Corrections

How to file a correction

There are two ways, depending on what you're flagging.

A specific figure. On any nation or leader page, use the Spot an error report attached to the value. It records the entity, the indicator, and your note, so the correction arrives with the exact figure it concerns.

Anything else — a source, the methodology, a broken link, a question about how a score is built — email corrections@nationshelm.com. It is a monitored inbox, read by the editor who signs the corrections.

How a correction is handled

A flagged value is checked against its source. If it is wrong, the value is corrected, the correction is dated and signed, and the entry is added to the log below. If a change to how a number is made is involved, the scoring model version is bumped and the change is added to the methodology changelog (currently v5.5.0) — so every score can name the exact version that produced it.

Predictions resolve the same way. Every published prediction carries a resolution date and resolves in public, on the card that made it. A miss stays on the record, stamped MISSED, not quietly removed.

The corrections log

DateCorrectionSigned

2026-08-18

Undervalued Nations was ORDERED by the canonical Undervaluation Gap while almost every published number read something else. The ranking row carried each nation's Opportunity Score rather than the gap that had ordered it, so the deterministic facts layer — which re-ranks on whatever the row carries — re-sorted the result by that score, and the headline, leader card, gap-to-#2, field spread, analyst summary and share card then named whichever nation held the highest Opportunity or Overall Rating instead of the widest gap. A screen whose first row was one country announced that a different one "shows the widest residual" at a number that was in fact its Overall Nation Rating and could not be a residual at all; the same substitution produced a negative "gap to #2", which is impossible when both sides are gaps. The row now carries the Undervaluation Gap itself, so every surface inherits the correct quantity from one canonical result: Opportunity Score, Overall Nation Rating, Strategic Position and the Fundamental Floor (the higher of the latter two) are published beside it as the values it is built from, and the identity Gap = Opportunity Score − Fundamental Floor is verifiable on the page. The Comparison Matrix now centres that decomposition instead of a country-profile board; the Data Table exposes all four quantities under their own names; the share card shows the undervaluation ranking rather than an Overall-score deck. Three counts are kept distinct — eligible candidates, nations with a positive gap, and rows displayed — so a filtered pool cannot read as a global one, and the requested result count is a maximum: a pool with fewer positive gaps returns fewer rows, and one with none returns an explicit empty result. "Strongest opportunity vector" replaces "top driver": the Opportunity Score is a mean of its vectors, so no vector is shown to have caused the gap. The underlying methodology in lib/opportunitySemantics.ts is unchanged.

Undervalued Nations (/intelligence)

The ranking row carried the Opportunity Score while the order came from the Undervaluation Gap, so the leader card, headline, gap-to-#2, spread and OG were computed from a different quantity — publishing an Overall Nation Rating as a "residual" and producing a negative gap to second place→The Undervaluation Gap is the published ranked quantity across every output, with the Opportunity Score, Overall Rating, Strategic Position and Fundamental Floor shown beside it as the values it is built from; qualifier, eligible and displayed counts are stated separately and only positive gaps qualify

C·B

2026-08-18

Two corrections, both the mirror of ones already logged. (1) Rising Countries carried the same defects Declining Countries did: a requested Top N was filled out with countries that had held flat or moved the other way, and a selected Focus never reached the ranking basis, so "Rising Countries · Focus Stability" ranked the aggregate all-category change and omitted every country whose stability rose while its overall score fell. Only strictly positive changes on the active basis now qualify, the requested count is a maximum, a pool where nothing rose returns an explicit empty result, and a selected Focus sets the ranked quantity end to end. Causal wording went with it: the category that gained most is not shown to have caused the ranked change, so the "Top change driver" card reads "Largest category gain", and the share-card footer states how many of the eligible pool actually rose. (2) The Executive Brief's timeframe framing made claims about the PATH between a window's endpoints that no two-point comparison can support — a five-year read asserted that "a single-year wobble wouldn't be enough to produce this picture", and a ten-year read that a change of that size "reflects systemic change, not a cycle". Both now state what the window does establish: direction and magnitude, or size, but not whether the change accumulated steadily or in a single period. This line is shared, so the correction applies to every insight whose Brief carries it; the momentum insights state the limit in their own words instead.

Rising Countries (/intelligence) · Executive Brief timeframe framing

Rising Countries returned exactly the requested number of rows, topped up with flat and declining countries, and ranked the aggregate change even when a Focus was selected; the Brief's five- and ten-year framing asserted that the change could not have come from a single period→Only strictly positive changes on the active basis appear in Rising Countries, a selected Focus is the ranking basis, and the requested count is a maximum; the timeframe framing states direction, magnitude and size without claiming how the change accumulated

C·B

2026-08-18

The provenance taxonomy was reduced to its true two buckets, because the site was publishing three descriptions of it at once. The chrome and the "Transparent by design" block counted two classes (110 live + sourced / 82 derived = 192); the metric catalog and the methodology FAQ named three and stated "every value is one of these three"; and internal specs still carried a fourth (15 beta, excluded from scoring) that no longer existed and appeared nowhere in the 110/82 arithmetic. The distinction between "live" and "sourced" never described the value — both are figures ingested verbatim from a citable public source; only the delivery differed (a scheduled pull versus a published index entered by hand). Publishing that difference as a provenance class implied a difference in the number's standing that did not exist. The 79 indicators previously labeled LIVE are now labeled SOURCED, joining the 31 already so labeled: 110 sourced. No value, score, formula, or count changed — 110 / 82 / 192 are the same figures under one name each. Delivery mechanism is retained where it is operationally real, on the internal data-status page, renamed automated/manual so it cannot be read as provenance. The beta tier, retired 2026-08-07, was removed from the remaining specs, tokens, and engine vocabulary. The public API's provenance enum narrows to [sourced, derived].

Provenance tiers · Live / Sourced / Derived / Beta

Provenance tiers Live / Sourced / Derived (and Beta in internal specs); 110 "live + sourced" / 82 derived→Provenance tiers Sourced / Derived only; 110 sourced / 82 derived = 192

C·B

2026-08-18

Rising and Declining Countries published more countries than the query asked for. Thin-coverage countries are shown unranked alongside the ranked field, in whatever space the result limit leaves; the space available was computed as the ranked shortfall, falling back to "every unranked row" when that shortfall was zero. A zero shortfall means a FULL field, not an absent limit, so the fallback fired exactly when it should not have and let every thin-coverage row through: a requested Top 10 published 12 rows, a Top 5 published 7, a Top 20 published 22. The canonical limit invariant caught it on most real queries and recorded the violation, but the surplus rows still reached the page. Thin-coverage rows now take only the room the ranked field leaves, so the published field never exceeds the requested count. No score, change, rank, or gap is affected — the ranked field and every quantity derived from it were already computed over the ranked rows alone; what changes is how many rows are displayed.

Rising / Declining Countries (/intelligence)

A requested Top 10 could publish 12 rows — the two extra being thin-coverage countries admitted because the ranked field had already filled the limit→Unranked thin-coverage rows occupy only the space the ranked field leaves, so the published field is never larger than the requested result count

C·B

2026-08-18

The published method for Current Challenges did not describe the flags the card actually shows, so a reader could not reconstruct a leader's challenge list from the stated rule. The catalog claimed "any leadership dimension below 58" — in fact only three of the eight dimensions raise a challenge (Economy, Institutional Integrity, Communication); Governance, Diplomacy, Politics, Crisis Response and Vision never do, at any value. Five firing rules went unstated entirely: Crisis Exposure above 60 (severe above 80), Continuity & legitimacy below 40, Stability Trend below 50, adverse External Conditions below 50 (severe below 25), and the Stable Trajectory / Mixed Signals reading when nothing fires. Government Stability's severe cut of 40 was published nowhere and was hardcoded in the flag code rather than read from the policy object, as was the crisis-exposure severe cut of 80; both now live in the policy the code executes. The card's cap of four flags, most severe first, was also unstated. The method string is now interpolated from that same policy object — the discipline already applied to Strengths & Weaknesses — so the stated rule and the running rule cannot drift apart again. The stale "(v5.4.0)" stamp on the catalog cell is gone: the thresholds it named (58 / 48 / 50) were and remain correct on the single public scale, but the cell was the only one in the catalog carrying a model version, and it read as stale next to a v5.5.0 model. No flag, threshold or displayed challenge changed — the correction is to the published description of the rule.

Current Challenges (leader)

"Rule-based flags aligned to the public interpretation bands (v5.4.0): any leadership dimension below 58 (below the benchmark band; severe below 48); Stability <50 (its own scale); or governing conditions Severely Constrained / Extreme"→The full stated rule: Economy / Institutional Integrity / Communication below 58 (severe below 48) only; Government Stability below 50 (severe below 40); Crisis Exposure above 60 (severe above 80); Continuity & legitimacy below 40; Stability Trend below 50 (severe below 25); External Conditions below 50 (severe below 25); governing conditions Severely Constrained / Extreme; four most severe shown; Stable Trajectory / Mixed Signals when none fire

C·B

2026-08-17

The Similar Countries Comparison Matrix used one fixed layout regardless of what the similarity was actually measured on. Two consequences. (1) When the match basis was a SINGLE selected dimension, the profile radar was suppressed — correctly, since a one-axis polygon has no shape — but nothing replaced it, leaving a value grid in which the selected dimension's own sourced indicators (for Strategic Position: demographics, resources, stability, education, technology, geopolitics) sat unlabelled beneath it and read as six further dimensions that had produced the match. A single-dimension match basis now gets a scalar distance-from-anchor comparison instead: every country on one shared scale, the anchor's value marked on each bar, and each candidate's signed distance from it beside the similarity it earned; the value grid states that indented rows are the indicators behind the dimension above them, not additional match dimensions. (2) The matrix's countries were a generic first-six slice of the result, so a tie group crossing that boundary was cut in half — five of seven countries sharing an identical published similarity were shown, implying the two dropped ones matched less well. The matrix now takes a tie-aware peer set that expands through a split tie, and discloses the count when peers remain off screen. The similarity computation, candidate eligibility and every other output are unchanged.

Similar Countries · Comparison Matrix (/intelligence)

One layout for every match basis: a single-dimension query showed that dimension's sourced sub-factors as if they were six more match dimensions and offered no comparison of distance from the anchor, and the matrix's six countries could show five of seven co-closest countries without saying so→The visualization follows the dimensionality of the match basis — a profile radar on the match-basis axes when it is multi-dimensional, a scalar distance-from-anchor comparison when it is one dimension — with the anchor marked as the reference point and a tie-aware, disclosed peer set shared by every surface of the matrix

C·B

2026-08-17

A candidate comparable on only part of the requested Focus could top the match ranking. Similarity is a distance normalised over the dimensions the two nations actually share, so a nation with data on one of five requested categories was scored on that single dimension and could tie a nation compared on all five — while the result described both as closest 'on 5 default categories'. Searching from Canada across the default five-category Focus, Gibraltar ranked joint-first at similarity 97 on Quality of Life alone. A candidate must now be comparable on at least 60% of the requested match basis to be ranked; below that it is excluded and counted under its own reason, kept distinct from the existing exclusion for candidates with no comparable Focus data at all. Retained candidates compared on part of the basis carry that fact inline. The same gap also removed the profile comparison: the radar can only draw an axis every displayed nation has a value for, so one thinly-covered match collapsed four of five axes and the chart disappeared — restoring the coverage floor restores the chart. Single-dimension Focus queries are unaffected, since coverage there is all-or-nothing. Separately, where the result limit splits a similarity tie, the number of eligible countries sharing the last displayed score is now stated rather than described as 'additional countries'. No published similarity formula changed.

Similar Countries (/intelligence)

Candidates scored on a fraction of the requested Focus ranked against candidates compared on all of it (Canada, 5 default categories: Gibraltar joint-first at similarity 97 on one dimension; 199 eligible); profile comparison chart suppressed; split ties described without a count→Candidates must cover at least 60% of the match basis to rank (Canada, 5 default categories: 173 eligible, 26 excluded for partial basis, closest tier of six at 97; Hong Kong and Netherlands unchanged at 96); profile comparison restored; split ties state how many countries share the last displayed score

C·B

2026-08-17

Rising Countries is a momentum ranking — the change in a nation's all-category score over the selected timeframe — but several surfaces ranked, headlined and measured it on CURRENT scores instead. (1) The ranking facts read each row's current score as the ranking value, so the highest-scoring nation was named the mover and its level was printed wearing the change unit: 'Gibraltar moved the most at 84 Δ' stated an 84/100 current score as a delta, when Gibraltar's five-year change was +13 and the fastest riser was Tuvalu at +25. Rank, leader, gap-to-#2, field spread and the analyst summary now derive from the published change; a current score is never rendered with a Δ. (2) The selected timeframe did not reach the computation: 'Last 5 Years' was displayed while every ranking compared readings ~3 years apart. Last 1 / 5 / 10 Years now set the actual comparison window (nearest valid readings when exact dates are unavailable, with the observed span disclosed, e.g. 2018–2023). (3) 'Gap to #2' and 'Field spread' were computed on current scores while the rank order came from change, which could publish a negative gap; both are now measured in momentum points (South America services-led, five years: Suriname +2 leads Paraguay +1 by 1 point, spread 6 points from +2 to −4). (4) 'Top Driver' named a nation's strongest CURRENT category; it now names the category contributing most to the move over the window, or 'Mixed' when no category explains it. (5) The top snapshot, comparison matrix and share card ranked by current level — the snapshot now leads with change and baseline → current, the matrix's primary rows are category Δ with current levels labeled as context, and the share card summarizes the momentum result rather than an unrelated analyst finding. (6) Two-point trend visuals labeled 'illustrative trend, not verified historical data' in fact plot two real scored readings, and are now labeled as such; no intermediate path is drawn. Current scores remain visible throughout as context. No indicator value changed; overall confidence stays capped at Medium by the change-based method.

Rising / Declining Countries (/intelligence)

Ranked, headlined and measured on current scores: 'Gibraltar moved the most at 84 Δ' (an 84/100 level), 'Uruguay Leads the Overall at 69' as the momentum summary; 'Last 5 Years' shown while comparing ~3 years; gap and spread from current scores; Top Driver = strongest current category→Ranked and measured on change over the selected window: Tuvalu +25 leads Gibraltar +13 (gap 12 momentum points); Suriname +2 leads the services-led South American field (gap 1, spread 6, +2 to −4); timeframe drives the comparison window with the observed span disclosed; Top Driver = driver of the change, or 'Mixed'

C·B

2026-08-17

Three defects in the momentum ranking, found by running the published queries against the indicator record. (1) Thin-coverage nations set the field's story. A nation's momentum is only a broad national signal when enough multi-indicator categories back it — lib/momentum.ts already computed that `sparse` flag, but the ranking ignored it, so at the five-year window Gibraltar led the published field at +13 on 6 indicators across 2 categories, and at ten years Gibraltar +22 tied Vatican City +22 on 5 indicators across 1. Thin-coverage nations are now published with their real change but UNRANKED, tagged with the coverage behind them, and excluded from leader, gap, spread, headline, change matrix and share card; when every nation in a field is thin the ranking is shown with that stated. (2) The result limit cut through tie groups silently: nations sharing the last published change were dropped without trace, and the field spread was measured over whatever the arbitrary cut left behind. The count of equal-momentum nations beyond the cutoff is now disclosed. (3) The window disclosure overstated the comparison. The observed range printed the OLDEST baseline and NEWEST reading across every metric — an envelope, not the window — so a ranking whose typical comparison is exactly five years was labelled 'observed window 1992–2025'. Measured across 5,614 nation-metric pairs at the five-year setting, 94% of comparisons land on target. Surfaces now state the representative (median) comparison plus how many readings fall outside it. Separately, the selected timeframe is now bound to the computation: every momentum reading carries the window it was built at and is verified against the window it is labelled with, so a 'Last 5 Years' label can no longer be published over a shorter comparison. No indicator value changed; confidence stays capped at Medium by the change-based method.

Rising / Declining Countries (/intelligence)

Ranked and headlined by thin-coverage nations (five years: Gibraltar +13 on 6 indicators / 2 categories; ten years: Gibraltar and Vatican City tied at +22 on as few as 5 indicators / 1 category); equal-momentum nations dropped at the cutoff without disclosure; window shown as the outer envelope of all readings ('observed window 2000–2025' for a five-year query)→Ranked on broadly-sourced change only — five years: Liberia +10 leads, Tuvalu and Belize +9 (gap 1, spread 3); Gibraltar, Vatican City and Taiwan published unranked with their coverage stated; nations sharing the last published change are counted and disclosed; window shown as the representative comparison ('typical comparison 5 years; 2 of 18 readings fall outside that window')

C·B

2026-08-17

Declining Countries published results that were not answers to the question asked, in two ways. (1) The requested result count was treated as a QUOTA rather than a maximum: when seven countries in the eligible pool declined and ten were requested, three flat countries were appended to fill the list. Every figure derived from that field then described countries which had not declined — the published field spread read −5 to 0 (five points) instead of −5 to −1 (four), and the share card's second panel presented improving countries under a Declining headline. Only countries whose change is strictly negative on the ranking basis now appear; a field with fewer decliners returns fewer rows and states how many of the eligible pool declined, and a field with none returns an explicit empty result instead of the least-improving countries. (2) A selected Focus never reached the ranking basis: the result was ranked on the aggregate all-category change whatever the Focus said, so "Declining Countries · Focus Stability" omitted every country whose stability fell while its overall score rose, and included countries whose stability had improved. The Focus now sets the ranked quantity end to end — admission, rank, baseline, current, change, gap, spread, headline, title, matrix, data table, methodology and share card all read that category's own change, with the aggregate carried alongside as labelled context. A Focus with no comparable historical series (Strategic Position, Energy, the Opportunity Score) returns an explicit statement rather than an aggregate ranking wearing the Focus's name. Separately, causal wording was removed: no contribution arithmetic shows that the category which fell furthest CAUSED the ranked change, so "Top change driver" reads "Largest category decline", and the brief no longer claims a single-year wobble could not have produced the picture — two endpoint readings establish direction and magnitude, not the path between them. The same two corrections were applied to Rising Countries — see the entry of 2026-08-18.

Declining Countries (/intelligence)

A Top 10 request always returned ten rows, padded with flat and improving countries whose values then entered the published spread and the share card; and a selected Focus was displayed but never ranked — the order came from the aggregate all-category change regardless→Only strictly negative changes on the active basis qualify, the requested count is a maximum, and a selected Focus IS the ranking basis across every output — with the aggregate demoted to labelled context, un-rankable Focus selections refused outright, and category moves described as measured moves rather than proven causes

C·B

2026-08-17

The Data Table published every country-level insight as the same four columns — Rank, Country, Score, Tier — so one anonymous header carried quantities that are not comparable: a current level, a change over a window, and closeness to an anchor. A reader could not tell which. Three consequences were substantive rather than cosmetic. (1) On Rising / Declining Countries the “Score” column held the CURRENT rating while the order came from the change, so the table's own numbers did not reproduce its ranking, and the baseline against which the change was measured was absent. (2) On Similar Countries the column held a similarity score — closeness to the anchor — under a header that reads as performance, and the anchor's own value was nowhere on the table. (3) On Undervalued Nations the column held each nation's Opportunity Score while the ranking basis is the Undervaluation Gap (Opportunity Score minus the higher of Overall and Strategic Position), so the quantity that ordered the table was never shown. The columns are now derived from the canonical ranking basis: every numeric column is named for the quantity it holds, a change is shown as current, prior and change, similarity is shown beside the match-basis values with the anchor stated outside the ranked rows, a contextual value is labelled “(context)”, and a gap-to-leader is published only when the column it is measured on reproduces the published order. Scores, ranking calculations and every other output are unchanged.

Data Table (/intelligence)

One “Score” column for every insight type: a momentum ranking showed the current rating (never the change or its baseline), a similarity search showed closeness to the anchor under the same header as a performance score, and Undervalued Nations showed the Opportunity Score while ranking on the Undervaluation Gap→Columns derived from the ranking basis and named for what they hold — “5Y Prior / 5Y Change”, “Similarity” beside the match-basis values with the anchor stated separately, “Undervaluation Gap” beside the Opportunity Score it is computed from — with contextual values marked “(context)” and a gap published only when its column reproduces the published order

C·B

2026-08-17

The Comparison Matrix's per-entity statistics counted a “No data” cell as a score of zero. Those statistics — an entity's average across the selected metrics, its standard deviation and its coefficient of variation — decide the matrix headline, the executive summary, which entity leads on average, and the most balanced / most specialized calls. Reading an absent value as zero therefore published a coverage gap as a low, uneven profile: an entity missing one selected metric could be named the most specialized profile, or placed behind peers on the average, on a zero it was never given. Absent cells are now excluded from every per-entity statistic, matching the per-metric statistics, which already ignored them. An entity with no measured value in any selected metric has no comparable profile and is left out of these reads rather than entering them at zero, and the profile counts state how many entities were actually compared instead of how many were selected.

Comparison Matrix (/intelligence)

A “No data” cell entered the average, standard deviation and coefficient of variation as 0 — an entity measured on four of five selected metrics was averaged over five values, one of them a zero it never scored — so missing coverage read as an uneven profile and could carry the “most specialized” call or sink the entity's average→Per-entity statistics are computed only over measured cells (four measured values averaged over four); an entity with no measured value in any selected metric is excluded from the average, balance and specialization reads, and profile counts report the entities actually compared

C·B

2026-08-16

A five-part correction to the Signal Radar record. (1) Three signals were retracted: Israel and Australia were published as 'Election' and a United States 'Conflict' signal bundled unrelated stories — the collector counted syndicated copies of one wire story as independent articles and classified on a first-match theme order, so a single syndicated story could manufacture a dominant category. The collector now folds syndicated copies to one story before counting, scores every category and takes the strongest, and matches keywords as whole words. (2) Placeholder titles ('Ukraine: coverage activity') were replaced across the record by a deterministic cascade: a corroborated Event Registry headline when one exists, else the category and measured coverage intensity; publishing now rejects placeholder titles. (3) 'Confidence' is now 'Data confidence' and always shows its components (record coverage 0–60, corroboration 0–30, freshness 0–10); corroboration counts independent voices — one story syndicated across N domains is one voice — and the three live values computed under the old counting were restated (Turkey 99 → 84; Ukraine 91 → 84 and 83 → 84). (4) Signal URLs and datelines: slugs were re-minted from the event's own date (they previously encoded the internal freshness-expiry date, which read as a future date), every date label is ISO, and the page now labels its dates — Event and Published — explicitly; old URLs redirect permanently. (5) The 'What changed on the record' block showed the momentum series' own level as a second N/100 beside the structural score in the prose (Ukraine 44 vs 52), reading as a data error; it now shows the structural score with the momentum delta labeled as 3-year momentum. No underlying nation score changed in any of the five.

Signal Radar (/news)

Signals categorized by syndication-inflated counts; placeholder titles; bare 'Confidence — 99'; slugs dated by freshness expiry; momentum level rendered as a second score→Story-folded corroboration; synthesized titles; 'Data confidence' with components (99→84, 91→84, 83→84); slugs and labeled datelines on the event's own date (ISO, 308 redirects); structural score + labeled 3-year momentum

C·B

2026-08-16

The opportunity vocabulary was corrected to one rigid meaning per term, and the homepage card it affected was restated. (1) The homepage 'Hidden potential' card showed an 'Unrealized potential score' that was in fact a nation's single highest Opportunity Index vector, and labeled that nation 'undervalued' on a best-vector-minus-overall gap — while the platform's canonical Undervaluation Gap (Opportunity Score minus the higher of overall and strategic position, the same computation behind the rankings' 'Undervalued Nations' count and the Most Undervalued Countries ranking) was negative for that same nation. The card now reads the canonical gap, shows the Opportunity Score /100, and by construction names the same nation as the rankings board. (2) The 'Hidden Potential' insight was renamed 'Undervalued Nations' and now ranks by the canonical gap, positive gaps only; the strongest single vector remains as narrative context, never the ranking basis. (3) The vocabulary is now single-sourced in code: Opportunity Index = the three-vector instrument, Opportunity Score = the mean of the vectors that resolved (minimum two), Undervaluation Gap = the signed point difference, with 'undervalued' reserved for a positive gap. (4) The indicator registry's 'Start a Business' row claimed a live World Bank B-READY 2024 source with 145-nation coverage; the sub-score is in fact derived from Governance Effectiveness (0.55) and Economy Growth (0.45) group scores. The row now states the real inputs, and all four Opportunity rows carry coverage measured against the active snapshot (172 / 153 / 169 nations; composite 169). No formula changed — the corrections are to names, labels, one homepage card's basis, and registry provenance text.

Opportunity Score / Undervaluation Gap

Homepage: 'Hidden potential — Unrealized potential score' (best single vector, own gap vs overall labeled 'undervalued'); registry: Start a Business 'live · World Bank B-READY 2024 — 145 nations'→Homepage: 'Undervalued nation — Opportunity Score /100' on the canonical Undervaluation Gap; insight renamed 'Undervalued Nations'; registry: Start a Business 'derived · Governance Effectiveness (0.55) + Economy Growth (0.45)', measured coverages 172/153/169

C·B

2026-08-16

Country Comparison's interpretation of published scores was corrected to respect ties, in five parts. (1) The tie policy is now published precision: two scores are tied exactly when their published integers are equal. The previous ±1-point tolerance manufactured ties a reader could not reconstruct from the visible scores (Uruguay 55 and Peru 54 reported as 'tied on Economy at 55'). (2) Equal published Overall scores were ordered by selection position, naming one arbitrary country the leader — a comparison whose top scores were 69 and 69 rendered '{A} Leads Overall By A Narrow Margin', a 'Gap to #2' of 0 points, and the co-leader as Runner-up. Co-leaders now share rank 1 under competition ranking (69, 69, 56, 56, 46 reads 1, 1, 3, 3, 5) and the gap runs to the next distinct published tier. (3) Category win counts credited a shared first place as an outright win ('wins 4 of the 5 selected metrics' where one of the four was a tie); sole and shared firsts are now counted separately, and the five compared dimensions are named categories, with 'metric' reserved for indicator rows. (4) Balance and specialization are now judged on the published profile range — a nation's highest minus lowest selected category score — instead of an unpublished standard deviation, so the call is reconstructable from the visible numbers; ties keep every holder. (5) The share card now leads with the comparison result itself rather than a secondary trajectory finding, and never splits a tie group across its Highest/Lowest columns; the derived-findings block states its universe (the selected nations, with the statistical context pool labeled explicitly); and the Attribution confidence axis no longer renders for nations — their scores are descriptive measurements, and a causal-attribution level was being inferred from the sourced share alone. No published score changed — the corrections are to ranks, ties, gaps, win counts and labels derived from them.

Country Comparison (/intelligence)

±1-point ties ('Uruguay and Peru tied at 55' over 55 vs 54); an arbitrary sole leader over an exact tie ('leads by 0 points', co-leader as Runner-up, ranks 1,2,3,4,5); shared category firsts counted as outright wins; balance judged on an unpublished dispersion statistic→Published equality is the only tie; co-leaders share rank 1 with competition ranks (1,1,3,3,5) and the gap runs to the next distinct tier; sole vs shared category wins separated; balance and specialization by published profile range

C·B

2026-08-14

The first published bilateral-trade batch queried UN Comtrade by ISO-numeric country codes, but Comtrade publishes six economies only under special M49 codes (USA 842, France 251, India 699, Norway 579, Switzerland 757, and Taiwan as "Other Asia, nes" 490). Those six were therefore missing both as reporters and — more damagingly — as trade PARTNERS in every nation's rows, which distorted partner and bloc shares: no nation showed the United States, France, or India among its partners (Mexico's top export partner displayed as Canada instead of the United States). The crosswalk now maps both code families and the batch was re-ingested in full: 170 reporters, with the United States present as a partner for 169 of them.

Bloc Exposure · trade partner shares

Partner shares computed without US/France/India/Norway/Switzerland/Taiwan (e.g. Mexico top export partner "Canada")→Full partner sets (e.g. Mexico top export partner United States, 84.5% of recorded exports, 2024)

C·B

2026-08-07

A nation with no gated risk previously emitted a hardcoded "Regional spillover risk" threat — a non-signal with no datum behind it, laundered as the nation's sharpest risk. It was removed. When no absolute gate fires, the sharpest risk is now the nation's own relatively-weakest measured dimension (real, sourced); a nation with no scored dimension at all emits no threat rather than a fabricated one. The scored-deficit band was also generalized so any category whose sharpest input sits below the concern threshold surfaces — not just the two hand-named narratives — so a genuine sub-50 weakness can no longer be hidden behind the fallback.

SWOT · Threats

"Regional spillover risk — Exposure to neighboring instability" whenever no other threat fired→The nation's relatively-weakest measured dimension, or no threat when nothing is sourced

C·B

2026-08-07

The nation-page "Predictive Insights" module was rebuilt as "Advanced Diagnostics". The six heuristic cards (Transition Risk, Geopolitical Risk Score, Stability Erosion, Institutional Bottleneck, Strategic Opportunity, Policy Gap) computed 0-100 scores from arbitrary hardcoded coefficients over one or two composites, advertised input factors they did not actually use, and — for Geopolitical Risk and Policy Gap — always subtracted a fabricated "Regional spillover risk" threat. They were removed. In their place are deterministic second-order diagnostics that classify from real evidence and disclose coverage: Stability Erosion Detection, Fiscal Trend, Institutional Bottleneck Mapping, Transition Readiness, and (behind a data gate) Bloc Exposure. No card produces a 0-100 score from invented coefficients, and every result is reconstructable from the evidence shown.

Nation dossier · Predictive Insights module

Six heuristic cards scored 0-100 from hardcoded coefficients (e.g. Transition Risk 69, Geopolitical Risk 52), with advertised inputs the formula ignored and an always-present fabricated threat→Deterministic diagnostics with named classifications, honest coverage states, and no invented coefficients or fabricated threats

C·B

2026-08-07

The v5.4.0 weakness-band alignment (a weakness requires scoring clearly below the 58–69 benchmark band, ≤57) was applied incompletely. Two documentation surfaces still stated the retired rule — the Strengths & Weaknesses tooltip said weaknesses score "below 70", and the methodology catalog said the columns were "computed from top/bottom 3 stats" (neither the cut nor the count of up to 5 was right) — so the displayed columns could not be reconstructed from the stated method. Both now interpolate their thresholds from the same policy object the ranking code executes, so the stated rule and the running rule cannot drift apart again. One display surface was also still on the old cut: the former/historical leader Record module gated its "Relative Weaknesses" at under 70, branding near-benchmark dimensions in the 58–69 band as weaknesses. It now gates at ≤57, identically to the active-leader column; a former whose lowest dimensions sit in the benchmark band lists no weaknesses.

Strengths/Weaknesses · Former-leader Record

Tooltip: weaknesses "scoring below 70"; methodology catalog: "computed from top/bottom 3 stats"; former-leader Relative Weaknesses listed dimensions scoring 58–69→One rule stated and applied on every surface: strength ≥70, weakness ≤57, near-benchmark 58–69 in neither column; thresholds interpolated from the executing policy

C·B

2026-08-07

The BETA provenance tier was retired. Fifteen "Premium / Intelligence" placeholder indicators (Transition Risk Prediction, Geopolitical Risk Score, Active Threats, a Bloc Exposure stub, Defense Posture, Shock Scenarios, Policy Gap and eight others) were never wired to a real pipeline and never fed any score; they have been removed from the indicator registry. Every metric is now labeled live + sourced or derived only, and the Beta tag was removed from the footer, header trust ticker, provenance legend, About page, and methodology.

Provenance tiers · Beta indicators

15 beta placeholder indicators; provenance tiers Live / Sourced / Derived / Beta→0 beta; provenance tiers Live / Sourced / Derived

C·B

2026-08-07

Three Advanced Diagnostics modules were renamed to describe their inputs more precisely: “Stability Erosion Detection” is now “Structural Erosion Detection” (it reads per-category structural deltas, not only the Stability category), “Fiscal Trend” is now “Macro-Fiscal Trend” (its inputs span debt, fiscal balance, and unemployment — macro-fiscal conditions, not the fiscal ledger alone), and “Transition Readiness” is now “Transition Resilience” (it classifies structural resilience entering a transition; it does not assess preparedness or predict that a transition will occur). Methodology, inputs, classification logic, and every computed result are unchanged — the correction is to the names only.

Advanced Diagnostics

Stability Erosion Detection · Fiscal Trend · Transition Readiness→Structural Erosion Detection · Macro-Fiscal Trend · Transition Resilience

C·B

2026-08-02

A nation's defining asset — the strength named first in the dossier SWOT, the persona brief, and the strengths-and-risks FAQ — was the indicator with the highest normalized score. On saturated indicators most of the world also maxes, that selection carries no information: Barbados read 'Electricity Access' (an indicator whose world-median score is 100 across 201 nations) as its defining asset, and roughly half of all nations led with a similarly universal metric. The defining asset now requires distinction, not just a high score: scored strengths rank by their lift over the world-median score on the same indicator, computed from the pinned snapshot; a strength the world broadly shares falls behind the nation's structural assets (strategic resources, the leading CIA World Factbook industry, low-carbon generation share), and an indicator sitting below the world median never surfaces as a strength at all. Barbados now leads with Ease of Resolving Disputes; the United States, Japan, South Korea and Germany with Patent Applications; Trinidad and Tobago with Petroleum; Namibia and South Africa with Mining. No score changed — the correction is to which sourced strength is presented as defining.

Defining Asset (SWOT)

Defining asset = the nation's highest-scoring indicator (Barbados: Electricity Access, world-median score 100)→Defining asset = the strength with the largest lift over the world median, else a structural asset (Barbados: Ease of Resolving Disputes)

C·B

2026-08-01

Four corrections from the post-calibration methodology audit (scoring model v5.4.0). (1) Legacy anchoring: the tenure and luck factors multiplied the absolute performance base, which silently moved the benchmark — a leader who performed exactly at the world trajectory drifted below it on a short tenure or above it under a headwind. The factors now scale the distance from the benchmark (50 + (base − 50) × tenureFactor × luckFactor), so a benchmark performance stays a benchmark legacy; per-leader shifts ran −3 to +7 raw points (e.g. Merkel 54 → 51 raw: her at-benchmark contribution no longer gains points from the headwind multiplier; Karzai published 82 → 83). (2) Crisis Response now publishes through the same calibration as every dimension — its internal 50 is the median outcome among comparable-severity crises, a benchmark point — so displayed cri values moved (raw unchanged, composite unchanged). (3) A rating now requires all four core dimensions (Governance, Economy, Politics, Institutional Integrity) sourced; 18 of 885 previously-rated mandates now honestly read No data, and every rating carries a visible coverage note (n/8). (4) Strengths/Weaknesses and challenge flags are aligned to the published bands: a weakness now requires being clearly below the benchmark band (≤57), so above-benchmark dimensions in the low 60s are no longer branded weaknesses, and challenge flags no longer fire on values that had beaten the world benchmark.

Legacy Rating · Crisis Response (published) · Leadership Rating coverage · Strengths/Weaknesses

Legacy = base × factors (benchmark drifted with tenure/luck); cri displayed uncalibrated; ratings possible without a core dimension; every dimension under 70 labeled a weakness; Economy under 64 flagged as a challenge→Legacy = 50 + (base − 50) × factors; cri published on the calibrated scale (60 = matched comparable crises); all four core dimensions required, coverage n/8 shown; weakness ≤57 / strength ≥70 with a near-benchmark middle; challenges fire only below the benchmark band

C·B

2026-08-01

Semantic calibration of the published leader scale (scoring model v5.2.0). The contribution engine is internally centred on 50 — a leader who matched the world trajectory scored exactly 50 — which the public 0-100 scale made read as mediocre or failing: competent stewards of already-excellent, mature economies displayed in the high 40s and low 50s. The published scale now remaps that neutral point to 60 via one strictly monotonic map (raw 0 to 0, raw 50 to 60, raw 100 to 100), applied only after each raw dimension or composite is fully computed. The underlying contribution estimates are unchanged: rank order is provably identical (871 scored leaders, 0 rank moves — see docs/data-quality-fix/leader-calibration-diff-report.md), snapshots and rankings stay on the raw scale, and the transparent ledger retains both values. Crisis Response is not remapped (an absolute response-quality scale, not benchmark-relative). A separate Stewardship lens was added alongside — whether strong inherited institutions were preserved — never blended into the rating.

Leadership Rating · Legacy Rating · leader dimensions

Published leader scores displayed the engine scale directly: benchmark-matching performance read as 50 (e.g. Frederiksen 57, Store 48, Kristersson 46, Orpo 46)→Benchmark-matching performance publishes as 60 on the calibrated scale (Frederiksen 65, Store 58, Kristersson 55, Orpo 56); raw values retained in the ledger; ranking order unchanged

C·B

2026-08-01

Active leaders' Party field concatenated every open-ended Wikidata P102 membership with ' / ', presenting a leader's whole party history as their current affiliation (Shavkat Mirziyoyev showed five parties including the Communist Party of the Soviet Union; Bosnia's record blended the parties of all three presidency members into one string). The identity refresh now resolves the ONE party held during the mandate via the same term-overlap picker the former-leader path uses (statement rank + start/end qualifiers), groups Wikidata officeholders per person so collective presidencies and duplicate items can never blend affiliations across people, and refuses to write party or traits when Wikidata resolves a different person than the record (identity coherence gate). Exactly two simultaneous affiliations render as a dual membership (Labour / Co-operative in the UK, Broad Front / Movement of Popular Participation in Uruguay); a set of three or more is genuinely a party history, not a dual, and shows the house null mark, never a concatenation. Two records carry reviewed manual values with locked provenance: Uzbekistan (Uzbekistan Liberal Democratic Party, the nominating party in the 2016/2021/2023 elections), Cayman Islands (prior Caymanian Community Party retained; Wikidata carries no P102 for the premier and absence is not evidence of no affiliation). 22 active-leader party values changed; no score changed except Hungary, whose refresh also grounded a Wikidata-derived dynast trait.

Party affiliation

Party = every open-ended Wikidata P102 joined with ' / ' (party history, and in collective presidencies other people's parties, presented as current affiliation)→Party = the single mandate-overlapping P102 statement per resolved person, with identity coherence gates; exactly two simultaneous memberships show as a dual, three or more show the null mark

2026-08-01

A leader's office was stored as a single flat vocabulary bucket, which collapsed the constitutional structure a head's title carries: collective and rotating heads of state, provisional (acting/interim) holders, co-heads sharing one office, and dual-hat holders all rendered as the plain bucket. Bosnia and Herzegovina, Libya and Yemen — each led by a chair of a multi-member body — displayed as 'President'; Saudi Arabia showed 'Crown Prince', omitting the executive office (Prime Minister) through which the record ranks the leader; Nicaragua's constitutional co-presidency, Venezuela's acting presidency and Togo's head-of-government office were flattened away. A structured office layer now records the exact title(s), any collective body (with seat count), provisional status and co-heads; the generic bucket stays for scoring, and a derived display title carries the truth. Seven current records corrected. No score changed — this is a titles/labelling correction.

Role / office title

Collective, acting, co-, and dual-hat heads all rendered as the flat role bucket (Bosnia / Libya / Yemen shown as 'President'; Saudi Arabia as 'Crown Prince', omitting the executive office)→Office-derived display titles: 'Chairman of the Presidency' (Bosnia), 'Chairman of the Presidential Council' (Libya), 'Chairman of the Presidential Leadership Council' (Yemen), 'Co-President' (Nicaragua), 'President of the Council of Ministers' (Togo), 'Acting President' (Venezuela), 'Crown Prince & Prime Minister' (Saudi Arabia)

2026-08-01

Four current-officeholder slots named the wrong person. Myanmar's record carried the name of the deposed civilian president over a profile whose traits and scores are the junta chief's — corrected in place to Min Aung Hlaing, retaining the profile that is genuinely his. Peru named a superseded interim (José María Balcázar) whom Keiko Fujimori replaced on assuming the presidency on 2026-07-28; Balcázar is preserved as a former. Haiti named a person the source does not carry as head of state — the presidency is vacant (the Transitional Presidential Council has ended) and the executive is the Acting Prime Minister, Alix Didier Fils-Aimé. Lithuania named Inga Ruginienė, whom Mindaugas Sinkevičius replaced on appointment by presidential Decree 1K-733 (in force 2026-06-30); Ruginienė is preserved as a former. Identity for the first three is sourced from Wikidata; Lithuania is a reviewed manual correction with locked provenance because the upstream head-of-government statement still lags at the prior premier. A live incumbency audit (scripts/audit-incumbency.mjs) and offline roster invariants were added to catch this class of drift going forward.

Current officeholder

Current officeholder: Myanmar = Win Myint; Peru = José María Balcázar; Haiti = a non-sourced name; Lithuania = Inga Ruginienė→Myanmar = Min Aung Hlaing; Peru = Keiko Fujimori (assumed 2026-07-28); Haiti = Alix Didier Fils-Aimé, Acting Prime Minister (presidency vacant); Lithuania = Mindaugas Sinkevičius (Decree 1K-733, 2026-06-30)

2026-08-01

Defense was retired as a leader dimension (scoring model v5.3.0). The axis overlaid the nation's Military Score — a log-scaled composite of real force counts — onto the person, which is a property of the country rather than something the leader's tenure moved. Under the contribution model (v5.0.0, which measures what changed under the leader net of the world) the dimension was never produced: every leader read No data on it, and its 0.07 weight never participated in any composite because the overall renormalizes over sourced dimensions only. The leader stat set is now eight dimensions (Governance, Economy, Diplomacy, Politics, Crisis Response, Vision, Communication, Institutional Integrity). No leader's score, rank or archetype changed — the republished snapshot's diff gate ran with zero value findings. National military capability remains scored on the nation's own Military Score.

Defense (leader dimension)

Leader radar and ledger carried a ninth Defense axis that always read No data→Eight-dimension leader stat set; Defense removed from radar, ledger, comparisons, OG cards and the indicator registry; scores, ranks and archetypes unchanged

C·B

2026-07-27

The per-category momentum trend arrows (rankings, nation dossiers, and the Intelligence nation filters) were computed from a LIVE database query at request time — the last live-DB reading left in the page render, which both contradicted the pinned-snapshot guarantee (scores are served from an immutable snapshot, never a live DB overlay) and, on a cold cache with a slow connection, hung the Intelligence page while it waited for the database. Trends now read from the same committed, pinned momentum-history batch the snapshot itself uses, through the same deterministic builders — so the arrows match the pinned vintage exactly and the page no longer depends on a live connection. Values are the same underlying data (the batch is the export of those observations); an arrow only shifts where the live database had drifted ahead of the pinned batch, in which case it now agrees with the snapshot rather than a newer live reading. No score changed; overall momentum and the snapshot content hash are unaffected.

Category Momentum

Category trends from a live DB query (could drift ahead of the pinned snapshot)→Category trends from the pinned committed batch (agree with the snapshot)

C·B

2026-07-27

Scoring v5.0.0 shipped the leader Vision dimension unscored (No data) while the contribution model was worked out. v5.1.0 restores it as an innovation CONTRIBUTION, consistent with the other dimensions: each year, WIPO patents-per-million and UNESCO tertiary enrolment are ranked across all countries, and Vision is the change in that innovation rank over the tenure, net of the world (50 = neutral, above = the country climbed the field, below = it lost ground). Ranking — not the absolute level — is used because the global innovation frontier rises steeply, which would otherwise read every leader as declining. Because WIPO patent data lags about three years, active leaders have no fresh endpoint and read No data (a stale reading is never presented as current); only former leaders whose mandate sits inside the patent window are scored — about 347 of them (e.g. Angela Merkel 30, as Germany's innovation rank slipped while Asia surged; José Mujica 79, as Uruguay climbed).

Vision

Vision — No data for all leaders (deferred in v5.0.0)→Vision — innovation-rank contribution for former leaders in the patent window; No data for active leaders (patents ~3y stale)

C·B

2026-07-27

On /intelligence, a comparison focused on a single metric still reported the overall Leadership Snapshot instead of the selected metric. The share/OG card printed the Snapshot scores under the focused-metric heading, and the panel could name the Snapshot leader as the winner of a metric they did not lead — the 'wins every / wins 0 of 1' contradiction. A comparison of Burnham, Merz, Macron and Meloni focused on Economy showed Burnham 74 (the Snapshot) as the top number, when on Economy Macron leads at 69 and Burnham is last at 54. All intelligence surfaces now read one canonical facts object: the overall Snapshot and each selected metric are ranked separately (competition ranks, ties shown), the share card follows the selected metric, and where a metric winner is not the Snapshot leader the panel states it outright. Confidence is now reported on three distinct axes — provenance, temporal comparability, and attribution — rather than one collapsed badge that could mask a weak axis. No score changed; the fix closes a display leak that conflated the overall rating with a single-metric ranking, consistent with the single-snapshot guarantee.

Intelligence Panel

Single-metric focus showed the overall Snapshot scores and leader→Snapshot and each selected metric ranked separately; the card follows the selected metric

C·B

2026-07-27

The Intelligence findings engine presented the freshest STORED reading of a series as the current state without checking that it was actually recent. Two detectors did this: the historical-record detector (a series hitting a new high/low, or a run of consecutive moves, ending at the latest point) and the trajectory-crossover detector (who leads whom 'now'). When a feed had stopped years ago — WIPO patent applications last observed 2021, learning outcomes 2020, both read against a 2026 snapshot — a years-old reading was laundered as a live record or a current standing, the stale-as-current trust violation the house rules forbid. Both detectors now gate the latest reading through lib/freshness against the snapshot's publish year and each metric's own freshness window (three-year default for annual indicators); past the window the finding is suppressed rather than shown as current. No score changed. The visible effect is that stale records drop out and a fresh finding takes the slot — e.g. in the fragile-state pool a genuinely current GDP-per-capita record surfaces where a stale reading previously sat.

Intelligence Findings

Findings read the latest stored point as 'current' with no recency check→Present-tense findings gate on the metric's freshness window vs the snapshot publish year

C·B

2026-07-26

Quality of Life scored three pillars — Safety, Healthcare, Education — while its public description claimed it covered 'health, safety, environment and living standards.' Environment and living standards were never in the formula, so the copy overstated what the number measured. Both are now scored for real: QoL is five equal pillars (20% each) — Safety, Healthcare, Education, Basic Living Conditions, and Environment. Basic Living Conditions is a weighted group of basic sanitation, electricity access, and clean-cooking access (World Bank / WHO-UNICEF JMP), floored at ≥2 of 3 present inputs. Environment is PM2.5 mean annual exposure (World Bank / WHO EN.ATM.PM25.MC.M3), inverted so cleaner air scores higher — the signal that registers, e.g., that air quality across India and the Gulf is among the world's worst, a QoL fact the three-pillar model missed. Coverage floor is unchanged in spirit, now ≥3 of 5 groups. Safely-managed sanitation and extreme-poverty headcount were considered but stay display-only (too sparse to score without imputing or serving stale readings). Reorders the Highest Quality of Life ranking and shifts Overall for nations whose sanitation, electricity, cooking, or air-quality readings diverge from their prior three-pillar QoL.

Quality of Life

METHODOLOGY v2.10.0→METHODOLOGY v3.0.0

C·B

2026-07-26

No leader score is authored any longer. About 20 historical 'ICON' leaders (Lincoln, Churchill, Gandhi, Napoleon, Mandela and others) carried a Legacy Score and Peak-Career Rating computed from a curated editorial model, not from sourced indicators — an authored number the record should never contain. That model was removed (scoring v4.0.0): a former or historical leader is rated only from sourced, tenure-windowed data, or is shown Unrated. Leaders whose mandate ended before 1990 (13 records) cannot be sourced — the indicator panels begin in the mid-1990s — and were removed. Nelson Mandela and Václav Havel became rated from real tenure data; Lee Kuan Yew, Deng Xiaoping and Margaret Thatcher read Unrated, their tenure predating the data window. (Leader dimension scores were subsequently re-derived again under the contribution model, v5.0.0.) Nation scores and rankings are unaffected.

Legacy

METHODOLOGY v3.0.0→METHODOLOGY v4.0.0

C·B

2026-07-26

A leader's card and radar showed the trait-adjusted Diplomacy dimension, but the FAQ, persona brief, OG signal card, and Diplomatic Signal breakdown read the raw sourced Diplomatic Conduct score directly — bypassing traits. The same leader carried two Diplomacy numbers on the same page: Narendra Modi read 64 on the card and radar but 67 in the FAQ, brief, and OG card. Every surface now quotes the same post-trait value (Modi 64), and the breakdown shows the trait step explicitly — conduct 67, traits −3, shown 64 — so the decomposition reconciles instead of contradicting itself. No score changed; the fix closes a display leak that undermined the single-snapshot guarantee. Diplomacy was the only exposed dimension — the only one with a standalone raw sourced object that downstream surfaces re-read.

Diplomacy

Two Diplomacy numbers per leader (raw vs trait-adjusted)→One post-trait Diplomacy number on every surface

C·B

2026-07-26

Leader Governance, Economy, Politics and Institutional Integrity read the country's ABSOLUTE level — active leaders overlaid the live World Bank / WGI level, formers the term-window mean — so the score measured the nation, not the person. It read Angela Merkel at 82 ('Germany is rich'), and let a new leader inherit the predecessor's numbers the day they took office: a successor and the departed leader showed the same country figures. Scoring v5.0.0 replaces the level with a CONTRIBUTION — how each indicator moved over the tenure, net of the world's move over the same window (50 = neutral, above = improved the country's standing, below = declined). Active and former leaders now use the same measurement and are directly comparable; a successor's score starts from their own inauguration, so nothing is inherited; a leader under a six-month tenure, or below the two-dimension coverage floor, is Not yet rated. Every leader was re-derived (e.g. Merkel 82→55, Mujica 69→50); nation scores are unchanged. The difficulty multiplier and the Context-Adjusted Leaders insight were retired because contribution is already country-neutral. Vision is temporarily unscored for leaders pending a per-capita contribution pass.

Leadership Rating

Leader dimensions = the country's absolute level (a successor inherited the predecessor's numbers)→Leader dimensions = the leader's contribution over their tenure, net of the world (50 = neutral)

C·B

2026-07-24

Bloc membership now filters on two axes, not one. The Wikidata P463 fetch dropped only memberships a country formally ended (end-date P582), so an organization that had dissolved entirely still read as current — the Southeast Asia Treaty Organization, dissolved in 1977, was scored as a diplomatic strength for the United States, United Kingdom, France, Australia, New Zealand, Pakistan, Thailand and the Philippines, and surfaced in Thailand's and the Philippines' SWOT. A past organizational dissolution date (P576) now drops SEATO and ten other defunct bodies (League of Nations, Organisation of African Unity, G8, …) for every nation. Ukraine is removed from the Commonwealth of Independent States Free Trade Area on the record of its exit from the CIS framework — the May 2018 presidential decrees ending its participation in CIS bodies and agreements, after the free-trade arrangement with Russia was suspended in 2015. Brexit/EU, Russia/Council of Europe and Mali–Niger–Burkina/ECOWAS were already correct — Wikidata carries those end-dates. 192 nations carry at least one scored bloc; overall headline rankings are unaffected.

Geopolitics

Defunct & withdrawn memberships scored as current→Active memberships only (P582 / P576 + sourced exits)

C·B

2026-07-23

The Geopolitics 'blocs' signal no longer scores a hand-picked ~30-org allowlist — an arbitrary list that silently omitted the Council of Europe, OSCE and WTO, so Ukraine read 'No bloc memberships on record' despite deep Euro-Atlantic integration. The taxon is now open and derived from the source: all Wikidata P463 memberships are ingested and the alignment blocs are selected by the org's own type within a size band plus a small explicit supplement (35 scored blocs), scored on a diminishing curve. Ukraine Geopolitics 13 to 63, China 53, North Korea a real 0; universal UN-system and technical bodies are shown in the total but not scored. Overall headline rankings are unaffected.

Geopolitics

METHODOLOGY v2.9.0→METHODOLOGY v2.10.0

C·B

2026-07-23

An empty tracked-bloc list was scored as a real 0 inside the Strategic Position Geopolitics component (blocs.length × 20) — a fabricated fallback that, because [] is a structurally-valid value, slipped past every missing-data guard and entered Geopolitics at 40% weight. Empty is now null and renormalised out over the signals actually present. Ukraine Geopolitics 13 to 22, Strategic Position 46 to 48, with similar honest lifts for Georgia, Serbia, Bosnia, Taiwan and North Korea. Overall is unaffected.

Geopolitics

METHODOLOGY v2.8.0→METHODOLOGY v2.9.0

C·B

2026-07-23

Government Stability's Violence & Safety sub-signal re-anchored to UCDP conflict thresholds. It no longer inverts per-capita deaths against the worst country in the cohort; it's now min(a population-normalized curve to a fixed 50 deaths/M anchor, a UCDP band cap at the 25 armed-conflict / 1000 war thresholds), taking the worse of the finalized-annual (GED) and current-year (CED) readings — so an active war can't read 'safe' in a large state and one country's war no longer moves another's safety. Stability fell for conflict nations: Iran 44 to 31, Syria 34 to 20, Russia 48 to 34, Mexico 53 to 39, CAR 45 to 23.

Stability

METHODOLOGY v2.7.0→METHODOLOGY v2.8.0

C·B

2026-07-22

Leader Diplomacy re-scored under the Diplomatic Conduct model. The old Diplomatic Signal rewarded engagement volume, so a leader projecting force abroad scored high; a restraint component — net material force initiated, not received — is the fix. Trump 90 to 75, Netanyahu 81 to 59; Zelenskyy holds at 86 (invaded, restraint 89).

Diplomacy

METHODOLOGY v2.6.0→METHODOLOGY v2.7.0

C·B

2026-07-19

Quality of Life is now hard-capped at 40 for nations that are active organized-violence parties. Population-scaled safety metrics had let a large nation at war read as safe — Iran scored 80 — so 11 conflict nations were corrected; Overalls fell 2 to 10 points (Russia 58 to 48, Iran 49 to 41, Mexico 51 to 45).

Quality of Life

METHODOLOGY v2.5.0→METHODOLOGY v2.6.0

C·B

Who is accountable for every number, and how the numbers are made, is on the about page. How each figure is sourced and scored is on the methodology.