Army Encyclopedia MAPS

About the method

Every figure of the Analysis tab on the map is derived from the equipment cards and from the table of published norms below. Nothing is estimated per country by hand, and the same engine serves every country, My Army and VS. Words such as equivalent, could and by published norms are deliberate: these are models of what an inventory could form or cost, not the real order of battle.

Status of a norm: sourced = taken directly from a published table of organisation or an official report; derived = arithmetic on sourced figures; low-confidence source = only a weak source was found; assumption = no source, a stated working value. Norms table dated 2026-10-19.

What each block shows

Formation equivalents

Equivalents = units ÷ the norm of the category (for example tanks ÷ 36 = tank battalion-equivalents). Manoeuvre brigade-equivalents = (tank + mechanised battalion-equivalents) ÷ 3; division-equivalents = brigades ÷ 3. Motorised (protected mobility) battalions (MRAP and protected 4×4) are shown but not added to brigades. Surface-to-air systems are converted with a norm per family (Patriot 24 per battalion, S-400 8, Buk 9, Tor 4 per battery …); families without a line use the fallback 8 per long-range battalion or 4 per battery. Navies are listed by class with the crews they need. The result is an estimate of what the inventory could form — not the actual order of battle, training, doctrine or losses.

Manning & readiness

Direct personnel = Σ units × (crew + maintainers) from the operating-cost table of the map, with three rules: light 4×4s, MRAPs, trucks and ambulances have pooled crews (crew × 0.3, maintainers × 0.5 — their drivers are the infantry and support soldiers already on strength); mortars, anti-aircraft guns, ATGM, MANPADS and anti-tank guns are manned by line infantry (crews × 0.25); crews of auxiliaries, patrol vessels and landing craft follow the size of the hull (from the unit price). Support personnel = direct × 0.6 (staffs, logistics, medical, training — derived from tooth-to-tail ratios); required = direct + support. Manning = people on strength ÷ required, calibrated so that the United States — the best-documented inventory with a known personnel figure — comes out near 100 %. Effective fleet = units × availability × min(1, manning), with availability 0.65 for aircraft (USAF mission-capable average), 0.36 for ships (US Navy share at sea) and 0.70 for ground equipment (assumption). In My Army the readiness setting low / normal / high replaces the defaults.

Budget realism

Inventory value = documented unit prices × units (My Army: estimates too); the share of priced types and units is shown next to it. Yearly operation & support per unit = the unit price × the category rate of the cost table — the documented price where there is one, otherwise the same estimate that My Army uses (shown as «O&S from estimated unit prices», never used for the inventory value), otherwise the category default — or the light-platform default of the cost table when the type is a turboprop, a boat or a light truck; a drone type with a unit price under $1 M (or, when no price is known, recorded in quantities of 200 or more) is a small tactical or loitering drone ($20,000 a year, 1 operator), and the equipment-family overhead is capped at 30%. Personnel cost = people on strength × an assumed all-in cost per service member — a model tier (low $15k / mid $45k / high $120k, never a salary) fitted to the budget: the highest tier at which personnel stays within 45% of the annual budget; if even the low tier exceeds it the army is flagged «budget too small for the personnel at any tier». Implied procurement share = 1 − O&S share − personnel share; years to re-buy = inventory value ÷ (budget × procurement share). Benchmarks: US FY2025 request and the NATO guideline. Where rear categories are not documented the figures are a lower bound.

World standing

Ten metrics ranked among the documented countries — those whose inventory is loaded on the map; the denominator grows as countries are added. Closest armies: every inventory becomes a vector of the square roots of its category-group shares (about thirty groups, mass categories damped as in the index), normalised to length 1; the three armies with the highest cosine similarity are shown — the shape of the inventory, not its size.

Inventory Strength Index

Five blocks — Land, Air, Sea, Air defence & missiles, Enablers (with active personnel). Block score = Σ weight × units over its categories, with quality factor 1 for every type in this stage (price is a poor proxy of quality; a modifier comes only after the year of introduction and characteristics are collected). Mass categories — light vehicles, MRAPs, trucks, ambulances, patrol boats, auxiliaries, landing craft, mortars, anti-aircraft guns, ATGM, MANPADS and small drones — enter on a square-root scale above 100 units (1,000 count as 316, 10,000 as 1,000), so that 160,000 light vehicles cannot outweigh tanks and artillery. A documented type without a published quantity counts as 1 unit, a lower bound. Each block is divided by the best documented country and scaled to 0–100; the index is the mean of the five blocks. Nuclear warheads are shown separately and never enter the index. Enablers are computed only over documented categories; when none is documented the block is n/a and the index is the mean of the other four.

Why publish the weights: the well-known rankings either do not disclose their formula (Global Firepower, more than 60 undisclosed factors and unexplained jumps between years) or weight population and industry as heavily as forces (Correlates of War CINC, six equal shares). This index is a transparent arithmetic score: move a weight and the result moves, and everyone can see by how much. It measures inventory, not training, doctrine, readiness, allies, geography or will — like every such index.

A gap in the data is not a gap in the army

Open sources rarely list the rear: trucks, engineering, medical, command and CBRN vehicles. For a real country such a category with no card is shown as not documented in open sources — it enters neither the structure score nor the Enablers block, and the manning and cost figures are marked as a lower bound. Combat categories (tanks, artillery, air defence, aircraft, ships) are always listed by open sources, so a zero there is a real absence. In My Army every zero is real: the user did not add it.

Stage 2 — per-type figures, coverage and reach

Stage 2 adds one documented figure per type where a source states it: engagement range of air-defence, anti-tank and coastal systems, detection range of radars and AEW aircraft, combat radius of combat aircraft and helicopters, range of ballistic and cruise missiles and rocket artillery, endurance and speed of major warships, and the year each type entered service. The figures were collected by a research pass over the Wikipedia infobox of every type (a reputable defence source where the infobox is silent) and checked by an independent second pass; a type without a documented figure keeps null and is listed on the tab as not counted — nothing is estimated. Combat radius is kept as the source gives it: an explicit radius, a published combat range, or a plain range, which the map divides by three (the sourced radius-of-action rule) and labels as such.

Coverage of the figures at this build: engagement range 218 types, detection range 115, combat radius 104, missile range 78, ship endurance 51, year of introduction 585 — out of 832 types with at least one documented figure.

Air-defence coverage

Systems are grouped by engagement range into short (under 30 km), medium (30–100 km) and long (100 km and more). Batteries = launchers ÷ launchers per battery (the family norm of sam_units where defined, otherwise 4 for short-range and 6 for longer systems). Area covered by a tier = batteries × π × range² × 0.6 — the 0.6 stands for overlap between batteries, terrain and gaps (assumption); coverage = min(1, area covered ÷ area of the territory), the area being computed from the country outline on the map (spherical polygon area) or from the My Army territory. Against low-flying aircraft and drones the useful radius is limited by the radar horizon, 4.12 × (√h_radar + √h_target) km with a 10 m mast and a 30 m target ≈ 36 km, whatever the missile range — so the tab shows every tier twice: at altitude and against low-flying targets. MANPADS, anti-aircraft guns and counter-UAS systems defend points and columns, not territory; they are counted as point defence and never added to the covered area.

Strike reach

The tab lists the systems with the longest documented ranges: combat aircraft by combat radius, ballistic and cruise missiles and rocket artillery by missile range, coastal systems by the range of their anti-ship missile; with tanker aircraft on record a second, dashed ring shows the radius × 1.5. For a real country the rings are drawn from the border — a buffer of the outline by the range, read as from any point of its territory the system reaches this far — because the markers on the map stand in formation, not at bases; the rings are never drawn from the markers. In My Army the rings are drawn from the markers the user placed, so a self-built army gets a real coverage map (air-defence rings too, with a dashed low-altitude ring). At most five systems get a ring (two per kind); the rest are listed.

Fleet age

For every category with a service-life norm the tab shows the average year of introduction of the types on record (weighted by units), the share of units whose type entered service more than 30 years ago and the share that will be past the service life of its category by 2036 — destroyers 35 years, carriers 50, submarines 33 and helicopters 25 are sourced, the rest are assumptions. Only types with a documented year count, and the line says how many types of the category have one. Upgrades and life-extension programmes are not tracked: past service life means due for replacement or modernisation, not unusable — the M1 and the F-16 show that airframes and hulls serve 40–70 years with upgrades.

Two rules added after the stage-2 review: a helicopter figure above 400 km is a range, not a radius, and is halved (jets: a published combat range above 2,500 km is a total range and is divided by three); the fleet-age table shows the year the TYPE entered service, not the airframes — fleets built long after the debut of the type (Mi-8, F-16, Tu-95) look older there than they are.

Transport helicopters are not strike systems: their combat radius is collected for the fleet-age table but is never listed under strike reach or drawn as a ring; attack helicopters stay (Eduard, 30.09.2026).

Stage 3 — consumption and days of stock (My Army only)

Shown only in My Army: for a real country the figures would be read as its actual stocks, which nobody publishes. Inputs: intensity (low = peacetime training, medium = sustained operations, high = major combat) and days of stock (30 by default — the NATO stockpile planning period). Fuel per day = Σ vehicles × litres per 100 km of their category × km driven a day; the MBT figure (300 l per 100 km on roads) is published for the Leopard 2, the other classes are scaled from it, and cross-country consumption is 1.5–2 times higher, so the result is a lower bound. Artillery rounds per day = self-propelled and towed guns × rounds per gun (6 = the Ukrainian 2025 fleet average, derived from 5,000+ rounds a day over about 900 155 mm guns; 30 at high intensity, 2 at low). Sorties per day = combat aircraft × sorties per aircraft (1.0 at high intensity — the Desert Storm average; 0.5 in peacetime; 0.75 at medium). Stock for N days = daily burn × N, shown in tonnes of diesel (0.84 kg per litre), rounds and sorties. Mortars, rocket artillery, tank guns, helicopters, drones and ships are not modelled. The three intensity assumptions (50 km a day, 30 rounds at high intensity, 0.5 / 1.0 sorties) were confirmed by the author on 30.09.2026; the low and medium interpolations are working values of the same kind.

High is the tempo of a protracted major war (Ukraine 2022–25), not the doctrinal peak of the first days of combat: US planning norms for intense combat give several times more rounds per barrel. The three values were approved as they stand by Eduard on 30.09.2026; the status stays assumption.

Live War model

In VS the button «Live War» runs an attrition model for N days (30 by default): both inventories wear down day by day, the markers of exhausted types fade from the two outlines, each side shows the share of its inventory that remains. No winner is declared: the result is «remaining inventory after N days — A x %, B y %», a model of how fast two inventories would wear down at a given tempo, computed from the cards and the Analysis norms alone.

Composition and block sizes

A side is the index list of its cards (a type without a published quantity counts as 1 unit, except a card marked «counted elsewhere» — its units sit in another card's figure, such as the missile types of Iran whose launchers are counted by family) — for My Army, the types the user added. Only types with a block in the index enter the model; personnel, reserve stocks, orders, nuclear and space assets stay out. S_g = Σ weight × damped units over the categories of block g (the Inventory Strength Index arithmetic without the personnel term, mass categories on the root scale). Remaining = Σ_g 100 × S_g(t) ÷ Wmax_g divided by the same sum on day 0, over the blocks the side has — i.e. the index score that remains; per block simply S_g(t) ÷ S_g(0). Units lost = Σ (q(0) − q(t)) over the types.

Power of a block

P_g = S_g × A_g × M × I: A_g is the availability norm (ground 0.70 for land, air-defence and enablers; 0.65 for air; 0.36 for sea; My Army with a readiness setting uses its value for every block), M = min(1, manning) from the manning block (1 when the strength is unknown), I = the tempo multiplier (sustained 0.03, campaign 1, sprint 1.85), multiplied by stock_cut once the day count passes the side's days of stock (30 — the NATO planning period; My Army uses its own figure). P_sam is the same product over the SAM categories only — the launchers of ballistic and cruise missiles neither shoot at aircraft nor enter it; the missiles themselves are a stock (below).

P_cmb is the same product over the combat aviation only — fighters, attack aircraft, bombers and attack helicopters (cmb_cats): it is the air power that strikes and fights in the air. Drones, transport, trainer, patrol, AEW and tanker aircraft stay in the Air block for the losses but do not strike (tanker aircraft still extend the reach × 1.5). Air superiority sup = P_cmb(s) ÷ (P_cmb(s) + P_cmb(d)), recomputed every day from the combat aviation that remains, scales the strikes of side s on land, air defence and sea; air → air stays symmetric.

Gates — who can reach whom

dist = the shortest distance between the two outlines (0 when one lies inside the other), from the outline points sampled every 25 km. Ground contact: automatic when dist ≤ contact_km, or forced on / off by the switch. Aviation: g_air = clamp(1 − dist ÷ reach, 0, 1), where reach is the longest documented combat radius of the side (× 1.5 with tanker aircraft on record); a side without a documented radius reaches only an adjacent opponent. Missiles: a ballistic or cruise missile type launches when its own documented range covers dist (a type without a documented range follows the side's longest one; rocket artillery belongs to the land block and fires only in ground contact). Fleets meet only when the outlines lie within sea_km (500 km): navies of distant countries do not meet in the model — beyond it sea → sea = 0 and ships are reached by aviation alone (g_air × sup). The defender's air defence cuts incoming air strikes by g_ad = 1 − ad_k × cov, cov = the larger of its long-range and medium-range SAM coverage at altitude, recomputed every day from the SAM quantities that remain — so SEAD opens the sky gradually.

Daily losses

For the defender d and each block g the share lost in a day is f_g = min(cap, Σ terms ÷ S_g(d)) with a ceiling of cap (10 %) a day. The ground-fire and the fleet terms are k × gate × P_attacker — a Lanchester-type linear law; the air terms are the results of the attacker's sorties (the subsection «Sorties» below) and the missile and drone terms the damage of what gets through. A type of the block loses the block share × its exposure (part I); a marker leaves the outline when fewer than 0.5 units remain. Both sides are computed from the state at the start of the day and applied together. Enablers wear at f_en = en_k × f_land plus their share of missile damage. One term is special: SEAD divides by S_sam(d), the defender's SAM strength without the missile launchers — the launchers lose the same share as the block, so a side whose SAMs are gone loses its launchers at the cap when aircraft reach them. The table below lists the rates that remain; all are assumptions (status A), published so that anyone can argue with them — a change is one line of norms.json.

Sorties

Since part K the aviation strikes and loses through sorties, not through a weighted power. Every combat-air type — fighters, attack aircraft, bombers, attack helicopters and the reusable strike UAVs that are not mass tactical drones — flies units × avail_air × manning × gate_t × sortie_rate sorties a day: the gate of the type is the one of parts G and H (the theatre radius with tankers, the global bombers, the carrier part, the forward bases), the rate depends on the class and the tempo — fighters and attack aircraft 0.3 / 1.2 / 2.0 sorties per aircraft a day at the sustained / campaign / sprint tempo, bombers 0.2 / 0.5 / 0.8, attack helicopters 0.5 / 1.5 / 2.5, strike UAVs 0.5 / 1.0 / 1.5. The sorties that reach the opponent are split over four missions by priority × what the opponent has (mission_target × the size of its block, the rule of part H): counter-air (fighters and attack aircraft only), SEAD (× sead_x 3 while the opponent's long / medium SAM coverage at altitude is above cov_sead 0.5 — the air defence is suppressed first, 1991 and June 2025), strikes on the land block (× land_nocontact 0.2 without ground contact — a strategic air war strikes the missile force, the air defence and the fleet before the army in its barracks) and on the sea block. The fighters that cannot reach fly defensive counter-air at home — (all − reach) × the air share × the side's air superiority: a side that has lost the air keeps its fighters on the ground.

The result of a sortie is pk[mission] weighted units of the opponent — the expected result of one sortie of index weight 1 (a fighter): land 0.06, SEAD 0.06, sea 0.02, counter-air 0.03. A sortie counts × the index weight of its type (a bomber 2.0, an attack aircraft 0.8, a helicopter 0.6, a strike UAV 0.2) × gen_f, the land strikes × g_ad of the defender as before, and × pk_tempo at the sustained tempo (0.25: the glide-bomb war of 2024–25 strikes positions, not vehicles in the open). A counter-air sortie that reaches finds the opponent's aircraft in the air (presence = its sorties ÷ its combat aircraft) and on the ground (oca_ground 0.5 of the parked fleet), a defensive one only in the air; the kill is × the quality ratio (gen_s ÷ gen_d)^air_q, clamped to 0.1 … 2 — a generation-3 fleet against a generation-5 one scores a fifth. The sums by block replace the former air → land / air defence / sea / air rates; the cap of 10 % a day stays.

The losses of the attacker's own sorties: att = att_base ÷ gen_f × (1 + sam_x × the defender's surviving SAM share × its coverage at altitude × (1 − the attacker's air superiority) × the share of the engagements the defender's interceptor stock covers) × (1 + air_x × the share of the defender's counter-air among the aircraft in the air) for the strike sorties, without the air term for the counter-air sorties that reach (the sweeps meet the opponent's fighters through pk.air), the base rate alone at home — att_base 0.00025 (the coalition of 1991: 38 combat losses in about 65,000 combat sorties, 0.0006; the model's base is half of it, the SAM term carries the rest), sam_x 3, air_x 2. The surface-to-air units draw int_shots.aircraft 0.1 interceptors per sortie over the covered share of the territory (fixed-wing from the long / medium pool first, helicopters from the short-range pool first — part I); an air defence whose interceptors are gone no longer kills sorties. The air block of a side loses its own sorties and the opponent's counter-air kills on top of the missile and drone damage; the panel shows «sorties: N today · N since day 1 · aircraft lost N», the window of the line the sorties by type and mission with their results and their losses, the CSV adds the sorties and the aircraft lost by day.

The Ammunition switch of the pair (part K): published — the artillery rounds and the surface-to-air interceptors are the stocks of part I and run out; unlimited — no stock runs out, the model of parts H and J. The corridors of the calibration are checked at unlimited (the regression check of the core); the published figures are reported on their own, with the Ukraine – Russia campaign at 45–55 % instead of 28–38 % — the published flows of shells feed the campaign tempo for 3 (Ukraine) and 9 (Russia) days.

Calibration of part K (01.11.2026, once, after the two steps of the part; the knobs sortie_rate, pk, att_base, sam_x and air_x within × 0.5 … × 2 of the brief's start, the Ammunition switch at unlimited for the corridors of parts H and I; the figures by step and pair are in the project report): moved pk.ad 0.04 → 0.06 (× 1.5 — the SEAD of a sprint must take half of Iran's launchers and SAM strength in 40 days, the June 2025 claims), att_base 0.0005 → 0.00025 (× 0.5 — the 1991 rate holds the SAM and gun losses too; with the SAM term on top the full rate doubled the losses) and sam_x 6 → 3 (× 0.5 — Ukraine's aviation would otherwise lose 7 aircraft a month against the 1–3 of Oryx); not moved sortie_rate, pk.land / sea / air, air_x; added by the executor (status A) pk_tempo, land_nocontact, oca_ground, air_q and int_shots.aircraft. After: Ukraine – Russia sustained 94.5 / 99.4 % at unlimited (94.3 / 99.5 published), the drone share 57 / 78 %, Russia 227 fixed-wing sorties a day (100–300) and 3.3 aircraft a month (Oryx 2–5), Ukraine 36 and 2.9 (1–3); campaign 30.8 / 96.0 % at unlimited (28–38 / ≥ 94) and 48.5 / 98.0 % published (45–55); Russia – United States campaign 71.4 / 94.9 % (68–74 / 93–97); United States – Iran campaign: Iran's air defence −46 %, land −8 %, the United States 100 % — its aviation −25 % against the corridor −30 … −60 % (the former air → air term struck a 532-unit air block with the whole reaching power of the United States; the sorties find Iran's aircraft on the airfields at oca_ground); Israel – Iran sprint 40 days 95.1 / 77.3 % — Iran in the corridor 76.2 ± 1 with its launchers and SAM strength −60 %, Israel 1.0 above the upper edge 94.1: its aviation loses 14 aircraft in 40 days (93 %) where the former SAM → air term took 90 (84 %); the Desert Storm reference with forward bases, 43 days at the campaign tempo: 0.68 sorties per inventory aircraft a day (about one per deployed aircraft), 0.0003 losses a sortie (≤ 0.001), Iran's army −63 % (1991: 32–47 % from the air). Not reached: Israel without a loss in 12 days — at the floor of att_base the model loses 5 aircraft (2 on sorties, 3 to Iran's counter-air); zero needs att_base below the band and breaks the 1991 reference.

Missiles — stock, launches, interception

Ballistic and ground-launched cruise missile types are a stock, not a rate. Stock M(0) = the missile figure of the card (ms) where one is published, else launchers × msl_per_launcher. Launches a day U = min(M, launchers × msl_rate), where msl_rate depends on the intensity; the launchers are the card's quantity and wear down with the air-defence block. Interception = int_max × the defender's long / medium SAM coverage at altitude (recomputed every day), so H = U × (1 − interception) missiles get through; each removes msl_dmg weighted index units from the defender, split by msl_target over air (airfields), air defence, land and enablers: f_g += H × msl_dmg × share_g ÷ S_g(d). The panel shows «missiles: launched · intercepted · stock left» for every side that has missiles, the feed «−N missiles in stock», and the end table the launches and interceptions.

Since part H the damage of the missiles and drones that get through falls on the defender's blocks in proportion to priority × what the defender has: share_g = target_g × S_g ÷ Σ_h target_h × S_h, so f_g += dmg × target_g ÷ Σ_h target_h × S_h — msl_target and uav_target stay the priorities (a priority weighted by availability), a block the defender does not have takes nothing, and a small block no longer takes a fixed share of every missile (before part H a 76-unit block of enablers took a tenth of every Tomahawk's damage, and «Safir Jeep −16,612» led the losses of United States – Iran). The cap of 10 % a block and day stays. The five types that lost the most in the end table are ordered by the index weight of the loss (units lost × the weight of the category); the figures shown are units. The reusable strike UAVs take the missile share of the air block like the rest of the block.

Air-launched cruise missiles (since part H): there are no cards for them — the norm alcm holds, per country and missile family, the carrier types as they stand on the map, the range, the stock for a 30-day window (on hand plus a month of output; 0 where nothing is published — the family then launches nothing), the warhead and a confidence, with a source per entry. A family counts when at least one carrier has aircraft on the side's cards; the side's stock is the sum of those families; launches a day = alcm_rate × the 30-day stock × the gate of the family (the drone rule: 1/30 at the sustained tempo, 1/15 at the campaign, 1/10 at the sprint), never more than what is left; the gate of a family is the best of its carriers — clamp(1 − dist ÷ (radius × tanker factor + range)), or the carrier's global or carrier-borne gate; interception as for the other cruise missiles, the damage msl_dmg × (warhead ÷ 450 kg, within 0.5 … 1.5), split as above. The missiles line of the panel counts them in the stock («incl. N air-launched»), the end table shows «of which air-launched», the feed says on day 1 «N air-launched cruise missiles on M bombers and fighters (model)»; in the cinema their arcs start from the carrier's marker. The Kinzhal is aeroballistic and is intercepted like a cruise missile — a known simplification.

Stocks for a 30-day window (2026-10-27; family, stock, carriers): Egypt — SCALP-EG 100 (Dassault Rafale); France — SCALP-EG 300 (Dassault Rafale, Dassault Mirage 2000); Germany — Taurus KEPD 350 150 (Panavia Tornado IDS); Russia — Kh-101 / Kh-555 185 (Tupolev Tu-95, Tupolev Tu-160), Kh-22 / Kh-32 250 (Tupolev Tu-22M), Kh-59 / Kh-69 24 (Sukhoi Su-34, Sukhoi Su-35, Sukhoi Su-30, Sukhoi Su-57, Sukhoi Su-24), Kh-47M2 Kinzhal 50 (Mikoyan MiG-31); Ukraine — Storm Shadow / SCALP-EG 30 (Sukhoi Su-24, Mirage 2000-5F); United Arab Emirates — Black Shaheen (SCALP export) 600 (Dassault Mirage 2000); United Kingdom — Storm Shadow 400 (Eurofighter Typhoon); United States — AGM-158 JASSM / JASSM-ER 2,950 (B-1 Lancer, B-52 Stratofortress, B-2 Spirit, McDonnell Douglas F-15E Strike Eagle, F-16 Fighting Falcon, F-15EX Eagle II, F/A-18 Super Hornet), AGM-158C LRASM 450 (B-1 Lancer, F/A-18 Super Hornet, B-2 Spirit). No published stock — launches nothing: Azerbaijan, China, Iran, Israel, Netherlands, Turkey.

Drones — one-way-attack stocks, tactical swarms, reusable strike UAVs

Every UAV type has a role in uav_roles.json, collected by research agents with a source for each type (a country card can override it with ur and a source us): one-way attack (owa — loitering munitions, FPV strike drones, long-range strike drones such as Geran-2, Shahed-136, Liutyi, FP-1), reusable strike (TB2, MQ-9, Orion, Mohajer-6, Hermes 900 …), reconnaissance (Orlan-10, Raven, ScanEagle, the mini and micro drones) and other (cargo, targets, tankers, decoys); a type nobody could identify is reconnaissance and only takes losses. One-way-attack drones are a consumable, not a unit: the card carries ds — the stock for a 30-day window, on hand plus one month of production, with its source (Russia's Geran-2 6,000 a month: the Ukrainian Air Force counted 6,297 Shahed / Geran and decoys in July 2025; Ukraine's FPV output about 200,000 a month in 2025; Iran's Shahed-136 1,500; the US Switchblade 4,400) — a card without ds uses its quantity. The stock falls with the launches, apart from the missiles, and is never replenished; the Inventory Strength Index, Balance and the remaining % do not see it (the aggregate FPV cards carry no quantity).

Long-range drones (range ≥ 500 km) are launched like missiles: uav_rate × the stock a day (0.0333 at the sustained tempo — a month of output spread over 30 days; 0.0667 at the campaign, 0.1 at the sprint tempo), never more than what is left; a type reaches when its range covers the distance between the outlines (Geran-2 reaches Alaska across the Bering Strait, the Shahed-136 Israel, the FP-1 Russia's European part); the defender intercepts min(uav_int_max, int_max × its long / medium SAM coverage) — the Ukrainian Air Force reports 85–90 % of the Shahed launches of 2025 intercepted or lost; each drone that gets through removes uav_dmg_long weighted units, split by uav_target over land 0.5, air defence 0.25 and enablers 0.25 — a 50 kg warhead, a tenth of a Tomahawk-class missile; most of the real damage falls on power plants and refineries, which are outside the model.

Tactical drones (FPV, Lancet, Switchblade, Phoenix Ghost, Molniya — under 500 km) fly only where the ground forces meet: the same launch rate from the tactical stock, uav_int_tac of them stopped by electronic warfare, misses and weather (plus uav_cuas when the defender has counter-UAS systems with units left — in 2024–25 only one to three FPV launches in ten reached their target), and each drone that gets through removes uav_dmg_tac weighted units of the land block. With the drones in the model the land → land rate is split: land_split.fire of the former 0.030 stays with artillery and direct fire (k.land_land = 0.024), the rest is the drone term — a pair without tactical drones loses that share. Oryx counts 60–70 % of the visually confirmed vehicle losses of 2024–26 as drone strikes.

Reusable strike UAVs enter the combat-air power P_cmb at uav_k per unit of index weight — a TB2 is not an F-16 — with the generation factor of the year of the variant (the band of gen_f.air: ≤ 1975 → 3, 1976–99 → 4, 2000–14 → 4.5, ≥ 2015 → 5), and since part K fly their own sorties (the class uav of sortie_rate) and lose through them — a small mass drone (the index's mass rule) stays out of the sorties. Reconnaissance and other drones stay as before — in the air block for the losses only.

On the screen: a line «drones: launched · intercepted · stock left» under the missiles for every side with one-way-attack drones, «−N drones in stock» in the feed, a card «drone stock exhausted», the sentence of the day in its drone form when the day's drone launches outnumber the missiles («Russia launches 200 drones at Ukraine; Ukraine intercepts 180 …»), and two rows of the end table. In the cinema the long-range drones are a stream of dashed arcs — slower than the missiles (2.5 s), a 3 px head, from the attacker's UAV markers, at most six in the air per direction, a «✕» on interception; the tactical drones are the swarms at the front.

Types with a role at this build: 246 — one-way attack 42 (long-range 10), reusable strike 46, reconnaissance 147, other 11; cards with a sourced 30-day stock: 29.

Calibration of part F (20.10.2026, Ukraine – Russia at the sustained and campaign tempos, Israel – Iran at the sprint, Russia – United States at the campaign; before → after): the specification's start — a fire share of 0.4, uav_dmg_tac 0.005, uav_int_tac 0.5, uav_k 0.3, uav_dmg_long 0.03 — gave Ukraine – Russia 92.7 / 98.7 % at the sustained tempo and Ukraine 52 % at the campaign tempo against 94.3 / 99.5 and 33.2 % before. The campaign corridor (28–38 %) cannot be reached with a fire share of 0.4 whatever the drone coefficients: the drone stock is one month of production and is spent within the window at every tempo, while the fire term grows 17 × with the tempo — so the fire share went to 0.8 (land_land 0.024, the limit of the corridor), uav_int_tac to (nine FPV launches in ten fail), uav_k to 0.15 (Iran's Mohajer-6, Saegheh and Shahed 129 would otherwise lift its combat-air power by 22 % and take Israel to 91.3 %) and uav_dmg_long to 0.015 (a tenth of msl_dmg, proportional to the warhead). After: Ukraine – Russia sustained 93.9 / 99.3 %, campaign 35.8 / 95.6 %, Russia – United States 71.1 / 94.9 %, Israel – Iran sprint 92.2 / 76.8 %, My Army – Israel unchanged. At the sustained tempo the drones now do about a quarter to a third of the land attrition of Ukraine – Russia; at the campaign tempo the fire carries it, as a production-limited stock cannot grow with the tempo.

Long-range aviation — strategic bombers, carrier aviation, forward bases

Since part G the air gate is set per type, not per side. P_cmb(s) = Σ over the combat-aviation types (and the reusable strike UAVs × uav_k) of weight × n × gen_f × gate_t, where gate_t = max(theatre, global, carrier): the theatre gate is the former gate of the side — clamp(1 − dist ÷ reach) with the tanker factor — or 1 when the side's switch «Forward bases: in theatre» is on; the global gate is 1 for a strategic bomber with tanker support; the carrier gate is the naval gate for the part of a type that flies from ships. Attack helicopters and strike UAVs take the theatre gate only. The strikes on land, air defence and sea use the power that reaches, × sup as before; air → air and SAM → air take the share of P_cmb that reaches where the side's gate stood; the losses of the air block and the rates k are unchanged, and sup keeps the whole P_cmb.

Strategic bombers: a type of the Bombers category whose published unrefuelled range (perf.json cr with crk «range», reviewed in perf/range_overrides.json with a source per type) is bomber_global_km = 9,000 km or more reaches any opponent while the side has aerial refuelling tankers with units left — B-52H 14,200 km, B-1B 9,400, B-2 11,000, Tu-95MS 15,000, Tu-160 12,300; the Tu-22M3 (6,800) and the H-6K (6,000) stay below the threshold and reach by radius (range × ⅓, × 1.5 with tankers) like the fighters. Without tankers the former radius rule holds for every bomber.

Carrier aviation (norm cv): a combat-aviation type whose name contains one of cv.names flies from ships (pool cv = aircraft carriers; both = carriers and amphibious assault ships, for the STOVL types F-35B and AV-8B), and a mixed card listed in cv.share flies its carrier share only (the US «F-35 Lightning II» 0.3 — the F-35B and F-35C among the three variants; France's «Dassault Rafale» 0.2 — the Rafale M). The air-wing capacity of a side = Σ q × cap over its carriers (cap by name — Nimitz and Gerald R. Ford 48, Queen Elizabeth and Charles de Gaulle 24, Liaoning and Shandong 24, Fujian 36, Admiral Kuznetsov —, the light carriers and amphibious ships 12; else by category: aircraft carriers 48, amphibious assault ships 12; the helicopter carriers Mistral and Type 075 carry no fixed-wing aircraft) × the sea availability (avail_sea: about a third of the carriers are at sea at any time), and it falls with the carriers of the sea block. The carrier part of a type = min(n × share, the free capacity), served in the order of cv.names (the Super Hornets fill the US carriers first, the F-35B take the amphibious ships); it strikes through the naval gate — from the day after nav_delay when the outlines lie beyond sea_km, never at a landlocked opponent — while the rest of the type keeps the theatre gate.

Forward bases: a switch per side on the panel — «none» (default) or «in theatre»: the side's aviation and attack helicopters fly from bases next to the opponent, so every type reaches (the United States from the Gulf bases against Iran); the hash keeps it as the seventh and eighth fields of &war= (none | theatre), an old link without them means none. The feed says on day 1 which types of a side reach — the bombers with tanker support, the carrier wing and the day it starts, the forward bases.

Since part H the theatre gate is set per type by its own radius: gate_t = clamp(1 − dist ÷ (radius_t × tk_t)), where radius_t comes from perf.json (a combat radius, or a range × the radius factor) or from the country's own card (cr with its kind crk and its source crs — Israel's F-16I Sufa 1,500 km with conformal tanks against the 550 km of a clean F-16C) and tk_t is the tanker factor for fighters, attack aircraft and bombers while the side has tanker aircraft with units left, 1 for helicopters; an attack helicopter without a perf radius takes heli_radius_km (300 km — the Ka-52 and the AH-64 fly 250–300 km), a reusable strike UAV its range from uav_roles, a combat type without any figure the median radius of its category on the side, and only a side without any radius keeps the former gate of the side. The global and carrier gates of part G are unchanged (gate_t = max of the three); the «reach: air» figure on the panel stays as a reference — the longest radius of the side. Israel – Iran at 896 km: the F-15 at 0.68, the F-16I at 0.60, the F-15I at 0.53, the F-35I at 0.46 — 56 % of Israel's combat-air power reaches instead of the 68 % of the side gate; China's J-10 against the United States 0 (the H-6K no longer opens the gate for the fighters).

Calibration of part G (27.10.2026, United States – Iran at the campaign tempo, 30 days, mutual, no forward bases — a pair whose aviation gate was closed before: the outlines lie 7,529 km apart, beyond the longest radius × 1.5, so the United States struck with the fleet's missiles alone and Iran kept 95 %): with the bombers (140 B-52H / B-1B / B-2 from day 1) and the carrier wing (about 230 Super Hornets and F-35 at sea from day 8) Iran keeps 76.4 % — air defence and missiles −48 %, aviation −39 %, land −2 %, the United States 100 % (its aircraft meet Iran's SAMs: a few units). The reviewer's corridors for air defence (−35 … −60 %) and aviation (−30 … −60 %) hold; the land corridor (−5 … −15 %) is not reached — with k.air_land 0.012 and a land block of 6,400 weighted units the whole US combat aviation at full reach would wear it by about 5 % in 30 days (forward bases in theatre: −12 %), and the model has no air → land term beyond that rate; the rates were not moved. Israel – Iran, Russia – United States, Ukraine – Russia and My Army – Israel are within the corridors of the brief (± 1 / ± 2 / unchanged / unchanged).

Breaking point

A block that has lost a large part of itself does not fight like the whole: the studies of Dupuy and the TNDM put the loss of combat effectiveness of a unit at 30–50 % casualties, and the withdrawals of March – April 2022 left columns of abandoned vehicles behind. The model gives every block a breaking point as a share of its day-0 size — break_pt: land 0.5, aviation 0.4, air defence and missiles 0.3, fleet 0.4. Above it the block fights at full cohesion; below it the block's power is multiplied by cohesion_f, linear from 1 at the threshold to cohesion_min (0.2) at zero — the remnants fight on, they do not vanish. The ground fire, the sorties, the SAM fire and the fleet strikes of a broken block fall with it. A land block below the threshold in ground contact loses × break_loss_x (1.5) — the abandoned and captured vehicles whose share in the Oryx lists rises during a withdrawal.

On the screen: a card «ground forces below the breaking point (48 %)» with the flag and the name of the side, once per block on the day it crosses, the same chip in the feed, a row «below the breaking point since day N» in the end table, the multiplier in the «why» window of the block and of the attacker's power, and the front line drawn shifted into the territory of the side below the threshold — 20 px at the threshold, 60 px at zero. No block is ever «defeated»: the words of the screen are the threshold and the multiplier.

Saturation of the air defence

A salvo of 500 missiles is not intercepted at the rate of a salvo of 50. Every launcher of the air defence engages a number of targets a day — ad_channels of its family (engagement channels × firing cycles a day): Patriot 6, S-400 8, SAMP/T 6, Buk 6, Tor 12, Pantsir 16, Iron Dome 20, NASAMS 8; a family not in the list takes the value of its tier — long 6 / medium 8 / short 12. The capacity of the defence a day = Σ launchers × channels × availability; the targets of the day = the missiles + the long-range drones × sat_uav_ch (0.3 of a channel each) + the opponent's sorties over the coverage. The ballistic missiles are engaged by the long- and medium-range pool (A) alone, the cruise missiles, drones and aircraft by the short-range pool (B) first and then by what pool A has left. The interception of the day = the full rate × min(1, capacity ÷ targets)^sat_k (1.0); the excess over the capacity passes at the floor.

References: Ukraine's record nights of 2025 (700 + Shaheds and 20–40 missiles — about 85 % of the drones and 50–70 % of the missiles intercepted, the Ukrainian Air Force), the twelve-day war (salvos of 100–200 ballistic missiles — 86–90 % intercepted, the IDF), the Red Sea of 2024 (the Houthi salvos against the destroyers — about 100 % at two or more SM-2 a target). On the calibrated pairs the capacity is never reached — Ukraine about 4,300 targets a day, Russia about 14,700, Israel 500–650, Iran 700–1,000, the United States about 7,400 against at most 1,500 targets a day — so the saturation changes none of the corridors; it shows for a small air defence against a large salvo. On the screen: «AD capacity N targets a day — today N» under the missiles, a card «air defence saturated: N of N targets engaged» with the flag and the name of the side on the first day the targets exceed the capacity, the table of the pools in the «why» window of the interceptors.

Targets outside the model and the drone chain

Most missiles and long-range drones are not fired at the equipment the map counts: power plants, cities, refineries, factories and command posts are outside the model. strike_outside is the share of the launches at such targets — 0.4 by default (an assumption), and for the pairs with a source: Russia → Ukraine 0.65 (the Ukrainian Air Force, 2024–25: most Shaheds and Kh-101 go to the energy system and the cities), Ukraine → Russia 0.7 (refineries, airfields without equipment in the model, depots), the United States → Iran 0.3 (nuclear sites, command, industry), Israel ↔ Iran 0.4. The share is taken out before the damage; the interception is counted on every launch, so the screen's «launched · intercepted» stays the fact and the damage to the equipment comes from the rest alone. msl_target went back to the neutral priorities and was recalibrated with the share (part N.4).

The tactical drones are one chain with the same result as before and honest links: launches × (1 − uav_void 0.89: misses, jamming, weather and repeat strikes on a vehicle already hit — RUSI 2024–25: 10–30 % of the FPV launches reach the target; × (1 − uav_cuas 0.1) when the defender has counter-UAS systems with units left) = the drones that reach; × uav_hit_kill 0.05 = the vehicles destroyed (most FPV hits immobilise, the kill takes a second or third drone — about one destroyed vehicle per 180 launches, Oryx against the launch volumes of 2025); × the average weighted size of a vehicle of the defender's land block (the vehicles alone — the ATGM systems, mortars, anti-tank guns, MANPADS and personnel of the block are not vehicles; Ukraine 0.27, Russia 0.47 index units) = the weighted units lost. The calibration keeps the drone share of Ukraine's land losses in its corridor (50–70 %). The screen says «6,970 launches → 836 reached → 84 vehicles» in the «why» window and «N vehicles destroyed» under the drones.

Scenario, tempo, supply, the fleet and the strategic launchers

Scenario: a mutual war, or «A attacks B» — the defending side strikes with its land power × def_k (2.0: the 3 : 1 rule of thumb for an attacker, FM 3-0 / Dupuy) in the land → land term, the attacker's strikes carry no multiplier; aviation, air defence, sea and missiles do not change. The feed says on day 1 who defends.

Tempo: sustained is the observed 2022–2026 rate — a fire intensity of 0.03 plus the tactical drones, which launch a month of output within the window whatever the tempo (Oryx, visually confirmed losses of 2025 — a floor: Russia about 53 tanks and 344 vehicles of all kinds a month, Ukraine 27 and 283; the model's 30 days at the sustained tempo wear Russia's land block by about 3 % and Ukraine's by about 8 %). Campaign (1) is about 33 × and sprint (1.85) about 62 × the fire intensity of the sustained tempo, and the panel says so: «Tempo: sprint ≈ 62× the observed 2022–2026 rate»; the land losses of the 30 days differ less than that, because the drone term does not grow with the tempo.

External supply per side: moderate 5 % or high 10 % of the initial weighted land, air and air-defence park a month (deliveries to Ukraine in 2022–24 were of that order for armour), credited every day to every type of those blocks — q += q(0) × supply ÷ 30, never above q(0); a type that had vanished returns to the outline with «+N». Missiles are not replenished: the stock ms does not grow.

The fleet strikes the coast with cruise missiles: a ship whose class carries land-attack missiles (nav_cells — cells per ship × the share loaded with land-attack missiles; Ticonderoga 122 × 0.4, Arleigh Burke 96 × 0.3, Ohio SSGN 154, Virginia 12, Yasen 32, Oscar II 72 × 0.5, the Russian corvettes and frigates with a universal launcher 8) adds its cells to the naval missile stock of its side; launches a day = Σ q × nav_strike × nav_rate (0.1 / 2 / 4 per ship-weight at sustained / campaign / sprint — about one Kalibr a day is the observed 2022–2026 rate; cruisers weigh 1.0, destroyers 0.8, frigates and corvettes 0.5, SSGN 3.0, other submarines 0.3), from day nav_delay + 1 (7) when the outlines lie beyond sea_km, never at a landlocked defender (landlocked_iso); interception and damage are those of the land missiles. The missiles line counts them together, the end table shows «of which naval». United States – Russia at the campaign tempo, 30 days: about 3,700 cruise missiles from the US fleet and about 700 Kalibr from Russian submarines and corvettes; Iran has no strike-capable ship.

Strategic launchers — a card of the missile categories whose name is in icbm_names (Yars, Topol, Voevoda, DF-5, DF-31, DF-41, Minuteman, Trident …) or that carries nuc: 1 — stand outside the conventional model: they neither launch in the missile exchange nor enter the air-defence & missiles block (the card says «not in the model: strategic nuclear»); the Inventory Strength Index and Balance keep them as before. Russia's launches at the United States in 30 days fall from about 4,000 (2,500 of them ICBMs) to about 1,700 — 1,000 from the land launchers and 700 from the fleet.

The ballistic-missile submarines are strategic launchers as well: a Submarines card whose name is exactly one of ssbn_names (Ohio class SSBN, Borei, Delta IV, Type 094, Vanguard, Triomphant) carries nuc: 1 and stays out of the sea block — on deterrent patrol it takes no part in the conventional exchange and it carries no land-attack cruise missiles in nav_cells; the Ohio class SSGN with Tomahawk and the attack submarines stay in.

Generation of the types

The generation of a type multiplies its combat power in Live War — P_cmb of the combat aviation (gen_f.air), P_sam of the area air defence (gen_f.sam) and the main-battle-tank part of P_land (gen_f.mbt; the other land types at 1.0): a 1960s F-4 strikes and fights at 40 % of an F-35, a 1980s F-14 or MiG-29 at 70 %, an I-HAWK battery of 1971 defends at 40 % of a David's Sling. The inventory, the remaining %, the Inventory Strength Index and Balance do not change — a type counts as one unit however old it is; only how hard it hits and how much it shoots down does.

The value comes from perf.json (gen — collected by research agents with a source for each type: a stated fighter generation 3 / 4 / 4.5 / 5, or for bombers, attack helicopters and air-defence systems the band of the year the variant entered service, ≤ 1975 → 3, 1976–99 → 4, 2000–14 → 4.5, ≥ 2015 → 5; main battle tanks 1 / 2 / 3 / 3.5) or from the country's own card (gn with its source gs — Israel's F-16I Sufa 4.5, Iran's F-5-derived Saeqeh 3), which takes precedence. A type without a value counts at 1.0; the panel says «N types without generation data» and lists them in a tooltip.

Combat types of the documented countries: 263 · with a generation: 261 · without generation data (counted at 1.0): 2

Calibration with the generation factor (Israel – Iran, sprint, 40 days, October 2026): Iran's air superiority falls from 0.37 to 0.22 — its F-4, F-5 and the F-5-derived Saeqeh / Kowsar count at 0.35, its F-14 and MiG-29 at 0.7, its HAWK, Rapier and S-200 batteries at 0.4; Israel keeps 93 % of its inventory, its aviation 88 % and its air defence 82 % (85 / 75 / 53 % without the factor), Iran's launchers and SAM strength −51 %. The specification's air multipliers {3: 0.5, 4: 0.8, 4.5: 0.9} gave Israel's aviation 83 %, an air superiority of 0.71 and Iran −45 %, so 3 → 0.35, 4 → 0.7 and 4.5 → 0.95; air_ad stays 0.030. Ukraine – Russia at the campaign tempo: Ukraine keeps 33 % (26 % without the factor — Russia's T-72B3 strike at 0.6, its Su-34 and Su-35 at 0.95, Ukraine's Leopard 2A6 at 1.0); Russia – United States 71 / 95 %.

Calibration and what the model leaves out

Reference points of the rates (v2 with combat aviation and air superiority, October 2026): two equal neighbouring armies at the campaign tempo each lose about 30 % of their inventory in 30 days — the land block about 45 %, the air block 15–25 %, the air-defence block up to 40 % (the United States against itself: 29 %; Israel 42 %; at the sustained tempo about 2 %); the air blocks of two equal sides no longer wear each other down fast, because each strikes with half the air superiority and only with its combat aviation. A 3 : 1 opponent is worn down to about a tenth in 30 days while the stronger side loses about 7 % — the linear law keeps the attacker's power while the defender's blocks shrink; two armies that do not touch and have neither long-range aviation nor missiles lose nothing on the ground — only in the air, by the gates, and at sea only within sea_km.

Calibration on Israel – Iran 2026 (high intensity, 40 days, no ground contact), against the published figures of the war of 28 February – 8 April 2026: Iran's ballistic-missile launchers −55 % (IDF, 10 April 2026: about 200 of 470 destroyed and 80 disabled), its SAM strength −55 % (IDF and CENTCOM: about 80 % of the air-defence systems hit; the model counts launchers that survive, not radars), its stock of missiles able to reach Israel −54 % (650 launched at Israel in 40 days and about 1,400 at Gulf states), its fleet −28 % (CENTCOM: more than 155 vessels damaged or destroyed, most of them small craft of low index weight), about 70 % of the launches intercepted as Israel's coverage is worn down (about 90 % in 2026). With the combat aviation and air superiority of October 2026 the model gives: Iran's launchers and SAM strength −43 %, its missile stock −61 % (about 1,500 launches, 90 % intercepted), its fleet −27 % — reached by Israel's aviation alone, because the two outlines lie 900 km apart and the fleets do not meet; Israel keeps 85 % of its inventory, its aviation 75 % and its air defence 53 % against the real 97–98 % and a few batteries hit. The coefficients were moved within × 0.5 … × 2 of the specification: air_ad 0.030 (the upper limit — the strikes now come from the combat aviation alone, scaled by air superiority), sam_air 0.008 and sam_eff 0.25 (no Israeli aircraft was lost to Iranian SAMs), msl_rate at high 0.25 (the average over 40 days, not the day-1 rate), msl_dmg 0.15, air_sea 0.006, land_land 0.030. What remains open is the quality of the types: a quantity-only index gives Iran's 200 fighters of the 1960s–80s the same weight per unit as Israel's F-16, F-15 and F-35, so Iran's combat aviation keeps 37 % of the air superiority and wears Israel's small SAM force at the same share per unit as Israel wears Iran's; a generation factor is the next step and is not part of v2. In 2026 most of the strikes on Iran were flown by the United States: the model reproduces the direction and the order of magnitude of the 2026 figures, not the war.

The cinematic layer of the war screen is decorative: it draws the day's figures and never feeds back into the arithmetic; it is off under a reduced-motion setting of the system and can be switched off in the panel (Cinematic). Each panel has its own layer clipped to the panel — nothing is drawn over the middle column; a flight from one outline to the other leaves at the edge of its panel and enters at the edge of the other panel on the same curve. The action runs in real seconds whatever the speed of the run — the day is only a counter: the sorties and the missiles are a continuous stream drawn from the current day's figures. A sortie takes 3 s: the aircraft or helicopter sprite takes off from the attacker's marker nearest to the target (1.2 s out), strikes (a burst of three rings, smoke over the hit marker for 6 s), flies back along the same arc (1.2 s) and lands with a flash on its base; the share lost that day, at least one when one was lost, does not come back and falls with a trail of smoke. A new sortie leaves every 0.8 s while the day has strikes, two to six are in the air per direction. Missiles fly 1.5 s on high ballistic arcs with a bright head and a tail; the interceptors start from the defender's nearest SAM marker (at most min(6, interceptions ÷ 20)) and burst on the approach with a «✕». Where the ground forces meet, the front line is drawn along the common border on both outlines, with artillery flashes along it and the tanks and fighting vehicles nearest to it pushing 20–30 px towards the front and back. At most four «−N» labels a panel, 3 s each; at most six bursts a panel at once; the sentence of the day under the plate changes at most every 3 s; a type that reached zero stays two days as a grey wreck. Sprites are 40 px at zoom 6 and closer, 32 px at zoom 5 and 26 px further out. At most 60 elements move at once in a panel.

Record is a clean scene for a video: 4 s a day, the camera on, the pickers and the buttons hidden (Esc brings them back); an intro of 4 s and a finale of 8 s frame the run (2 s and 4 s at 1 s a day). The camera is a real zoom of the map (Leaflet flyTo), on by default at 1, 2 and 4 s a day, and works like a film: a shot is one camera position held for 6 s with a single slow movement inside it — a push-in of 0.7 zoom level (× 1.6 of scale, never above zoom 8.5, so the outline stays recognisable) or the pull-out back to the fit; the camera never follows the day. A cycle of five shots takes about 30 s: the overview of both panels (where the event cards show — stocks exhausted, a block below 75 / 50 / 25 %, the fleets engaging, the first supply, a status line every tenth day; one card at a time, 3 s), the attacker's panel pushing in towards the cluster its sorties and missiles start from, the defender's panel pushing in towards the cluster of the strikes, the attacker's panel pulling out while the sorties return and land, the defender's panel pulling out. The roles swap every cycle, so both panels get the same close-ups whoever dominates; a push-in that would leave fewer than three markers in the frame stays at the fit. On Day 0 and on «Fit each» both panels fit the equipment — the bounding box of the markers plus 10 % (Russia: its European part, the United States: the continental part); a zoom or a drag by the user stops the flight and keeps the camera off that panel for 10 s; at 0.5 s a day there is no camera. The figures are the same as without the cinema.

Not modelled, by design: production, mobilisation, allies, terrain, training, morale, repair of damaged equipment, weather, intelligence (external supply and the generation of the types are the only quality terms — above). Reserve stocks are never committed (a «reserve enters on day X» option is in the backlog). Personnel is not «killed» — it enters only as manning. Nuclear weapons are left out entirely. The feed names the day's losses by category and the end table shows initial → remaining per block, total units lost, missiles and drones launched and intercepted, and the five types that lost the most on each side.

Validation and limits

Calibration of part H (27.10.2026, once, after the three changes of the part — the damage split by priority × availability, the theatre gate per type, the air-launched cruise missiles; every coefficient within × 0.5 … × 2 of build 2026-10-19k; the figures by step and pair are in the project report): three corridors from before were kept — Ukraine – Russia sustained 94 ± 1 / 99.5 ± 0.5 and campaign 28–38 / ≥ 94, Israel – Iran sprint 40 days 93.1 / 76.2 ± 1, Russia – United States campaign 68–74 / 93–97, United States – Iran campaign without bases: air defence −35 … −60 %, aviation −30 … −60 %, land −5 … −15 % — and one was added: at the sustained tempo the tactical drones carry 50–70 % of Ukraine's land losses and at least half of Russia's (Ukrainian claims of 2025: 66 % of the Russian equipment disabled in January by strike UAVs; RUSI, February 2025: 60–70 %), against 22 % and 40 % in the model before. Moved: intensity_f.low 0.03 (× 0.5, the floor of the corridor — the observed losses are now fire plus drones, so the fire half is smaller), uav_dmg_tac (× 2, the ceiling), uav_int_tac (× 0.95: with counter-UAS systems 14 launches in 100 reach a target), intensity_f.high 1.85 (× 1.23 — the per-type gates cut the reaching share of both sides of Israel – Iran, Israel 0.68 → 0.56 and Iran 0.48 → 0.21 of P_cmb, and the 40-day sprint had risen to 94.3 / 81.0 %), msl_target 0.6 / 0.25 / 0.1 / 0.05 (the air priority up, the land priority down: a cruise-missile campaign opens on airfields and air defence — June 2025: 15–16 Iranian aircraft destroyed on the ground at six airfields). After: Ukraine – Russia sustained 94.8 / 99.0 % with a drone share of 59 / 78 %, campaign 31.8 / 95.2 %; Israel – Iran sprint 93.5 / 76.7 %; Russia – United States campaign 72.4 / 95.1 %; United States – Iran campaign without bases: Iran 73.5 % — air defence −52 %, aviation −38 %, land −8 %, the United States 100 %. Russia's sustained figure stands at the edge of its corridor: the three sustained targets meet only in a narrow band at the floor of intensity_f.low — reported, not forced.

The sprint tempo is anchored to a single reference — the twelve-day war of June 2025 (Israel – Iran, 40 days at the sprint in the model); the sprint figures of every other pair are an extrapolation of that one case, not a calibration.

Three published cases against the model — the facts with their sources, the model's figures for the nearest pair and tempo, and what the comparison says. The facts are counts where a count exists (Oryx, the imagery-based GWAPS tables) and claims where only claims exist (the IDF's and CENTCOM's figures); the model is run as it stands, with no setting changed for the comparison.

Sensitivity (part K.5, re-run at build 2026-10-19s after part N): doubling or halving any single knob of the norms war, sam_load and exposure — 42 knobs: the rates and tempo multipliers of parts B–H, the sortie rates by class, the mission priorities, pk by mission, pk_tempo, land_nocontact, oca_ground, air_q, att_base, sam_x, air_x, int_shots, rounds_per_tube, the reloads, reserve_rate, reserve_days, the exposure of tanks and IFV and of the artillery, and the knobs of part N (uav_hit_kill, cohesion_min, break_loss_x, sat_uav_ch, strike_outside) — changes the 30-day remaining of Ukraine – Russia at the sustained tempo by at most 2.0 points (uav_hit_kill × 2 (Ukraine)) and at the campaign tempo by at most 26.0 points (k.land_land × 2 (Ukraine)), Israel – Iran at the sprint by at most 13.9 points (sortie_rate.fighter × 2 (Iran)), the United States – Iran campaign by at most 8.9 points (pk.ad × 2 (Iran)); the breaking point and the saturation move none of the calibrated pairs (their thresholds are not crossed there). The full table of all 84 runs is in the project report (test/war_sens.js --mode sens; war_figures.js --k name=value runs any single set).

Uncertainty band

Since build 2026-10-19t every run carries its own uncertainty band: when a war opens (and after every change of its settings) the model is re-run silently with each of the six largest movers of this sweep — k.land_land, intensity_f, sortie_rate.fighter, rounds_per_tube, pk.ad, pk.land — at × 0.5 and × 2, twelve runs from day 0 to the last day, in idle chunks beside the real run; the band is the lowest and the highest remaining share by day over those runs, with the base run inside it. It is drawn as a shaded band around each curve of the chart, written as «41–63 % · at ± 50 % on the key coefficients» under the remaining row of the end table, and the «remaining» window names the two coefficients that gave the edges. The band says how far the figure moves when one key coefficient is halved or doubled — not a confidence interval of the outcome: the data gaps, the tempo and the scenario are not in it.

Known limits

What part N did not close and the model does not claim — the reader should hold these against every figure of the war panel:

• The loss ratio of an ambush war: the defender of a one-sided scenario strikes with its land power × def_k (2.0), so the attacker's losses can reach about twice the defender's — March 2022 (3–4 : 1, anti-tank missiles and a logistic collapse) is beyond it, and so is any war decided by terrain, training and logistics, none of which is a term.
• The aviation losses to low-altitude air defence: the SAM term of the losses per sortie is the long- and medium-range coverage × (1 − air superiority); MANPADS, guns and short-range systems are not a channel, so an air force flying low over a dense front loses fewer aircraft than in 2022.
• An uncontested drone war: pk is calibrated on contested air, so a strike UAV against a side without fighters and without counter-UAV air defence (Karabakh 2020) removes 50–100 times less than it did; the model has no term for the defender's counter-UAV capability.
• The sprint tempo rests on a single reference (June 2025); every other pair at the sprint is an extrapolation of that case.
• Targets outside the model: strike_outside takes the launches at energy, cities and industry out of the damage, but their effect — a country without electricity, a refinery on fire, a command post lost — enters nothing; the model counts equipment and nothing else.
• The breaking point is a norm, not an observation of the pair: break_pt 0.5 / 0.4 / 0.3 / 0.4 and cohesion_min 0.2 are the executor's reading of Dupuy and of the withdrawals of 2022, not a calibrated value — a side's morale, command and politics are not in the data.
• The saturation capacity assumes every launcher is in the right place: the channels of the whole park count against the whole salvo, with no geography inside the country and no radar horizon — the salvo of the real night hits one city.
• Thirty days of stock and the published quantities: beyond the window the exhausted stocks stay exhausted (no production beyond the cards' per-month lines), and a card without a published quantity counts as 1 unit — the readiness figure of the country says how much of the model rests on such cards.
• Production, mobilisation, allies, terrain, training, morale, repair, weather and intelligence stay outside, as before; nuclear weapons entirely.

Data readiness

Every documented country carries one figure, «data for Live War: N %» (the Analysis tab, this page's analysis of the country and the war panel show the same one): the share of the fields the war model needs that are on record, weighted by their importance — the combat radius of the combat-aviation types (weight 3: without it the theatre gate of the type is 0), the published quantity of the combat cards (3: a card without one counts as 1 unit), the generation, the UAV role, the 30-day stock of the one-way-attack drones, the missile stock of the launcher cards and the artillery-rounds line of the country (2 each), the bomber range, the interceptor stock of the SAM cards, the missile range, the strike cells of the ships, the air-launched cruise-missile family, the production and the storage lines (1 each). A check with nothing to close does not count; where a field is missing the norm or the quantity stands in, as the parts E–K describe. nc/nc_check.py --war <Country> lists the gaps type by type with the field that closes each; --task writes the research task for the agents, nc/war_data_merge.py merges their results.json, test/test_war_all.js runs the war of every country against three reference opponents and test/war_figures.js --check compares the calibrated pairs with analysis/war_ref.json after every build.

The per-country checks apply only where there is something to close (part L.6): the artillery-rounds line from 20 artillery tubes, the production and the storage lines from 200 cards or with a documented storage, the air-launched cruise-missile family with a carrier aircraft of a cv or alcm family on the cards; a check that does not apply is neither in the denominator nor among the gaps — a small army does not lose 30 % for what it does not have.

Reading the war panel

Every figure of the war panel opens a window that says where it comes from (part J): click a block row (Land / Air / Sea / Air defence & missiles / Enablers), the «remaining N %» of a side, the missiles or drones line, a number in the sentence of the day, or a row of the end table. The block window shows the weighted units of the block on day 0 and today (Σ over its categories of units × idx_w, the mass categories damped above 100 on the root scale), the loss of the last day as the sum of its terms — ground fire, aviation, tactical drones, missiles, long-range drones for the land block; air-to-air and SAM fire for the air block; SEAD for the air-defence block; the fleet and aviation for the sea block; the land share × en_k for the enablers — each with its formula and the figures of the day (k.land_land 0.024 × g_land 1 × P_land 7,833 ÷ S_land 5,336 …), the name and status of the norm behind it and a link to its row in the table of norms, the cap when it was reached, and the power of the attacker behind the terms (P = S × availability × manning × tempo; strike = sup × the reaching combat-air power). The sum of the terms is the loss of the block (above the cap, the cap).

The window of «remaining N %» lists the blocks with their weighted units on day 0 and today, the Wmax of each block (the largest block of the documented countries — the scale of the Inventory Strength Index) and the points 100 × S ÷ Wmax, whose sum today divided by the sum on day 0 is the figure. The missiles window splits the day's launches into land launchers × msl_rate, the fleet (ships × strike weight × nav_rate) and the air-launched families (alcm_rate × the 30-day stock × the gate of the family), shows the interception as min(1, int_max × coverage) with the coverage of the defender's SAM at altitude (batteries × π × range² × overlap ÷ the area of the territory — the Analysis tab) and the damage of the missiles that got through, split by priority × what the defender has; the drones window does the same for the long-range and the tactical drones. The window of the sentence of the day lists both sides' blocks with their terms and the missiles and drones of the day; a card popup during the war shows the type's initial → remaining figure and the causes of the day as shares of its block's terms.

Before the start the panel «Assumptions of this pair» lists what the model has decided from the two inventories and the map: the distance between the outlines, whether the ground forces meet and why, the share of each side's combat-air power that reaches the opponent (the theatre gate by radius, the bombers with tanker support, the carrier wing and the day it starts, the forward bases), which missile types reach, whether the fleets meet and when the fleet strikes, the drone stocks by class, the SAM coverage of each side and the interception it gives, the combat types without generation data, the scenario, tempo, supply and bases. Under it the data line says how many types of each side are in the model, how many have a published quantity (the rest count as 1), how many carry a source link, how many lack generation data and how confident the drone stocks are; the calibration line says whether the pair stands on a direct historical reference — Israel – Iran 2025 at the sprint, Ukraine – Russia 2025 by Oryx at the sustained tempo — or is an extrapolation of the calibrated rates. The end table repeats the panel collapsed and offers the table by day as a CSV (day, side, the blocks, the losses by term, the launches, interceptions and stocks of missiles and drones, the event cards; the header carries the assumptions and the data version) — built in the browser, nothing is sent anywhere. In the Record scene the windows and the button are hidden.

Since build 2026-10-19t the day plate carries the plates of the run: CALIBRATED with the years of the reference when the pair and tempo stand on a historical case (Israel – Iran 2025 at the sprint, Ukraine – Russia 2022–2026 at the sustained tempo), CORRIDOR when the owner accepted a corridor without a historical case (the three campaign corridors), EXTRAPOLATION otherwise; INCOMPLETE DATA with the readiness figure when a side's data readiness is under 60 %; «hypothetical tempo × N» beside the campaign and sprint tempos; a note beyond the 30-day stock window with published ammunition (exhausted stocks stay exhausted — no production beyond the cards' lines); and in Record the permanent «MODEL · not a forecast». The plates warn, they never block a run. The percentages of the panel, the plate, the end table, the cards and the PNG are whole numbers (the top-5 losses with «≈»); the CSV and the «why» windows keep the decimals.

The cinema has three channels of cards: the centre card of an event (with a second line that carries its number — «29 of 58 weighted units · fleet strikes since day 1»), and a compact card in the corner of each panel — the flag and one or two lines, 2.5 s, at most one per 4 s per side, never while the centre card is up. The threshold lines come once (the air defence saturated, the long-range pool on reserve, interceptors for N days, shells for N days, shells exhausted, the drone stock exhausted, the reserves from storage, the breaking point, the first carrier strike, the fleet strike); the status lines rotate about every 6 s (intercepted of launched, sorties · lost · SEAD, FPV launches → vehicles, the long-range drones and their share at targets outside the model, the reach of the combat aviation, the air block and its losses by cause, produced · from storage) — all from the figures of the «why» windows and the assumptions. The director shows the side card of the panel it pushes into. The war code itself (about 330 KB with its styles) is loaded only on the first entry into Live War or by a &war= link, so the map page is a third lighter for everyone who does not open a war.

Munitions, reserves and replacement

Exposure (part I.0): a block still loses its daily share f_g, but its types do not lose it evenly — a type loses f_g × exposure × n_g, where exposure is the loss rate of its category relative to the block (norm exposure: tanks 1.0, IFVs 1.3, APCs 0.5, MRAPs 0.1, self-propelled guns 0.4, towed guns 0.25, rocket launchers 0.25; fighters 1.0, attack helicopters 1.2, transports 0.3, tankers 0.1, UAVs 1.5; boats 1.0, submarines 0.3; SAM 1.0, MANPADS 0.2; trucks 0.3 — the Oryx 2025 additions against the parks of the cards, and the logic of what stands at the front) and n_g the norm of the block, Σ m·q ÷ Σ m·q·exposure over its types, so that the weighted loss of the block stays exactly what the calibration gives; m is the marginal index weight of a unit. The block figures of part H do not move; the losses by type, the top-5 and the fact → model table do. A card's popup shows its share of the day: block share × exposure × n.

Interceptors (I.1): the surface-to-air missiles of a side are a stock. A SAM card with a published 30-day stock carries it (is, with its source iss — Israel's Iron Dome and Arrow, Ukraine's Patriot, S-300 and SAMP/T, the US Patriot, THAAD, Stinger pool and Iron Dome, Iran's Tor, China's S-300, all low or medium confidence); every other card takes launchers × ready missiles per launcher of its family (norm sam_load) × (1 + reloads in the window: 3 at the sustained tempo, 2 in a campaign, 1.5 in a sprint), and the panel says «estimated by norm». Each interception spends int_shots missiles — 2 per ballistic missile, 1.2 per cruise missile, 0.3 per long-range drone (the tactical drones fall to electronic warfare and mobile groups, not to missiles). The stock is drawn in two pools: the long- and medium-range types alone engage ballistic missiles; cruise missiles and drones draw on the short-range types first. When a pool does not cover the day, the interception falls in proportion to int_floor 0.15 for missiles and uav_floor 0.5 for drones (guns, electronic warfare, aviation); without the long- and medium-range pool the coverage no longer cuts the air strikes (g_ad = 1), without any interceptors the SAM units stop firing at aircraft (P_sam = 0). The panel line «interceptors: fired · stock left», the end table, the event card «air-defence interceptors exhausted», the sentence of the day and the «why» window carry it.

Shells (I.2): the ground-fire term of a side is limited by its artillery ammunition. Norm ammo gives the rounds for a 30-day window where they are published — Russia 475,000 (its 2025 output plus the North Korean deliveries, no stockpile added: the intelligence services say it was mostly spent), Ukraine 138,000 (its own output plus the Czech initiative and the German deliveries) — and null where they are not (the United States, Israel, Iran, China): then the fire is not limited and the panel says «not limited — the stock is not published», honestly, without an estimate. The need of a day = the tubes that fire (the park of self-propelled and towed guns and rocket launchers × availability × manning, as in P_land) × rounds_per_tube at the tempo — 3 a tube a day sustained (Russia 2025: 10,000–15,000 a day on about 4,000 guns), 20 in a campaign (the Desert Storm averages), 50 in a sprint (one Soviet unit of fire); nothing is fired without ground contact. The fire term is multiplied by the smaller of 1 and stock ÷ need, and by fire_floor 0.2 when the stock is gone (direct fire and mortars). At the campaign tempo the published flows last 9 days (Russia) and 3 (Ukraine) — a campaign on the 2025 flows runs out of shells within two weeks, which is what the stockpile assessments say.

Storage and production (I.3): equipment in storage judged restorable sits on the cards (rs, with its source rss — Russia 8,354 units on 26 cards from the 2025–2026 satellite counts of the storage bases and the IISS in-store lines, the United States 6,616 on 12 cards from the AMARG and the Army depots, Israel's M113) or as a category total in norm storage (Israel's 400 tanks); from day reserve_delay + 1 (day 8) every stored type returns reserve_rate × its stored stock a day — 0.3 % at the sustained tempo, 1 % in a campaign, 1.5 % in a sprint — into its quantity and its block, never into the losses, and never above the quantity it had on day 0 (storage replaces losses, it does not grow the army). New production (norm prod, units a month by country and category, to the named types where the source names them: Russia T-90M 12, BMP-3 40, BTR-82A 50, artillery 25, Su-34 / Su-35 / Su-57 one each; Ukraine Bohdana 40; the United States F-35 8, AMPV 8, HIMARS 4; China J-20 10, J-16 8, ZTZ-99A 5, PCL-181 12 …) adds per_month ÷ 30 a day to its types, with no ceiling — a side with no losses shows 100.1 % remaining. The panel line «replacement: +N from storage · +N produced», the end table, the feed and the «why» window show the arrivals; the external supply switch of part E stays separate.

What the model does not have: mobilisation (the storage return and the production are fixed flows, nothing accelerates with the tempo); the stockpiles behind the published monthly flows of shells and interceptors (where they are not published the model assumes none, or no limit); allied resupply in the war (the interceptors the United States fired for Israel are not on Israel's cards); the rationing a staff applies when a stock runs low (the model fires at the full rate to the last round); repair (a damaged vehicle is lost, while Oryx counts damaged and abandoned too); command, electronic warfare, intelligence and training (a generation factor and a manning factor are the only quality terms); terrain and logistics (the front is a line of contact, the rear is never cut); personnel and morale (people enter only as manning and never break); stealth and stand-off weapons (an F-35 and an F-4 take the same SAM share per unit of power); dispersal, shelters and decoys (the same share for every unit of a block); allies and coalitions (one country against one); manoeuvre — it is a linear attrition model: the sides wear each other down in place, nobody encircles, withdraws or surrenders.

Table of norms

Formation equivalents · 18

Manning & readiness · 8

Budget realism · 10

World standing · 2

Inventory Strength Index · 12

Stage 2: air-defence coverage, strike reach, fleet age · 15

Stage 3: consumption and days of stock (My Army) · 7

Live War attrition model (VS) · 15

The notes to the norms are given in English together with the sources — a technical appendix; the labels, values and statuses of the table are translated.

What it does not measure

Training, doctrine, command, morale, readiness of individual units, combat experience, allies, geography, industry, stockpiles of ammunition and spares, the age and condition of every vehicle. The inventories themselves are open-source estimates: sources disagree, fleets change, and quantities marked + are lower bounds. Read every number as an order of magnitude and compare like with like.

Planned

Stage 3 (My Army only): fuel and ammunition consumption by published norms and days of stock; ranges and years for the remaining categories (tanks, artillery, patrol boats) as the research passes continue.

Open the interactive map