not shared purpose.

The numeric gates do not catch it. On dataset 9.0.0 the value passes the wealth screen at 0.200 against log GDP per capita and the redundancy screen at 0.457 at most (interpersonal trust, n = 72). On the 8.1.0 frame, adding it would have raised Shared Purpose's own correlation with income from 0.202 (n = 50) to 0.406 (n = 51, with Haiti). The decision stands on construct and not on correlation.

What is published. The value sits on every country page under Shared Purpose as a check, not scored, with this reason attached, and the capability page lists it. behaviouralChecks in diagnostics.json reports its correlation on the published value as -0.200 against log GDP per capita across the 125 and -0.043 against the Shared Purpose score (n = 123). No Shared Purpose score, confidence or coverage count moves.

What remains. A reader can still take a low value for a closed regime as a finding; the attached reason is the only mitigation. The fix that would let it score is a reading conditioned on competition existing at all, or a behavioural row (civic participation, volunteering, voter turnout) that agrees with the item outside the closed regimes. Neither is wired, and the gap stays open.

A14: Half of Agency's scored rows are frozen at 2019

Severity: medium.

Agency is scored on four rows: new business density, perceived control over life (D127), and the time and the number of procedures to start a business. The last two come from Doing Business and stopped in 2019 (A6). Internet users, account ownership and private credit sit beside the score as conditions (D122). So half the scored evidence is a 2019 reading of registration rules, and the one current row that is not a business count is a perception (A15).

Where new business density is missing, Agency is the two frozen rows and perceived control. On dataset 9.2.0 that holds for the United States (82.1), Nicaragua (79.8), Lebanon (47.1) and Venezuela (26.1), each at confidence 0.367. Sudan has neither new business density nor perceived control, so its 53.6 is the two 2019 rows alone, at confidence 0.222. Haiti and Malawi are in the same position: their new business density is from 2010 and 2009, older than D159 lets a value count, so their 21.1 and 56.6 are the two 2019 rows alone, at confidence 0.222.

The Doing Business rows themselves are inside D159's 15 years until the 2034 release, so the rule does not touch them yet.

Read Agency through its confidence. The fix is a behavioural row that observes people acting, which the O1 triage sweep did not find with frame coverage.

A15: Two survey rows read regime and question format

Severity: medium. Both rows are scored (D127, D128). On the 125 neither row's income correlation is a finding: perceived control has none, and civic participation's negative one carries the question format.

The figures in this entry are from dataset 9.0.0 unless they name another run.

perceived_control (Agency, A173) is a perception, and a closed or electoral autocracy can read high on it. Vietnam reads 8.1 out of 10, third of 72 with Colombia, and Nicaragua 8.0, fifth, surveyed in 2019-20 after the 2018 crackdown; Venezuela reads 7.7, ninth. Response style on a 10-point scale moves the ordering too: Mexico, Uruguay and Colombia sit at 8.1 to 8.2 and Japan, on a mail survey, at 6.0, tied 70th with Greece. The row correlates with log GDP per capita at -0.006 (n 71), so what it adds to Agency is unrelated to income and partly a reading of regime and response style.

civic_participation (Shared purpose, A080_01) harmonises two questions. EVS shows a list and asks which organisations the respondent belongs to; WVS reads each type aloud and counts active and inactive members. The published values do not name the programme behind a single-row country, so the format split is quoted from dataset 8.0.0, where the eight EVS countries in the frame averaged 9.9% and the 29 WVS countries 19.4%, and the row tracked income at 0.83 inside the EVS group. Among the ten countries the release surveyed under both programmes, WVS reads higher in eight, by a median 1.9 points; all ten are now in the frame, and D64 holds their survey values out of the score. On 9.0.0 Azerbaijan (0.5%) and Estonia (1.5%, EVS) are the floor and Kenya (38.4%) and Indonesia (37.9%, both WVS) the top, and across the 72 the row runs against income at -0.271 (n 71).

What is published. Both values, with the programme's question, the survey year and the valid-answer base in each observation's note. The release stamps every value 2022, though fieldwork ran from 2017 to 2023.

What remains. Neither row is adjusted. A reader can take Vietnam's perceived control or the income correlation of either row as a finding; read them through this entry first. A behavioural row with frame coverage is the fix. A format term needs pooled respondent-level microdata, which sits behind a registration the open-data rule excludes (D154), so none is in view.

A16: Adaptability reads low unemployment unchecked in 16 countries

Severity: low.

The figures in this entry are from dataset 9.0.0.

Adaptability scores four rows: the unemployment rate, the long-term unemployment share (D120), export concentration (D119) and new export products (D149). 15 countries have no long-term share. ILOSTAT publishes none for India, China, Haiti, Uzbekistan, Guinea, Tajikistan, Papua New Guinea, the Republic of the Congo and Trinidad and Tobago, and the plausibility gate holds every year it publishes for South Korea, Mexico, Peru, the Philippines, El Salvador and Uruguay. On dataset 9.2.0 Cuba joins them: its only share is from 2010, which D159 sets aside, and it reads 55.8 (84th) on a 1.7% unemployment rate (7th lowest) and new export products ranked 123rd. Those 16 are scored on the other three rows, so the unemployment rate is a third of each score with nothing beside it that tells a fluid labour market from one where nobody can afford to stay unemployed.

Ranks are of 125: lowest unemployment first, most new products first. Informal employment is a condition beside the score (D150) and is not ranked here. On dataset 9.2.0 all 16 carry Adaptability confidence 0.459; the other 109 run from 0.478 to 0.617, mean 0.598.

What is wrong. In India, Mexico, El Salvador, Tajikistan and Peru most work is informal, and a person who loses a job there takes any work within weeks, so the unemployment rate reads low because unemployment cannot be afforded. On four rows that low rate is checked by the long-term share: Ethiopia (3.3%), Nigeria (3.1%) and Burundi (0.9%) carry long-term shares of 53.9%, 56.1% and 95.5% and rank 92nd, 107th and 114th. For the 16, the check is the new export products row, which observes reallocation rather than slack, and the informality figure printed beside the score, which a reader has to put together with the rate. India's fifth place rests on a low rate and a high entry rate; its informal share is 87%, so the low rate is the weaker half of that reading. Uzbekistan's sixth rests on the highest entry rate in the frame and a rate in the middle of it. China, Uzbekistan, the Philippines, Trinidad and Tobago, Papua New Guinea, Guinea and the Republic of the Congo have no informality value either.

Fix. A long-term share for the 16: a labour force survey that records the length of a search, passed through the same gate. ILOSTAT's duration tables give the same questionnaire signature at every band for six of the gate-held countries (see the A16 fixes memo), so for them this waits on a survey change. The two declared gaps would add rows for every country, but neither has a source.

A17: Venezuela's Building has no manufacturing row

Severity: low.

The figures in this entry are from dataset 9.2.0.

The World Bank publishes Venezuela's manufacturing value added (NV.IND.MANF.ZS) as exactly 0% of GDP for every year from 1991 to 2011, its last year. The ingest drops those zeros as placeholders (D157), and the one value before them, 14.2% in 1990, is more than 15 years old and does not count (D159). So the row a reader would most expect to describe a petro-state's industrial base is empty, and Venezuela's Building, 8.9 and 124th of 125 at confidence 0.251, rests on two rows: the 2019 Doing Business electricity connection score (A6) and economic fitness.

Nothing in the frame's sources says what Venezuela's manufacturing share is today. The score reads low on the two rows it has, and the coverage part of its confidence says how much is missing.

Fix. A national or regional source for Venezuela's manufacturing value added, published and comparable with the World Bank series.