Max-level stats are the wrong number to compare

Ask which of two units has more attack and you will get an answer measured at maximum level. It is the number on the card, the number in the tier list, the number every database leads with — including, by default, this one.

It is also the number that describes your roster at a point most of it will never reach, and it is not merely imprecise. For some pairs it gives the opposite answer to the level you are actually at.

A worked case

Two Arknights 6★ operators, both fully promoted, compared on ATK. On this site the level axis runs continuously through the promotions: Lv.1–50 is Elite 0, Lv.51–130 is Elite 1, Lv.131–220 is Elite 2. So Lv.50 is "E0 max", Lv.130 is "E1 max", and Lv.220 is the number the card shows.

At E2 max, Blaze leads by 77 ATK. At E0 max, Rosmontis leads by 32. At E1 max they are within three points of each other. Three completely different answers to "which one hits harder", all correct, all about the same two operators.

The curves do not merely start at different heights. They cross.

Wiš'adel against Blaze runs the same way: 460 against 429 at E0 max, 687 against 765 at E2 max. Gavial the Invincible finishes 111 ATK clear of Nymph and spends the whole of Elite 0 behind her.

Across the 126 six-star operators whose curves span the full axis, 452 pairs invert somewhere between Lv.1 and E2 max. That is not a handful of curiosities; it is a routine property of how the growth tables are written.

Why this happens

A growth curve in Arknights is a set of breakpoints — a value at Lv.1, a value at each promotion boundary, a value at the cap — with the levels in between interpolated. Two operators can be given the same endpoint and very different slopes, or the same slope and different starting points. Nothing forces the ordering to be stable along the way, and the designers have no reason to make it so: they are balancing the operator at the levels players actually use it, not at a single point on a spreadsheet.

The practical consequence is that an operator can be a good early investment and a mediocre finished one, or the reverse, and the card tells you nothing about which.

The second problem: "max" is not one level

Comparing two units at their respective maxima is only meaningful if the maxima represent the same amount of investment. In Arknights they do not. Sorted by rarity, the top of the continuous axis is:

A 4★ at its cap and a 6★ at its cap are not the same experiment. The 6★ has 45 more levels of runway and a far more expensive promotion behind it. Ranking them against each other at "max" quietly bakes in a cost difference that never appears in the comparison.

This is the reason the percentile figures on unit pages are computed within a rarity cohort rather than across the whole roster — a 4★ measured against 6★s would look uniformly terrible and tell you nothing. See how to read a stat percentile for what that cohort is and where it stops being the right one.

What to compare instead

• Compare at the level you will actually reach. If your account promotes to E1 and stops, E1-max is your number and E2-max is a fantasy. Every game we track with a level system has a Stat Calculator that takes a level and returns exact values — not interpolated between tiers, computed from that unit's own curve.
• Compare at the level you are at today when you are choosing what to spend materials on next. The question "who should I promote" is a question about the next 30 levels, not the last one.
• Check the shape, not just the endpoints. Two units with the same maximum can differ by 30+ points through the whole middle of the curve, which is where your account spends most of its life.
• Stay inside a rarity when you rank. A cross-rarity stat comparison is mostly a comparison of promotion costs.

None of this makes the max-level number useless. It is the right number for "what is the ceiling on this unit", and that is a real question. It is simply not the same question as "which of these two should I build", and the two get conflated constantly.