The most quoted figure in technology geopolitics is a count of published papers. It is accurately produced, its methodology is stated plainly by the people who built it, and for 3 years it has been read as a measure of capability rather than as a measure of what capability publishes.

Nine economies sit on that map, not two, and the instrument reading them sees 4 of the 9 badly — each dark for a different reason. What sorts the map is not who leads. It is which positions are real, which are shadows cast by a metric that counts only what is disclosed, and which have already crossed out of the research record and into the physical economy, where this publication normally finds them from the other side.

Two things happened while this edition was being written. The monopoly inventory, which had never contained a position held by anyone but China, took its first crack. And a Chinese laboratory shipped a frontier model into the open, in the field where the argument for hidden Western capability was strongest.

Where the next 10 years live has an answer. It is not the one the headline number gives, and it is not the one its critics give either.

China led 69 of 74 critical technologies in the March 2026 refresh of ASPI's Critical Technology Tracker. In the June refresh it remains unchanged.

The number not moving is interesting in itself. A figure that stops climbing because it has run out of room to climb is not a moving front but a settled position, and the two behave differently under policy. A front can be halted. A settled position has to be dislodged, and nothing announced in the past year is scaled to dislodge anything. The 20 year transition this publication traced back then — 60 of 64 down to 5 of 74, crossover around 2016, the strongholds tipping one by one to compounding Chinese output rather than to American collapse — has arrived somewhere and stayed there. That trajectory is no longer the story, because the transition was never a duel.

Outright leadership, though, is the rarest thing in the Tracker. China's 69 and the United States' 5 exhaust all 74 — every tracked technology has a Chinese or an American leader, and no third country holds a single one. What the rest of the field has is placement: top 5 presences, second place finishes, and strategic niches.

India is the Tracker's standout riser: top 5 in 56 of 74, second in 16 of the categories China leads, third in the world by raw output at roughly 180,000 papers a year — a volume China itself passed around 2011. It holds no monopoly risk position and funds research at 0.64% of GDP — against China's 2.43%, America's 3.4%, Korea's 4.8%. More tellingly, it cannot spend even that: its Anusandhan National Research Foundation spent none of its budget in FY2023-24 and none in FY2024-25, then 61% of Rs 2,000 crore in 2025-26, and only 15% of the previous year's wider R&D allocation was utilised. The 2026-27 Budget has raised the allocation to Rs 20,000 crore regardless. India's own Chief Economic Adviser has diagnosed the cause as structural and cultural rather than financial — a protected domestic market that never forced its firms to need frontier science. That is a diagnosis no cheque resolves.

Korea has climbed to 36 top 5 placements on the highest national R&D intensity in the world, above 5% of GDP, and has committed a 19.3% increase in government research funding for 2026 with 28.6% of it directed at AI. Japan has fallen to 3 from around 10, and its own national survey now ranks it 16th in international basic research standing, down from 1st 15 years ago; Japanese researchers report spending roughly 10% of their working time on research. The United Kingdom, at 40 and falling from 48, is the steepest decliner among established powers. Germany, Europe's strongest, sits at 27 and slipping. France has collapsed to 2.

Summed as a bloc, the European Union is a top 5 presence in nearly all 74 and second in around 30, and on that aggregated basis it leads 3, down from 4. The qualifier is doing more work than the number: those 3 are technologies China or the United States leads country by country, and the bloc figure is the same 74 re-run on a different unit rather than a third slice of the same pie. No European country leads anything. Which is the whole European problem in a line — the bloc is a research power only in aggregate, and no actor corresponds to the aggregate.

At the margins sit two countries with 8 top 5 placements each, for opposite reasons. Neither leads a technology or holds a monopoly risk position. Iran's 8 are what sanctions selected for — defence and dual use, ahead of the United States in smart materials, fourth in drones and swarming, third globally in air-independent submarine propulsion, a capability ASPI itself notes would matter in the Strait of Hormuz. Australia's 8 are allied relevant and alliance sustained, and Australia built the Tracker.

Then there is Taiwan, which makes over 90% of the world's most advanced logic chips and does not appear in the country rankings at all. TSMC ranks 20th in the citation record for the category it physically owns.

Nine economies; seven structurally different research models, none of which can be transplanted into another. Distributed American pluralism cannot reproduce Chinese state direction. Fragmented European policy cannot reproduce American private R&D scale at $700 billion a year. None of the large economies can reproduce the applied node concentration that ties Korean, Taiwanese and Japanese research directly to a production line. Each has to compete on its own structure, which is why the policy conversation — conducted almost entirely in the currency of money — keeps missing.

But the deeper problem is not that the field is misread. It is that the instrument reading it cannot see 4 of the 9 properly, and a table in which the maker of the world's most advanced semiconductors is absent is not a table that has finished telling you something.

The Critical Technology Tracker measures one thing, and measures it precisely: each country's share of the top 10% most cited published papers in a given technology, taken as a lead indicator of capability. The methodology is sound, the citation data comes from the same Web of Science and Scopus sources Western institutions use, and across most of the map it does what it claims to do.

The word carrying the weight is published.

Before that observation is allowed to do any work, the objection to it has to be put in its strongest form, because it is a serious one. An argument that a declining power's real capability is hidden in classified vaults, corporate laboratories and unpublished process knowledge is, on its face, indistinguishable from cope. It is the argument every fading incumbent makes. Worse, it’s a classic trap: if you can’t see the capability, it’s used as 'proof' that it’s successfully hiding.

Two things rescue it from that. One of the four blind spots is public, quantified and published annually by patent offices. And the argument has since been tested against a live event, which is where it will end.

The four blind spots are these. Classified and military research never enters open literature at all. Corporate research is deliberately withheld, and since roughly 2022 that has included the methods work of the leading American AI laboratories, who largely stopped publishing what matters most. Patent route invention is protected as property rather than disclosed as science. And tacit process knowledge, the kind embodied in a fabrication line rather than a paper, was never written down in the first place.

What makes this structural rather than incidental is that the blindness is not randomly distributed. It falls, with something close to precision, on the economies that concentrate their most strategic work outside the open literature.

Sort the nine on that axis and they split cleanly. The Tracker reads accurately the open publishers: China, whose prestige and promotion incentives push even strategically sensitive research into open journals; India and Iran, whose output is academic and whose citation position matches their reach; Korea; and Australia, which is open by disposition and, reflexively, built the instrument. It under reads the United States, whose national laboratories, intelligence research apparatus and black budget programmes generate decades of capability relevant work producing almost no American attributed citations, under reads Europe, whose applied and industrial invention lives in patents the count does not read, under reads Japan, a former citation superpower whose record collapsed while its industrial capability held, and it does not read Taiwan at all.

The economies the metric sees least are the West and its closed research allies. The economy it sees most completely is China.

Korea is the case that stops this being a convenient story. If the argument were that applied, industrial, production tied research is invisible to citations, then Korea — the most research intensive economy on earth, above 5% of GDP, run through the chaebol and tied directly to the production line — would be invisible. It is not. It is top 5 in 36 of 74 and rising. Korean applied research passes through its universities and into the citation record, and the metric picks it up without difficulty.

So the variable is not applied against basic. It is disclosed against not. Korea publishes and is seen; Taiwan does not and is not.

Which brings it to the patent record, the measurable part.

On raw volume China is overwhelming: roughly 1.8 million applications a year and around 70% of all patent families in the world. On the gold standard measures the picture reverses. Triadic patents — those filed at the American, European and Japanese offices together, the standard filter for inventions their owners think worth defending globally — are led by the United States and Japan, with China lagging. On patents pursued internationally at all, the United States and Japan account for nearly half the global total. And the single number that resolves the volume paradox: only around 3.3% of Chinese patent families are ever filed at more than one office. China files most of the world's patents and internationalises almost none of them.

The sector map lands where the citation record is weakest. Pharmaceutical patents concentrate in the United States, Switzerland, the Netherlands and Israel. Transport technologies belong to France, Germany, Italy and Sweden. Electrical machinery and energy to Germany, Japan and Korea. European and Japanese applied strength is patent visible precisely in the categories where their citations place them second or absent. In semiconductors, TSMC and Samsung sit among the world's leading filers — Taiwan's decoupling from the citation record made measurable in the one register that can see it.

Then, on 16 July, the argument met a test it did not choose.

Moonshot AI released Kimi K3: 2.8 trillion parameters, natively multimodal, a context window of 1 million tokens, and the weights released openly to anyone who wants them. Whatever its position against the best American systems — a ranking worth nobody's confident assertion this early — a Chinese laboratory shipping at that scale is deployed capability, not a citation artefact. And it arrived in natural language processing, the category China took from the United States most recently, and the one where the closed laboratory argument was doing most of its work.

This publication wrote in April that the gap between research lead and deployed capability flatters the American position, and named commercial leadership in AI products as something the Tracker misses. 3 months later that gap is narrower than it was, in the fastest moving of the fields named.

But the more interesting thing K3 does is invert what disclosure means in the argument.

The visibility axis has so far treated openness as a measurement problem — Chinese research is over represented in the count because Chinese incentives push it into journals. K3 was released with open weights, into the same ecosystem, by the same disclosure culture. That is not a laboratory accidentally revealing itself to a metric. It is a distribution strategy: give the artefact away, and the developers, the tooling and the downstream dependencies accrue to you. The American laboratories bought secrecy at the price of the ecosystem. Moonshot bought the ecosystem at the price of secrecy.

Which means Chinese openness is not merely why China scores well on a citation count. It is a competitive instrument, and the citation lead and the deployment strategy are the same behaviour observed twice.

The honest limits follow from that, and they have moved.

The correction still cuts in one direction and still does not do so everywhere. It holds for classified defence work, where the American and allied blind spot is structural and unmeasured. It holds for tacit process knowledge, where Taiwan's absence from the rankings is the clearest case in the dataset. It does nothing whatever to the industrial monopolies: batteries, photovoltaics and critical mineral processing have already converted into commercial dominance, and there the patent record corroborates the citation record rather than offsetting it. And in frontier AI, where the closed laboratory argument was strongest 3 months ago, it now carries considerably less weight than it did. The blind spot is real; the size of what is hiding in it is smaller than the argument assumed.

Of the four blind spots, only one is measurable. Patents can be counted. Classified output and tacit process knowledge cannot, and any statement about their scale — including the ones made above — is inference from structure rather than measurement. That is a real weakness and it should be named as one.

There is a fifth blindness that has nothing to do with disclosure. The Tracker's current window is 2021 to 2025. In 2026, more than 30 Iranian universities were bombed and Sharif University of Technology, the elite engine of the entire Iranian system, was damaged, with power and telecommunications outages crippling what remained. Every Iranian figure in the record describes a research base that no longer exists in the form that produced it. The same lag runs the other way: Korea's 19.3% budget surge, Japan's pledged doubling and the American talent shock are all invisible to a record that is, structurally, a photograph of the recent past.

The strongest evidence that the visibility problem is real does not come from us. ASPI has announced it is developing patent based analysis to complement the Tracker — the instrument moving to see what it was built blind to.

That also supplies the remaining test, and it should be stated before the result is known rather than after. If the patent module lands and shows the patent map broadly tracking the citation map, with China leading quality adjusted invention as it leads citations, then the correction argued here is wrong and the raw count was the right reading throughout.

The Tracker's second measure is stricter than its first, and less discussed. A technology is marked at high monopoly risk when a single country produces more than 3 times the high impact research of its nearest competitor and hosts the bulk of the world's top 10 institutions in the field. This is not a measure of who leads, but one of how alone the leader is.

42 categories of the 74 now carry that marking. China holds 41 of them.

The June refresh deepened the inventory in two places that were chosen carefully, whether or not anyone chose them: AI hardware and accelerators, and biological manufacturing, both upgraded from intermediate to high, joining advanced robotics and autonomous systems operation from the March update. One category moved the other way — brain computer interfaces, still Chinese led, stepped down to intermediate.

And in neuroprosthetics the United States strengthened its lead to high monopoly risk: the first such position ever held by a country other than China.

In April this publication reported that China held every one of the 41 monopoly risk positions and that no other country held any, and that neuroprosthetics was the only tracked field with zero Chinese institutions in the global top 10. Both statements were accurate when they were published and neither is accurate now. The inventory is 41 to 1 rather than 41 to nothing, which barely moves the balance of power and substantially disciplines the claim; an absolute statement has become a proportional one, and the difference between those two is most of what separates analysis from advocacy.

The 41 are not distributed evenly, and grouping them is where the previous section starts to earn its keep. Roughly 15 sit in industrial anchoring: batteries, photovoltaics, advanced materials, critical mineral processing, hydrogen and ammonia. Roughly 10 are defence adjacent: hypersonics, drones and swarming, radar, satellite navigation, advanced robotics, autonomy. Around 9 cover sensing and communications, including photonic and quantum sensors and 5G and 6G. The remaining 7 or so are AI and biotechnology: AI hardware, generative AI, computer vision, biological manufacturing.

Apply the visibility correction to those four groups and it does not apply evenly to any two of them.

The industrial 15 are the group the correction cannot touch, because they have already left the citation record and entered the physical economy. Chinese dominance in batteries, photovoltaics and critical mineral processing is not an inference from publication counts; it is visible in refining capacity, installed manufacturing and export volumes that I have traced repeatedly through the physical layer. Where a research monopoly and a commercial monopoly describe the same thing, the research metric is not overstating anything. It is confirming what the loading docks already show.

The defence adjacent 10 are the group most vulnerable to it, and the vulnerability is mechanical rather than rhetorical. The monopoly rating is a ratio: more than 3 times the nearest competitor. Classified American and allied work produces no citations at all, so in exactly those categories where the West conducts its most serious research behind clearance, the denominator of that ratio is artificially small. The rating does not merely miss Western capability in hypersonics and radar; it converts the missing capability into evidence of Chinese concentration.

Which does not mean the defence positions are illusory. Chinese hypersonic research is genuinely formidable, and the physical evidence — test cadence, deployed systems, shipbuilding throughput — corroborates a great deal of it. The claim is narrower: in this cluster the number overstates the gap by an amount nobody outside a classified compartment can quantify, and the honest position is to hold the direction while declining to trust the magnitude.

The AI and biotechnology 7 have just been tested in public, and they came through in China's favour. AI hardware and accelerators was upgraded to high monopoly risk in June; Kimi K3 shipped in July at 2.8 trillion parameters with open weights. A monopoly rating in the compute layer and a frontier model release in the application layer are the same finding arriving through two different doors, roughly 30 days apart. Of the four clusters, this is the one where the count and the deployed reality currently agree most closely — and it is the cluster the closed laboratory argument was built to defend.

The sensing and communications 9 sit between the two poles and resist a clean verdict, which is itself worth stating. Some of it is commercially confirmed, in the 5G and 6G equipment base; some of it is dual use and partially classified, in quantum and photonic sensing. The cluster is a mixture and there is no principled way to say in what proportion.

The American position deserves the same scepticism applied in the other direction. Neuroprosthetics is a genuine monopoly, resting on NIH funded clinical research networks, DARPA programmes and a concentration of institutional depth that no other country matches. It is also one category out of 74, in a field adjacent to a Chinese led category that stepped down in the same refresh; brain computer interfaces moving from high to intermediate while neuroprosthetics moved from medium to high should make any reader cautious about how much load a single categorical boundary is bearing. The crack in the inventory is real and it is one position wide.

There is a limit that applies to everything above. The monopoly rating is derived from the same publication data as the leadership count, which means it inherits every blind spot named previously rather than providing independent confirmation of anything. It is a second reading of one instrument, not a second instrument. Nothing in the 42 constitutes corroboration of the 69; the two numbers are the same measurement asked a different question.

And 32 of the 74 tracked technologies carry no monopoly risk marking at all, which is the part of the map that gets least attention and may deserve the most. Those are the categories where leadership exists but concentration does not — where more than one country retains real institutional depth, and where the outcome is therefore still open. They are where policy has purchase. The 42 are, for the most part, already decided.

The outbound numbers are real, and they should be put at their strongest before anything is done to them.

Nature's jobs board analysis found American based scientists submitting 32% more applications to positions abroad, and 41% more to positions in Canada. The hardest single metric belongs to the European Research Council: applications from American based researchers for early career grants rose from 60 for the 2024 call to 116 for 2025 to 169 for 2026, while senior researcher applications rose from 23 in 2024 to 114 in 2025 — close to a quintupling at the end of the market where each departure takes a laboratory, a grant history and a training pipeline with it. Surveys of intent run higher still; a Nature poll found more than 75% of American researchers considering leaving, a later Elsevier poll 40%.

That is the case for the brain drain, and on its own terms it is a strong one.

What it does not survive is the distinction between an application and a move. Weighing the option is far easier than executing it, and the receiving ends are capacity constrained in ways that cap the flow regardless of how many people want to make it. One French programme received 300 American applications for 15 funded places. Europe's Choose Europe instrument requires host institutions to co-fund and to offer permanence, which is money the bloc cannot compel its members to match while national research budgets are being cut; Aix-Marseille has recruited around 60 researchers and Vrije Universiteit Brussel 12. India is bidding with research chairs worth up to Rs 14 crore, into a system whose apex research foundation disbursed nothing for 2 consecutive years. Realised departures run in the hundreds to low thousands, not the tens of thousands the intent surveys imply.

The Chinese channel is the one most often cited and the most misread. China is drawing named senior figures home from Harvard and Intel, backed by a new visa for foreign STEM researchers, and the recruitment is aggressive and well funded. But the flow is dozens of prominent people rather than thousands, and it is disproportionately ethnically Chinese — which makes it the recapture of a specific cohort rather than a general victory in an open talent market. A country reclaiming its own diaspora is doing something narrower than winning the competition for the world's researchers, and the distinction gets lost every time the two are reported as one story.

Narrower is not smaller. Yang Zhilin took his bachelor's at Tsinghua and his doctorate at Carnegie Mellon in 2019, co-authoring XLNet and Transformer-XL inside the American system; he returned to Beijing, co-founded Moonshot AI in March 2023, and shipped Kimi K3 in July. The recapture channel, run once, at full scale, produced a frontier laboratory in a field the United States assumed it owned. Dozens is not thousands, and dozens was enough.

The inbound side is where the real damage sits, and it is measured far less often because a magnet that stops attracting produces no departures to count.

Applications to American positions from European and Chinese scientists have fallen steeply. New international student enrolment dropped 17% in autumn 2025. The $100,000 H-1B fee remains in force pending appeal — struck down on 8 June 2026, stayed by the same judge on 12 June, now before the First Circuit with parallel challenges pending — which means the deterrent operates continuously whether or not it ultimately survives. A directive reviving the Justice Department's China Initiative compounds the chilling effect on researchers of Chinese descent, including those with no intention of leaving.

On 18 July a doctoral student in artificial intelligence at Oxford set out the mechanism in public, and the post has since been seen more than 2 million times. Junde Wu had turned down a Stanford offer because Proclamation 10043 barred him from American graduate school; he qualified subsequently for an EB-1A, the extraordinary ability green card, and found himself in a country specific backlog he expects to run another 4 to 5 years. Meanwhile, by his account, 3 or 4 approaches arrive from China in a typical week, offering salary, housing, research funding, and money simply to come and have the conversation. His closing question is the one the inbound data cannot phrase as sharply: why spend the energy trying to enter a country that does not appear to want you, when another is doing everything it can to bring you back.

He is one person, and one testimony is illustration rather than evidence. What makes it worth quoting is that every mechanism in it is independently documented — the proclamation, the country cap, the recruitment intensity — and that he is not a departure at all. He never arrived. He is the inbound collapse with a name attached, and the drain framing has no column for him.

The largest number in the entire talent picture involves no migration whatever. More than 10,000 people holding STEM doctorates had lost or left their jobs through federal workforce cuts by early 2026 — attrition out of science altogether rather than movement to a rival. They are not in Shanghai or Marseille; most are not in research at all. A talent base can be depleted without a single departure abroad, and this one primarily has been.

Which leaves the timing, and it disciplines every claim above including the ones made here.

None of this can appear in the Tracker before roughly 2028 to 2030. A shock to the talent base in 2025 and 2026 propagates into the citation record on the lag at which graduate students become authors and laboratories become productive. The current American citation position is therefore flattered by a talent base whose stress has not yet registered — and, in the same breath, any claim that the data already shows the damage is premature. Nobody arguing either side of this has evidence from the metric, because the metric cannot yet contain it.

The test runs on the same clock. If the American share of top decile output holds through the 2028 to 2030 windows, then the talent shock was absorbed and the outbound signal was noise in a system with more depth than its critics credited. If it declines in the fields where federal funding was cut hardest and international enrolment fell furthest, the mechanism argued here was operating all along and simply had not surfaced. That is 4 years away, which is an uncomfortable thing to say and better than pretending the answer is already visible.

Research position converts into commercial and military position on a lag of 5 to 15 years, and that conversion is the most useful forecasting mechanism available in the open record — more useful than capital expenditure, which reveals intent rather than capability, and more useful than policy announcements, which reveal neither.

Two of China's research monopolies have already made the crossing, and they are no longer forecasts of anything. Chinese firms account for roughly 68.9% of global battery installations and China refines about 91% of the world's rare earths. The research leadership in electrochemistry, materials and mineral processing that the Tracker recorded a decade ago is now visible as loading dock throughput, and this publication has traced that physical position repeatedly from the other direction — through refining capacity, shipping and the chokepoints that gate them. Where a research monopoly and a commercial monopoly describe the same thing, no conversion remains to be predicted.

The interesting cases are the ones mid crossing, and one of them crossed while the ink was drying on the refresh that recorded it.

ASPI upgraded AI hardware and accelerators to high monopoly risk in June. Kimi K3 shipped on 16 July at 2.8 trillion parameters with open weights. Roughly 30 days separated the research finding from the deployed artefact, in a field where the conventional lag would have suggested years. Biological manufacturing was upgraded in the same refresh and has produced no equivalent public event; robotics and autonomous systems, escalated in March, are converting through industrial installation rather than a single release. The clock is not uniform, and software compresses it in ways that batteries and refining capacity cannot.

The counter case has to be priced at the same weight, because the mirror is a tendency rather than a law, and the exceptions are substantial.

Japan is the clearest of them. Its citation record collapsed — 3 top 5 placements against roughly 10, and a self assessment that ranks it 16th in international basic research standing against 1st 15 years ago — while its industrial capability in materials, precision equipment and machinery held, and its triadic patent position held with it. Two decades of citation decline have not produced a corresponding commercial decline. The United States is the second exception, and the one that cuts closest to the argument made here: 5 leads on the open record, alongside deployed AI systems, pharmaceutical output and semiconductor design capability that no citation count would predict from 5. If research position determined commercial position on a fixed schedule, neither of those states could exist.

Both exceptions have the same explanation, and it is the visibility axis arriving for the third time. Japanese and American capability did not vanish from the world when it vanished from the citation record; it moved into patents, into classified compartments, into corporate secrecy and into process knowledge embodied on production lines. The mirror still holds — it just cannot be read off the published record alone in the economies where the published record has gone dark. Which is a considerably weaker claim than the one the raw count invites, and the correct one.

The semiconductor chokepoint is where that reading matters most, because all three of its layers sit in economies the metric misreads. Korean memory research passes into the citation record and is visible; Taiwanese logic capability is almost entirely absent from it, with TSMC ranked 20th in the category it physically owns; Japanese materials strength sits behind a collapsed citation record and an intact patent fortress. The commercial capability in two of those three layers is precisely the part the count cannot see, which means the single most consequential industrial chokepoint on earth is the one where the open record is least useful as a guide to who holds what.

Europe's narrow exception layer maps to the same pattern from the other end: ASML, Airbus and the pharmaceutical cluster are commercial positions of genuine global consequence held by a bloc that leads 3 technologies on aggregate and none at country level, with the deeper transport and machinery strength showing up in patents rather than papers. India's mirror is the reverse and the most honest of the nine — breadth and volume in the research record, breadth and volume in the physical economy, depth in neither, with the same root cause its own Chief Economic Adviser named: a protected domestic market that never forced its firms to need frontier science.

Which leaves the part of the map that gets least attention.

42 of the 74 tracked technologies now carry a monopoly risk marking; 32 do not. Those 32 are categories where leadership exists but concentration has not, where more than one country retains genuine institutional depth, and where the arithmetic is exact rather than estimated — 28 of them Chinese led without monopoly risk, 4 American led without it. They are the categories in which the next decade has not yet been decided, and they are therefore the only ones where policy, funding and talent decisions taken now can still change an outcome. The 42 are, for practical purposes, already resolved; committing resources to reversing them is committing resources to a position that was settled while the commitment was being debated.

The next 10 years live in the 32.

That is a claim with a date on it. If the coming refreshes show monopoly risk positions migrating out of the 42 — a category China holds at high concentration stepping down and staying down, as brain computer interfaces has just done once — then concentration is more reversible than argued here and the settled reading is too strong. If instead the 32 erode into the 42 while the 42 hold, the mechanism described here is operating and the window for policy is closing rather than open. Both are observable within two refresh cycles, which is roughly a year, and neither requires waiting for the talent shock to surface in 2028.

The transition itself is not in doubt and has not been for some time. What remains genuinely uncertain is its size, and the instrument that measures it is currently being rebuilt by the institution that built it — which is the strongest available evidence that the number everyone quotes was never the whole of the thing.