Cold War Science Programs

The claim under test is that the Cold War is the model for how a state mobilises science, and that it worked. Testing it requires separating two questions that the invocation runs together: what the apparatus was, and what it did. On the first, there was no design. Vannevar Bush's Science, the Endless Frontier of July 1945 proposed a single National Research Foundation with divisions for medicine, the natural sciences, national defence and scientific personnel, funded at “approximately $10 million at the outset” rising to about $50 million a year. Congress did not act for five years. What filled the vacuum is what people now admire.

On the second, the record is unusually good, because the money left an audited trail and most of the programme documents have been declassified. That is why this brief carries more Established than most on this map: the budget series are official, the failure costs are GAO's, and the one clean measurement of what secrecy did to innovation is a published identification study. Where the record is genuinely contested — whether the model reproduces, whether conversion was ever evaluated — the brief says so rather than rounding it into the history.

Three arguments deliberately do not appear here. The deadline-and-redundancy argument belongs to the Manhattan Project; the mandate-and-termination argument belongs to Apollo; agency design and the programme-manager mechanism belong to DARPA. This brief owns the apparatus those programmes left behind: the budget shares, the laboratory system, the secrecy instrument, and the priced record of what was cancelled.

Established Start with the number that kills the framing as usually stated. From NSF's Science & Engineering Indicators (NSB-2025-7): total US R&D was 2.53% of GDP in 1960 and 2.62% in 1968, the Cold War peak. In 2023 it was 3.39%, and it has been above 3.0% every year since 2019. The United States is not spending a smaller fraction of its economy on research than it did at the height of the Cold War. It is spending a substantially larger one. Established What collapsed is the federal share of R&D funding: 62.9% in 1957, 66.5% at its 1963 peak, 25.1% in 2000, 18.3% in 2023, with business now performing 75% of all US R&D. Frontier So “we used to mobilise and now we don't” is a claim about who pays and who directs, not about how much research happens — two different arguments with two different remedies, and the second one is false.

Established The composition series carries a break that any long comparison has to declare. NSF's annotated history of federal R&D obligations FY1951–2024 tracks DOD, NASA and NIH as the three primary categories and warns that development obligations were redefined from FY2016, so pre-2016 totals “are not directly comparable” with later ones. Established A chart running 1951 to today through that discontinuity without a note is showing a level shift as though it were a trend.

Established The one durable institutional artefact is the laboratory system, and its interesting feature is a contracting form rather than a scientific one. DOE currently lists 17 national laboratories — nine under the Office of Science, three under NNSA, four others — with founding dates spanning Lawrence Berkeley (1931), Los Alamos and Sandia (1943–45), Brookhaven (1947) and Livermore (1952). Almost all are government-owned, contractor-operated: the state owns the plant and a university or consortium runs it, with NETL the stated exception. Frontier That is the mechanism by which the state acquired permanent large-instrument capability without acquiring a permanent civil-service research workforce, and it is the part of the apparatus least often meant when people say “mobilisation.” Established DOE's own description defines the system by scale rather than by subject — work that “extend[s] beyond the scope of academic and industrial institutions,” addressing “large scale, complex research and development challenges.” The labs are not a field. They are the only place a $2bn instrument can live.

Established The Cold War's distinctive research instrument was not money. It was secrecy, and its cost has now been measured. Daniel Gross's study of the wartime patent secrecy programme finds roughly 11,200 patent applications placed under secrecy orders 1940–45 — “roughly 1 in every 25 patents filed” in the period. Secret patents filed in 1940 were significantly less likely to receive any future citations; among later filings, patents held secret for shorter periods were about 15% more likely to be cited than those held longer. Established The effect concentrated on the extensive margin — whether a patent got any citations at all — and fell hardest on small firms without government contracts. Affected firms reduced patenting in treated technology classes through at least 1960 and shifted the direction of their research away from them, with Du Pont's product catalogues showing restricted commercialisation. Established Gross's own summary is that secrecy “interfered more broadly with the functioning of the innovation system by restricting commercialization and impeding follow-on innovation” while being effective at its stated job. Established And the instrument did not end with the war: on USPTO data obtained under FOIA, 6,543 secrecy orders were in force at the end of FY2025. Frontier The honest reading is that secrecy works as secrecy and costs as innovation policy, with the cost falling on firms outside the defence contracting system — a line the advocates of mobilisation never enter on the ledger.

Established Did defence money crowd private research in or out? Two independent studies say in, and both say the magnitude is small. Moretti, Steinwender and Van Reenen find that a 10% increase in government-financed R&D generates a 5–6% additional increase in privately funded R&D — preferred OECD elasticity 0.518, French firm-level 0.119. Frontier They also estimate that a permanent one-percentage-point rise in the predicted defence-R&D-to-value-added ratio is associated with a 0.08 percentage-point rise in annual TFP growth, about 8.3% of baseline. Established The scale check is the part that matters and is almost never quoted: applying their own model to the entire post-9/11 defence R&D increase yields 0.005 percentage points of extra annual TFP growth. Real, positive, and tiny.

Established The strongest pro-mobilisation evidence carries its own refutation of the strongest version of the claim. Gross and Sampat's study of wartime R&D records OSRD running 2,254 contracts with 461 distinct contractors, over US$(2022) 9 billion, producing 3,137 identified patented inventions; at the wartime peak government funding touched nearly one in every eight US patents, “more than five times prewar and modern levels.” Doubling OSRD treatment intensity in a cluster is associated with 20% higher patenting by 1960 and 30% by 1970, top-quartile treated clusters 60% higher by 1970, with effects persisting “until at least the 1970s.” Established And the authors' own qualification, which the invocation always drops: the effects were “entirely driven by counties that were already among the most inventive” in the 1930s. Frontier Mobilisation amplified existing innovation centres rather than creating new ones — which is precisely the thing the mobilisation story promises and this evidence does not deliver.

Established The failure record is the best-documented part of the apparatus and the least cited, and the audited numbers are large. On the Superconducting Super Collider, terminated in October 1993, GAO's post-mortem records that “during fiscal years 1988–93, the United States expended about $1.6 billion” on construction, with approximately $2 billion spent in total before the termination vote; about 15 miles of tunnel and shafts excavated; Texas contributing $279 million plus land and services and claiming a $539 million refund DOE had not agreed; and DOE's March 1994 termination-cost estimate of $568 million excluding any Texas settlement against roughly $735 million available. A separate GAO testimony puts the total cost to completion above $11 billion. Established Why it died is the transferable part. The American Institute of Physics' own bulletin records Rep. Sherwood Boehlert conceding it was “good science” but “simply not affordable science”; 81 of 113 freshman representatives voted against; and the floor arguments were the deficit, the absence of foreign co-funding, and management, with Rep. John Dingell calling it “among the worst projects we have seen in terms of contract mismanagement.” Established Michael Riordan's account in Physics Today adds what an outsider would not have said: a succession of “acting or ineffectual project managers” until 1992, no cost-and-schedule control system of the kind already mandatory on defence projects, imported military-industrial contractors whose culture deterred physicists from relocating — and, decisively, the physics community split in public, with Philip Anderson and Theodore Geballe attacking the overruns in 1989–91 and undermining the congressional coalition. That is the physics community's own magazine publishing a physicist's account of physicists helping to kill the largest physics project in American history, and it is weighted up for exactly that reason.

Established Two more cancellations complete the pattern. Aircraft Nuclear Propulsion ran from 1946 to 1961 at nearly $1 billion — roughly $12bn today — and produced the NB-36H, which flew 47 test flights between September 1955 and March 1957 carrying a reactor that never powered it. Kennedy cancelled it holding that “the possibility of achieving a militarily useful aircraft in the foreseeable future is still very remote,” with GAO finding the programme “at least four years away” from flight on a reactor-engine combination. Established The National Aero-Space Plane absorbed roughly $1.7 billion appropriated FY1986–93 for Phase II plus about $450 million in NASA personnel, facilities and utilities FY1986–94, against a cost-to-build estimate that rose from $3.1 billion in 1986 to $13.2–15.1 billion by January 1992, with total programme cost potentially $17 billion. Its flight-test and ground-test argument belongs to Frontier Aerospace Programs; the cancellation arithmetic is here. Frontier None of the three failed on scientific consensus. All three had it. They failed on cost growth, on the absence of a downstream customer willing to pay, and on political durability across budget cycles.

Established The model has been reproduced in peacetime once, and GAO audited it. Operation Warp Speed obligated about $13 billion across DOD and HHS by 31 December 2020, of which roughly $8.8 billion ran through other transaction agreements — outside normal procurement. It supported six vaccine candidates across three platform technologies in parallel. The named acceleration mechanisms were starting large-scale manufacturing during clinical trials, combining or running trial phases concurrently, and relying on prior platform data rather than repeating studies. Established The binding constraints GAO found were physical rather than scientific: facility capacity, supply chains where lead times moved from one week to “4 to 12 weeks,” and workforce gaps met with expedited visas and DOD personnel. As of 31 January 2021, 63.7 million doses had been released — about 32% of the 200 million contracted for delivery by 31 March. Frontier Warp Speed satisfied the Manhattan precondition set exactly — one well-posed objective, elastic budget, a hard external deadline, and prior knowledge that the thing was physically possible — which makes it evidence for the preconditions rather than for the model's generality.

Frontier H1, the strong mobilisation thesis: Cold War state direction produced the postwar technological order and could be re-run. This is the working assumption of most new-industrial-policy advocacy and of the ARPA-replication movement. For it: Gross and Sampat's persistent regional patenting effects; Moretti et al.'s crowding-in. Against it: those effects were concentrated where invention already was, the TFP magnitudes are small, and the federal share fell for thirty years without total research effort falling. Genuinely contested, and the contest is about magnitude rather than sign.

Frontier H2, the accident thesis: what we call the model is the residue of Bush losing, and its virtues are unintended. NSF's own 75th-anniversary volume concedes the sequence in its foreword — the five-year gap “gave other federal agencies — notably the Atomic Energy Commission, the National Institutes of Health, and the Office of Naval Research — the opportunity to expand their portfolios,” and therefore “the new NSF did not become the funder of all basic research across the federal government that Bush had envisioned.” NSF's first grant was $10,300. Established That is the agency conceding its own founding defeat, and nobody in the policy literature defends the apparatus as designed, because nobody designed it.

Frontier H3, the secrecy-cost thesis: classification was the period's characteristic instrument and it was net-negative for innovation. Gross supplies the estimates. The thesis is contested only in the sense that nobody has measured the security benefit on the same scale — there is no counterpart study pricing what compartmentalisation bought. Speculative H4, the classification-as-cover thesis, which runs the other way: secrecy is what makes a high-risk portfolio politically survivable, because the failures are invisible. It is argued in the DARPA literature and has no measurement behind it at all; it is also an argument about institutional tolerance rather than about research productivity, which is a different dependent variable.

Frontier H5, the reproduction thesis: Warp Speed shows the model still works in peacetime. GAO's audit supports the components. The limit is structural rather than evidential: Warp Speed met the precondition set exactly, so it is evidence about preconditions. Speculative H6, the squandered-peace-dividend thesis: post-1991 conversion failed and the capability was lost. Directionally supported by the federal-share series and by the weapons laboratories' re-missioning into stockpile stewardship — but this research pass could not locate a measured evaluation of any defence-conversion programme, and that absence is reported as a finding rather than filled with a judgement.

Speculative H7, the residual anti-framing position: the Cold War “worked” only in the sense that a very rich state buying a great deal of anything gets a great deal of it, and the interesting variance is in what was bought. Nobody holds this as a named position; it is what remains after H1's qualifications are applied, and it is stated here because leaving it out would make the hypothesis space look tidier than it is. Frontier The testable version is whether the composition of federal purchasing — weapons, health, space — predicts downstream civilian capability better than the level does. Nobody has run it.

Frontier The live question the whole space turns on: is the federal share a policy variable or an accounting artefact? A denominator that grows because business research grew tenfold produces a falling federal share with no federal retreat at all. Established Federal R&D obligations in constant terms and the federal share of a growing total are different series answering different questions, and the mobilisation argument is almost always made with the second while meaning the first.

Established The FY2016 development redefinition is a real break in the only long federal series available, and NSF flags it: obligations before and after “are not directly comparable.” Any argument resting on the shape of federal R&D since 1951 has to say what it did about that discontinuity, and almost none do.

Established Classification is a measurement bottleneck as well as a policy instrument. With 6,543 secrecy orders still in force at the end of FY2025, an unknown quantity of federally funded invention is invisible to exactly the patent-based measures the field uses to evaluate federal funding. Frontier The direction of the resulting bias is not obvious — suppressed patents understate output, while the concentration of secrecy in defence classes may overstate the civilian yield of what is visible — and nobody has bounded it.

Established The identification problem is severe and the field is honest about it. Gross's secrecy design works because the secrecy orders were assigned on a technology-class rule that plausibly separates treated from untreated filings. The macro literature has no equivalent. Frontier Moretti et al.'s instrument is the differential exposure of countries and industries to defence procurement shocks, which is credible for the OECD panel and thin for any single-country claim; the elasticities differ by more than a factor of four between the OECD and French firm-level estimates, and the paper does not resolve which is the transferable one.

Established The largest bottleneck is that the counterfactual has never been constructed for the period as a whole. There is a measured effect of wartime OSRD on postwar patenting; there is a measured effect of secrecy; there is an estimated elasticity of private on public R&D. There is no estimate of what American science would have looked like without the Cold War apparatus, and the invocations all assume one. Speculative A synthetic-control design over the OECD around 1950 is conceivable and nobody has published it.

Established And the conversion question has no evaluation literature at all. The weapons laboratories found a substitute mission in stockpile stewardship — certifying warheads without nuclear testing through simulation, large facilities and subcritical experiments. Lawrence Livermore's own page describes ensuring “the safety, security, and effectiveness of the U.S. nuclear deterrent without testing” and improving “the predictability of physics and engineering simulation codes.” Frontier Note what the laboratory's public page does not claim: that simulation fully substitutes for testing. The same substitution question, with a documented answer of “not yet, and later than promised,” recurs in flight test and is worked through in Frontier Aerospace Programs.

Established Nothing on this map produces a result this brief waits on. This is history with an audited paper trail; what it waits on is archival and statistical rather than scientific, and the three binding items are recorded as typed constraints in section 13. Established They are: a declassification clock that would let the 6,543 standing invention-secrecy orders be studied as a cohort the way the wartime orders were; a published evaluation of post-1991 defence conversion, which does not exist; and a programme budget line that survives more than one appropriation cycle, which is the condition every cancelled programme in section 2 failed.

Frontier What it waits on from research is narrow and specific. A federal R&D obligations series reconciled across the FY2016 definitional break. A measure of federally funded invention that does not vanish under a secrecy order. And a crowding-in elasticity estimated on a design that separates the defence shock from the industrial composition of the countries exposed to it. Established None of the three is a discovery; all three are data-construction jobs that a statistical agency could commission.

Established Two adjacent briefs supply arguments this one deliberately does not restate. Scientific Funding Models owns the allocation-mechanism question — peer review, lotteries, block grants — and Innovation Ecosystems owns the regional-agglomeration literature that Gross and Sampat's already-inventive-counties finding sits inside. Frontier The finding matters more in that second context than in this one: it is a statement about where mobilisation lands, not about whether it works.

Established The cheapest high-value study is to run Gross's design forward on the standing secrecy orders. The wartime cohort was measurable because the orders were eventually released and the technology classes are on the record. 6,543 orders are in force now, released rolling as review permits; a design that compares released filings against matched unreleased ones would price the current instrument rather than the 1940s one. Frontier It requires a release schedule that is systematic rather than discretionary — which is why it appears as a typed institutional requirement below rather than as a research proposal.

Established Second: evaluate a conversion programme, any conversion programme, retrospectively. The United States ran defence-conversion initiatives through the 1990s with named appropriations and identifiable recipient firms and localities. This pass located no published evaluation of any of them. Frontier The design is unremarkable — treated and comparison localities, employment and patenting outcomes, a staggered rollout — and the reason it has not been run is not methodological.

Frontier Third: pre-register the evaluation of the next large cancellation before it happens. Every case in section 2 was evaluated by auditors after the fact, which is why the record is a cost ledger rather than a counterfactual. A programme with a published termination-decision rule and a pre-specified comparison would produce the first evidence on whether cancelled programmes leave capability behind. Speculative The X-57 reframing in Frontier Aerospace Programs — an agency booking the paperwork as the deliverable of a vehicle that never flew — is what happens when nobody specified the test in advance.

Established Fourth: reconcile the federal obligations series across the FY2016 break and publish the bridge. This is a week of work at a statistical agency and it would retire a recurring class of bad chart. Frontier Fifth, and harder: build a synthetic control for American research intensity over 1945–1990 from OECD comparators, and see whether the apparatus shows up at all in the aggregate. The prior should be that it does not, because the aggregate did not move much and the composition did.

Established The transferable engineering of the Cold War apparatus is contractual, not scientific, and it has three parts. The first is GOCO: government-owned, contractor-operated laboratories, which let the state own $2bn instruments without owning a permanent research civil service, and which is why the lab system outlived every mission it was built for. Frontier The second is a standing category that permits a build without a production commitment — the X-designation series is the aerospace instance and is worked through in Frontier Aerospace Programs; the national laboratory is the instrument-scale instance.

Established The third is the contracting vehicle that lets money move faster than procurement law normally allows. Warp Speed put $8.8 billion of $13 billion through other transaction agreements. That is not a scientific mechanism, it is a legal one, and it is the single most reproducible feature of the whole record. Frontier It is also the feature that carries the least oversight, which is the trade the model actually makes and rarely states.

Established Parallel redundancy is the fourth, and it is worked out on the Manhattan Project page rather than here. The Warp Speed instance is the modern one: six candidates across three platform technologies, funded simultaneously, with the expectation that most would not be used. Established The condition that made it rational is the same one Manhattan had — not knowing which approach works while knowing that at least one will.

Established What the apparatus could not engineer is a downstream customer. SSC had a scientific constituency and no buyer; ANP had a buyer who concluded the product was not militarily useful; NASP had neither by 1993. Frontier Every cancellation in the record is a customer failure with a technical story attached, and the technical story is usually the one that gets told.

Established And the secrecy apparatus has an engineering cost that shows up in the innovation data. Compartmentalisation is cheap to impose and expensive to lift: the wartime orders suppressed follow-on citation, pushed affected firms out of treated technology classes through at least 1960, and hit hardest at small firms without government contracts — the part of the economy least able to absorb it.

Within this map: Scientific Funding Models, which owns the allocation-mechanism question this brief treats only as a budget share; Innovation Ecosystems, where the already-inventive-counties finding belongs as a statement about agglomeration; Scientific Governance Models and Scientific Advisory Institutions, which own the advice-to-power relationship the SSC's collapse ran through; Civilization Scale Investment, the general form of the cost-growth problem priced here; Long-Term Institutions, on surviving more than one appropriation cycle; and Megaproject Governance, which owns the contract-management failure mode that killed the collider.

Within the Cold War group: Soviet Frontier Science is the mirror image and the only available control on what a differently organised state got for its money; Operation Paperclip owns the personnel-transfer question; and Frontier Aerospace Programs owns flight test, including the half of the National Aero-Space Plane record that is not cancellation arithmetic.

Outside it: the economics of science and the patent-based measurement literature; public-administration work on other transaction authority and procurement law; the history of the national laboratory system; and defence economics, which supplies the crowding-in estimates and the identification arguments about them.

Established The evidence base here is unusually good for one specific institutional reason: an adversarial auditor with subpoena power published while the programmes were dying. The SSC, NASP and Warp Speed figures in this brief are GAO's, and GAO is adversarial to the programmes it audits by design. Established The interested-party structure is worth stating explicitly, because in this subject the most valuable sources are the ones conceding against themselves. NSF's anniversary volume concedes that NSF did not become what Bush proposed. Physics Today published a physicist assigning part of the SSC's failure to physicists. The Smithsonian's air and space museum — a museum of the field — is candid that the nuclear bomber never had a path. All three are weighted up for it.

Established On the other side, the Federation of American Scientists is an advocacy organisation for openness, and the 6,543 figure is theirs to publish though not theirs to produce — it is USPTO data obtained under FOIA, which is the right provenance to state. Established DOE's laboratory pages describe DOE's own system, and Lawrence Livermore's stockpile-stewardship page describes Livermore's own mission; both are used for what the institution says about itself and not for whether it is working.

Established The binding institutional constraint the record actually identifies is durability across appropriation cycles, not money and not consensus. SSC had scientific consensus, elite backing and a well-posed objective, and lost 81 of 113 freshman representatives. Frontier A programme whose completion horizon exceeds the attention span of the body that funds it is exposed to a risk no engineering practice addresses, and the model has never solved it.

Established The second constraint is a declassification regime that operates on a clock rather than on discretion. The wartime secrecy programme is measurable because its orders were released; the current one is not, because 6,543 of them have not been. Frontier That is a policy choice with a measurable research cost, and it is recorded as a typed requirement below rather than as a complaint.

Established The distributional finding is the one the mobilisation story never carries. Compulsory invention secrecy fell hardest on small firms without government contracts — firms that bore the cost of a national security programme while being excluded from the contracts that compensated everyone else for it. Frontier That is a transfer, not a side effect, and no accounting of the Cold War apparatus prices it.

Established The regional finding has the same shape. If wartime mobilisation's long-run effects were “entirely driven by counties that were already among the most inventive,” then federal research money as deployed was concentrating. Frontier A policy defended on the grounds that it builds national capability, whose measured effect is to deepen capability where it already existed, is doing something other than what it says — and the case for it has to be made on those terms.

Established The secrecy trade is the sharpest ethical question in the subject and it is genuinely two-sided. Secrecy worked at its stated job. It also suppressed commercialisation and follow-on invention for two decades in the affected classes. Frontier Nobody has measured the benefit on the same scale as the cost, which means every existing statement of the trade-off compares a measured number against an asserted one. That asymmetry should be stated whenever the trade is invoked, in either direction.

Established And there is a question about who is conscripted. The apparatus recruited scientists into classified work under conditions where the alternative was often no funding at all in their field. Operation Paperclip is the extreme case of the same logic applied to foreign nationals with war-crimes exposure; Soviet Frontier Science is the extreme case of coercion. Frontier The American version is milder by orders of magnitude and it is the same variable: state direction of science works partly by narrowing the researcher's outside option, and that is a cost borne by individuals rather than by budgets.

Established The terminal position is that the framing is answering a question the data do not pose. Research intensity did not fall — 3.39% of GDP in 2023 against 2.62% in 1968. Federal direction did — 66.5% in 1963 to 18.3% in 2023. Those are different facts with different implications, and the entire “we used to mobilise” literature is built on conflating them. Frontier If the concern is basic research, long horizons and public-goods spillovers, the federal-share collapse is a real problem and should be argued as one. If the claim is that America stopped doing research, it is false on the official series.

Frontier What the record actually supports is narrow and useful. Parallel funding of multiple approaches works when you cannot know in advance which one works. Contracting vehicles that bypass normal procurement move money faster and cost oversight. A laboratory system that owns instruments rather than fields survives its founding mission. Established All three are institutional design, and none of them is about how much a country spends.

Established What the record does not support is the transferability claim. The one clean peacetime reproduction, Warp Speed, satisfied every precondition the Manhattan analogy usually omits. The failures — SSC, ANP, NASP — each lacked one: a customer, a feasible product, or political durability. Frontier The model is not a general-purpose instrument that can be pointed at a hard problem. It is a specific instrument that works on a specific class of problem, and the class is defined by conditions that are not chosen by the funder.

Frontier The long-run implication is about direction rather than level. A country whose research is 75% performed by business is a country whose research agenda is set by expected private return. That is not obviously worse — private R&D grew because it pays — but it is different, and the categories where private return is weakest are exactly the ones the federal share used to cover. Speculative Whether that gap is being filled, and by whom, is not answerable from the aggregate series, because the aggregate series is dominated by the sector that is not filling it.

Speculative And the strongest sceptical reading deserves a sentence of its own. It may be that the apparatus “worked” only in the sense that an extremely rich state buying a great deal of anything gets a great deal of it, and the interesting variance is entirely in what was bought. Nobody argues this as a named position, the estimates in this brief are compatible with it, and it is unfalsifiable with the measurements that exist.

These horizons track declassification schedules, statistical revisions and appropriation cycles rather than technology:

• 10 yr: Established The federal share is a slow series and there is no mechanism visible that would reverse it inside a decade; business performance of R&D has been above 70% since well before 2019. Frontier The standing invention-secrecy caseload is the variable that could move: 6,543 orders in force at the end of FY2025 is a policy number, not a physical one, and a review schedule could halve it. Frontier Expect at least one ARPA-model replication — ARPA-H, ARIA and their successors — to reach an evaluable age inside this window, which is the first real test of H1 since Warp Speed.
• 25 yr: Speculative The plausible split is that the measurement improves where the archives open and stays absent where they do not: a released secrecy cohort would price the current instrument, and a conversion evaluation would settle H6. Speculative Neither is technically hard and both wait on a decision. Handwave Forecasting whether either decision gets made is asserting a political outcome, not extrapolating a measurement.
• 50 yr: Speculative If the federal share stabilises near a fifth, the interesting question stops being mobilisation and becomes whether the categories private return does not cover are being funded at all — which is a question about funding mechanisms rather than about budgets. Speculative The laboratory system is the component most likely to still be recognisable, because GOCO has already survived three mission changes. Handwave Both are extrapolations from a single institutional form.
• 100 / 250+ yr: Handwave Beyond useful forecasting. The only observation at that horizon is that the institutional residue of a bureaucratic defeat in 1945 — a weapons commission, a navy research office, a medical institute network and a small civilian foundation — is still the shape of American science eighty years later. Handwave One accident, in one country, is a story rather than a base rate.

• Depends on Nothing on this map. This brief is history with an audited paper trail; it waits on no result another brief produces, and the constraints that bind are archival, statistical and political rather than scientific. No typed depends-on edge is claimed.
• Requires (not on this map) Three things this subject waits on, none of them a research result. First, a declassification clock on standing invention-secrecy orders: the wartime cohort of about 11,200 orders is measurable precisely because it was released, and the 6,543 orders in force at the end of FY2025 are not, so the one instrument whose innovation cost has been cleanly identified cannot be priced in its current form. A systematic release schedule would make that cohort study possible; discretionary review will not. Second, a published evaluation of post-1991 defence conversion. The United States appropriated money for conversion programmes with named recipients and identifiable localities, and this research pass located no measured evaluation of any of them — which means the claim that the peace dividend was squandered and the claim that it was not are both unevidenced. The design is routine; the data exist; nobody has run it. Third, a programme budget that survives more than one appropriation cycle. Every cancellation priced in this brief had scientific consensus and failed on political durability: the Superconducting Super Collider lost 81 of 113 freshman representatives after roughly $2bn had been spent and about 15 miles of tunnel dug, and Aircraft Nuclear Propulsion consumed nearly $1bn over fifteen years before a president concluded there was no militarily useful aircraft in prospect. None of the three is a discovery. All three are choices a legislature or an agency could make and has so far not made.
• Enables In principle, every argument on this map that invokes state mobilisation of research rests on what this record actually shows — Scientific Funding Models, Civilization Scale Investment and Innovation Ecosystems most directly. No typed enabling edge is claimed, because the relationship is evidential rather than productive: this brief supplies a corrected premise, not a result.
• Adjacent The economics of science and patent-based measurement; defence economics, which supplies the crowding-in estimates; procurement law, which supplies other transaction authority; the history of the national laboratory system; and within this map Soviet Frontier Science, Operation Paperclip and Frontier Aerospace Programs, together with the method studies of the Manhattan Project, Apollo and DARPA.

Handwave “The Cold War was a golden age of research funding that America has abandoned.” Half true and stated wrongly, which makes it useless. Established Total research intensity is higher now (3.39% of GDP, 2023) than at the Cold War peak (2.62%, 1968). What has gone is federal direction: 66.5% of funding in 1963 against 18.3% in 2023. Say “the federal share collapsed” and you have a defensible and important claim. Say “we stopped doing research” and you have a false one.

Speculative “Classified research is more productive because it is unconstrained.” The only clean measurement points the other way: 11,200 wartime secrecy orders reduced follow-on citation, suppressed commercialisation and pushed affected firms out of the treated technology classes for two decades, with the damage concentrated on small non-contractor firms. Speculative The serious counter-argument — that classification provides political cover for failure and therefore permits riskier bets — is made in the DARPA literature, is a claim about institutional tolerance rather than about research productivity, and has no measured support of any kind. Both halves of that sentence matter.

Handwave “The mobilisation model can be pointed at any hard problem.” The failure record answers it, and the answer is not technical. Established SSC had consensus, a well-posed physical objective and elite backing, and died on cost, contract management and a public split among physicists. ANP had fifteen years and a billion dollars and no path to a militarily useful aircraft. NASP's build estimate quadrupled while its scramjet “does not produce sufficient thrust at lower speeds.” Three programmes, three different missing preconditions, and the science uncontested in all three.

Handwave “Vannevar Bush designed the postwar American science system.” He designed a system that was not built. Established NSF's own history says the five-year delay let the Atomic Energy Commission, NIH and the Office of Naval Research expand into the space, and that NSF never became the funder Bush proposed; its first grant was $10,300. The admired features of the “Cold War model” are downstream of a bureaucratic defeat, and the agency that lost is the source for saying so.

Handwave “Defence spending pays for itself in civilian technology.” Crowding-in is real and measured, and it is small. Established A 10% rise in government-financed R&D generates a further 5–6% of privately funded R&D; applying the same authors' model to the entire post-9/11 defence R&D increase yields about 0.005 percentage points of extra annual TFP growth. Handwave The strong version — that military R&D is a superior general industrial policy — is asserted far more often than it is estimated, and where it has been estimated the number is a rounding error.

Frontier “Wartime mobilisation created American technological leadership.” The best evidence for mobilisation says otherwise about the mechanism. Gross and Sampat's persistent effects — 20% higher patenting by 1960, 30% by 1970, 60% in top-quartile treated clusters — were “entirely driven by counties that were already among the most inventive” before the war. Established The apparatus concentrated American invention where it already was. It did not create capability from nothing, which is the thing the story promises.

Speculative “The peace dividend was squandered and the capability was lost.” Directionally plausible and entirely unevidenced. The federal share fell and the weapons laboratories were re-missioned onto stockpile stewardship rather than converted; but this pass could not locate a measured evaluation of any defence-conversion programme, so the claim and its denial are equally unsupported. Established Naming the absence is the honest terminal position, and it is a finding rather than a gap.

Handwave And the framing itself: “the Cold War model.” There was no model. There was a proposal that lost, a weapons commission, a navy research office, a medical institute network, a defence department, and last and smallest a civilian foundation stripped of the defence division its author wanted inside it. Established Every structural feature people admire is downstream of that accident, and the first step in any argument that begins “we should do what the Cold War did” is to say which of those five things is meant.