How grades work
The inputs are the game’s own numbers: displayed DPS, cost, range, hitbox, farm, and placement limit. Never estimated, never typed by hand.
The weights are ours. They are editorial, not derived from the game, and they are the reason two jobs can grade the same unit differently.
Weights per job
Each column is one job. Overall’s column is the fallback for a unit that is not on a job list. The numbers are the same weights the job grades use.
Editorial weight per score component, for each job.
| Component | Overall | DPS | Support | Utility | Crowd-control |
|---|---|---|---|---|---|
| Output | 12 | 50 | 0 | 20 | 0 |
| Efficiency | 8 | 10 | 0 | 5 | 0 |
| Reach | 20 | 20 | 10 | 15 | 30 |
| Utility | 40 | 15 | 80 | 40 | 55 |
| Reliability | 12 | 5 | 5 | 15 | 15 |
| Economy | 8 | 0 | 5 | 5 | 0 |
What each score is
6 axes, each in its own units, then min-maxed against this roster.
- Output
- Displayed DPS at max tier, multiplied by how many copies you can place. Only attacking units have this axis.
- Efficiency
- Output divided by the yen spent to reach max tier. Zero when there is no output, or when the cost is missing or zero.
- Reach
- Hitbox shape weight times inner-normalised size (when size applies, otherwise 1) times inner-normalised range at max tier. Support and Farm still contribute: their reach is real even when they do not attack. Shape weights are in the table below.
- Utility
- A weighted count of capabilities, not a flat count. Stun, Block and TeamBuff are worth more than AbilityUptime; Drawback subtracts. On a job list, only that job’s capabilities count. The weights are in the table below.
- Reliability
- One minus the unit’s dependency count over the longest dependency list on the roster. A unit that needs a whole team assembled is worth less than the same unit that needs none.
- Economy
- Farm yen at max tier, or zero if the unit does not farm.
Reach · hitbox shapes
Shape weights for reach, strongest coverage first.
| Shape | Weight |
|---|---|
| Full | 5 |
| Circle | 4 |
| Cone | 3 |
| Line | 2 |
| Radial | 1 |
| Support | 1 |
| Farm | 1 |
Utility · capabilities
Editorial weight per capability. Drawback subtracts.
| Capability | Weight |
|---|---|
| Stun | 3 |
| Block | 3 |
| Team Buff | 3 |
| Shield | 2.5 |
| Anti Shield | 2.5 |
| Slow | 2.5 |
| Summon | 2 |
| Copy | 2 |
| Do T | 2 |
| Mark | 2 |
| Aura | 2 |
| Follow Up | 2 |
| Ramp | 1.5 |
| Crit Scaling | 1.5 |
| Status Scaling | 1.5 |
| Boss Amp | 1.5 |
| Execute Scaling | 1.5 |
| Max Health Damage | 1.5 |
| Economy | 1.5 |
| Transform | 1.5 |
| Cleanse | 1.5 |
| Mobility | 1.5 |
| Ability Uptime | 1 |
| Drawback | -3 |
Axes a unit does not use
Units that do not attack, and units that deal no damage, are excluded from output and efficiency. The support and crowd-control jobs never score those axes either — a buffer is not a bad carry, and stall is not a second DPS list. Those axes are removed from the blend and their weight is redistributed across the remaining components — not scored as zero.
Utility is per job
On a job list, only capabilities that belong to that job count toward utility. Copy does not help crowd-control; a stun does not help support.
How overall is formed
Overall is composed from the unit’s job grades, not from a separate count of kit tags. The strongest job leads; a second job adds a smaller share. Extra tags on the same kit cannot win overall. Farms, which are not on a combat job, use a fixed overall score.
How much each job grade contributes to overall.
| Job | Weight |
|---|---|
| DPS | 1 |
| Support | 0.85 |
| Utility | 0.4 |
| Crowd-control | 1 |
How bands are cut
Bands are proportional. S is the top tenth of this roster, not an absolute bar. A flat meta still produces some S units.
Capability classification
The classification behind utility is hand-authored and maintained per update. A new unit is not graded until its kit has been read into that vocabulary.
What this cannot see
The model cannot see team synergy, map geometry, or player skill. Printed DPS is this unit’s own attack, not the in-game map-total, and a printed ability is graded as written. Job grades blend printed numbers; overall is composed from those grades, not a prediction of a specific run.