top of page

On Misses, Accuracy, and Dodge

A recent message posted in the Suggestions forum of the TRV Discord, generated quite a bit of debate on the nature of misses, accuracy, and dodge. So, rather than enjoying the sunny Saturday, I figured I would return to my cave to write a quick blog post outlining some random thoughts on the topic.


Damage Mitigation vs. True Misses

First, a bit of background. When we query the >10 million exchanges in which champs have dealt blows in The Red Village, we observe numerous examples in which the outcome of a champ’s attack is 0 damage. This outcome can occur as the result of two distinct phenomena:

  1. When an attack is fully mitigated by a defender.

  2. When an attacker “misses” an opponent (or the defender “dodges” the attack; more on this later).

How do we know that these are two distinct processes? First, there is a field in the TRV API titled “missed_hit”. Although 100% of these entries return the value “FALSE”, there was a period of several days in April 2022 in which the API actually was properly demarcating misses as “TRUE”. Not all 0s were designated a “Miss”, attesting to the assertion that Phenomena #1 and #2 are, indeed, distinct processes.

Note that this update has since been rolled back (the “Miss” field is now an artifact), because the TRV Team made the decision not to denote Misses in the Public API. However, we do also see evidence of these distinct phenomena in the hit-by-hit damage data.

Have a look at the distributions of damage dealt by different types of attacks facing BASE DEFENSES, represented in the chart below (posted on my Twitter back in April 2022 and accessible on TRVtools here.

Green designates base attacks that were met by base defenses. We see that the median of the distribution falls around 124 damage, while the lower tail of the distribution drops off at 13 damage. Only 25 out of 10 million Base v Base interactions yielded less than 20 damage, and 60% of those occurred on nerfed first strikes.

To this day, there are zero instances in the Base v Base data in which 1-12 damage has been dealt. However, 0 damage occurs relatively frequently (~15% of Base v Base interactions). This bimodal distribution indicates that 100% of 0 damage observations in the Base v Base data are the result of True Misses. In other words, there is strong evidence for Phenomenon #2 in the damage data.

[As an aside, it’s also interesting to note that Special Attacks (in red) never produce a miss/zero damage.]

So, what about Phenomenon #1? Do we ever see evidence of attacks that are fully mitigated? Yes, Phenomenon #2 is observed when Base attacks are met with Special Defenses. This phenomenon is evidenced by the Special Defense data. Have a look at the chart below, depicting damage dealt by BASE ATTACKS facing different types of defenses (posted on my Twitter back in April 2022 and accessible on TRVtools here.

[Note that The Median of the Base v Base (green) distribution above should read 124 damage; it is the same distribution as on the previous plot.]

The Base Attack v Special Defense distribution (blue) abuts 0. Accordingly, it is very likely that the 0 damage outcomes that we observe in Base v Special interactions are the result of both Phenomenon #1 (full damage mitigation) and Phenomenon #2 (true misses).

Also note that it is impossible to distinguish which Phenomenon is the cause of a 0 damage outcome in any given Base v Special interaction. Thus, to maximize signal relative to noise, TRVtools Accuracy analytics focus exclusively on 0 damage outcomes that occur during Base v Base attacks.


Accuracy vs. Dodge

Okay, so if we accept the premise that all Base v Base attacks are the result of True Misses, we can ask the question, “How are these misses generated by the battle algorithm?” Without knowledge of the inner workings of the algo, we can only use inference, but I think the data do provide some hints.

In principle, “True Misses” could be attributable to the action of the attacker (accuracy), the action of the defender (dodge), or a combination thereof. Thus, we can make some predictions as to what we would expect to observe in either case.

For example, if [Accuracy/Dodge] is part of the battle algorithm, one might predict [Accuracy/Dodge] to be:

  1. Correlated with champ outcomes (e.g., expect High Accuracy (few misses when attacking) and/or High Dodge (many misses when defending) to be positively correlated with Elo)

  2. Influenced by stance (e.g., ‘Accurate Stance’ may elevate accuracy, whereas ‘Defensive Stance’ may elevate Dodge)

  3. Heritable (i.e., Accuracy and/or Dodge should be passed along from parent to offspring with some probability)

There is insufficient Summoning data to investigate item #3 above, and SEER models have found mixed evidence for the relative importance of Accuracy and Dodge, suggesting that item #1 requires further investigation. With regard to item #2, Champion Scouting Reports (available on TRVtools champ pages) demonstrate that most champs *generally* exhibit an increase in Accuracy in the Accurate Stance, an increase in Strength in the Aggressive stance, and an increase in Defense in the Defensive Stance. By contrast, variation in Dodge amongst stances seems to be the result of statistical noise, rather than a consistent stance effect.

So, what does all of this mean? In essence, there are still a lot of question marks as to role that Accuracy and Dodge play in the TRV battle algo. This topic is still very ripe for debate, and I welcome any thoughts or further discussion that this post may generate.


Recent Posts

See All
bottom of page