SIOP//62

Grid/84 is not fundamentally a mapping application. It is a world-state playback engine: a clock, a set of entities whose state changes over that clock, and a globe that renders the state at any instant. The Terminal Atlas is that engine idling with a single entity, the camera. Routing adds moving entities on real networks. SIOP//62 adds thousands of entities, each with a documented origin, a reconstructed path, and a modelled consequence, and asks the engine to play the most consequential timetable ever written.

The distinction from DEFCON is that this is not an abstract strategy game: it is a specific historical plan, on real geography, with real order of battle. The distinction from NUKEMAP is that it does not show one isolated detonation: it attempts to make the plan unfold globally, with readiness states, launches, converging trajectories, designated ground zeros, overlapping effects and the steadily changing state of the world.

The subject is the transformation of apocalypse into systems engineering: targets as records, populations as estimates, cities as overlapping effect radii, annihilation as a progress bar. The study renders that with bureaucratic calm because the calm is what the documents actually sound like.

PLAN EXECUTION            00:17:42
OPTION                    1 · ALERT FORCE
WEAPONS COMMITTED         1,685 (JCS 1961) · >1,700 (JCS est. 1961)
DESIGNATED GROUND ZEROS   1,060 · BY COUNTRY: WITHHELD
ASSURANCE                 75% MINIMUM · 85% ACHIEVED · 95% TOP PRIORITY
OUTCOME CALCULATION       IN PROGRESS

The United States has never declassified any complete SIOP. Historians work from the briefing given to President Kennedy on 13 September 1961, from the planning staff's own excised history, from later reviews, and from an earlier target study that shows the method. The complete plan cannot be reconstructed, and the study must never pretend otherwise.

That incompleteness is the visual language. Every entity carries an evidence tier and a source reference, and the renderer derives its line grammar from the tier. Nothing is drawn without provenance, and provenance is one hover away.

Two rules follow. Redaction is a datum. A blacked-out column on a released page proves that a record exists at that position; it renders as a positive mark with its own citation, never as an empty slot. Disagreement is carried, not resolved. Where documents give different totals, the readout shows both with their sources. Picking one would be a claim the record does not support.

Primary and archival sources

Figures, with their disagreements

The 14-versus-16 options and 1,060-versus-1,043 ground zeros discrepancies are between the briefing as delivered and the planning staff's own history. Both are primary. The study shows both.

Order of battle, 1961 to 1962

All of this is documented tier from published unit histories and weapon records, and it is what makes the launches real rather than decorative.

Open: a per-base bomber and tanker count for mid-1961. The Norris and Cochran NRDC tables and SAC unit histories will supply it.

Literature

The academic record is deep, and the study should stand on it rather than on the archive alone. Titles marked "core" are the ones the evidence contract is built against.

The National Security Archive keeps a select literature list that should be read before this table is considered complete.

The globe is 2026 OpenStreetMap. In 1961 the frontiers were different and Gorky, Sverdlovsk, Kuibyshev and Stalingrad existed under those names. Three things fix this:

• The 1956 study carries its own coordinates, in degrees and minutes, alongside period names. A target transcribed from it needs no geocoding. Its coordinates are documented tier; the modern name is an annotation.
• historical-basemaps has world_1945.geojson and world_1960.geojson under GPL-3.0, with a BORDERPRECISION field and the author's own warning to verify against other sources. The 1960 borders are drawn as a documented-with-caveat layer, precision shown.
• HYDE 3.3 (Klein Goldewijk, Utrecht University, 2023) supplies gridded population at 5 arc minutes in ESRI ASCII format. The Utrecht publication page states the licence as CC BY-NC-SA 4.0, not the CC BY that a mirror claimed; the study is non-commercial and share-alike is acceptable, and the attribution goes on the readout. The download sits behind a bot check that a browser passes.

Section 1 of the released study, OCR-read and quoted with its spelling corrected, states the method the target lists serve:

• The mission is executed "through the execution of the following tasks, in the order indicated: (1) Win the Air Power Battle by destroying SovBloc Air Power. (2) Destroy systematically SovBloc war-supporting resources." The airfield list is the first task; the city list, "systematic destruction", is the second.
• Damage criteria: "a 90% probability of collapse of all above-ground structures, whether on airfields or in areas containing Air Power Battle targets"; "a 70% probability of damage on individual targets if the total average for any complex amounts to 90%"; and "the criteria in the preceding sub-paragraph are less in Satellite areas (50%–70%)".
• On yields: "every effort has been made to restrict the use of high yield weapons on Air Power Battle targets in the European Satellites in consideration of political, psychological and offensive and defensive implications, provided damage criteria and economy of force are maintained."
• The listings are organised as a cross-reference list, the category code list (section 3, now transcribed to data/siop62/categories-1956.json), an airfield list with weapons, and a complex list with weapons, each in an unrestricted and a restricted allocation, the latter of 1,203 DGZs.

This is the question that decides scope, and the answer is favourable.

The city list is a 306-page PDF of typewritten, monospaced, cleanly scanned pages with no text layer. The row grammar is visible on the page: a priority number, a complex number, the period name in capitals, coordinates as DDMM-DDDMM, then indented rows of three-digit category codes with Bombing Encyclopedia numbers. Population rows carry category 275. On every page a box on the right covers the weapons column, printed with its statutory exemption under the Atomic Energy Act. That box is the first withheld-tier datum in the study.

• First pass, 9 September 2026. All 306 pages are transcribed to data/siop62/city-list.jsonl, every record with its raw OCR line and a confidence flag. Totals after three passes: 1,697 complex rows, 9,065 category rows, 1,025 additional ground-zero rows, 2,787 unparsed lines still to work through, and the weapons-column redaction read on 217 of 306 pages. The checker geocoded 30 random period names through Photon: 22 landed within 10 km of the 1956 coordinate, the rest were ambiguous or renamed places or OCR damage. Nineteen complexes fall outside a Sino-Soviet-bloc bounding box, all recognisable digit confusions in the scan (5 read as 9, 1 as 3, 0 as 9). None of this is corrected automatically; the verification queue is the unparsed lines, the low-confidence category rows, the out-of-box coordinates and the spot-check flags.
• Trial. Tesseract 5.5 at 300 dpi with page-segmentation mode 6 reads the typewritten rows cleanly. Complex rows such as 1045 0230 ANADYR 6444-17728 and 131 0249 AN SHAN MANCH 4107-12257 parse with their coordinates; category rows parse with occasional digit confusion in the Bombing Encyclopedia number; margin noise from the scan is the main source of rejected lines. Some complexes carry no priority number. The list is alphabetical by name, not by priority. The script is scripts/transcribe-1956-city-list.py; it writes JSONL with the raw line and a confidence flag on every record and records the redaction box on every page as a withheld datum. Sanity checks to add: coordinates must fall inside the named country's 1960 borders, and a hand-checked sample of rows against the scan.
• The Future of Life Institute digitised about 1,100 city targets from this release and Alex Wellerstein rendered them with NUKEMAP in 2016. No download or licence for that transcription was located. Ask before reusing; never scrape the visualisation.
• The 1956 study is a proxy. It shows the method and the categories of the 1961 target list, not the list itself. Targets from it render as documented-1956 and reconstructed-1961, never as SIOP-62 fact.

The models are public even though NUKEMAP's code is not. NUKEMAP's own FAQ names them, which is the best possible validation set.

The outcome readout carries two numbers: the blast-only figure by the planners' own OTA/DCPA method, and the mass-fire bound after Postol, in which everyone inside the third-degree-burn radius is counted dead. The gap is the subject. Casualty fractions by band come from OTA 1979 figure II-1: 98 percent dead above 12 psi, 50 percent dead and 40 percent injured from 5 to 12, 5 and 45 from 2 to 5, none dead and 25 percent injured from 1 to 2. The 12 and 2 psi radii are log-log interpolations between Sublette's tabulated constants.

Most of the study is subsystems, and each subsystem earns its place through a lab before it touches the execution study. A lab is a self-contained specimen page with a pure model module, a visual grammar, a validation table against published values, and a declared fidelity ceiling. Fidelity ceilings are declared, not discovered. The point is to establish the lines of the possible, not to build a weapons-effects code.

The labs are also where the register is tuned. A plume that grows in silence over Sverdlovsk while a counter ticks up is the whole study in miniature, and it should be tested as a specimen before the full plan is attempted.

Absolute bureaucratic calm. Fixed-width readouts, upper case, no exclamation, no explosion sprites. Things happen on schedule. Numbers carry their source. One line from the 1956 study, if it survives verification, is shown in the study's own words rather than paraphrased: the category list entry for population. It needs no help.

The first deliverable is not the plan. It is one of everything, across every tier:

1. One documented base: the 564th SMS at F.E. Warren, six Atlas D, W49.
2. One documented sortie: a Chrome Dome northern-route B-52 with its refuelling track.
3. One transcribed 1956 target with its coordinates and category rows, rendered documented-1956 and reconstructed-1961, with the weapons column shown as a withheld datum citing the page.
4. One modelled consequence: prompt effects and a plume for one weapon at that target, with the two-number exposure readout.
5. One ghost: a ground zero known only to city level, inside its uncertainty ring.

Exit: the proof plays from H-hour on the Terminal Atlas clock, every mark on it can be traced to a page, and the redaction is visible.

Built, 9 September 2026. The proof runs. The Atlas D launches at H+15:00 on the documented reaction time, flies a minimum-energy arc of 21.6 minutes to apogee 1,026 km, and detonates at H+36:38 with the prompt-effects rings drawn on the Anadyr estuary; the camera descends to the target in the terminal phase. The Chrome Dome B-52 flies its reconstructed loop throughout, passing Thule half an hour before H-hour, its SIOP target logged as withheld. Every mark opens a provenance panel with its facts badged by tier. Population exposure is shown as not computed. Population exposure now computes at impact from the HYDE 1961 grid: the Anadyr weapon gives about 7,000 dead by the blast-only 1961 method and 8,500 with mass fire, both labelled modelled with their sources. A burst switch makes the weapon a contact surface burst: the overpressure rings shrink to the contact-burst radii (Glasstone §3.34 with the Kinney & Graham free-air fit), the blast-only count falls to about 6,600, a plume grows downwind on the clock under an assumed 15 mph wind, and four days after impact the log reports about 39,000 people under the 1 rad/hr contour and 2,900 acute fallout deaths with no shelter, badged inferred. Not yet built: the Soviet response.

The engine is general and SIOP//62 is the jumping-off point. Grid/84 as a whole is an interactive exploration of the history and doctrine of strategic nuclear weapons, and the candidate studies below each need their own brief before any code. Each has a documentary spine already.

The rule from the manifesto holds for all of them: impractical, never fake. A study that cannot name its sources on the readout does not get built.

#/study/siop62-alert plays option 1: the force that could launch on fifteen minutes' warning, from the mid-1961 order of battle to the 1956 study's highest-priority complexes.

Tiers. The launch sites and the totals are documented; the order of battle is built by scripts/build-order-of-battle-1961.py from Wikipedia's B-47 unit list and strategic-wing list, the Atlas, Jupiter and Polaris records, and the theatre totals of Table 1, with positions read from the modern map. The strength on each base is reconstructed by stated rules: 45 aircraft for a heavy wing, 15 for a dispersed strategic wing, 45 for a B-47 wing, two weapons per B-52 and one per B-47, and one alert fraction across the force chosen so the SAC bases carry the documented 1,212 aircraft weapons. That fraction comes out at 0.48, which is the briefing's "approximately half" recovered rather than assumed. Every weapon-to-target assignment is inferred by one rule, highest priority first and nearest launcher in range with missiles before bombers, and the study's subtitle says so.

What it shows. The documented pathology becomes a picture: the missiles land within the hour and the bomber stream is still crossing the pole eight hours later, so half the targets are struck after H+8 and the last after H+14. The sum over the 1961 grid, computed in four workers as the clock runs, brackets the planners' own figure: 53 million by their blast-only method as a union over the grid, 61 million summed per target, against their 80 million, and 110 million with fire, 140 summed. The enacted figure covers the whole bloc including China, North Vietnam and Eastern Europe, while the JCS figure is Soviet dead only, so the comparison is a bracket, not a match.

Attrition is statistical and calibrated, not invented: each system has a reliability, Atlas's documented and the others inferred beside it, and bombers also have a penetration probability, the one free number, solved so the weapon-weighted average delivery equals the documented 85 percent. It comes out at 0.95. Each sortie's fate is a deterministic hash of its identity, so the same sorties are lost every run; reliability losses happen at launch and penetration losses in the last part of the route, where the track simply ends. The readout states the reliabilities and their tiers.

The Soviet response is drawn from documented force levels and inferred everything else, and says so on every mark. Sagan's Table 4 for September 1961 gives 10 to 25 ICBMs with none on alert and one to three hours to fuel (the R-7 needed twenty hours' preparation), about 78 to 90 submarine missiles that had to close to 150 to 350 nautical miles and surface, and about 200 bombers that could be put over North America, of which the NATO planning material put 10 percent on ground alert with Schelling's caveat that the figure may be an artefact of the game. So the missiles and submarines cannot launch within the day, and the log says so; the bombers can. The 200 are spread over six inferred Long Range Aviation fields; the alert tenth launches thirty minutes after the first US launches, the rest generate in three hours and fly only if no delivered US weapon has landed within 30 km of their base by then, which couples the response to the strike above. Their targets are the SAC launch sites of the order of battle and the ten largest cities of the 1960 census; their yield is assumed at 3 Mt; US air defence, for which Sagan notes no official estimate exists, is assumed to stop half.

Lemnitzer's warning is the finding: the response comes anyway, from the bombers that were already alert or at fields the missiles did not reach in time.

What it omits, on the readout: refuelling and routing; Soviet air defence and any Soviet response; about six in ten of the 1,100 airfields, which the transcription does not yet read; per-weapon yields, with Mk-28 class assumed for bombers; and overlap between targets, which the summing counts more than once. Airfields sit inside or beside the cities whose complexes are also struck, so the airfield weapons add little to the death count and much to the picture: the Air Power Battle is the first hour of the day.

The fourteen options. #/study/siop62-alert/<n> plays any execution option. Each launcher grows on a straight line from the alert tables to the fully generated tables of 15 July 1961 by a stated rule per system: every SAC bomber generated by fourteen hours, 24 of 78 ICBMs on alert (Sagan p. 29), two of five Polaris boats on station, the Jupiters already on alert, and the theatre aircraft and cruise missiles in proportion to Tables 1 and 3. Option 14 comes to 3,267 weapons with 188 ballistic missiles, which are Table 3's figures exactly: 78 Atlas, 80 Polaris and 30 Jupiter. The finding the briefing hints at and the enactment makes plain is that the options barely differ. Every option strikes the same 1,312 targets, the first detonation is the same Atlas at H+21 minutes and the last the same B-47 at H+17 hours, because the target list and the bombers' flight set the day's shape, not the size of the force. Preparation time adds weapons per target, from two to three, and megatons.

The megatonnage is the visible disagreement: the documented 7,420 Mt for the generated force implies the Mk-36 and Mk-41 class weapons the non-alert bombers carried, which the enactment does not assign, and the readout says so at every option above the first. The Soviet response stays the tactical-warning one; a Soviet force generated during the same hours of strategic warning is not modelled, and that omission is stated too.

The sum. Each side's headline is a union over the population grid rather than a sum over targets: every grid sample is claimed once, by the most severe blast band any detonation reaches it with, and flagged if any fire zone covers it; once a plume has fallen, the sample carries the highest dose any plume gives it. Deaths are taken in sequence per sample, blast by the 1961 bands, then fire as Postol's bound, then acute fallout among those the first two left alive, and the readout's last cell is all three together, one minus the product of the survivals. The per-target lines in the log and the provenance panel are still summed separately, and the readout shows what the sum would have said, so the size of the overlap is visible: for the alert force, 61 million by blast summed against 53 million in the union.

Rendering. A study with more than a hundred tracks draws them through a WebGL custom layer rather than GeoJSON sources (src/map/track-layer.ts). Every track is densified once into a static vertex buffer with the time it reaches each vertex; a study tick rewrites only an index list of the flown segments, one head segment per moving vehicle, and one point per vehicle. Projection is MapLibre's own shader prelude, so the lines follow the globe and its transition to mercator. The evidence grammar survives the change: each tier keeps its colour, vehicle marks keep their dark fill, tier-coloured ring and halo, and aircraft draw as a silhouette turned to the track's heading so a bomber stream reads apart from a missile salvo. Ballistic sorties are drawn as trajectories rather than ground tracks: each carries the height of the minimum-energy arc at every point, an Atlas to Anadyr rising past a thousand kilometres, and the layer projects the height through MapLibre's three-dimensional projection under both the globe and the plane, so an arc seen from a pitched camera leaves the surface and comes back to it, and the vehicle's label rides up with it. On the alert force this costs 1 to 6 ms of main-thread time per update against 65 to 95 ms just to build the GeoJSON features it replaces.

• Which 1956 targets can be tied to 1961 ground zeros, and by what stated rule? Airfields and the largest cities are the safest reconstruction.
• Does the Future of Life Institute transcription exist as data with a licence, or is a fresh transcription the cleaner path?
• The Soviet response is now enacted from documented force levels with inferred bases, targets and air defence; the open question is whether any documented Soviet targeting or basing for 1961 can replace the inferred parts.
• Publication review. The study is history and its inputs are public, but the register is deliberately disturbing. Decide the framing text before anything is public.

• Sagan, Scott D. "SIOP-62: The Nuclear War Plan Briefing to President Kennedy." International Security 12, no. 1 (1987): 22–51. Archive.org copy.
• National Security Archive EBB 236, 538, 638 and 798, linked above.
• Wellerstein, Alex. "Mapping the US nuclear war plan for 1956." Restricted Data, 9 May 2016, and the NUKEMAP FAQ.
• Future of Life Institute, "1100 Declassified U.S. Nuclear Targets."
• Klein Goldewijk, K., et al. HYDE 3.3, PBL / Utrecht University, CC BY 4.0.
• Ourednik, A. historical-basemaps, GPL-3.0.
• Glasstone, S., and P. J. Dolan. The Effects of Nuclear Weapons, 3rd ed., 1977.
• Miller, C. F. Fallout and Radiological Countermeasures, SRI, 1963.