The General Staff figure has risen in three steps: about 200 a day in mid-2022, 600–1,000 through 2023, and 1,000–1,500 from 2024. In 2026 it climbed again, averaging 1,750 a day in October and 2,277 a day in early November, the highest sustained level of the war. Fighting in Ukraine is known to follow the ground. Mud in spring (rasputitsa) and late autumn slows armour and assaults, and summer and frozen winter ground favour them.

No source outside Ukraine publishes a measured length for the active front. Ukraine's command gives 1,200–1,250 km (reported by AP, September 2025, unverified). Independent analysts' maps give about 1,000–1,100 km. The model below works with any figure in that range, and the answer barely changes across it. A long front works against the "too consistent" argument. A daily total built from hundreds of separate sectors should be fairly smooth, in the same way a national daily birth count is smoother than one hospital's. The real question is how smooth. The model below adds up independent sectors and lets you set how clustered the losses are and how much front-wide events (weather, big offensives) move them.

Daily weather comes from five front-line towns (Pokrovsk, Bakhmut, Kupiansk, Orikhiv and Kherson) via the Open-Meteo historical archive. Each day's claim is compared with its own 29-day median, so the test looks only at day-to-day movement. Bars show how far above or below a typical day the claim sits in each weather band, with 95% confidence intervals.

A real casualty count should jump when a set-piece battle peaks, and dip when a truce holds. Battle dates come from ISW, CNN, Al Jazeera, AFP, Meduza and other sources outside Ukraine. For each battle's most intense phase, the claimed daily average is compared with the six weeks before it, then ranked against every other stretch of the same length since June 2022.

Two military explanations could make an honest count look strange. Firepower: Russian losses should rise when Ukraine's ability to kill rises. That ability went from a shell shortage in early 2024 to drone dominance in 2025–26. Posture: the old rule that attackers lose more is weaker than it sounds. Across hundreds of historical battles, the Dupuy Institute found attackers' losses are on average similar to defenders'. Only failed attacks cost the attacker clearly more, about 1.7 to 1. The famous 3:1 rule is about how many troops an attack needs, not about who loses more.

Russia mobilised 300,000 reservists in autumn 2022 and has relied on paid contract recruitment since. Official figures come from Dmitry Medvedev and the defence ministry. Independent estimates come from regional enlistment-bonus spending (Janis Kluge, SWP; iStories). Vladimir Putin has put the Russian force in the war zone at 617,000 (December 2023), almost 700,000 (June 2024), 700,000 (September and December 2025) and more than 700,000 (June 2026). Since mid-2024, then, Russia's own figure has the force roughly flat.

Mediazona and BBC Russian confirm deaths one name at a time from obituaries, social media and leaked registries. It is a hard floor, and it lags by months. Separately, Mediazona and Meduza estimate total deaths from Russia's probate registry, which shows excess deaths among men of fighting age. Neither counts the wounded. Ukraine's figure does.

Ukrainian commanders and ministers regularly give their own figures for Russian losses, for a single battle, a month or a year. These statements are the thing being tested, so they come from Ukrainian officials by definition. Each is checked two ways: against the General Staff's own daily totals for the same dates, and against figures for the same battle from governments and researchers outside Ukraine.

In January 2026 Ukraine's defence minister Mykhailo Fedorov named 50,000 Russian casualties a month a strategic goal, as reported by Business Insider and Al Jazeera. Under a points system, drone units earn credits, spent on new equipment, for each kill confirmed on video. Video confirmation is a real safeguard. A public target plus a reward for each counted kill is still the textbook setting for Goodhart's law: a measure that becomes a target stops being a good measure.

Since 2025, most of the figure rests on drone footage. Outside sources put drones behind 70–80% of casualties, and Ukraine's own video-verified drone count equals 63–107% of the monthly total in 2026. An FPV drone's video ends when its camera dies. That happens on a direct hit, but also when the drone strikes the ground a few metres short, is jammed in the last seconds, or the target dives clear. From the strike drone's own feed you cannot tell these apart, or tell killed from wounded. Only a second camera watching the aftermath settles it.

Ukraine's defence ministry says every counted strike is verified on video inside its own systems. It has not published the standard, and no outside party audits it. The one outside measurement comes from RUSI, which found that only about half of Ukraine's publicly claimed drone kills of Russian artillery were real destructions.

Russian losses are estimated by three kinds of source: governments and institutions that back Ukraine, independent researchers, and Russia itself. Each estimate below is set against the Ukrainian MOD's running total on the same day. Casualty estimates and death estimates are kept on separate charts, because mixing them is the most common error in this debate.

All statistics were computed from the raw series: 1,716 daily General Staff figures (24 Feb 2022 – 6 Nov 2026), Open-Meteo daily weather averaged over five front-line towns, and AFP's monthly analyses of ISW maps (2024–2026). The General Staff series and officials' statements are the subject under test, so they are necessarily Ukrainian. Every other input comes from outside Ukraine; DeepState appears only as a comparison. Purported Russian documents released by Ukrainian intelligence (an alleged Kremlin loss assessment, March 2026, and an alleged Russian loss sheet, October 2025) are excluded, because their provenance runs through a party to the war. Daily tests use 2023 onward, after the war of movement ended. Correlations are Spearman rank correlations. Recruitment, death and battle figures are as published by their sources and carry their sources' uncertainty.

Each month is compared with the average of the three before it. These are the largest moves in the armour series, with what was happening on the ground.

Oryx, run by Dutch analysts and volunteers, counts only Russian equipment losses shown in photos or video, each linked to its source. It is a floor: it misses whatever goes unfilmed, and its pace has slowed since its main author stepped back in 2023. It is still the only item-by-item count of Russian equipment losses outside Ukraine's government.

LostArmour is a Russian pro-war project that logs Ukrainian equipment losses from photos and video, each with a date and a link, much as Oryx does for Russia. As a pro-Russian source it has every reason to count generously, so its totals are checked against Oryx's own independent list of Ukrainian losses. On armour the two agree closely.

Ukraine's General Staff equipment claims come from the same daily bulletins as the manpower series. Oryx totals come from its live page and Wayback Machine snapshots from July 2022 to November 2026. Months are compared with the average of the three before them. Research findings come only from sources outside Ukraine; figures that trace back to Ukrainian units or officials are left out.