Methodology for Unit Evaluation - Intrinsic Rating Criteria

One of the approaches commonly used to evaluate units in Fire Emblem: Heroes (and by extension in most unit-selection-based games) is tier lists. At its core, a tier list is a description of the metagame - an analysis of what is or isn't popular among people playing the game at a high level.

A tier list may or may not reflect actual mechanical strength - sometimes, weaker units will sit near the top of most tier lists for a while because their weaknesses haven't been completely identified yet. Likewise, stronger units will occasionally be overlooked until someone actually does research on them and discovers their strength.

What this means is a tier list is less interested in hierarchising units based on their natural potential, than they are in hierarchising units based on their usefulness in the current metagame. It's a pragmatic approach that tries to tell you how to best use your box based on the current state of the world.

Intrinsic unit evaluation (and therefore intrinsic rating) takes the opposite approach and looks at each unit from a strictly mechanical perspective, discarding metagame and (current) unit popularity in the process. This is about trying to see exactly how mechanically good each unit's individual assets are, regardless of whether they'd be useful in the current metagame.

Tier lists are useful to figure out what to use in Arena. Intrinsic ratings aim to help with figuring out what is worth building for long-term box improvement.

Establishing the scales

Our first impulse was to establish a single rating procedure to evaluate each unit against. Unfortunately, this led to a lot of inconsistencies - which in retrospect is normal. Any single rating procedure will naturally favor some kinds of units while disfavoring others (for instance: nukes getting better ratings than tanks), which makes it necessary to have at least two distinct procedures for evaluation to account for unit diversity.

We have arbitrarily and unilaterally decided to work on exactly two rating procedures - the attacker scale, and the supporter scale, detailed below. For the short version: an unit's attacker rating represents its ability to kill things, whereas an unit's supporter scale represents its ability to help attackers do their job and keep the team alive.

Furthermore, every unit gets a budget rating on both of these scales - budget ratings correspond to the unit's potential in limited-resource, no 5star inheritance scenarios.

We went for two rating procedures (instead of, say, three or four) because although a single procedure definitely didn't make sense, having specialized scales for every kind of unit would simply defeat the purpose of ratings altogether. Working with two scales is sufficient to give an accurate evaluation of most units in the game, which is what we're after here.

About values

Relax, 7/10 is a great grade. -- Hector, to virtually anyone not named Hector.

Each unit gets four ratings - budget attacker, budget supporter, attacker and supporter. For each rating, it's possible to obtain scores between 1 and 10, in 0.5 increments. (Budget ratings, by definition, will always be inferior or equal to their corresponding non-budget rating.)

Not many units actually have a shot at scoring 10 in any of these ratings. This is on purpose - the rating procedure is meant to be tough, and most heroes will probably score between 4 and 7. A rating of 3 or below is abysmal, and a rating of 8 or above is excellent.

A large part of the rating procedure is entirely mechanical. For instance, in the attacker rating procedure, a unit with at least 35 in atk automatically gets two points no matter what.

Although it's minimized, there is still a part of subjectivity in the rating procedure. Whether a unit's apotheosis ("best potential build") is excellent or just great is at least partly a matter of opinion, and although guidelines are provided they aren't (and can't be) absolute.

In order to account for this, 0.5 point increments are used to represent nuance. The rating procedure only grants full points, so when you see a 8.5 rating, this means the unit either scored 8 with a positive nuance, or 9 with a negative nuance.

Attacker rating procedure (Attacker scale)

Go through this list, adding one point whenever the condition for it is fulfilled. Maximum grade is 10. Budget attacker rating is obtained by going through this list again while discarding any inheritance cost that is 5star-only.

  • Unit nature (1 point total):
    • Is the unit's class primarily offensive? Sword users, lance users, axe users, bow users, dragon stone users and tome users are considered primarily offensive. (That's every colored class plus archers.)
      • If the unit's class is primarily offensive, 1 point, else 0 points.
      • (This is to account for weapon availability.)
  • Spread Quality (3 points total):
    • Is the unit's attack at least great (i.e. at least 30)?
      • If it is, 1 point, else 0 points.
      • If it's excellent (at least 35), 2 points.
    • Among the units spd, def and res, is at least one of these great (i.e. at least 30)?
      • If it is, 1 point, else 0 points.
  • Legend Strength (2 points total):
    • Can the unit wield a unique weapon?
      • If it can, 1 point, else 0 points.
      • If it's an exceptionally strong offensive weapon (i.e. built-in Distant Counter, Vantage, Desperation, etc.), 2 points.
  • Unit Apotheosis (4 points total):
    • Does the unit have at least one synergetic attacker build (i.e. a build that works really well with its stat spread)?
      • If it does, 1 point, else 0 points.
    • Is the unit's best build good, great or excellent?
      • The build is excellent if it doesn't have more than a couple strong counters, and is able to tinker around to defeat these with appropriate support.
      • The build is great if it needs help to deal with its counters, but doesn't have too many of them.
      • The build is good if it has many counters it can't handle at all.
      • 2 points if excellent, 1 point if great, 0 points if good.
      • 1 more point if the unit has at least two excellent synergetic builds (this is about rewarding high-level versatility).
  • Adjustment point (1 point total):
    • If the unit's total score so far is 3 or less and you feel this score is undeserved (i.e. the unit brings something to the table that isn't being properly represented by this rating procedure), you may add 1 adjustment point to its score.
      • This is to account for stat spread oddballs such as RobinM or RobinF.
Supporter rating procedure (Supporter scale)

Go through this list, adding one point whenever the condition for it is fulfilled. Maximum grade is 10. Budget supporter rating is obtained by going through this list again while discarding any inheritance cost that is 5star-only.

  • Spread Quality (3 points total):
    • If the unit is a healer:
      • Among the unit's spd, def and res, how many of these are great (i.e. at least 30)?
        • 1 point if at least one, else 0 points.
        • 2 points if at least two.
        • 1 more point if at least one of these is excellent (i.e. at least 35).
    • If the unit isn't a healer:
      • Among the unit's atk, spd, def and res, how many of these are great (i.e. at least 30)?
        • 1 point if at least one, else 0 points.
        • 2 points if at least two.
        • 1 more point if at least one of these is excellent (i.e. at least 35).
  • Legend Strength (2 points total):
    • Can the unit wield a unique weapon?
      • If it can, 1 point, else 0 points.
      • If it's an exceptionally strong support weapon (i.e. built-in Hone Atk, Threaten Atk, etc.), 2 points.
    • (Adjustment point) Is the unit a healer or a dagger user?
      • If it is, 1 point, else, 0 points.
      • This adjustment point is there to account for the wide array of support effects available to these classes.
      • This adjustment point cannot push the Legend Strength rating of an unit above 2 points total (case in point: Jaffar).
    • (Adjustment point) Is the unit a dancer/singer?
      • If it is, 1 point, else 0 points.
      • This adjustment point is there to account for the huge support usefulness of dance/sing.
      • This adjustment point cannot push the Legend Strength rating of an unit above 2 points total.
  • Unit Apotheosis (5 points total):
    • Does the unit have at least one synergetic support build (i.e. a build that works really well with its stat spread)?
      • If it does, 1 point, else 0 points.
    • Is the unit's best build good, great or excellent?
      • The build is excellent if it improves any team it's used with and isn't a significant weak defensive spot.
      • The build is great if it improves most teams but also needs to be shielded from a couple specific threats.
      • The build is good if it improves some teams but represents a very weak defensive spot.
      • 3 points if excellent, 2 points if great, 1 point if good.
      • 1 more point if the unit has at least two excellent synergetic builds (this is about rewarding high-level versatility).
  • Adjustment point (1 point total):
    • If the unit's total score so far is 3 or less and you feel this score is undeserved (i.e. the unit brings something to the table that isn't being properly represented by this rating procedure), you may add 1 adjustment point to its score.
A few comments

From a general point of view, these rating procedures are designed with simplicity and straightforwardness in mind. The assumption here is that most of the units that are actually good will naturally obtain good scores because their strength also stems from their mechanical properties, and that there isn't much to gain in going for a complex procedure.

The scoring thresholds we're using for stat spread evaluation (30 and 35) may raise some eyebrows. Although they're obviously arbitrary, they seem to produce relatively accurate data and modifying them doesn't seem to improve the ratings significantly. We've decided to accept that RobinM and RobinF simply don't get any point for being average (even though three of their four stats are exactly one point below the scoring threshold, at 29), and adjust their score with a dedicated adjustment point instead of looking for values friendlier to them.

This rating procedure does not specifically take movement type and class-buff-reliance into account. Naturally, armors get better stat spread scores, but their lowered mobility has a good chance of lowering their apotheosis potential which ultimately balances their scoring ability out when compared with other movement types.

A given unit's rating is relatively independent from its position in the various tier lists. This is by design - tier lists and intrinsic ratings aren't looking at the same things at all.

The rating procedure is the same for everyone (except for healers in supporter scale, as they get a special atk penalty and can't obtain rating points from atk as a result). This means color variance isn't taken into consideration, and this is also by design.

Taking color variance into consideration would require fine tuning the criteria for each individual class. Fine tuning per-class would introduce a much greater part of subjectivity in the rating process, which isn't desirable - we want something as simple and straightforward as possible.

A few examples
  • Ephraim:
    • Attacker rating: 1 (Unit Nature) + 3 (Spread Quality) + 1 (Legend Strength) + 3\~4 (Unit Apotheosis) = 8\~9
    • Supporter rating: 3 (Spread Quality) + 2 (Legend Strength) + 4\~5 (Unit Apotheosis) = 9\~10
  • Jaffar:
    • Attacker rating: 0 (Unit Nature) + 1 (Spread Quality) + 2 (Legend Strength) + 2\~3 (Unit Apotheosis) = 5\~6
    • Supporter rating: 1 (Spread Quality) + 2 (Legend Strength) + 2\~3 (Unit Apotheosis) = 5\~6
  • Gwendolyn:
    • Attacker rating: 1 (Unit Nature) + 2 (Spread Quality) + 0 (Legend Strength) + 3\~4 (Unit Apotheosis) = 6\~7
    • Supporter rating: 2 (Spread Quality) + 0 (Legend Strength) + 4\~5 (Unit Apotheosis) = 6\~7
  • Azama:
    • Attacker rating: 0 (Unit Nature) + 1 (Spread Quality) + 0 (Legend Strength) + 1\~2 (Unit Apotheosis) = 2\~3
    • Supporter rating: 1 (Spread Quality) + 1 (Legend Strength) + 4\~5 (Unit Apotheosis) = 6\~7
  • Ike:
    • Attacker rating: 1 (Unit Nature) + 3 (Spread Quality) + 2 (Legend Strength) + 3\~4 (Unit Apotheosis) = 9\~10
    • Supporter rating: 3 (Spread Quality) + 1 (Legend Strength) + 1\~2 (Unit Apotheosis) = 5\~6
  • Olivia:
    • Attacker rating: 1 (Unit Nature) + 1 (Spread Quality) + 0 (Legend Strength) + 2~3 (Unit Apotheosis) = 4\~5
    • Supporter rating: 1 (Spread Quality) + 1 (Legend Strength) + 4\~5 (Unit Apotheosis) = 6~\7
  • Odin:
    • Attacker rating: 1 (Unit Nature) + 1 (Spread Quality) + 0 (Legend Strength) + 0\~1 (Unit Apotheosis) = 2\~3
    • Supporter rating: 1 (Spread Quality) + 0 (Legend Strength) + 2\~3 (Unit Apotheosis) = 3~\4
Thank you for helping us help you help us all!

Although a lot of thought has gone into these rating procedures, they may still be perfectible. If you notice something that seems off (especially units scoring higher or lower than seems rational), please point it out and we'll take a look.

-- Article and procedures written by takaminacchan.