Aion 2 Builds
Menu

Methodology: how we sample and label data

Where each number comes from, how the sample is drawn, how often it changes, and where it stops being reliable.

Updated By Zane Game version: Global launch build Sources

Editorial

On this page
  1. Where the data comes from
  2. How the armory sample works
  3. How often it updates
  4. Confidence labels
  5. Item pages and indexing
  6. What the data cannot tell you
  7. Corrections

Where the data comes from

  • Official character pages. Public NC / PlayNC character and search pages for AION 2, read without logging in. This is the base of every class statistic on the site.
  • Official item interface. The item detail that NC publishes for equipped items gives base stats, item level and enchant limits. We publish item base values only, never one player's roll.
  • Official notices and patch notes. NC and PlayNC announcement pages come first for release facts, dates and rules.
  • Community sources. Used only when they carry a source link and a label from the list below. When community sources disagree with each other or with an official page, we show both side by side instead of picking one.

The full source list with the date we last opened each one is on the sources page. We rewrite facts in our own words and do not copy third party text or tables.

How the armory sample works

The latest sample is dated 18 October 2026 and covers 1,600 characters (8 classes, 5 Global regions). Its label is Official-sample.

  1. For each class we collect level 45 candidates (about 400 per class) from the public character search, using an equal quota per region.
  2. We read every candidate's combat power and keep the top 200 per class for a full read of equipment, skills, stigmas, titles, wings and pets.
  3. Daevanion boards are the same for every player, so their structure is read from two characters per class.
  4. Item base values come from the official item interface, at enchant 0 and at the most common enchant level.
  5. Only aggregated numbers are kept. Player names, character IDs, profile images, server names and guild names are not stored in published data. A privacy check runs before any sample is accepted. Raw files stay on our own machine and are not published.
  6. Combinations such as stigma sets or full gear sets are only published when they were seen at least 3 times.

The search only returns the first few pages of results, so this is not a random sample of all players. Requests are plain GETs with no login and no browser automation, at most 1 request per second, with a hard cap of 6,000 requests per day. The User-Agent names this site and a contact address, and the job backs off when the server returns errors.

How often it updates

  • The armory sample is rebuilt on a schedule, aiming at daily. Every sample-based figure shows its own sample date, which is the date to trust.
  • Because each run draws a fresh sample, a difference between two dates can be sampling noise. We call it a change between samples, not a trend.
  • We also check NC notices and patch notes for new releases and redeem codes, and update the affected pages and their dates when something changes.

Confidence labels

Every data block carries one of six labels. We never upgrade a label without new evidence. The full rules are in the editorial policy.

  • Official: Given directly by an NC page or interface.
  • Official-sample: Aggregated by us from sampled official armory pages.
  • Client-data: Extracted from the Global client by a third party and checked by us.
  • Observed: Three or more independent community sources agree.
  • Editorial: Our own judgment, with the method explained.
  • KR-only: Only confirmed in Korea or Taiwan. Not a Global guide.

Item pages and indexing

Every item with official base stats keeps a page (459 records in the current sample). A page is offered to search engines only when the item was worn at least 70 times in the sample. Less common items stay reachable and searchable on the site but are marked noindex, so thin pages do not crowd the index. The split is recomputed from each new sample, not fixed in a list.

Light and Dark versions of one piece share a name and an identical base stat line. We count them as one piece: wearings are added together for comparisons and guides, and the Dark page points to the Light page as its canonical version. The indexing threshold uses the representative record's own wear count, not the merged total.

What the data cannot tell you

  • Combat power is not a published formula. NC does not document how combat power is calculated, so we use it only to rank characters inside the sample and make no claim about what drives it.
  • The sample leans toward strong characters. We keep the top 200 by combat power from each class's candidates, so results describe strong level 45 characters, not every player.
  • Regions are a quota, not a census. The region mix is set by our equal quota. It does not show where players actually are.
  • Equal sampling per class hides class popularity. Class shares show relative strength among the sampled top characters, not how many people play each class.
  • Changes between samples are not trends. Each run is a new draw.
  • Not enough data means no conclusion. Small counts are shown as small counts, and we do not draw conclusions from them.

Corrections

If a page is wrong or out of date, write to [email protected] with the page link and, if you have one, an official source. See the contact page for what we can answer. The page's updated date moves when a fact changes.