Redactions

The unredacted General Advisory Committee report on the hydrogen bomb

by Alex Wellerstein, published November 20th, 2026

After the detection of the first Soviet atomic detonation in September 1949, a small group of scientists and administrators began a push for a weapon they thought would preserve American strategic superiority: the hydrogen bomb, or “Super” as it was frequently referred to at the time. The idea of the “Super” had been around since 1942, but it had always been treated as a long-term project. While it played a part in early discussions about the potential effects of nuclear weapons (as I recently wrote about on my other blog, Doomsday Machines, which you should subscribe to if you haven’t already), it wasn’t treated with any particular urgency — the atomic bomb was understandably the problem to be solved during World War II, and the problem of increasing and improving the atomic stockpile took up the lion’s share of resources and attention even several years after the end of the war.

What this group of people, which included scientists like Edward Teller and Ernest Lawrence, as well as political movers like Lewis Strauss and Sen. Brien McMahon, wanted was an “all-out” or “crash” effort to build a weapon that was then understood to be theoretically without limit. Practically, they spoke of it as a weapon between 10 and 100 megatons of TNT equivalent, with many analyses at the time assuming it might be around 40 megatons. This suggestion was not well received by certain others within the US Atomic Energy Commission, including its chairman David E. Lilienthal, and, famously, the scientist and “father of the atomic bomb” himself, J. Robert Oppenheimer.

The General Advisory Committee to the United States Atomic Energy Commission in 1947. At left, the chairman, J. Robert Oppenheimer, on the cover of LIFE magazine. At right, the rest of the GAC, from left-to-right: (sitting) Enrico Fermi (University of Chicago); Glenn Seaborg (University of California, Berkeley); Hartley Rowe (Vice President, United Fruit Company); James B. Conant (President, Harvard University); (standing): Hood Worthington (Dupont & Co. — not a member in 1949); Cyril Stanley Smith (University of Chicago); and I.I. Rabi (Columbia University). Missing in the photo on the right for 1949 is Lee A. DuBridge (Caltech) and Oliver E. Buckley (Bell Labs).

Ultimately, the Atomic Energy Commission, in October 1949, requested that its General Advisory Committee, a body of scientists chaired by Oppenheimer, examine the question. This led to a famous report on the “Super,” with a “Majority” and “Minority” appendix, that the GAC sent back to the AEC on October 30, 1949. The report argued, for a variety of reasons, that no “all out” effort should be made for the “Super,” and in fact the United States should publicly declare that it would not build such a weapon at all, in a perhaps vain hope that this might also cause the Soviet Union to foreswear it. 

This entire debate was entirely secret until it leaked in November 1949 under strange circumstances that I have written about previously. Once made public, the debate quickly came to a head, as the idea of the H-bomb was outrageously popular with the American public and political establishment. Polls from the spring of 1950 suggest that nearly 80% of Americans approved of building the H-bomb.1 Truman ended up approving that the weapon be built, even though he expressed personal reservations about it. (For more on that, and the exact circumstances that led to Truman’s order, check out chapter 16 of my new book, The Most Awful Responsibility.)

The arguments against the H-bomb fell into a few different categories. I would summarize these as being based around strategy, resources, or morality. For strategy, the argument was that there were few targets in the Soviet Union that warranted weapons in the megaton range (multiple kiloton-range bombs could do the trick), the plans about H-bombs at the time suggested they would be incredibly hard to deliver by air (and the Soviets had few coastal targets worth using a boat-based bomb on), and, ultimately, the US was more vulnerable to H-bombs than the Soviets were (because of its own large coastal cities), and thus wouldn’t it be in the US interest to see if such a weapon could be avoided in the first place.

The resource arguments were about detracting from the existing atomic energy program. Nobody knew whether an H-bomb could actually be built in late 1949–early 1950, and so spending finite manpower and money on a “crash” effort for a speculative weapon was, for many, a foolish risk. Furthermore, the ideas for H-bombs at the time suggested that fairly large amounts of tritium would be necessary to make it work. Tritium is produced in nuclear reactors by bombarding lithium with neutrons, neutrons that might otherwise have been used to make plutonium, which was still considered scarce in the US atomic program. The ratio at the time assumed that every gram of tritium produced was 70-80 grams of plutonium that was not produced. Depending on how much tritium you needed in a “Super” — which, again, was unknown — it might mean the production of many fission bombs’ worth of plutonium for every “Super” developed. 

And, lastly, there were the moral arguments. Without getting into the details (I am writing something more extensive on thermonuclear morality for Doomsday Machines, which I hope to be able to post next week), the basic argument was that moving from kilotons to megatons constituted an entirely different level of moral hazard than was already embodied by fission bombs. H-bombs, in this view, were necessarily indiscriminate, and were weapons of civilian extermination — even genocide. The moral argument was made both on its own merits, and as an argument about psychology and diplomacy: whether you or I agree think that it is a weapon of genocide, the perception of the immorality of the weapon would mean that there could be a perceived loss of American moral superiority.

Herbert York’s transcribed version of the first page of the GAC’s October 1949 report on the “Super.”

The strategic and the morality arguments were part of the public phase of the debate; the resource arguments were much less so, because they involved highly classified information at the time, like ideas about how much tritium might be needed in a “Super” and the relative difficulties the United States was having in generating sufficient quantities of fissile materials. The fact that the GAC had recommended against the building of the “Super” had become known in the 1950s, and featured in Oppenheimer’s security hearing, but the details of the report were classified for a long time. The first version of the report to be declassified and released came as an appendix in Herbert F. York’s The Advisors: Oppenheimer, Teller, and the Superbomb (Stanford University Press, 1976). This has largely been the text that historians, educators, and everyone else has been using since then, and the basis of the transcribed version on Atomic Archive, which is the version most people are likely to use today.

As mentioned, I have been working on a post about thermonuclear morality for Doomsday Machines, and was going over the GAC report very closely as a consequence. Two things jumped out at me that I hadn’t really thought a lot about before. The first was that the version in York’s book has some indicated redactions. Could those be uncovered? The second was that I ran across some text that it occurred to me might be a transcription error. Could that be checked? And the third was just that nagging desire I have to see original copies of documents — not just because transcription can introduce error all too easily, but because, gosh darn it, I just really like the “reality effect” of seeing these historical documents, as opposed to clean transcriptions. It gives such a better sense of the historicity of the thing, something a web page just cannot do.

There are three redactions in York’s version. 

  • In an early part of the report on increased neutron production, there is a sentence that reads: “With regard to the scale of this project the figure * * * per day may give a reasonable notion,” where the asterisks indicate a redaction. 
  • In the discussion of the “Super,” there is a section on the “steps which need to be taken if the super bomb is to become a reality.” The first of these has a redacted figure: “The provision of tritium in amounts perhaps of several * * * * * per unit.” This would be interesting to know, as it would give the order of magnitude for how much tritium was being imagined as necessary in the “Runaway Super” design that was favored at the time.
  • The third of these steps also has a missing word: “The engineering of designs which may on theoretical grounds appear hopeful, particularly with regard to the * * * * *  problems presented.”
  • In that same discussion of “steps,” there is a final “step” that reads: “Carefully instrumented test programs to determine whether the deuterium-tritium mixture will be ignited by the fission bomb, * * * * *” It does not indicate how much text has been redacted, nor if there were even more “steps” that were redacted.

Note that none of these are reproduced correctly in the Atomic Archive version, which just elides over the missing parts without signaling them as such. Now, none of these redactions are so crucial that knowing them would radically change how we read the report, probably, but, still, wouldn’t it be interesting to know? Obviously you wouldn’t have read this far if you weren’t the kind of person who thought so.

The transcription error I noticed was more subtle. In the discussion of the neutron production figure (the same one redacted above), the GAC make clear that they think these neutrons should be going towards several things other than just tritium production for the “Super.” Specifically, one of them is “The conversion of U-235 to plutonium,” while another is “A secondary facility for plutonium production.” The question that immediately came to mind for me was, do they mean U-235, or U-238? Because those are different situations and imply a different use of the reactor. All plutonium-239 is produced from U-238. So if they said the conversion of U-238 to plutonium, that would just mean an effort to make more plutonium. To say they wanted to convert U-235 to plutonium would suggest to me that they were thinking of putting enriched uranium into their plutonium production reactors, which would increase the neutron flux and thus the plutonium yield. This would be a very expensive way to make plutonium, but if your weapon supply was limited by your plutonium production, and you could more easily produce large stockpiles of enriched uranium, then this might be a smart move from the standpoint of the stockpile size. And indeed this kind of bad U/Pu production “ratio” was exactly what the United States found itself in after the war.

And the “secondary facility” — also plutonium production? Is that… redundant? Or just something that emphasizes how much plutonium they want? Or, perhaps, some kind of error?

I was able to find the redacted information fairly quickly. Chuck Hansen’s Swords of Armageddon (1995) contains the text from the three “steps” of the “Super.” He cites some sources that I was unable to access directly, but I trust that he would not have just made it up. In his telling of it, those three redacted lines were:

  • “The provision of tritium in amounts perhaps of several hundred grams per unit.” This is which are moderately interesting: several hundred grams per unit is actually a pretty low-end of the “Super” estimates in this period. It would rise over time to 1 kilogram, and ultimately at least 3-5 kilograms. If we assume that the composite bombs they had in mind (like one tested at Operation SANDSTONE in 1948) needed 2.5 kg of plutonium, that means that 100 g of tritium would translate to each Super being a “loss” of 3 fission bombs’ worth of plutonium. At 1 kg of tritium, it rises to 32 fission bomb. And at 5 kg, it rises to 160 fission bombs. So the GAC argument about tritium is actually quite optimistic.
  • “The engineering of designs which may on theoretical grounds appear hopeful, particularly with regard to the cryogenic  problems presented.” This is not all that surprising, as it is well-known that they were thinking about liquid deuterium as their main fusion fuel for the “Runaway Super,” and that requires cryogenic refrigeration. But still, nice to know exactly what engineering issues they were talking about.
  • “Carefully instrumented test programs to determine whether the deuterium-tritium mixture will be ignited by the fission bomb, whether this ignition in turn ignites the deuterium, and, if possible, to determine what, in the event that such ignition did not take place, went wrong with the calculations.” To me this just indicates, again, that there were many uncertainties in the making of the Super. It is frankly not entirely clear to me why it would have been deleted at all, except as a result of excessive rule-following. It tells one nothing about how to design a “Super.” (The people in 1949, again, did not know how to do it, at least not how to do it correctly.)

What about that earlier one, about neutron production? I have a copy of an executive session of the Congressional Joint Committee on Atomic Energy from January 1950 in which the committee (quite derisively and skeptically) read the entire GAC report out loud. It’s a lovely document, and one I’ve used for years while teaching, because it’s very revealing about how the GAC’s argument was received by these particular Congressmen several months later, and I will talk about it some more in my forthcoming post on Doomsday Machines. (Are you picking up the pattern? If you’re not subscribing to Doomsday Machines, which is free, you are missing out!) Anyway, the transcript preserved those other two redactions, but list the neutron production figure outright: “one gram per day.”

Well, there you go. They also read the sections that I was curious about transcription errors: they say it was “U-235 to plutonium,” but the “secondary facility” was for polonium production. (Polonium was used in neutron initiators.) Neat. Now, of course, the JCAE transcript isn’t necessarily as accurate as an original version, because it is being read out loud by a non-expert (McMahon) and then transcribed by a non-expert (whomever made the transcript). So is it possible that McMahon could have misread, or the transcriber have misheard? Sure. So while useful, I still wanted an original.

Where to find one? I have gone through the “thermonuclear” files of the Joint Committee on Atomic Energy for my research, and I have looked through Hansen papers (although not exhaustively) at the National Security Archive, and of course I could search my computer and the web and so on. I did this over the course of a few days, looking through all of the sources I had available to me, including my own extensive archival photographs, and I found… Bupkis. Nothing. Nada. No original document. Hrm.

And then I had a little insight: Glenn Seaborg, the Berkeley radiochemist, published an edited version of his diaries in the early 1990s with Lawrence Berkeley Laboratory. Seaborg was a member of the GAC in 1949, but did not take part in the discussions about the report, because was in Europe when they were happening. This is a point of minor interest in the official history of the H-bomb debate, because the GAC constantly depicted their conclusions against the H-bomb as unanimous, whereas Seaborg had registered his view with them that perhaps the US ought to build the “Super.” Seaborg would later go on to be Chairman of the Atomic Energy Commission in the 1960s.

Anyway, Seaborg’s diaries sometimes have reproductions of documents in them. Might I be able to check those? I had many volumes of Seaborg’s diaries on my computer, but not the 1949 volume. These LBL publications are not the easiest to get ahold of as they were a relatively small print run and not exactly best selling books; Seaborg’s diaries, in physical form, practically take up an entire bookcase, as I can remember well from my undergraduate days at Berkeley (a copy of them used to be stored in the Office for the History of Science and Technology, where I started working as a freshman in 1999, and was my introduction into this whole world). 

But today we have lots of ways of finding obscure books. The volume I wanted wasn’t on Archive.org, but it was on Hathitrust. And sure enough, it contained a full reproduction of Seaborg’s copy of the GAC report for October 1949! It still had the same redactions as York’s, but now that I knew what that text was, I could fill those in, and make sure they fit the space correctly (which they did). They also had Seaborg’s annotations on them, which might be interesting for people studying Seaborg’s reception of the document, but for me just got in the way, so I removed them when I cleaned up the document (sorry, Glenn Seaborg!). 

The first page of the GAC’s October 1949 report on the “Super,” as taken from the Seaborg journals, cleaned up by me.

And so I present to you, for the first time ever, a totally cleaned-up, unredacted, slightly-annotated (just to indicate the sources of the added material) version of the General Advisory Committee’s Report on the Super from October 30, 1949. Behold! A historical-looking document! It even includes the signatures on the “majority” and “minority” annexes — which, interestingly, are not at all individually labeled.

I also took great pains to make it so that there is an underlying text layer that perfectly reproduces the text in the report, so you can search it, highlight it, copy and paste it, read it with a screen reader, and so on. I am fairly pleased with it and happy to share it. You can download it here. Enjoy. I hope you have also enjoyed this glimpse into my process, by which I mean, how I get pathologically obsessed with finding something and then just keep looking for it until I am satisfied.

  1. Hazel Gaudet Erskine, “The Polls: Atomic Weapons and Nuclear Energy,” Public Opinion Quarterly 27, no. 2 (Summer, 1963), 155–190, on 181. [↩]
News and Notes

NUKEMAP roadmap

by Alex Wellerstein, published February 10th, 2026

I’ve been working on upgrades to the NUKEMAP for several years now, but I actually would like to get them implemented this year. I figured that one way to both make this “invisible” work visible to the rest of the world would be to write about it on here, and that this would also, perhaps, encourage me to Get It Done.

NUKEMAP as it looks in Chrome on a Mac in February 2026. Sorry, New York. Note the use of Protomaps and not Mapbox or Google Maps, both of which became prohibitively expensive.

Over the past few years, I’ve done some major “behind the scenes” upgrades to NUKEMAP that are basically invisible unless you are paying very close attention. Some of these are basic bug fixes and quality-of-life improvements, but the more substantial relate to migrating a lot of the functionality of the site away from Amazon AWS — which is relatively expensive — and to Cloudflare. The site still does require AWS for its PHP and RDS (database) functions, as migrating those to Cloudflare’s Worker and database systems would both be a lot of work and not necessarily be cost-effective (Cloudflare and AWS RDS have very different pricing models, and for some things Cloudflare is better, and for others it is much worse), but I’ve managed to cut the monthly costs down considerably.

Most importantly, the map UI is now run off of a self-hosted (Cloudflare hosted) Protomaps instance, which is much, much, much cheaper than Google Maps, Mapbox, whatever. If you are a developer or organization that uses maps and do not require them to be at-the-moment constantly up-to-date and are looking to shave a lot of cost off of your bottom line, check out Protomaps. It is a pain to set up the full “stack” of things you need for it to work (I will write up how I did it at some point, but 80% of the difficulty is just getting the style sheets migrated correctly to Protomaps’ format) but once you do it costs almost an insignificant amount compared to those other systems even with very high map usage. It’s not perfect, in part because it is reliant on OpenStreetMaps data for tagging (and sometimes someone will make a mistake that like, eliminates Pennsylvania, and that ends up in the “build”) and because I’ve made some tweaks so that it will render even on older (non-WebGL supporting) browsers that disallow me from using the latest versions of Protomaps data builds, but the cost savings are so significant that it makes up for it.1

So what lies ahead? Here are the major things for NUKEMAP I would like to implement in 2026:

  • Enhanced casualty calculations and updated population data. The NUKEMAP’s casualty calculations work on LandScan population data, and that underlying dataset is by now somewhat out of date. I have migrated the new data into a database, but am experimenting with alternative modes of casualty calculation as well before I roll it out. These include looking at alternative calculation models, e.g., not just pegging them to blast as a proxy for mortality, but looking at thermal and conflagration models, and allowing the user to change which models they use. I also have integrated an urban land-use dataset into my casualty dataset so that there are possibilities of using building data for these calculations (e.g., being able to take into account whether a given grid square is “high-density urban,” “medium-density urban,” etc. or not, which can affect various estimations). I would say this is 90% done, and just waiting on some tweaks to the underlying effects model (covered below) that I am implementing. 
  • Humanitarian impacts. A very long time ago, when Google Maps was actually cheap/free/affordable, I used to have a “Humanitarian impact” button that would look up information about schools, churches, and hospitals that were within the area of heavy blast damage. I have been meaning to return a similar functionality, and have put together a database OpenStreetMaps’ tagged metadata (basically an extracted database of these kinds of institutions, along with a “cultural” category that includes libraries, theaters, and museums). I also have added, as part of my enhanced casualty calculations, estimates as to what component of your casualties are specifically children, which I think is an effective way to make these numbers a bit more impactful. This is basically ready to go and just waiting on my updates to the casualty calculations above, and integrating the output into the user interface.
  • Upgrading the effects models. I have for some time now been creating a new Javascript framework that I call the Atomic Weapons Effects Library (AWEL.js, you see what I did there?) which is a “universal” nuclear effects library that allows me to integrate different effects models and data into one place and use them in a way that is much easier than the current NUKEMAP effects library. While doing this I have been going over every aspect of the modeling and improving it where I can, either by drawing on additional or better sources when they are available, or adding new functionality when I find it. So, for example, I am not only using Glasstone and Dolan’s The Effects of Nuclear Weapons (1977), but also bringing in data from the previous (and formerly classified) Capabilities of Nuclear Weapons (1960 and 1972 editions) where it is available. So this means that the new effects library not only will be able to do a number of things the current one cannot do (like go into the “very low” overpressure region, down to 0.25 psi), but also will be open-source and on Github for people to use and critique and add to as they see fit. It is also a very useful modular framework that is unit agnostic and will make the life of anyone who does these kinds of calculations much easier, I think (it has made my life easier, anyway). Anyway, while this is undoubtedly a Quixotian, never-ending project (since it is meant to be able to accommodate as many effects models as one wants to include into it), I will have a “basic” version done and implemented in the new NUKEMAP fairly soon. I would say this is 80% done, with just a few things needed to finesse and fix before it is “production ready.” 
  • Adding more fallout capabilities. Aside from wanting to allow the user to more easily understand the impacts of fallout (like being able to instantly generate “exposure over time” graphs for places downwind of a detonation), I have also been working on new implementations of WSEG-10 and, possibly, DELFIC, that can run in the browser, including as WebGL shaders, which are shockingly fast. So I am planning to allow the user to choose between fallout models. An advantage of WSEG-10, for example, is that you can easily do things with it that are not so possible with the existing Miller model, like showing the changing dose-rate over time after the detonation. This is maybe 70% done for WSEG-10 and maybe 20% done for DELFIC. (I have not gotten far enough with DELFIC to confirm that it can render in a reasonable amount of time, even as a web shader, so it may be impossible for this implementation. With WSEG-10, the web shader version renders maybe 100X faster than a normal in-browser version, and feels very plausible.)
  • A total UI overhaul and recode. This is a big job, but the NUKEMAP interface and underlying code was originally made almost a decade and a half ago and has been added to incrementally since then. My own Javascript programming experience has expanded a lot since then, and what a web browser is capable of has changed a lot since then. Moreover, the modes in which people use websites have changed since then — mobile users represent a significant component of NUKEMAP users — and NUKEMAP is not really built around non-desktop modes (I made a half-assed mobile stylesheet port a few years ago, just so it isn’t impossible to use on a mobile phone, but it is very half-assed). I have been experimenting with this for several years now, unwilling to commit to one framework or another (they all have their pros and many cons, and a major “con” for all of them seems to be that once you commit to one, you’re committed to it), but I’m in a place now where I’m pretty ready to do this. This will involve rewriting basically all of the site in a much smarter way than it is currently written. Lest you worry, the essential aesthetics are not going to change that much — I am, if anything, stuck in my ways, so it is not going to turn into something “trendy” — but I think there are ways to streamline it without removing any existing “advanced” functionality. This will also involve making it more accessible for those with visual impairments, and building in a native framework for translating its components into other languages, something I have wanted to do for a long time now. 

OK, so that’s a lot for 2026, but most of those tasks, as you can see, are already pretty advanced, as I have been working on them for some time now. The UI recode is the one that is the most nebulous in terms of “how much has been done, how much is there to do,” because I’ve started and re-started this many times over the years, and had students do mockups and proofs-of-concept, and so on, but I probably am going to just restart it again from scratch, because it’s that kind of project. But I’ve been thinking about it for so long that my mind has definitely formed some very firm opinions about how to go about it.

So that’s the game plan! I’ll be posting updates on here as bits of it get completed. At the moment I am thinking that the order of completion will be: effects model, casualties, humanitarian impact, fallout, UI overhaul. But we’ll see how that shakes out in practice.2

I would be remiss not to acknowledge that this upgrading work, and NUKEMAP’s general costs, have been for the past few years supported by a grant from Ploughshares, and some aspects of this work have been supported by a grant from the Future of Life Institute.

  1. It is a general theme for me that NUKEMAP has had to change many times because of the ways in which map API companies “squeeze” their users over time with very high costs for a website as popular as NUKEMAP. For a while Mapbox gave me a steep discount because of the not-for-profit/educational nature of NUKEMAP, but a) in periods of very high use (like the Russian invasion of Ukraine) that still was very expensive, and b) they started requiring me to essentially re-apply for that discount on a regular basis and started wanting me to “do things” for them, like give them positive reviews on certain websites, and I just really don’t appreciate that kind of thing. Right now I am of course at the whim of the AWS and Cloudflare pricing schemes, which undoubtedly will change over the next decade or so, but at least my core “stack” is one that is inherently more portable because it is self-hosted. Protomaps for the win. [↩]
  2. Note: before someone chimes in (and someone always does these days) and suggests that I let AI do all of this work for me: a) how dare you, have some pride for your craft and labor; b) while I am generally pretty skeptical of AI making life easier for anyone other than the oligarchs who are trying to push it onto us, I am actually not totally opposed to using it in the same way I would Stack Overflow, which is to say, to produce well-defined functions that have long been “solved” (and are largely uninteresting to me) but are not at my fingertips (“write me a function in Javascript that inputs an array of bounding boxes and outputs a common bounding box for them all in Web Mercator”), but even there, my experience is that you have to be very careful because the first thing they put out often has bugs or incorrect assumptions (if you don’t specify “Web Mercator” it can give you a function that gives results that are quite incorrect when used in a web mapping API, for example), and I certainly am not interested in “vibe coding” an entire web framework (I have seen some other NUKEMAP-like knockoffs that were clearly “vibe coded” and they sometimes are very strange in their implementations). So, yeah, this isn’t what I am interested in doing, except sparingly. “Convert this function from Python to Javascript” is something I am happy to let a bot do for me, although even then I feel ethically uneasy about using these products, and one still needs to bulletproof-check the output. [↩]
News and Notes

New book: The Most Awful Responsibility

by Alex Wellerstein, published December 9th, 2025

This is just a brief post to make sure I have announced on here formally that my new book from HarperCollins is now on sale!

Cover of The Most Awful Responsibility by Alex Wellerstein

The Most Awful Responsibility: Truman and the Secret Struggle for Control of the Atomic Age (HarperCollins, 2025) is a new and provocative and possibly controversial take on Harry Truman and his role in US atomic policy, from Hiroshima through the Korean War (and everything in between). The research for it took well over a decade, and some of the things I’ve previously written about on this blog regarding Truman have been part of that process. But the book pulls together quite a lot of new source material, new interpretations, and covers a much wider scope than I’ve posted on here. So if you’re interested in these kinds of things — or have someone in your life who might be — check it out!

I’ve written up more about the argument of the book on my other blog, Doomsday Machines, which I will just link to here:

https://alexwellerstein.com/writing/books/the-most-awful-responsibility/

And if you’re not already a free subscriber to Doomsday Machines… why not? What more do you want, mermaids?

OK — that’s all for now! Thank you for your interest, attention, tolerance, support, et cetera!

Redactions

Manhattan Project fissile material inventories

by Alex Wellerstein, published October 10th, 2025

Understanding how much fissile material — enriched uranium and separated plutonium — was produced, and at what time, by the Manhattan Project is one of those seemingly-obscure technical questions that comes with a lot of important historical implications. It is what determined the scheduling of the bombings, for one thing. For another, it is how you might answer questions about what would happen if the war had carried on. And for yet another, it is a way to debunk the various conspiracy theories that periodically emerge about either the United States having atomic bombs before July-August 1945, or being dependent on German uranium, or what have you.

Unfortunately for us, information that relates to the production rates of fissile material was considered among the most secret nuclear information in the postwar and Cold War period, because from that kind of information you can estimate stockpile production capabilities and limitations. So even in a secret project, this kind of data has historically been kept even more secret than the rest of it.

The late John Coster-Mullen wrote that this photo shows: “These are the three lead-lined steel cases used to transport the six U-235 target insert discs that arrived on Tinian August 28 and 29 [1945]. Each case contained two discs. The scientist is shown securing the lids.” He lists the source as Los Alamos National Laboratory.

The Manhattan District History, General Leslie Groves’ secret, official history of the Manhattan Project, contained this information in two different “Top Secret Appendices.” I’ll handle them both separately since one is very straightforward and the other is not.

Enriched uranium

The one for uranium stockpiles is part of the book on the electromagnetic enrichment method, as the Y-12 Calutron plant was the final step in the enrichment process before the material was sent to Los Alamos. When the Department of Energy released a major declassification of the MDH in 2014, they failed to release a version of the appendix volume. The volume was, however, included as part of the 1977 microfilm release of the MDH. 

Cumulative Shipments of U-235 from Electromagnetic Plant to Los Alamos, 1944-1946

“Cumulative Shipments of U-235 from Electromagnetic Plant to Los Alamos, 1944-1946.”

The graph above is a cleaned up version of the graph that is included in the relevant volume of the MDH (Book 5, Volume 6, “Electromagnetic Operation: Top Secret Appendix”), with the background grid reduced in intensity, in order  to make it easier to read the data.

It’s an interesting graph, but it doesn’t exactly read itself. Looking at it closely, and the numbers included in the relevant volume, it is clear that they are being very literal about “U-235” here: this only measures the actual content of U-235 in the shipments, not the total uranium. As the uranium shipped at different times had different levels of enrichment, the actual total amount of enriched uranium on-hand is a bit different than what the graph suggests. Which is somewhat clear when one notes that at the point marked “Hiroshima” on the map, they had only shipped a bit more than 50 kg, when the Hiroshima bomb itself contained 64 kg of uranium (~80% enriched), and had been sent from Los Alamos to Tinian some time before the operation took place. 

Here is a table I made which better indicates the details of the uranium production, based on the data in the volume and a little bit of math.

Date U-235 (kg) Enrichment HEU (kg) Rate (HEU g/day) LB Bomb Equiv.
10/7/44 1.7 67.5% 2.6   0.04
10/21/44 2.3 68.9% 3.3 52.8 0.05
10/28/44 2.6 69.3% 3.7 58.2 0.06
11/19/44 3.8 70.6% 5.4 76.8 0.08
3/4/45 20.4 77.2% 26.4 199.9 0.41
3/25/45 24.9 78.1% 31.9 261.6 0.50
4/22/45 30.5 79.0% 38.7 241.2 0.60
5/20/45 35.6 79.9% 44.6 212.3 0.70
7/17/45 52.5 81.6% 64.3 340.4 1.00
8/12/45 66.7 82.6% 80.8 632.5 1.26
9/9/45 88.2 84.5% 104.4 842.4 1.63
12/2/45 182.1 88.9% 204.9 1,196.5 3.19
12/30/45 231.4 90.1% 256.8 1,853.3 4.00
12/29/46 1,041.0 95.0% 1,095.8 2,305.0 17.08

Above, “U-235” refers to the total content of the isotope U-235 in the uranium sent — the same as the graph above. “Enrichment” is the percentage of U-235 in the material sent. It is a little unclear in the source whether this is “the uranium sent in the last batch” versus “the total average enrichment of all uranium sent to Los Alamos up until this point,” but I suspect they are pretty similar numbers either way. “HEU” is a derived value from the U-235 and the enrichment — how much actual enriched uranium metal had been shipped, regardless of the enrichment level. So this is the number that lets you see that they definitely had enough metal on hand for the Little Boy bomb (64 kg) to be dropped on Hiroshima in early August. “Rate” refers to what the production rate of HEU was, in grams per day, and is a nice way to see how the rate of production improved over time (just as the “Enrichment” column lets one see how much better they got at enriching it), as they further debugged their plants and more parts came online. Lastly, “LB Bomb Equiv.” is just a simple measurement of how many Little Boy bomb cores the HEU number represents — it is just that number divided by 64 kg.

My estimate, based on the above data, is that you don’t have enough HEU for a second Little Boy bomb until mid-September 1945. Which is only slightly relevant, since, if the war had gone on, they likely would have not made more Little Boy bombs, but instead would have made composite (uranium-235 + plutonium) implosion bombs. But it does impact the question of “what would the atomic bombing schedule have been like if they only had gun-type bombs” (about two months between each bomb). 

Plutonium

For plutonium, it gets more difficult. The relevant volume (Book 4, Volume 6, “Pile Project – X-10 – Top Secret Appendix”) was not included in the 1977 release, but was partially included in the 2014 release. The relevant graph, though, was totally redacted:

A totally redacted page from the Manhattan District History, with a DELETED stamp repeatedly put across a blank piece of paper that is obscuring the content underneath

So that’s the end of the story, right? This is a graph too spicy for people like you and me, right? Right? As people of my generation used to say: “As if.”

I have to admit, when I see something like this — data that I don’t feel has any legitimate reason to be classified, and would be interesting/useful for my historical study — I take it as a bit of a challenge.

Now, I happen to know that parts of the MDH have been declassified and scattered across various databases and archives. Depending on who reviewed it, and when they did so, their interpretations of what could be released varied. So if one can locate multiple versions of its sections, one can sometimes reconstruct a “master” version with more content than any individual one. This is what the version of the MDH that I have uploaded to Archive.org (linked to in this post) is, and I have been doing this over the last ten years as I come across multiple copies. 

The little insight I had in this case was that there might be copies of the MDH in the Hanford DDRS database, which has been at a strange URL for a while now. And… I was right! There are actually two different copies of the relevant volume in there, and they both include the “plutonium production” graph, from a release of these volumes back in 1991. The only problem is that they are terrible microfilm scans, and so are barely legible. The best one (document D8719897) is this one:

A very hard to read graph of plutonium production

“Plutonium Production.” But a bit hard to read.

One can, at least, see the relevant lines. But what are they of? What is the scale? Better than nothing, but it “reads itself” even less than the uranium graph. There’s no bit depth there; it’s essentially all black and all white, so there’s no hope of drawing out the missing data by playing with the contrast. 

Judging from other similar graphs in the file, the two lines are at different scales; one is “Scale x 1000” and the other is “Scale x 1,” and then when line “A” wraps around on the right, it becomes “Scale x 2000.”  So if the units for “B” are kilograms, that would make the units for “A” in tons. Or if “B” was grams, “A” would be kilograms. Other graphs in the file have pretty standard scales, with each “square” having a height of 10 (of whatever unit). In the above, the visible squares are really twice as high, so each horizontal line corresponds to 20 “units.” 

Here is the raw data, expressed just as “A” and “B”, with the scale differences applied, based on my pixel readings of the graph (so there is some error there). I am reading each “date” value as the rightmost value of the date range, as that seems to be where the person graphing them made changes to the slope:

Date Curve A Curve B
Dec-44 1,397.8 0.0
Jan-45 2,580.6 0.3
Feb-45 5,161.3 0.5
Mar-45 7,311.8 0.9
Apr-45 15,053.8 1.6
May-45 21,828.0 8.5
Jun-45 30,537.6 13.7
Jul-45 49,462.4 20.2
Aug-45 70,000.0 32.8
Sep-45 97,096.8 44.5
Oct-45 120,215.1 64.6
Nov-45 145,161.3 86.1
Dec-45 164,946.2 102.0
Jan-46 180,860.2 121.8
Feb-46 197,311.8 136.7
Mar-46 221,075.3 151.3
Apr-46 235,914.0 167.3
May-46 249,784.9 184.1
Jun-46 258,064.5 198.7
Jul-46 271,505.4 211.8
Aug-46 283,440.9 226.6
Sep-46 294,408.6 236.3
Oct-46 308,172.0 246.1
Nov-46 320,000.0 255.7
Dec-46 330,752.7 269.1
Jan-47 346,451.6 278.9
Feb-47 356,774.2 287.3
Mar-47 371,828.0 295.9
Apr-47 383,010.8 302.0

My pixel counts are based on an assumption of scale that translates to 1 pixel being about 1.2 of the units of value, which gives some sense of the error bars of this kind of eyeballing.

I have graphed the above data here (without the 1000X scale applied, but the 2000X area is plotted as 2X):

The replotted data from Curve A (blue) and Curve B (orange).

Which lets us think about what “A” and “B” most likely correspond to. Some observations:

  • The percentage of “A” that “B” is changes over the course of the graph. Starting in May 1945, “B” is some 40% to 80% of “A,” not taking into account the scale differences (0.04% to 0.08% if you do take them into account).
  • You could also read “B” as possibly being offset in time by “A”: whatever (unscaled) value of “A” repeats in “B” some several months later. The fact that “A” seems to be repeated (at 1000X less scale) in “B” after 30-50 days seems meaningful.
  • Both numbers only go up, and never go down, which suggests to me that they must be cumulative, since the actual rates of plutonium production fluctuated a bit over this time period. 
  • Curve A starts in December 1944, and Curve B starts potentially in January 1945 (but its value may be 0), definitely in February 1945. 

So my guess, based on other of plutonium production records kept at the time, is that Curve “A” is related to tons of irradiated uranium processed, and Curve “B” is kilograms of final “product” (separated plutonium) isolated.

We have some data points we can compare these assumptions to to check them. Another document in the Hanford database (says that in early 1947, Hanford shipped 9.2 kg of plutonium in January, 8.6 kg in February, 9.6 kg in March, and 8.6 kg in April.1 Our numbers for Curve B above, when rendered into a “rate of change” per month for those months in 1947, come out with 9.8, 8.4, 8.6, and 6.2 — all close-enough look plausible for Curve B.

Los Alamos produced several inventories of their plutonium in 1945 that have been at least partially declassified. I have one of these; the Center for Disease Control released a graph of 20 datapoints of plutonium as part of their Los Alamos Historical Document Retrieval and Assessment (LAHDRA) Project in 2010:2

“Cumulative amounts of plutonium received [by Los Alamos] from Hanford in 1945.”

They don’t provide the raw data (boo!) but using my pixel-picking method I basically came up with this:

Date Plutonium (kg)
4/18/45 0.7
4/25/45 0.7
5/2/45 1.1
5/9/45 1.6
5/16/45 3.1
5/23/45 4.5
5/30/45 6.4
6/6/45 7.4
6/13/45 8.8
6/20/45 10.1
6/27/45 12.5
7/4/45 13.8
7/11/45 16.2
7/18/45 16.2
7/25/45 17.1
8/1/45 18.7
8/8/45 22.1
8/15/45 24.3
8/22/45 27.6
8/29/45 27.6

Which seems about right. (The report I have, for 8/29/1945, has the total as 27.5 kg.)

If we put that graph up along with our graph for Curve B, we get this:

LAHDRA (blue dots) versus Curve B (orange).

Which is interesting. They both have basically the same slopes, but Curve B is definitely higher than the LAHDRA curve. 

I have basically two hypotheses about this. One is that Curve B is the mass of plutonium in a different form, like plutonium nitrate, that might be heavier than the isolated pure plutonium metal that Los Alamos might have been using. I’ve poked around with this a bit, but I’m not really that convinced that is the issue.

The other, which I currently favor, is that we’re really just seeing the difference between plutonium separated at Hanford (Curve B) and the time it took to send it to Los Alamos. So Curve B has a value of about 1.6 on 4/30/45, and the LAHDRA data has 1.6 on 5/9/1945. There are other such “correspondences”; in general, there is what looks like ~10 day delay between the two curves. A section in the MDH’s “Top Secret Appendix” says that the normal schedule was for, once a week, Hanford to pack “all the cans available for shipment” into wooden boxes and ship them by truck to Los Alamos in a modest convoy. The convoy itself took 36 hours to reach New Mexico. So 8-9 days or so. At which point it would need to be processed and then added to the Los Alamos inventory. So those numbers more or less add up.

All of which makes me think that “Curve B” is indeed plutonium produced at Hanford — which is not necessarily the same thing as the amount of plutonium that was on hand and ready to use Los Alamos.

So if that is correct, we can generalize Curve B’s data for plutonium produced by Hanford:

Date Plutonium (~kg) Rate (kg/Pu/mo.) FM equiv.
Dec-44 0.0   0.0
Jan-45 0.3 0.3 0.1
Feb-45 0.5 0.2 0.1
Mar-45 0.9 0.3 0.1
Apr-45 1.6 0.8 0.3
May-45 8.5 6.9 1.4
Jun-45 13.7 5.2 2.2
Jul-45 20.2 6.6 3.3
Aug-45 32.8 12.6 5.3
Sep-45 44.5 11.7 7.4
Oct-45 64.6 20.1 10.4
Nov-45 86.1 21.5 13.9
Dec-45 102.0 15.9 16.5
Jan-46 121.8 19.8 22.0
Feb-46 136.7 14.8 24.0
Mar-46 151.3 14.6 25.2
Apr-46 167.3 16.0 27.0
May-46 184.1 16.8 29.7
Jun-46 198.7 14.6 32.0
Jul-46 211.8 13.1 34.2
Aug-46 226.6 14.7 36.5
Sep-46 236.3 9.8 38.1
Oct-46 246.1 9.8 39.7
Nov-46 255.7 9.6 41.2
Dec-46 269.1 13.4 43.4
Jan-47 278.9 9.8 45.0
Feb-47 287.3 8.4 46.3
Mar-47 295.9 8.6 47.7
Apr-47 302.0 6.1 48.7

This data is much less accurate than the uranium graph, because I am eyeballing the graph values for each month. But it looks about right to me. The plutonium (kg) column is the eyeballed data — as you can see, it is not really very precise. The “rate” is the change in those values change per month. And “FM equiv.” is just assuming that each core for a Christy-type Fat Man bomb is 6.2 kg. It should be kept in mind for thinking about stockpiles that a) the US “expended” ~12 kg of plutonium in the July and August detonations, as well as another ~12 kg in the tests of Operation Crossroads in mid-1946, and b) that any stockpile assumptions are complicated by the fact that after World War II, the US was also developing composite cores, as mentioned before, which did not use the same amounts of plutonium as the original “Fat Man” bombs.

One thing that jumps out from all of the data above is how different the production rates were between the uranium and plutonium methods. The uranium method just got better and better, in terms of quantity and quality (enrichment level). But the plutonium production rates bounce around a lot more. This is unsurprising, knowing the history — they had severe problems at Hanford with the reactors in the postwar, and even shut down B Reactor in March 1946 so as to preserve its longevity. 

Bringing it all together

We can somewhat haphazardly combine our two datasets now into one master graph for fissile material production by the Manhattan Engineer District:

Fissile material production (highly-enriched uranium and separated plutonium, both in kilograms) by the Manhattan Engineer District, 1944-1947.

I have thrown out the plutonium production for early 1947, just because that was under the Atomic Energy Commission, and because we don’t have the enriched uranium data to go alongside it. We can see two things here. One is that when plotted against the same axis, the dramatic differences in production rates are really apparent. One can readily see why the postwar Manhattan District, and the early AEC, took the issue of plutonium scarcity and uranium abundance so seriously, and why they were very invested in developing composite cores that could allow them to “stretch” their much more constrained plutonium supply further. 

We can also take this same data and re-plot it as the rate of production, in kilograms per month:

Fissile material production rates (highly-enriched uranium and separated plutonium, both in kilograms per month) by the Manhattan Engineer District, 1944-1947.

Which is choppy data (it is also interpolated between the “known” datapoints so it is at a constant date scale), but is very interesting. We can see how the rate of HEU production by Oak Ridge grew rather quickly, with an interesting dip immediately after the war, but then burst up to a high and mostly constant rate of around 70 kg/mo. by the spring of 1946. Hanford, by comparison, peaks in November 1945, and then decreases and stagnates, undoubtedly the result of the aforementioned production woes. But it is interesting to see them side by side in this way. To my knowledge this data has never rendered like this before.

To show how one can make some interesting and practical conclusions on the basis of this data, we can also convert this data into “weapon equivalent amounts.” Here is a little graph I have been playing with, which maps both the Little Boy equivalents of HEU (64 kg each) and the Fat Man equivalents of plutonium (6.2 kg each), as well as various speculative “composite core” quantities of plutonium and HEU mixtures:

Weapon equivalents of fissile materials 1944-1947. For “composite” cores, it assumes a 50:50 mixture of each (e.g. 2 kg Pu + 2 kg HEU).

Chuck Hansen suggests that the composite cores tested in 1948 “probably contained about two to two-and-a-half kilograms of plutonium; in addition to this amount of plutonium, each composite bomb core also contained a comparable amount of uranium.” In the above, it turns out to be entirely dependent on how much plutonium you assume, because the HEU quantities are generally in such excess of the plutonium that you can treat them as essentially unlimited except in cases where the plutonium content is low (e.g. 2 kg) and the HEU content is high (e.g. 8 kg). Anyway, this graph is just one example of what you can “do” with this data: to show how the fissile material supplies are intertwined with questions of possible stockpile size, and how certain design choices affect those outcomes. 

What a lot of effort to reconstruct a rather unsurprising (but still interesting) graph of data from some 80 years ago! I cannot see how any of the above could have compromised national security in any way. But I will further note that all of the above was accomplished on the basis of records that the US government actually already judged, in the past, as being declassifiable. So if somehow it did compromise national security in some nebulous way, the fault is on the redactors in either 1977 or 1991 who decided — I think quite correctly — that it was innocuous data. 

  1. “SF Materials Shipped or Received from 12/31/1946 thru 05/01/1947,” (May 12, 1947), Hanford DDRS access number D8265435. See also this cover letter for a little more context. [↩]
  2. Center for Disease Control, Final report of the Los Alamos Historical Document Retrieval and Assessment (LAHDRA) Project (November 2010), chapter 4, page 5. [↩]
News and Notes

News and updates

by Alex Wellerstein, published September 29th, 2025

I realized I haven’t updated things here for a long while, and that it would be worth consolidating a few overdue news updates.

First and foremost, if you want to read things from me on a more regular basis, you should be reading Doomsday Machines, which is another blog of mine, and is much more frequently updated that this one. While it is not exactly the same content or approach, you’ll find a lot of things from this blog replicated on there. Because the same person writes both of them. (I still intend to update this blog occasionally, but one only has so much time.)

Cover art for Alex Wellerstein's The Most Awful Responsibility: Truman and the Secret Struggle for Control of the Atomic Age

Second, I have a new book coming out in December 2025 from HarperCollins: The Most Awful Responsibility: Truman and the Secret Struggle for Control of the Atomic Age. You can think of it as a sort of “atomic biography” of Harry Truman, with about half of the book focused on the use of the atomic bombs during World War II, and the rest about the non-use of atomic bombs during the rest of his administration. My basic argument is that Truman was, somewhat surprisingly, one of the most anti-nuclear presidents of the 20th century. In other words, it is provocative, but extensively researched. Keep an eye out! And if you’re the kind of person who hosts a podcast or book clubs or book talks or whatever, feel free to get in touch. 

Third, and last, for at least 2025-2026 I am living in Paris, France. I am currently a visiting researcher at the Nuclear Knowledges group, at the Center for International Studies, in Sciences Po, working with the wonderful Benoît Pelopidas. I am still an associate professor at Stevens (I am on leave). I’ve been here about a month. I’m just putting this out there because it takes time for news to diffuse, and also to let anyone in Europe know (who might, say, want me to give a book talk) that I’m in your orbit. And yes, Lyndon, my dear old dog, has come with us, and is enjoying it quite a lot. 

OK, that’s all for now! I am busy with a number of projects. Again, if you want more regular posts, subscribe to Doomsday Machines, which is free. If you want to learn more about what I’m doing from week to week (or just support me, or just get more photos of the dog), you can also sign up for the paid subscription, but it is absolutely not required!