The published record on missile defence, figure by figure: what is official, what is an analyst’s estimate wearing an official coat, how the test records are counted and by whom, and which of the most-quoted numbers measure something other than what they appear to.
A research dossier compiled for Grid/84, as the basis of the interception lab and the defence lab, and of the loop that flies one engagement after another until the pattern of misses is visible. It exists because a model of interception is a model of specific hardware, and the specific hardware turns out to be one of the least documented subjects in open defence literature.
The finding that shapes everything else. No official burnout or interceptor velocity has ever been published for the American ground-based interceptor, for any variant of the SM-3, for THAAD, for Patriot PAC-3, for Arrow 3, or for the Russian 53T6. Every speed in circulation for these systems is an analyst’s estimate or a hobbyist compendium figure, and the most-cited compendium carries an explicit disclaimer of its own accuracy. A model that wants a number here has to say whose number it is.
Limits of this pass. The Missile Defense Agency’s own domain was unreachable and four further American government domains refused the request throughout the research session. Agency and operational-test wording quoted here is quoted at one remove, through the Congressional Research Service, the Government Accountability Office or named trade reporting, and is labelled that way. The annual operational-test reports would settle several of the counting disputes below.
DocumentedStated in a primary or official source: an accounting office report, a research service product, a service release, a presidential transcript, peer-reviewed physics.
ClaimAsserted by a state or a manufacturer, with no independent verification.
EstimateAn independent analyst’s calculation or judgement, presented as such by its author.
ContestedCredible sources give materially different figures, or the same figure counts different things.
Not foundNo attributable figure located. Usually because the answer is classified, or was never published.
Each is a plausible published number that measures something other than the thing it is quoted for. Each is expanded in its own section below.
The phase where the missile is slow, bright and carrying everything it will ever carry — and the phase in which everything the United States has built has been cancelled.
DocumentedThe window is set by the burn and then eaten from both ends. A solid intercontinental missile burns for about 180 seconds and a liquid one for 240 to 300 (Wilkening, Science & Global Security 12, 2004); theatre-range missiles burn 60 to 150. Real vehicles: Minuteman III 61 + 65 + 61 = about 187 s to burnout at 190 km; Titan II 147 + 182 = 329 s at 315 km; MX 169 s at 196 km; SS-18 205 s (Fetter, A Ballistic Missile Primer, Table 2). From that, subtract detection, subtract the decision, and subtract the margin needed to stop the debris falling short of the target: the physical society’s 2003 study put the usable window at 5 to 20 seconds.
DocumentedThe Airborne Laser destroyed two missiles and then was scrapped. On 3 February 2010 it destroyed a solid-fuelled sounding rocket; on 11 February a liquid-fuelled short-range ballistic missile from a sea platform, within two minutes of launch with its motors still thrusting — the first such shoot-down. Under an hour later it engaged a second solid-fuelled target and did not destroy it. Cumulative record by October 2010: two kills in five attempts. Cancelled December 2011, flown to the boneyard in February 2012, scrapped.
Not foundThe engagement range on either February 2010 date, and the laser’s output power. Both classified. The agency’s contemporaneous language was only “a range of hundreds of kilometers” as a potential. The 600 km against liquid and 300 km against solid pairing widely attributed to the 2003 study could not be verified verbatim in that report; the arms control association’s 300–600 km band is not split by fuel type, and should not be quoted as though it were.
DocumentedWhat the Secretary of Defense said, which is the most useful sentence on the subject. Gates to the House Armed Services Committee, 13 May 2009: a useful weapon would need “a laser something like 20 to 30 times more powerful than the chemical laser in the plane… to be able to get any distance from the launch site to fire”, and the aircraft “would have to orbit inside the borders of Iran”. Independent modelling agrees on the shape: Stupl and Neuneck (S&GS 18, 2010) find the aircraft effective against liquid-fuelled missiles at 400 and 900 km scenarios and failing at 1,600 km, and conclude that “in all presented cases, the intensity provided by the ABL is not sufficient to destroy warheads”.
DocumentedBrilliant Pebbles: three flight tests, no successes, about $1.1 billion. The constellation went from a concept figure of up to 100,000 through about 4,000 before the 1991 restructuring to about 1,000 after it, at roughly 400 km altitude and 8 km/s orbital velocity (GAO/NSIAD-92-91; DoD Inspector General 94-084). The three tests failed in August 1990 (a separation bolt), April 1991 (target not acquired) and October 1992 (a first-stage nozzle, destroyed by range safety at 55 seconds). Stop-work order 1 December 1993. What survived was the sensor and propulsion work, and Clementine, which went to the Moon on it.
EstimateThe Kinetic Energy Interceptor’s six kilometres a second is a modelling parameter, not a specification. The agency published no burnout velocity, saying only that the missile would have “a high velocity at burnout with heavy payloads”. The figure comes from the 2012 National Academies committee, which modelled “a notional 6 km/sec fly-out velocity, 70-sec burn interceptor without a third stage” from an agency briefing of January 2010. The programme ran six years, cost about $4.2 billion, was stopped and rescoped in 2007, and never flew.
The window is short, and the interceptor has to already be close. The physical society’s conclusion in 2003 was that a boost-phase interceptor needs 6 to 10 km/s against a missile doing 7 to 8, and that effective defence against solid-propellant intercontinental missiles “would not be achievable”. An aircraft has to orbit over hostile territory; a ground interceptor has to be based where the geography usually does not allow; a space layer has to be everywhere at once, which is the absentee problem and the reason the constellation numbers run into the thousands. Nothing in the record contradicts that conclusion, and nothing built since has tested it.
The longest phase, the one the American national defence is built around, and the one where the physics is against the defender in a way no amount of engineering has addressed.
DocumentedThe canonical statement, and it is worth having exactly. From the Union of Concerned Scientists and the MIT Security Studies Program, Countermeasures (April 2000), study chair Andrew M. Sessler, with Cornwall, Dietz, Fetter, Frankel, Garwin, Gottfried, Gronlund, Lewis, Postol and Wright: decoys “take advantage of the fact that there is no atmospheric drag in the vacuum of space, so that lightweight objects travel on trajectories identical to that of a much heavier warhead. Because the decoys can be lightweight, the attacker can use a large number of them.” Their headline conclusion is that an attacker “could defeat the planned NMD system by enclosing the warhead in a metal-coated balloon… while at the same time releasing large numbers of similar, but empty, balloons.” The thermal countermeasure is equally specific: chosen coatings hold the balloons near the warhead’s own 300 K, so there is no thermal difference either.
DocumentedThe report supplies its own caveat, which is the whole difficulty. Discrimination becomes possible when the objects start to reenter and the light ones slow down — which is after the midcourse system’s engagement is over. The National Research Council put it flatly in 2012: “The hard fact is that no practical missile defense system can avoid the need for midcourse discrimination”, calling it “the most formidable” challenge and one that “must be addressed far more seriously if reasonable confidence is to be achieved”.
DocumentedAcquisition range is the whole game, and the arithmetic is short. At the 10 km/s closing speed the 2000 report assumes, an interceptor that acquires its target at 1,000 km has 100 seconds and needs 0.0004 g of average lateral acceleration to move 20 metres. At 10 km it has one second and needs 4 g. At 1 km it has a tenth of a second and needs 400 g. This is why the countermeasures argument bites: a decoy does not have to defeat the kill vehicle, only to delay the moment the defender knows which object to fly at.
| Source | Figure | What it is |
|---|---|---|
| UCS/MIT 2000, Table 9-1 | 10 km/s | “We assume a closing speed of 10 km/sec, and a lateral miss distance of 20 meters at kill vehicle acquisition.” Their kill-vehicle divert budget is about 1 km/s in total |
| Garwin, APS Forum, March 1999 | 11 km/s | “at a relative velocity typically of 11 km/s” |
| HOE-4, 10 June 1984 | ~6 km/s | A measured test closing speed — the Army’s “more than four miles per second” |
| ERIS-1, 28 January 1991 | 13.4 km/s | Over 30,000 mph at 270 km. The documented high end |
| IFT-4, 18 January 2000 | 6.7 km/s | Final homing began six seconds before predicted impact |
| UCS 2000, on the interceptor itself | 7–8 km/s | The booster’s speed, not the closing speed. The two are constantly confused |
Not foundA published burnout velocity or engagement altitude band for the ground-based interceptor. The agency describes the engagement only as “outside the atmosphere”. The 7 to 8 km/s in circulation traces to a compendium whose author writes that “figures given below may therefore be inaccurate”. The one figure the makers do publish — “more than 16,000 mph” — is a closing speed.
ContestedThe ground-based test record is three records. The standard tally is 12 of 21 through FTG-12 of December 2023, the programme’s last slated flight test. The Union of Concerned Scientists scores FTG-02 of September 2006 a failure, because the interceptor delivered a glancing blow without destroying the target — the operational-test office’s own phrasing was “a hit, but not a kill” — giving 11 of 21. George Lewis counts FTG-11 of March 2019, in which two interceptors fired a minute apart at one target, as two intercepts rather than one test, giving 13 of 22 — while arguing that the test demonstrated nothing about salvo doctrine, since a second shot only pays when the first fails. The honest range is 52 to 59 per cent, against targets whose trajectory was known in advance.
DocumentedThe kill vehicle that was going to replace the current one was cancelled after $1.21 billion. The Redesigned Kill Vehicle had its critical design review deferred in December 2018 “due to the failure of certain critical components to meet technical requirements”, was stopped in May 2019 and terminated on 22 August 2019, with the seeker publicly identified as the problem. The accounting office found the spend was $340 million more than the original estimate for the entire effort. Its successor, the Next Generation Interceptor, was awarded in April 2024 at about $17.7 billion for twenty interceptors, and has already slipped eighteen months.
DocumentedA ship hit an intercontinental-range target once, in 2020. FTM-44, 16 November 2020: USS John Finn fired a single SM-3 Block IIA at a target launched from Kwajalein and destroyed it north-east of Hawaii. It satisfied a congressional mandate to test the missile against an intercontinental-class threat before the end of that year.
EstimateThe caveats on that test are the substance of it. Ankit Panda, three days after: the target flew short of the roughly 8,000 km a North Korean missile would cover, so the reentry vehicle was slower than a real one; and the test “likely did not incorporate sophisticated, or even rudimentary, countermeasures… including multiple warheads”. He notes it was only the third successful American hit-to-kill intercept against a threat-representative target of any kind, and that a ship cannot reload its launcher at sea. The missile was, in the strategic studies centre’s words, “originally designed and built for the Intermediate-range Ballistic Missile threat set”.
EstimateThe correctly hedged version of the SM-3’s speed, which almost nobody quotes hedged. George Lewis: “Although no official figures have been released, the Block IA and IB interceptors are typically described as having a burnout speed of about 3.0 km/second, and IIA interceptors about 4.0–4.5 km/second.” The strategic studies centre hardens this to “4.5 km/s” without the hedge; Laura Grego notes estimates running to 5.5. Published range figures are similarly unreconciled — about 2,500 km in one place, 1,200 km with a 900 to 1,050 km ceiling in another, with no source explaining whether these measure flyout or footprint.
ContestedTwo published tallies of the SM-3 programme cannot both be right. The arms control association gives 40 intercepts in 50 attempts as of fiscal year 2024; the Center for Arms Control and Non-Proliferation gives 34 of 43 as of May 2025. Neither states its counting rule; the likeliest reconciliation is whether Japanese flight tests are included. The two documented Block IIA failures are 21 June 2017, when a sailor mistakenly triggered the missile’s self-destruct, and FTM-29 of 31 January 2018, a third-stage ignition failure traced to the arm-and-fire device.
DocumentedMoscow’s answer to the miss-distance problem was to not need to hit. The A-135 entered service in February 1995 with sixteen long-range and sixty-eight short-range interceptors; the long-range missiles were retired about 2006, and since then the system has been short-range only. The interceptors “were built to use nuclear warheads”. Pavel Podvig reports they are normally deployed without them, stored about 80 km south-west of the city and mated only in a threatening period — his source is a single Russian analyst report and he hedges it himself — and notes that such a posture “clearly undermines the effectiveness of the missile defense system”. Soviet estimates put the whole system’s capability at one or two intercontinental missiles.
Not foundThe 53T6’s speed. The widely repeated three kilometres a second is a specification for “target speed up to 3 kilometres per second”. The dedicated reference states: “Maximum speed: No verified data.” Figures above 5 km/s and of 4.5 to 6 circulate with no attribution at all.
The two systems that have actually been fired in anger at scale, and the two whose published performance figures are the least well sourced.
DocumentedThe research service says it in one line: “Official PATRIOT ranges and coverages are not available.” What is published, from a NATO fact sheet reproduced by that service, is a flight ceiling of about 20 km and a defended area of about 15 to 20 km against incoming ballistic missiles, with radar range in excess of 150. Lockheed’s own product card and its 2026 fast-facts sheet publish no speed, range or altitude at all; the Mach 4, 4.5 and 5 figures in circulation come from specification-aggregator sites and disagree with each other.
EstimateTHAAD’s numbers are all one lineage, and its floor matters more than its ceiling. The 2,800 m/s traces to the same self-disclaiming compendium; George Lewis independently puts burnout at “likely about 2.6–2.8 km/s”. The 150 km ceiling is from the same lineage. The lower limit — “not publicly available, but is generally taken to be about 40 km” — is the constraining one, and the number nobody publishes: this is an upper-atmosphere weapon, not a last-ditch one. The defended footprint, from Lewis’s modelling of 2007 briefing slides, is about 430 by 340 km per battery against 300 to 1,000 km threats — a diameter pair, not a radius.
Contested“Sixteen for sixteen” and “fourteen of eighteen” describe the same tests. The agency and the manufacturer count only tests where an interceptor flew against a target: sixteen successes in sixteen attempts as of FTT-23, 30 August 2019. The research service and the strategic studies centre report that “between 2006 and 2019 the Army and the MDA conducted 18 THAAD intercept tests. Fourteen of the tests were successful, and four were cancelled prior to launch due to target malfunctions.” Both are literally true. Neither counts the development run: the first attempt of December 1995 failed, and five successive flights from 1996 to 1999 failed after it, attributed to quality control in interceptor manufacture.
ContestedHow many interceptors Israel and the United States spent in June 2025 has four incompatible answers. 39, an explicit minimum from video triangulation of launches seen from Amman. More than 150, from officials, “nearly a quarter of the interceptors ever purchased”. “More than 100 and possibly as many as 150”, about a quarter of the stockpile. And 92 of an estimated 632, cited by the research service as “one study” whose identity could not be established — that last figure should not be used. What is well attributed: of roughly 500 missiles Iran fired, about 50 to 60 impacts were identified, and American THAAD procurement since 2010 totals about 650 interceptors, so the expenditure rate exceeded even the most aggressive production rate by half again.
DocumentedThe claim, and the escape hatch built into it on the same day. President Bush at the Raytheon plant in Andover, 15 February 1991: “Patriot is 41 for 42: 42 Scuds engaged, 41 intercepted.” The very next paragraph read: “Not every intercept results in total destruction.” The distinction between interception and destruction was present in the original claim and became the Army’s position over the following year.
DocumentedThe retreat, verbatim. From GAO/NSIAD-92-340, page 2: “According to an early report, the Patriot destroyed about 96 percent of the Scuds engaged in Saudi Arabia and Israel. As more information became known, the Army reduced its assessment to 80 percent successful in Saudi Arabia and 50 percent successful in Israel. Now, as a result of the April 1992 revised assessment, the Army believes the Patriot was 70 percent successful in Saudi Arabia and 40 percent successful in Israel.” The 96 per cent was a combined figure; the final pair is 70 and 40.
DocumentedWhat “successful” meant. The programme manager “defined an engagement as successful if Patriot destroyed or disabled the incoming warhead or if Patriot caused the incoming warhead to change course and land outside a protected area” (GAO/T-NSIAD-92-27). Warhead kills and mere deflections were counted together. Of the 158 Patriots fired, at least 45 per cent went at debris or false targets.
| Set | Share of engagements | Note |
|---|---|---|
| Assessed warhead kills, any confidence | 52% | Low, medium or high confidence combined |
| High confidence | 25% | The deputy programme manager conceded that high confidence “did not mean the Army was absolutely confident” |
| Supported by the strongest evidence | 9% | The remaining 16% of the high-confidence set was not |
DocumentedThe finding that matters more than any of the percentages. The accounting office’s title says it: Operation Desert Storm: Data Does Not Exist to Conclusively Say How Well Patriot Performed (22 September 1992). “The data that would be needed to conclusively demonstrate how well the Patriot performed during Operation Desert Storm does not exist and there is no way to conclusively determine how many targets the Patriot killed or failed to kill.” Why it does not exist, which is the part a model should care about: portable data recorders were removed in Saudi Arabia by the brigade commander after suspected failures; telemetry transmitters were “not connected to each missile because of the added weight”; the system “reported a probable kill” if the missile reached the computed point of closest approach and stopped communicating; and the fuze “can sense its target and detonate at up to six times the required miss distance, resulting in an extremely low or no probability of kill. However, the system would still record a kill.”
EstimatePostol, and the framing of his that is almost always dropped. Congressional testimony, 7 April 1992: “Patriot’s intercept rate could be much lower than 10 percent, possibly even zero”, from frame analysis of intercept geometry — the warhead sprays about a thousand pellets forward, so beyond roughly 100 metres of miss distance the kill probability is near zero. The mature version, Lewis and Postol in S&GS 4 (1993): “The videos we have been able to obtain contain no unambiguous evidence that Patriot destroyed even one Scud warhead.” His own framing: “The real issue here is not whether Patriot was a technical success or failure… The real issues are whether the performance of Patriot has been misrepresented.”
ContestedThe accounting office then contradicted itself, and both halves are consistent. Nine days after saying the Army’s data could not prove Patriot worked, it issued GAO/NSIAD-93-22R saying Postol’s videos “cannot be relied upon as a data source”: 30 frames a second is too slow, and relative position can change up to 70 metres in a single video field. Both statements express one position — nobody had adequate data. A physical society panel later reviewed the technical objections, found them “largely wanting”, and concluded the two scorings are “profoundly incompatible”.
DocumentedThe Israeli assessment is harsher than the American critique. Reuven Pedatzur’s testimony of 7 April 1992 reported the air force commander, Maj. Gen. Avihu Ben-Nun: “only one al-Hussein warhead was evidently hit by Patriot missiles. The data on two others are not clear… No other warheads were hit by Patriots” — and that the public claims “should be viewed within the realm of psychological warfare”. On one night, 27 Patriots were fired at seven missiles and not a single warhead was hit. Four Patriots struck the ground in Israel during the war, from a defective safe-and-arm mechanism. The technical explanation offered: the warhead was designed for closing speeds up to about 2,100 m/s, and at 12 km altitude the closing speed reached 3,500 to 3,600.
ContestedThe claim that Patriot increased the damage is weaker than its restatement. The physical society panel records that Postol found ground damage in Israel higher after Patriot’s introduction but that the difference was not statistically significant — his point was that it was inconsistent with a 96 per cent success rate, not that Patriot demonstrably made things worse. Fetter, Lewis and Gronlund (1993) found serious apartment damage per missile fell slightly, from 37 to 34, while light damage, deaths and injuries rose.
DocumentedTwenty-eight American soldiers were killed and ninety-nine wounded in total. The 14th Quartermaster Detachment accounted for thirteen of the twenty-eight killed and forty-three of the ninety-nine wounded — 81 per cent of its 69 deployed soldiers, the heaviest casualties of any allied unit in the war. All twenty-eight deaths are frequently and wrongly attributed to that one detachment.
DocumentedThe mechanism, from GAO/IMTEC-92-26. “Because of the way the Patriot computer performs its calculations and the fact that its registers are only 24 bits long, the conversion of time from an integer to a real number cannot be any more precise than 24 bits. This conversion results in a loss of precision… directly proportional to the target’s velocity and the length of time the system has been running.” The office’s own table gives a 687-metre range-gate shift at 100 hours of continuous operation, and the system “will not track a Scud when there is a range gate shift of 50 percent or more”. Alpha Battery had been running about 100 hours.
ContestedDo not conflate 570 metres with 687. The often-quoted 0.34 s × 1,676 m/s = 570 m is Douglas Arnold’s derivation, and 570 m is the distance the incoming missile travels — a different quantity from the accounting office’s 687-metre range-gate shift. The office does not publish the binary expansion of one tenth.
DocumentedIt was a distribution failure, not a development failure, and the dates are the argument. 11 February: the project office received Israeli data showing a 20 per cent range-gate shift after eight hours; Army officials “believed the Israeli experience was atypical”. 16 February: corrected software released — nine days before the strike. 21 February: a message to users warning that “very long run times” could shift the range gate, which “did not specify what constitutes very long run times”. 25 February: Alpha Battery, past a hundred hours, did not engage. 26 February: the modified software arrived in Dhahran, the next day. A reboot, which takes 60 to 90 seconds, would also have cleared it.
EstimateThe distribution matters more than the average. Through 24 October 2025, Russia fired 939 Iskander and Kinzhal missiles at Ukraine and 227 were intercepted, about 24 per cent. But 273 of 345 attacks saw no intercepts at all, and the best sustained city rate, around Odessa by 2025, was about 30 per cent. The compilers warn explicitly that the near-binary distribution “prevents reliable predictive modeling”. They examine steeper trajectories, improved decoys and coordinated salvos as causes and find the salvo hypothesis “ultimately unconvincing”.
ContestedThe “37 per cent falling to 6” figure is weaker than its circulation suggests. It originates in press reporting of October 2025 citing officials; the analysis that followed stresses these are not Ukrainian official statistics and that the reporting lacked engagement-level data — attempts as against launches. Three candidate explanations are offered: steeper terminal trajectories; fire shifted to undefended areas outside the coverage; and Ukraine dropping from two or three interceptors per target to one because of shortages, which lowers the rate mechanically. No leaked document giving per-engagement effectiveness was found, and none should be asserted.
DocumentedThe Kinzhal shootdown is on the record from the American side. The intercept occurred overnight on 3–4 May 2023 over the Kyiv region at roughly 10 km. Asked directly on 10 May whether a Patriot had downed a Kinzhal, the Pentagon press secretary answered: “Correct.” The claim of intercepting a salvo of six on 16 May is Ukrainian, with no equivalent American confirmation. Russia’s counter-claim to have destroyed a Patriot radar and five launchers that day was contradicted the same day by American officials, who said two components were damaged, the damage was minimal and the system “was never offline”.
EstimateThe falling rate is not a raw-velocity problem. The Royal United Services Institute observes that the PAC-3 “have been demonstrated against medium-range ballistic missile type targets which typically re-enter the atmosphere at speeds well in excess of anything which could be achieved by the 9M723”. Kinzhal is quasi-ballistic, derived from the ground-launched Iskander, with limited terminal manoeuvre — not a glide vehicle and not scramjet-powered. The strategic studies centre adds that calling it hypersonic “is somewhat misleading, as nearly all ballistic missiles reach hypersonic speeds at some point during their flight”.
A defensive category that exists because of a gap in the other three, and in which nothing has yet been built.
DocumentedWhy the phase is its own category, in the words of the paper that establishes it. Tracy and Wright, Science & Global Security 28:3 (2020), p. 159: “During their glide and terminal phases hypersonic weapons would fly at altitudes too low for interception by defenses designed to intercept ballistic missiles above the atmosphere… Endoatmospheric defenses such as the U.S. Patriot and THAAD systems operate within the atmosphere, but are designed to intercept warheads from missiles with shorter ranges and slower speeds than ICBMs.” The glide is flown below the floor of one set of systems and above the ceiling of the other.
EstimateThe glide corridor, and how fast it bleeds away. The same paper models a notional 1,000 kg vehicle entering glide at about 49 km at roughly 6 km/s, and finds the velocity “halved after 6,000 km of glide”, with altitude falling as speed does — a narrow altitude-velocity corridor rather than a free choice of trajectory. The drag scaling is cubic: a Mach 20 vehicle loses energy eight thousand times faster than at Mach 1, and a Mach 15 glider making a 30° turn spends about 450 km of range out of 3,000.
DocumentedThe Glide Phase Interceptor exists on paper and on a schedule that has moved five times. Concept awards November 2021; Lockheed dropped in June 2022; Japan joined for propulsion in May 2024; Northrop Grumman selected sole-source in September 2024, years earlier than planned because of budget constraints, in the agency director’s own words. The programme value rose to $1.31 billion in April 2026. Delivery has been 2034, then a congressionally directed 2029, then 2035, then 2032, and most recently 2031. No intercept test date has ever been published. The interim capability against hypersonic weapons is a terminal one: SM-6 Sea Based Terminal Increment 3, certified 7 August 2025. The March 2025 test that is sometimes reported as an intercept was a simulated engagement; no live intercept occurred.
ClaimAvangard, as its own government describes it. Putin, 1 March 2018: the system “is capable of intercontinental flight at supersonic speeds in excess of Mach 20… In moving to its target, the missile’s gliding cruise bloc engages in intensive manoeuvring — both lateral (by several thousand km) and vertical. This is what makes it absolutely invulnerable to any air or missile defence system.” The Mach 27 figure is not his: it is a deputy prime minister’s, from December 2018. Yield is claimed at “up to” and, elsewhere, “more than” two megatons, with no independent confirmation of either.
EstimateWhat Mach 27 probably describes. Michael Kofman: “The proposition that the vehicle can reach mach 27 is likely true only during the brief return phase, when it is falling back to earth like a rock from near earth orbit.” His overall assessment is that the system is “an expensive insurance policy that in no way alters the strategic nuclear balance”. Twelve are deployed, at Dombarovskiy in Orenburg, at about two a year, and deployment appears complete.
DocumentedThe counter-argument, from the people who modelled it. Tracy and Wright’s conclusions, in their own words. Not faster: “Claims that the advent of hypersonic weaponry will reduce the time necessary for warhead delivery between, for example, the United States, Russia, and China, are false.” Slower where it counts: “The hypersonic vehicle, because it has been slowed by drag throughout its glide phase, begins the terminal phase at a lower velocity than either” ballistic case. Not stealthy: “Low altitude flight also produces immense heating of glider surfaces, yielding IR signatures sufficient for detection by existing space-based sensors.” Their verdict is that these are an “evolutionary—not revolutionary—development”, and that the hype arises from “social processes by which the organizations developing these weapons construct erroneous technical facts favoring continued investment”.
ContestedTwo caveats before that paper is used as a reassurance. Its vehicle is a notional one — 1,000 kg, lift-to-drag 2.6 — and not either of the deployed weapons; the authors flag the sensitivity themselves, noting that a higher lift-to-drag ratio would improve performance. And their argument is glider against ballistic reentry vehicle, not “gliders are easy to intercept”. On defence they are agnostic to pessimistic: manoeuvre “might prevent the current generation of defenses from successfully intercepting these targets”, and improved future defences “would still be capable of defending only relatively small areas”.
ContestedThe Chinese orbital test of 2021 is a different system from the DF-17, and the two are constantly merged. The DF-17 is a road-mobile medium-range missile lofting a glide vehicle on a regional sub-orbital trajectory, estimated at about 60 km altitude on a depressed trajectory, assessed as fielded from 2020 and described in the 2024 defence department report as carrying a conventional warhead. The 2021 test used a space launch vehicle to place a glide vehicle into a partial orbit, from which it de-orbited and glided. Different booster, range class, trajectory concept and strategic function. The chairman of the joint chiefs, October 2021: “I don’t know if it’s quite a Sputnik moment, but I think it’s very close to that.” The glide-phase interceptor is sized against the first, not the second.
Where it comes from, who turned it round, and the finding attached to the day they did.
Not foundNo first coinage can be attributed. The phrase circulated through the Nike Zeus debates of the 1960s and the strategic defence debates of the 1980s as critics’ shorthand for infeasibility, and is often attributed to McNamara-era scepticism. No documented McNamara quotation using it could be located, and it should not be attributed to him.
DocumentedThe best-documented use is the one that flipped it from criticism to boast. Amoretta Hoeber, Principal Deputy Assistant Secretary of the Army, after the fourth Homing Overlay Experiment on 10 June 1984, as recorded in the Army’s own history of the programme: “We tried to hit a bullet with a bullet and it worked.” The interceptor launched from Meck Island in the Kwajalein atoll against a target launched from Vandenberg 4,500 miles away, and struck it more than a hundred nautical miles up at a combined velocity of “more than four miles per second”. It was the first exoatmospheric hit-to-kill intercept in history. Three of the four tests in the series failed.
DocumentedThe caveat is an accounting office finding, not an allegation, and it cuts both ways. GAO/NSIAD-94-219 (July 1994) investigated after the defence department acknowledged in September 1993 that a deception programme had existed: “The planned deception was to explode the target if the interceptor failed to hit it but passed close enough to support the appearance of a hit.” The office found the plan was in place for the first two tests, did not affect their outcomes, and was dropped before the third. On the fourth: “GAO found no evidence that DOD deceived Congress about HOE 4 intercepting the target… Analyses of HOE 4 test data are consistent with the Army’s conclusion that the interceptor and target collided.” But “steps were taken to make it easier for the interceptor’s sensor to find the target” — the target was flown side-on and heated to 100°F before launch, giving “an infrared signature closer to the high end of the range of expected threat signatures”. The office judged that reasonable for an early technology demonstration, and noted that the department’s statements of 1984 and 1985 “did not disclose the enhancements of the target’s infrared visibility”.
Hitting a bullet with a bullet is the part that works. It was done in 1984 and has been done repeatedly since. Every documented failure mode in this dossier lies before that moment: nothing was within reach when the missile lifted; the object the interceptor flew at was a balloon; the track was handed over too late for the divert available to null the error; or the system recorded a kill it had no way to observe. The endgame is an engineering problem that has been solved. The four problems in front of it have not been, and three of them are not engineering problems at all.
Figures with no attributable source. In most cases the real answer is that the number is classified or was never published; filling these with plausible values is the fastest way to make a model quietly wrong.
ContestedWhere to go next. The annual operational test and evaluation reports, the agency’s own fact sheets and its published intercept flight test record would settle the ground-based counting dispute, the forty-of-fifty against thirty-four-of-forty-three question on SM-3, and several of the altitude bands. All were unreachable from this research session; none is classified.