Missing Damage, Cancelled Dodges, and Desync

If you have watched blood spray off a target that took no damage, or dodged cleanly and been yanked back into a stagger, you are not imagining it and it is not your build. Bellring published its own explanation in DevNote #9 on September 24, 2026, and this entry is that document turned into a reference.

The One Setting That Matters

Client-Side Damage Prediction is enabled by default. With it on, your client plays impact effects and hit audio based on its own guess, before the server confirms anything. Turn it off and the phantom hits stop, because nothing is drawn until the server agrees. The cost is that every hit then feels a beat late, which in an action game is its own kind of unpleasant.

There is no third option. Hits are resolved on the server, deliberately: Bellring says trusting the client would feel better for the attacker but would wreck the target's experience under high latency and would widen the surface for cheats. So the choice is prompt-but-sometimes-wrong, or correct-but-delayed.

Why Damage Goes Missing

Two scenarios, both of them a prediction error rather than lost damage:

The reason this happens as often as it does is that the hit windows are short. The entire active window on the first strike of the Sword and Shield is 62 milliseconds. A window that tight leaves very little room for client and server to disagree, which is why fast attacks against a moving or dodging target are the worst case.

Bellring has since narrowed the gap: prediction used to run off the real hit registration hitbox, and now uses a separate, slightly smaller one, so the client has to be more confident before it draws a hit.

Why Dodges Get Cancelled

Your dodge animation starts the instant you press the key, locally. The command then has to travel upstream before the server executes it. Two ways that goes wrong:

• Your client thought you were un-hit when you pressed dodge, but by the time the command landed, the server had you already struck. Dodging was never possible.
• You had already been hit on the server, and the downstream update had not reached you yet, so your client let you dodge something that was already over.

Either way the server's version wins, you get pulled back to where you started, and you eat the stagger. This is not dodge-specific: per Bellring, any skill can play locally and then be rubberbanded the same way.

Why Brandrgarde Feels Worse Than Hallowgrove

Because it measurably is. The server has to run full animation and Inverse Kinematics for every character and monster so that hitboxes stay accurate, and it tracks them per body part: head, torso, upper arms, lower arms, thighs, and shins. That is a heavy load, and it scales with how much is on the map.

Bellring's own monitoring metric is the share of server frames whose latency exceeds 50ms, and it reports frame drops as more likely in Brandrgarde than Hallowgrove, because Brandrgarde runs a higher volume and density of monsters. Community suspicion confirmed by the studio, in other words.

What Has Already Been Fixed

Shipped in the September 15 live update, all aimed at this exact class of problem:

• More network fault tolerance on the follow-up attacks that come after a dodge, so high latency is less likely to eat them.
• A fix for Breakaway and the skills usable in its window being interrupted on the client while still running on the server, which was a direct cause of position rubberbanding.
• Better dodge-direction synchronisation, reducing cases where an opponent's dodge appears to go one way before they snap somewhere else.

On the infrastructure side, Bellring moved its AWS storage from EFS to EBS, which it credits with resolving occasional multi-second stutters, stopped its cloud servers running hyper-threaded workloads, batched and asynchronously wrote its logs, constrained monster AI when no player is nearby, spread heavy work such as mass monster spawns across multiple frames, and enabled high-defence protection against UDP flood attacks.

What Is Still Coming

Named in DevNote #9 without dates, so treat these as direction rather than schedule: migrating blueprint script logic to C++, stress-testing higher-tier AWS CPU models, an Unreal Engine garbage collection pass for map streaming micro-stutters (first wave live the week of September 24), and spreading player and monster instantiation cost across frames to soften the hitch when another team appears.

The Honest Summary

If your damage disappears, check whether damage prediction is on before you assume a bug, and expect it most against dodging targets with fast weapons. If your dodges get cancelled, that is upstream latency and no setting fixes it. And if it all feels worse in Brandrgarde, the studio agrees with you and has said why.

Source: DevNote #9, Bellring Games, September 24, 2026, published as a Steam community announcement for app 3282300. The 62ms and 50ms figures, the per-body-part hitbox list, and the Brandrgarde comparison are all the studio's own.