Between April and June 2026 we ran a limited beta for Ruinveil with a group of players recruited through roguelite communities and our mailing list. We ended up with 214 participants who submitted at least one structured feedback form, with a subset of about 40 who played extensively across multiple difficulty modes. This is the most data we have had about how the game performs with people who did not build it.
This post is our attempt to summarize what we heard honestly. Not all of the feedback was things we fixed. Some of it is things we are still figuring out. We think it is more useful to write this post than to either oversell what went well or stay silent about what did not.
What the Beta Covered
We gave beta players access to the full dungeon depth (all five tiers), Standard and Veteran difficulty modes, and approximately 60 percent of the planned ability selection. Cursed mode was not included because the difficulty calibration changes we described in April were still being finalized and we did not want feedback on a mode we knew was miscalibrated.
Feedback was collected through a structured form submitted after each session (players were asked to fill one form per two-hour block, not per run), plus an open-ended Discord channel where participants could report bugs and discuss the game. We also had six one-hour recorded sessions with players who gave consent to screen share and think aloud while playing.
The Main Feedback Clusters
Ability interaction legibility. This was the most consistent issue across the beta, appearing in feedback from players across all experience levels. Players frequently described selecting abilities whose interactions with other abilities or with specific enemy types they did not understand until several runs in. In some cases the lack of understanding was frustrating; in others, players described it as a discovery process they enjoyed. The divide seemed to fall along experience with the genre: players with extensive roguelite background tended to find the ambiguity rewarding; players newer to roguelites found it confusing.
We made two changes in response to this feedback. First, we expanded the in-run ability description to include a brief interaction note for each ability that has known interactions with other commonly selected abilities. The descriptions are factual rather than evaluative: "Augments Surge Bolt when active" rather than "makes Surge Bolt better." Second, we added a post-run ability summary screen that shows which ability interactions activated during the run. This gives players a way to understand what happened and why without front-loading information they did not ask for.
Dungeon tier five pacing. A significant portion of the more experienced players, particularly those who completed multiple runs on Veteran, reported that dungeon tier five felt rushed compared to tiers two through four. The structural feedback was consistent: tier five has a shorter critical path, fewer branch rooms, and encounters that feel less compositionally complex than tier four. Several players described it as an anticlimactic stretch before the final boss approach.
This is a real design problem. Tier five was the last tier we fully developed and the grammar rules governing it received less iteration time than earlier tiers. We have been aware that it was underdeveloped relative to the rest of the dungeon, and the beta feedback confirmed that players could feel the gap. We are actively revising the tier five grammar rules, adding more optional branch room types and expanding the encounter ecology pool at that depth. We do not have a timeline for when this will be production-ready because we want to iterate through at least two full rounds of internal testing before releasing it.
Performance on lower-end hardware. We received consistent performance complaints from a subset of players running older or lower-spec machines. Specific reports clustered around the enemy behavior assembly pass during dungeon generation and frame rate during rooms with large numbers of simultaneous enemy actions. The behavior assembly performance issue is not new; Omar wrote about it in the April 2026 post. What the beta revealed is that the worst cases are worse than our internal testing had caught, because our development machines are uniformly high-spec.
We spent about four weeks on performance optimization after the beta closed. The most impactful change was reducing the worst-case enemy behavior assembly time by approximately 40 percent through a refactoring of how trait subtrees are evaluated during the assembly pass. We also introduced a quality-of-life toggle that reduces simultaneous enemy action complexity at the cost of some behavioral richness, targeting players on hardware below a specified threshold. This toggle is opt-in for now; we may make it automatic based on detected hardware performance in a later build.
What the Beta Did Not Surface as Major Problems
Two things we expected to be more prominent feedback were barely mentioned.
The directional navigation indicator we added after the September 2025 playtest results received almost no negative feedback. Players on the beta were able to navigate the dungeon without significant disorientation. We consider that problem resolved for Standard mode, though we are intentionally removing the indicator in Cursed mode as part of that mode's design.
The seed system received consistent positive responses. Players who engaged with it described sharing seeds in Discord and comparing runs with other players on the same seed. The use patterns we saw were the ones we designed for: players sharing seeds of particularly challenging or interesting dungeons and using the speedrunning-adjacent value of known dungeon layouts. One player systematically posted annotated maps of interesting seeds in the Discord, which was unexpected and delightful.
What We Are Still Working On
The ability interaction legibility problem is partially addressed but not resolved. The changes we made help players after they have invested time with the game; they do not fully address the difficulty of the first few runs for players new to the genre. We know there is a category of player who will not persist through the learning curve, and we have not found the right intervention that reduces friction for that player without compromising the discovery element that experienced players value. This is an active design question with no settled answer.
The tier five dungeon pacing problem requires grammar revisions that we are still testing. The current beta version reflects the dungeon as it was at beta close; the revisions will appear in a future build. We will write about the changes in a devlog post when the work is complete enough to be worth describing.
There is also an enemy role-synergy modeling gap that Priya wrote about in April that the beta data reinforced. Anchor-harasser compositions at mid-dungeon depths continue to produce death rates above what the threat model predicts. We have not published a fix for this yet because the extended threat model we are building needs more calibration time. We are not comfortable releasing a model that we have not validated against a sufficient sample of generated encounters.
A Note on What the Beta Meant for the Studio
The September 2025 playtest gave us honest external feedback for the first time, and it was valuable in ways we described in that devlog post. The beta was a different experience: more players, longer engagement periods, and feedback that reflected people who had formed actual opinions about Ruinveil through repeated play rather than first-impression reactions.
It was also harder in some ways. Reading feedback from someone who put fifteen hours into your game and has detailed opinions about what is not working is more specific than early-stage feedback. We are grateful for it and we think the game is meaningfully better for having run the beta. We are carrying several open problems out of it that will keep us busy into the second half of 2026. That is not a bad position to be in.