Weapon Scoring

When we started, the easy road was right in front of us. Take the armor thickness. Take the gun caliber. Take the range. Give each one a weight. Add it up. Call the result a score. Running a formula off a spreadsheet is far cheaper and far faster than asking an expert whether this tank was actually any good. On paper it looks a lot more objective too. We genuinely tried it that way for a while.

Then we looked closely at our own data, and that is where we understood why the approach was wrong. The spec data we hold is nowhere near as reliable as it claims to be. Some weapons have the wrong armor thickness on file. Some carry a manufacturer figure that flatters the real number. Some are just simple recording errors, and we caught a good number of these ourselves while going back through the data. A formula takes those errors without ever questioning them, runs the math, and hands you a result that looks scientific but is wrong underneath. An expert does not work that way. Before trusting the paper in front of them, they ask whether it holds up. They weigh it against what they already know, and when the two disagree, they trust their own knowledge over the sheet.

That is exactly how we score weapons. We ask the AI what it genuinely knows about a given system, and we treat our own stored spec data as nothing more than a secondary hint. Specs are context. A starting point. Never the final word. A T-72's armor can look impressive on paper. But to anyone who knows the ammunition sits in a rotating carousel directly beneath the crew, that number means something else entirely. It means a single penetrating hit turns the crew compartment into an oven. That is why we score the T-72 so low on crew survivability. No spec sheet will ever show you that. The real combat record does.

A weapon being famous is not, on its own, proof of anything to us either. Sometimes a reputation is fully earned, a design proven in the field, trusted, feared for good reason. Sometimes that same fame comes from a good marketing campaign, a movie, or one striking piece of footage, and has little to do with real performance. We try to pull these two things apart when we score. What we are actually looking for is what stands behind the reputation.

We give every weapon two separate numbers, because the question of how good something is actually has two different answers. First, where does this weapon stand against everything ever built. Second, where did it stand against its real contemporaries at the time. Those two questions often produce very different numbers, and both are true. A T-54 fielded in 1946 is an unremarkable tank by today's standards, and it scores low on the absolute axis. But against the other tanks built in the same years, the T-54 was genuinely dominant, which is why it scores high on the era axis. The Soviet IS tank that entered service in 1943 looks weak from today's vantage point, but it led its era in firepower. The era axis captures that. The absolute axis does not. Reducing a weapon to one number sends the false message that anything old must be bad. We keep this split on purpose.

These two axes also say something else. We judge a weapon within its real contemporaries, not in a vacuum. Weapons that fall in the same category and the same period are weighed directly against each other, who was actually better, who was actually worse. That is not an abstract idea. It is a direct question we ask every time. Where did this tank stand against the other tanks built in the same decade.

Our price-performance score works on the same logic, and here too we deliberately stay away from a mechanical formula. Score divided by price sounds mathematical, but in practice it means very little, because most of the prices you see on this site are incomplete or uncertain. At this scale, our pricing data will never be complete. We still give a price-performance score anyway, because it does not rest on a stored number, it rests on what is genuinely known about that weapon. Not knowing the price does not make the judgment impossible. Did this weapon actually deliver value for what was spent on it. Was it genuinely preferred in export markets, or did it only sell because of political pressure. Was there something better for the same money. The answer to those questions becomes a single score.

On weapon pages you will also find a localization card. It shows what share of that weapon is domestically produced. But we do not count every part equally. Nobody cares where a tank's seat cushion came from. What matters is where the engine, the armor, and the fire control system come from, because that is what actually answers whether a country could keep producing or repairing this weapon on its own in wartime. In today's world that is not a simple sum. A single helicopter engine can be built from parts sourced across dozens of countries, the turbine blade from one, the electronic control unit from another, the alloy from a third. We decide, category by category, what counts as critical and what counts as secondary. For a tank that means the armor and the gun. For an aircraft it means the engine and the avionics. So the localization percentage is never a flat parts count. It is a summary of who that weapon actually depends on, and how much.

None of this is perfect, and it never will be. There are more than fourteen thousand weapons on the site right now, each one evaluated this way, and staying error-free at this scale is not possible. From time to time we find a score that is simply wrong, a component that came out unrealistically high, or too low, and when we notice we go back and fix it. This very page is an example. For a long time it described an old formula we no longer use. The system had changed, but the description had not, and once we noticed, we fixed that too.

We will not hide this from you either. Even for the most famous, most recognizable weapons, our current data accuracy sits somewhere around forty to fifty percent. We have said this before, and we are saying it again here. Every one of our fourteen thousand plus records sits at a different level of maturity. Some have been through many passes and are solid. Some are still a first draft, waiting to be corrected. We would rather show you that than hide it.

Do not assume there is one true number for a weapon's real armor thickness or range. What the manufacturer states and what turns up in real military reports frequently do not agree, and we sit right in the middle of that disagreement. That is why you can occasionally run into a value on the site that looks inconsistent. The verification work we have built to sort through this runs slowly, but it never stops. Day and night, seven days a week, it keeps weighing every new source we get against the old ones to work out which one deserves more trust.

Our long-term goal is clear. We want to turn this scoring into something far more scientific and statistical than it is today, replacing intuition-driven judgment with a repeatable method backed by verifiable data wherever that is possible. We do not offer an API right now. The reason is simple. Our data quality has not earned that yet. But we intend to get there. One day we want this data available to everyone, completely free, through an API. Maybe one day an army will look at our price-performance and scoring data while choosing a weapon for its own inventory. That may be a distant dream, but we think in the long term.

So read the scores here as the best judgment we could produce with what we know, open to challenge, not as settled fact. A score you see in 2026 can change substantially by 2027, because new evidence surfaces, because a weapon performs unexpectedly in a real conflict, or simply because we made a better judgment call the second time around. If a score looks wrong to you, you are probably not entirely off base. Either we have not found that mistake yet, or you know something we do not. Both are possible, and both are part of how this works.