Scored classification · eight regions · horizon 2050
Lead Bloc or
Strained Adopter
This is a forecasting exercise, not a policy proposal. It asks one question about eight major economies — the US, EU, UK, Japan, China, India, the Gulf States and Southeast Asia:
By 2050, which of these can still build and run energy-intensive industry on their own account — and which will be importing capability they can no longer produce themselves?
Each region is sorted into one of the two classes that answer it. Nothing here ranks countries in general; the question is narrow and it is about industrial capability under energy and demographic constraint.
Lead Bloc
Can build and run energy-intensive industry at scale on its own account: it has the firm power, the working-age people, and the ability to permit, fund and actually deliver.
Strained Adopter
Adopts capability rather than building it, importing what it cannot produce domestically — constrained by power, by people, or by its own capacity to execute.
The exercise is conditioned on one specific view of the future, carried in from prior work and not re-argued here: progress to 2050 is regionally bifurcated and uneven, under strain — held at 55% probability — rather than a single global trajectory. That conditioning is why the answer is a split rather than a ranking: the premise is that regions diverge, so the model's job is to say which side of the divergence each one lands on.
Two facts about 2050 are treated as substantially locked in rather than open — demographics, and climate stress — and both are scored. Committed is what makes a variable scorable, not what excuses leaving it out.
Each region is scored 1–5 on seven indices. Three are the originals: the firm power it can add by 2035, the working-age population it will have in 2050, and its capacity to execute. Committed climate stress enters inside those three rather than sitting alongside them, because climate is not a separate outcome — it acts through power, through labour and through fiscal room. Four more were added after an audit found them missing: supply-chain position in critical minerals and grid hardware, cost of capital, capacity to substitute automation for labour, and institutional continuity to 2050. The seven are weighted into one composite and the class boundaries are derived from the resulting distribution rather than set by hand. Then a single intervention is tested: cut every bureaucratic approval timeline in half. Physical procurement is deliberately held constant, because transformers, turbines and grid hardware run on multi-year global supply chains that no government's paperwork speed can accelerate. That distinction is the point of the test — it separates regions whose problem is genuinely red tape from regions whose problem is physics and industrial capacity, and those turn out to be different regions than the usual discussion assumes.
The outcome: three regions clear the bar — China, the Gulf and the United States. Five do not, and they include India and every mature democracy in the set. Halving every permitting timeline in the world changes none of it.
What a region cannot buy matters more than what it can build. The four added dimensions move the answer more than climate did and more than permitting ever could. India leaves the Lead Bloc — it is the most exposed region in the set to everything the original model was not measuring, scoring worst of eight on cost of capital and on automation capacity. China takes first place outright on a supply-chain score of 5.0, which is the uncomfortable corollary of this model's own firewall: if hardware is the binding constraint, then whoever controls the hardware holds the constraint.
Japan is the largest single revision in the model's history. It gains +0.748 from the new dimensions — the biggest move of any region — and rises from last-but-one to the strongest of the Strained Adopters. It scores highest of all eight on the four dimensions the earlier versions ignored. It is still a Strained Adopter, because 55 million working-age Japanese in 2050 is not a thing that automation, cheap capital and stable institutions fully offset. But the gap narrowed sharply, and that is a warning about how much the original three indices were leaving out.
Permitting reform is inert, again. Halving approval timelines flips nothing anywhere. Japan takes the largest gain available from precisely that intervention — +0.176 — and needs +0.326 to reach even Borderline. Its binding constraint was never paperwork.
Largest revision
Japan, from the four added dimensions — it scores best of all eight on what the model previously ignored
Lead Bloc
China, Gulf, US. India drops out at −0.493. The US is the marginal member — see the severity band
Permitting reform
largest gain +0.176 (Japan), against a 0.326 requirement to reach Borderline
Everything below is that model, opened up. The scoring table and threshold ruler show where each region lands and how far the sensitivity case moves it; methodology, threshold derivation and the hardware firewall set out the weights, how the class boundaries were derived, and exactly how procurement is held constant; region rationale gives the indicator evidence behind every score including its climate component; the sensitivity test back-solves each region's flip requirement; weight robustness, the climate severity band and the version decomposition stress the result, known limitations says plainly where it is weak; and what would prove it wrong lists the numeric tripwires that would falsify it before 2031.
Scoring table
Toggle between the baseline and the sensitivity case. Watch how little moves — the composite bars and class pills are live.
| Region | A · Firm power to 2035 |
B · Working age to 2050 |
C · Execution capacity |
D1 supply |
D2 capital |
D3 robots |
D4 continuity |
Composite 1 --- 5 | Score | Class |
|---|
Ticks on each composite track mark the two thresholds: 3.25 (Strained ceiling) and 3.50 (Lead floor). Index C shown here is the weighted roll-up of four sub-scores — approval timeline, fiscal room, veto-player density, delivery record — which are listed per region below.
Threshold ruler
Every region's baseline position, with an arrow to where the 50% approval-timeline cut takes it. The two vertical lines are the class thresholds. No arrow crosses one.
Methodology
Weights and thresholds were fixed before composites were computed, so the sensitivity test could not be rigged by tuning them afterwards.
The three indices
Each region is scored 1–5 in 0.5 increments. Half-point granularity is used because several regions sit genuinely between whole points, and because a threshold band this narrow needs the resolution to be defensible.
A · Firm-power growth to 2035
Grid expansion, dispatchable generation additions, transmission buildout, industrial power reliability. Weighted highest because it is the binding physical constraint of the 2025–2035 window and the hardest thing to fix late. Climate component: cooling demand raises the firm capacity a region must have, while hydro variability and thermal or nuclear cooling-water limits derate what it already has.
B · Working-age trajectory to 2050
Demographic projections, aging ratio, migration offset. Weighted second because it is a high-certainty variable — substantially locked in — setting the 2040–2050 denominator for growth and the fiscal burden. Climate component: heat stress means headcount overstates usable labour hours, and the ILO puts the heaviest losses in South Asia and Southeast Asia.
C · Permitting / fiscal execution
Decomposed into four sub-scores so the sensitivity test has a defensible mechanism to act on: approval timeline (0.35), fiscal and capital-market room (0.25), veto-player density (0.20), delivery track record (0.20). Climate component: adaptation capex and disaster recovery as a standing claim on C2, including where insurance markets retreat and losses land on public balance sheets.
Why climate is not a fourth index
Because it is not a separate outcome. Committed climate stress is a mechanism that acts on power, labour and fiscal room — the three things already being scored. Adding it as a fourth index would double-count those channels and imply a region could trade climate exposure against firm power, which it cannot. Scores are held as raw-judgement and climate-component pairs so both halves stay auditable, and the integrated score is the one that classifies.
Why C is weighted lowest
It is the most malleable, the most reform-responsive, and the noisiest to measure. Deliberately held at 0.25 so the permitting sensitivity — which touches only one sub-component of C, plus a narrow channel into A — cannot dominate the result by construction.
Weights
A stays the largest single weight — unchanged logic, it is the binding physical constraint. D1 takes the largest of the new weights because the hardware firewall makes it load-bearing. D4 takes the smallest because 25-year regime forecasting is the least evidence-grounded judgement here. Within C: C1 approval timeline 0.35 · C2 sovereign fiscal room 0.25 · C3 veto density 0.20 · C4 delivery record 0.20. The permitting sensitivity moves C1 only.
Climate severity
The climate components are marginal — an estimate of what the raw judgement failed to price, not total exposure. Some stress is already inside the raw scores: India's 6–8%/yr peak demand growth and the Gulf's brutal summer peaks partly reflect cooling load that is already arriving. That discount is genuinely contestable, so severity is a declared parameter and the model is re-run across a band from 0.5× to 1.5×, with thresholds re-derived at each point. The headline case is 1.0×.
Thresholds are not fixed by hand in v2 — see threshold derivation. The A-floor of 3.00 is carried over unchanged from v1 as a substantive rule: you cannot lead an energy-constrained era without firm power, however well you score on demography and execution.
The seven indices
Three original, four added after an audit asked what the model was not measuring. The four are not decoration — together they move the answer more than climate did.
Each of the four additions closes a specific gap, and two of them close internal inconsistencies rather than merely adding detail — the same class of error as scoring demographics while ignoring climate.
D1 · supply-chain position weight 0.14
Control of, or secure access to, critical minerals, refining capacity and long-lead grid hardware. This closes an internal inconsistency. The model's own hardware firewall asserts that procurement, not paperwork, is the binding constraint. If that is true, then who controls the procurement is decisive — and relying on the firewall while not scoring it was incoherent. China scores 5.0: roughly 90% of rare-earth refining, 60–70% of lithium refining, ~80% of the solar supply chain, and dominant cell manufacturing. The uncomfortable corollary is that the Strained Adopters' hardware bottleneck is substantially a Chinese supply bottleneck.
D2 · project cost of capital weight 0.10
The WACC a real project actually clears. Energy buildout is capital-intensity-dominated, so cost of capital moves levelised cost more than resource quality does — a 4% and a 12% project are different businesses on identical engineering. This is why India and Southeast Asia build more slowly than their resource endowment implies, and it suggests the constraint often attributed to land acquisition and permitting is substantially financial.
D3 · automation substitution weight 0.08
Capacity to substitute capital for labour: installed robot density, industrial automation base, compute and AI position. The earlier model named this as a structural falsifier and then did not score it — an acknowledged hole left open. It matters most exactly where index B is weakest, because aging economies have both the strongest incentive to automate and the deepest installed base. Japan scores 5.0.
D4 · institutional continuity weight 0.06
Probability the governing arrangement that produces these capabilities still exists in 2050. This closes the second inconsistency. C3 scores veto-player density — friction within a stable system — and the earlier model gave China and the Gulf near-perfect marks for having no veto players. But "no veto players" and "no institutional check" are the same property: it is a strength for building and a risk for persisting. The upside was scored and the downside was not.
Double-counting, resolved by redefinition rather than ignored
Two of the additions overlapped indices that already existed. Rather than charge the same thing twice, the originals were narrowed and reweighted:
C2 was "fiscal and capital-market room." Project capital now lives in D2, so C2 is narrowed to sovereign fiscal room only. The United States is re-scored from 3.5 to 3.0 accordingly — its private capital depth is counted in D2, not in both.
B's weight falls from 0.35 to 0.20, because D3 measures the substitutability of exactly the labour B counts. Combined labour weight is 0.20 + 0.08 = 0.28, now decomposed into supply and substitutability instead of proxied by supply alone. C falls 0.25 → 0.16 for the same reason; combined execution-and-capital weight is 0.26, close to the original 0.25.
What is not solved: an additive model cannot represent the interaction. High D3 should reduce how much B matters, not merely sit beside it — a region that can automate is less exposed to its own demographic decline, multiplicatively. Japan is the case where this bites hardest, and it means Japan's score here is if anything conservative. Stated, not fixed.
Threshold derivation
v1 set the class boundaries by hand at 3.50 / 3.25. Integrating climate lowers every composite, so reusing those numbers would measure a new distribution with an old ruler. v2 derives them.
The rule was stated before the result was inspected, and it is mechanical:
- Rank the composites and compute every adjacent gap.
- Discard any gap that merely isolates a single top or bottom outlier — such a gap describes one region's distance from the field, not a boundary between two classes.
- Take the largest remaining gap as the structural break.
- Centre the Borderline band on that gap's midpoint, with width equal to half the gap — so the band is proportionate to how cleanly the field actually separates.
| Adjacent pair | Gap | Midpoint | Status |
|---|
What deriving buys, and what it costs
It buys the difference between a finding and an artefact. Applying climate penalties while holding a hand-set boundary would have shown the Lead Bloc collapsing on climate alone — mostly the ruler moving, not the world. What it costs is stability: a derived boundary relocates when the distribution changes, and in v3 it does, taking the United States out of the Lead Bloc at light climate severity. That is the honest trade, and both halves are reported — the version decomposition separates what climate did from what the added dimensions did.
Two consequences of deriving rather than fixing the boundaries. First, they are not comparable across model versions — v3's 3.35/3.16 and v1's hand-set 3.50/3.25 are different rulers, so composites should only ever be compared within a version. Second, the boundary can relocate when the distribution changes, and in v3 it does: at light climate severity the break moves to Gulf / United States and the US falls out of the Lead Bloc. That makes the US the marginal member, and it is reported as such in the severity band rather than smoothed over.
The hardware firewall
How the sensitivity test is implemented, and why it cannot manufacture equipment.
Constraint held absolutely
The 50% compression applies only to bureaucratic, permitting and approval timelines. It does not apply to physical procurement. Transformers, switchgear, gas turbines, HVDC converters and long-lead grid hardware run on multi-year global supply chains that are indifferent to any region's bureaucratic speed. Where a region's bottleneck is hardware rather than paperwork, the model is required to show no material effect and to say so.
This is enforced by two parameters per region, not by narrative assertion:
Parameter 1 · paperwork share
The fraction of the region's binding delay attributable to bureaucracy rather than procurement, EPC labour or finance. It caps the C1 gain:
C1′ = C1 + 0.5 × (5 − C1) × paperwork_share
Parameter 2 · idle-firm unlock
Index-A points recoverable by approval speed alone — firm capacity that is already built or already ordered and is waiting on a signature. Hardware that has not been procured cannot appear here at all:
ΔA = 0.5 × idle_firm_unlock
The second channel exists because the firewall's own logic runs in both directions. Paperwork cannot create hardware — but where the hardware already exists and is idle, paperwork is the entire binding constraint. Japan is the clean case: roughly 19 constructed, paid-for reactors are not operating, and the gap is review time and local consent, with zero procurement content. Ignoring that channel would have understated the one region where the tested intervention genuinely bites.
Is the null result earned or baked in?
A sensitivity test that cannot flip anything tells you nothing. So the mechanic's ceiling was back-solved. A hypothetical region with the worst possible paperwork score, a wholly bureaucratic bottleneck, and a large idle firm fleet (C1 = 1.0, share = 1.00, unlock = 2.0) would move +0.372 — still comfortably wider than the 0.190 Borderline band. The test can flip a class; no region in this set has the profile to use it. Note the ceiling fell from +0.575 in v2: adding four dimensions cut A's and C's weights, which mechanically shrinks how far any A/C-channel intervention can move a composite. The margin over the band narrowed but did not close.
Region rationale
Ordered by baseline composite. Each card opens to the three index justifications, the four C sub-scores, and the bottleneck attribution that drives the sensitivity result.
Sensitivity test
Run on the climate-scored base. Baseline class → sensitivity class → flip or no flip → why, with the flip requirement back-solved for every region that has one.
| Region | Paper share | Idle-firm unlock |
Δ actual | Δ needed to flip |
Baseline → sensitivity | What a flip would require |
|---|
Japan — the instructive case
Largest move in the table (+0.176) because it has the highest paperwork share and by far the largest idle-firm unlock. It receives the maximum benefit the tested intervention can deliver, and needs +0.326 to reach even Borderline. Approval reform is genuinely Japan's highest-leverage lever on firm power; it is not a lever on 2050 at all, because B = 1.0 carries 0.35 of the weight and demography is locked.
United Kingdom — reform aimed correctly, arriving small
Second-largest move (+0.103). The connection queue and NSIP consenting genuinely were the primary sequencing constraint, and the 2025 reforms target exactly that — this is the model's most accurate reform diagnosis. But the firm-power gap itself is nuclear construction and gas-fleet economics. Needs +0.616 to reach Borderline — further than in v2, because the added dimensions moved the boundary up faster than they moved the UK. Still unreachable by paperwork at any parameter value.
EU — the mistaken-bottleneck case
+0.064, needing +0.722. The EU's constraint is routinely described as permitting. It is HVDC converter and cable supply booked into the 2030s, a missing investment case for dispatchable capacity, and an industrial power price level running 2–3× the US. Veto density scored 1.0 is structural, not procedural — halving approval clocks does not remove a member state's veto or a constitutional court.
India — the new closest call
India gains +0.067 against a +0.087 requirement — 0.020 short of returning to Borderline, the narrowest miss in the table. That is worth reading carefully: permitting reform very nearly recovers what India lost, but the loss was never about permitting. India fell out of the Lead Bloc on cost of capital (D2 = 2.0) and automation capacity (D3 = 1.5), and no approval-timeline cut touches either. The near-miss is a coincidence of magnitudes, not a mechanism. Southeast Asia, v2's closest call, is now +0.404 away and no longer near any boundary.
Explicit no-effect findings, as required
China: exactly +0.000. C1 already sits at the 5.0 ceiling and there is no approval slack to compress. The model's control case.
Southeast Asia: +0.084, no class change. v2's marginal 0.004 boundary crossing is gone — the added dimensions moved both its composite and the threshold, and it now sits 0.404 from the nearest boundary. The v2 "flip" was threshold placement, not substance, which is why it did not survive a change to the index set.
Gulf States: +0.002. State-priority approvals are already effectively instantaneous. The bottleneck is gas turbines and EPC labour, and no signature accelerates either.
United States: +0.073, no class change. Permitting drag on transmission is real, but the acute 2026–2032 constraint is turbine order books sold out to roughly 2030–32, large-power-transformer lead times of 3–5 years, HV cable, and skilled labour. Halving NEPA does not conjure a turbine.
European Union: +0.064, no class change. Hardware, price and political structure, as above.
And a limit underneath all of the above. Even under the sensitivity case, Southeast Asia's firm power reaches only 2.60, the UK's 2.65 and the EU's 2.25, all below the 3.00 A-floor. Under v3's higher thresholds the composite binds first, so the floor is not the operative constraint here — but it means that even if paperwork somehow carried their composites over the line, three of the five Strained Adopters would still be barred from Lead Bloc on firm power alone. Two independent reasons, not one.
Weight robustness
Does the classification survive being reweighted? Thresholds held fixed at 3.50 / 3.25 throughout.
The Lead / Strained split is stable across all five weightings. Only three cells move, and every one moves down: the United States falls to Borderline under execution-first and under equal weighting — which is itself the finding about the US, that it leads on inputs and lags on conversion — and China falls to Borderline under demography-first, which is the finding about China. No region ever crosses from Strained to Lead under any weighting. All four Strained Adopters are Strained under every scheme tested, and the four Lead Bloc members are never displaced by anything but their own known weakness.
Climate severity band
The climate components are the softest numbers in the model. This is the test of whether that softness matters — thresholds are re-derived at each severity, so the ruler moves with the distribution.
Severity scales every region's climate component together, from 0.5× (climate is mostly already priced into the raw scores) to 1.5× (it is substantially under-priced). The headline case is 1.0×. Because the thresholds are derived rather than fixed, they move too — which is the point: the question is not whether composites fall, it is whether the classification is an artefact of one severity choice.
| Region | A adjusted | Composite | Class at this severity | Note |
|---|
A stable core and one marginal member
This is where v3 is less robust than v2, and the honest reading has to say so. China and the Gulf are Lead Bloc at every severity — that is the stable core. The United States is not: at 0.5× severity the structural break relocates to Gulf / US, lifting the Lead threshold to 3.59, and the US falls out. Its membership therefore depends on the climate-severity assumption. v2's headline was that membership held across the whole band; adding four dimensions tightened the distribution enough that it no longer does.
The A-floor is untested again
In v2 the A ≥ 3.00 firm-power floor finally bound: at light severity Southeast Asia's composite cleared the Lead threshold while its firm power did not. Under v3's thresholds it fires nowhere — the seven-index distribution is tighter and the derived boundaries sit higher, so no region is ever in the position of clearing on composite while failing on power. The rule is back to being a stated principle that no case exercises. That is a mild weakness, listed as one: a rule that never fires is a rule that has not been checked.
Version decomposition
Three model versions, two deltas, one table. It isolates what climate did from what the four audit dimensions did — which turn out to be different things acting on different regions.
Each step holds everything else constant, so the two deltas are separable rather than tangled:
v1 · climate-blind
A, B, C at 40/35/25, no climate component anywhere. Thresholds set by hand at 3.50 / 3.25.
v2 · climate scored
Same three indices and same 40/35/25 weighting, with committed climate stress inside A, B and C2. The first delta isolates climate alone.
v3 · audit dimensions added
Four new indices, with A/B/C narrowed and reweighted to avoid double-counting. Thresholds derived at 3.35 / 3.16. The second delta isolates the new dimensions alone.
| Region | v1 climate-blind | Δ climate | v2 climate-scored | Δ new dims | v3 final |
v1 class ruler 3.50/3.25 |
v3 class ruler 3.35/3.16 |
|---|
Composites are only comparable within a column. v1's thresholds were hand-set and v3's are derived, so the two rulers measure different distributions — the classes either side are each correct against their own boundary, and the point of the table is the deltas, not a like-for-like composite race.
What climate did
Lowered everything, and resolved v1's one Borderline call downward: Southeast Asia to Strained Adopter. Its exposure is among the worst in the set across all three channels. Climate changed the level of every score and one classification.
What the new dimensions did
Reordered the table. Japan +0.748, the largest revision in the model's history, moving it from last-but-one to strongest Strained Adopter. India −0.493, out of the Lead Bloc entirely. Gulf −0.526, losing first place to China. Climate changed levels; these changed ranks.
What survived both
The EU and the UK are Strained Adopters in all three versions, by every ruler, at every climate severity, under every weighting tested. No revision has ever moved them. That is the single most robust finding in the model — and neither of them is short of paperwork reform.
Ranked by how much each actually moves the answer: what a region can source and finance first, committed climate stress second, bureaucratic speed a distant last. The intervention the original brief singled out for testing is the one that turns out to matter least.
Known limitations
Stated plainly, because the ranking is misleading if these are not read alongside it.
Mass is not scored
The brief defines three indices, and none of them is scale. So the model measures capability and rate, not weight — which is why the Gulf States rank first on a population of roughly 60 million. The Gulf leads per unit and cannot anchor a bloc at 0.7% of world population. Population, electricity consumption and GDP are carried in the table as annotation and excluded from the composite by design, not by oversight. Adding a mass term would be a legitimate v2; it would reorder the Lead Bloc without changing its membership.
Three of eight "regions" are aggregates
The EU, Southeast Asia and the Gulf are composites of members that diverge sharply — Vietnam and Thailand have opposite demographics; France and Germany have opposite firm-power positions; Singapore and the Philippines are barely comparable. This is in real tension with the model's own premise that divergence is regional rather than global: the same logic that separates the eight would separate several of them internally. Southeast Asia's Borderline placement is the least meaningful cell in the table for this reason.
Index C is the noisiest
A and B rest on published projections and physical inventories. C rests on judgement about state capacity, and its four sub-scores are the least reproducible numbers here. It is weighted lowest partly for that reason. Two of the four — veto density and delivery record — would benefit from being built from a coded case list rather than assessed.
Indicator vintage
Figures are drawn to a knowledge cutoff of May 2026 and are approximate — order-of-magnitude reliable, not audit-grade. Several are moving fast, particularly US net migration, gas turbine order books, and UK connection-queue reform. Re-check the specific numbers against primary sources before this circulates outside the group.
Single-branch conditioning
Everything here is conditioned on the bifurcation branch at p=0.55. The model says nothing about the other 45% of the distribution, and would not survive the systemic-disruption branch at all — that branch changes which variables matter, not merely their values.
The climate components are judgements
A, B and C rest on published projections and physical inventories. The climate components rest on an estimate of how much stress the raw scores already priced, which is the softest number in the model and cannot be read off any dataset. The severity band is the honest response — the classification holds from 0.5× to 1.5× — but a genuinely rigorous version would build each component from a coded indicator (cooling degree days, basin-level hydro variance, ILO heat-hour loss, insured-loss ratios) rather than assessing it.
Derived thresholds move with the data
Deriving boundaries from the distribution is more defensible than fixing them by hand, but it means the ruler is not stable across model versions — v3's derived 3.35/3.16 are not comparable to v1's hand-set 3.50/3.25 as absolute capability bars. Composites should be compared within a version, never across. Where an absolute reading matters, the climate-blind comparison reports both rulers explicitly.
The A-floor never binds
Stated in advance and it constrains nobody in the event. That is a mildly weak feature: a rule that never fires is untested. It would fire on a hypothetical region scoring strongly on demography and execution while failing on power — closer to how the UK or Spain might look under a different A assessment.
What evidence by 2030 would prove this model wrong
Each tripwire is numeric and observable before 2031. If none of them trip, the model has survived; if the ones marked structural trip, the model is not adjusted but discarded.
The OECD firm-power constraint turns out to be soluble on this timescale
The model's largest claim is that firm power is the binding constraint and cannot be fixed quickly. Hardware clearing faster than the supply chain implies would falsify the weight on A, not just the scores.
Europe reverses its dispatchable position
EU and UK A-scores of 2.5 assume the dispatchable investment case stays broken and industrial power prices stay punitive.
China's build rate breaks, in either direction
A = 5.0 for China is the model's anchor point; the entire scale is calibrated to it.
Migration policy reopens at scale
B is the most locked-in index, so its falsifier is not fertility — nothing observable by 2030 changes 2050 fertility outcomes much. It is migration policy, which is fast-moving and is doing all the work in the US, UK and EU scores.
India converts headcount into labour force faster than assumed
India scores B = 5.0 on headcount, with absorption flagged as the real question. If absorption improves, the score is understated in effect rather than wrong.
The UK's permitting reform actually lands
This is where I would expect the model to break first, because the UK is the one democracy here with low veto density, an executive that can act, and reform aimed at the correct target. The model says it will not be enough. That is a real prediction.
A veto-dense system delivers anyway
EU C3 = 1.0 and US C3 = 1.5 are the strongest structural claims in index C.
The climate overlay is mis-calibrated
The overlay's penalties are marginal estimates of what v1 failed to price, and that judgement is the softest number in the whole model. It is also testable well before 2050, because the mechanisms are already firing.
Supply-chain concentration unwinds
D1 carries the largest of the new weights and it is what puts China first. It is also the dimension most exposed to deliberate policy reversal, because every other region is actively trying to break it.
Cost of capital converges
D2 is what removes India from the Lead Bloc. If emerging-market risk premia compress, that removal reverses.
Automation substitutes for working-age population
The deepest falsifier, because it does not adjust a score — it makes index B the wrong variable and reallocates 0.35 of the weight. It would hit asymmetrically, and in the direction that reverses this model's result: Japan, Germany and Korea are the densest industrial-robot installers on earth precisely because they are aging. The model treats a shrinking labour force as a hard constraint. If capital substitutes for it cheaply, the three Strained Adopters are the regions most incentivised to do it and best positioned to.
The bifurcation premise itself fails
The model is conditioned on regional divergence. If the eight regions are converging rather than diverging, the scores may all be individually defensible while the framework is wrong — and weight should shift from the 55% bifurcation branch toward the 15% managed-acceleration branch.
The classification boundary is in the wrong place
Threshold placement is now derived from a 0.367 structural break rather than set by hand, which removes the arbitrariness but introduces a different fragility: the break can relocate. It already does — at light climate severity it moves to Gulf / United States and the US leaves the Lead Bloc.
Preserved prior
Carried forward unchanged as a fixed input. Not re-litigated here.
This model scores only the 55% branch. Background assumptions taken as fixed and not open for debate: demographics to 2050 are substantially locked in; climate stress to 2050 is substantially committed; the world is regionally divergent rather than following one global trajectory.