Build data · Guide 06 of 08 · PoE 1 and PoE 2
Sorting by class or narrowing to a keystone can flip which build looks strongest. Not because one view is broken and the other is honest, but because a filter quietly swaps the question you asked for a different one, and then answers that instead.
The short version
Every filter on a build overview subtracts characters from a sample that was already truncated by the ladder itself. It never adds any. Once the subtraction happens, the percentages are recalculated against whatever is left.
So a build showing 3% of the whole ladder can show 34% inside a keystone filter, and both figures are correct. What changed is the denominator. Read the population before you read the ranking, and a lot of apparent contradictions stop being contradictions.
It is tempting to imagine that a filter reaches into Path of Exile and asks it a question. It does not. By the time you click anything, three things have already happened, and your filter is the fourth and least powerful of them.
What happens before your filter does
Grinding Gear Games' developer documentation sets the window: roughly 15,000 entries in Path of Exile and 1,000 in Path of Exile 2, paged 200 at a time. Everyone below that rank is never returned to anyone, ever.
Cannot be filtered back inCharacters are fetched, their gear and trees parsed, their main skill inferred. This is a photograph taken at one moment, not a live feed. A character who respecced an hour ago is still wearing yesterday's build in the snapshot.
Fixed at read timeOne class, one ascendancy, one main skill, a list of keystones, a list of uniques. The labels are what filters match against, so anything the labelling got wrong becomes invisible rather than incorrect.
Where attribution errors enterYour click removes rows from that stored table and recomputes the totals. That is the entire operation. It is subtraction plus arithmetic, applied to a sample that three earlier decisions already shaped.
Your only moveThat order matters more than it sounds. Because filtering happens last, it cannot repair anything the earlier stages did. If your ascendancy is rare enough that only forty of its characters cracked the rank cap, filtering to it does not find you more. It shows you the forty and prints a confident percentage next to them. The same trap turns up in pricing, where a summary value with eleven listings behind it looks identical to one with four thousand.
Popularity on a build overview is a share, not a count: this skill divided by everything in view. Filtering changes what is in view, which changes the divisor, which changes the share. Nobody is hiding this, but nothing on screen announces it either, and the number is presented with the same authority before and after.
The console below runs one sample through four questions. Same characters throughout. Watch the ranking, not just the percentages.
Sample figures for demonstration · live build data at poe.ninja
Nothing in that console contradicts anything else in it. Forbidden Rite is a minor skill across the whole ladder and the dominant one inside the keystone that makes it work. A player who reads only the unfiltered view concludes it is fringe. A player who reads only the filtered view concludes it is the best skill in the game. Both have read a real number and reached a wrong conclusion, because neither read the population it was drawn from.
This is also why how build popularity is calculated is worth understanding before you use popularity to decide anything at all.
These are not equally common, and they are not variations on one problem. Each has a different cause and a different tell.
Composition
Skills are not evenly spread across classes. When one class is unusually popular, its skills inflate the unfiltered ranking, and everything belonging to a rarer class is pushed down by arithmetic rather than by being worse.
Filter to a class and that pressure disappears. What was fourth on the whole ladder can be first among the players who could actually run it.
Tell A skill whose share jumps by more than about three times when you filter to its class was never really competing in the unfiltered list.Survivorship
Filter to a demanding keystone and every character returned is, by definition, one who got far enough up the ladder to be sampled while using it. The players who took it, died repeatedly at level 74 and rerolled are not in the window.
The result reads as evidence the keystone is strong. It is closer to evidence that the survivors survived.
Tell Filters on high-risk keystones almost never look risky. The failure cases were removed before the data reached you.Sample collapse
Each filter multiplies the shrinkage. Class, then ascendancy, then keystone, then a level floor, and a five-figure sample can land in the low dozens without any warning on screen.
This bites hardest in Path of Exile 2, where the ladder window is a tenth the size before you filter anything. One organised guild running the same setup can visibly move a percentage.
Tell Under about 30 characters, treat every percentage as a description of those specific people rather than of the game.Sort field
A damage column is produced by calculating each character under an assumed setup: buffs running, flasks up, charges held. Sort by it and you rank characters by how well their build matches those assumptions.
Builds that hold everything permanently rise. Builds that need a two-second window to reach the same number sink, even where the real ceiling is higher.
Tell If a sort moves a build sharply while the filter stays identical, you are looking at an artefact of the sort, not a difference in the build.Attribution
Nobody tells the tracker what your build is. It is worked out from your gem setup, which is straightforward for a single six-link and much less so for weapon swaps, trigger setups, totems, minions or a Vaal version sitting beside its base gem.
A character filed under the wrong skill vanishes from the right filter and pads the wrong one. Both errors are silent.
Tell Open a few results. If several are visibly not running the skill you filtered for, adjust every number on that page downward in your head.Wealth
Filtering to an expensive unique returns the players who could afford it. They also tend to be the players who could afford everything else, so the whole filtered set looks stronger than the item alone can explain.
The item gets credit for the wealth around it. This is the same pattern behind expensive builds dominating rankings.
Tell Compare the rest of the gear in the filtered set. If it is uniformly excellent, the filter selected for currency, not for the item.Here is the same operation stated plainly. Nothing on the right is a correction of the left. Each pair is one dataset answering two different questions.
Before the filter → after it
Forbidden Rite sits around 3% and looks like a curiosity.
Inside Chaos Inoculation it is the most-used skill in view.
A well-known bow skill leads comfortably.
Slam and channelled builds take the top, on assumed full uptime.
Cheap league-start builds hold most of the share.
Level 95 and up
Endgame specialists replace them almost entirely.
A workable sample of a few thousand characters.
Forty-odd characters, still shown as tidy percentages.
Most people filter to find the best build. That never works, because the tool has no concept of best. It works far better in reverse: describe the character you are actually able to play, then read what the survivors of that description did.
Every figure is scoped to one league of one game, and the two games are tracked separately. A Standard reading against a fresh-league plan is not slightly off, it describes a different economy and a different meta.
Why first: get this wrong and every step below is precise about the wrong population.The class you are committed to, the ascendancy you have already taken, the one unique you own. These are facts about your character. Filtering by what you wish were true just produces a list of other people's shopping.
Why: constraints shrink the sample toward relevance. Ambition shrinks it toward wealth.This is the habit that separates people who use build data well from people who get burned by it. Somewhere on the page is a count of characters matching your filter. That count sets how much weight anything else deserves.
Comfortable above a few hundred. Directional in the low hundreds. Anecdotal below about thirty, whatever the percentages look like.
Why: percentages are formatted identically at n = 4,000 and n = 14.If dropping your level floor or your keystone reshuffles the ranking completely, the ranking was a property of the filter rather than of the builds. If the same two or three setups stay near the top across several views, that is a real signal.
Why: a conclusion that survives a changed denominator is worth acting on.Take three results running the same skill and compare their gear. What all three share is the build. What one has is that player's budget. Then export and import into Path of Building and match the configuration on both sides before comparing any damage figure.
Why: the filter narrows the field. Only the export tells you what the character actually does. See the Path of Building guide.Four situations where the number on screen is technically correct and practically useless. Recognising them takes about five seconds each once you know the shape.
Large shares in small samples are the normal behaviour of small samples, not a sign of consensus. Drop a filter and check whether the dominance survives a bigger denominator.
Almost always the rank cap rather than a bug, and in Path of Exile 2 the cap is only a thousand deep. The character listing eligibility checker and the character listing guide cover the specific reasons.
You changed the sort, which changes what the list is ordered by, not what it contains. Damage sorting in particular rewards permanent uptime, which is why the highest DPS build may not be the best one.
Defensive trade-offs do not appear in a popularity ranking at all. A filtered list can be entirely composed of builds that fold to one unlucky hit, and nothing on the page would say so. Compare them properly using the defence comparison guide.
No. Filtering only removes rows from a sample that was already collected inside the ladder's rank cap. Anything below the cap was never available to be filtered in, so no combination of filters can surface it.
Because the divisor changed. A share is that skill divided by everything currently in view, so narrowing the view raises the share of whatever remains. Both figures are accurate answers to two different questions.
As a working rule: a few hundred characters or more is reasonably solid, the low hundreds is directional, and under roughly thirty is anecdotal. There is no hard threshold, and the page will format the percentage identically at every sample size, so the count is yours to check.
They collapse faster. The PoE 2 ladder window is around 1,000 characters against roughly 15,000 in PoE 1, so the same two filters that leave a healthy sample in one game can leave a handful in the other. The mechanics are identical; the headroom is not.
Because everyone it failed for is missing from the sample. Ladder data only contains characters who climbed high enough to be recorded, so risky builds are represented exclusively by their successes.
Main skill, when you want to know how a build is put together. Unique item filters answer a narrower question about who owns that item, and the answer is heavily shaped by wealth rather than by the item's merit.
Filters, sorting and the export button, in the order that saves the most time.
FoundationsWhat counts as one character, and which characters never enter the count at all.
CostlyThe misreadings that cost the most currency, and how to avoid each one.
Also useful: build rankings, build comparison, how the data is assembled, the glossary, and the game-specific hubs for Path of Exile 1 and Path of Exile 2.
Poeninjaa.com is an independent educational website. This product is not affiliated with or endorsed by Grinding Gear Games or poe.ninja. All figures on this page are illustrative examples chosen to demonstrate how filtering behaves, not live data.
Path of Exile and Path of Exile 2 are trademarks of Grinding Gear Games. For current build and economy data, visit poe.ninja directly. Found something wrong here? Tell us and it gets checked, fixed and logged under our corrections policy.