This will be a brief article before I post another analysis of the lack of Signora’s return in 5.8 and why I still believe she will return in the full release of Mare Jivari or Snezhnaya.
Recently, I had the idea of digging through Genshin’s old closed beta testing (CBT) builds because I wanted to know whether Signora’s death had been planned since the beginning or if this was a last-minute addition when 2.x was already in development, and use the information I’ve got from those CBT builds to create a hypothesis on Hoyo’s original intent on what to do with her at the time.
I’ve seen some speculations out there that the reason why her death was so rushed is because Hoyo might have added it at the very last minute, and that she might have originally not been involved in the Inazuma arc given how she only showed up towards the end of act 3. Others say that she was really only there to promote Raiden at the expense of people wanting to collect Harbingers.
Given how she’s a character that was introduced since launch, then it would make sense to look at the prerelease builds to see when she first showed up and whether there are any assets that might point to one or the other. Obviously, this involves some educated guesses because we can’t really know for sure what happened behind the scenes, but it should give us an idea on whether this was indeed a last-minute decision or a preplanned death.
The three closed beta tests
Genshin’s betas were divided into three phases: CBT1 (aka CB1.0.0), CBT2 (0.7.1), and CBT3 (0.9.9). These betas were invite-only and randomly given to people who signed up for the waitlist. The naming was inconsistent, going from 1.0.0 down to 0.7.1 and back up to 0.9.9, so it might get confusing at first.
- CBT1 was released on June 21, 2019, but compiled 4 days earlier on June 17. It looked very different from the final release, with an entirely different font, and a less polished UI, lighting, and overall look.
- CBT2 was released on March 18, 2020, but also compiled 4 days earlier on March 14. As this was released much closer towards the game’s launch, predating it by six months, the game is starting to look much more like the final release, but a lot of things like the Paimon menu or Barbara’s voice are still different.
- Interestingly enough, Barbara’s voice in this version is much closer to her combat lines during 1.0 - 1.2, until they changed it in 1.3, which was a source of controversy as some preferred the more energetic voice. This suggests that the developers might have either overlooked the combat lines, or initially kept the old lines but ultimately decided that making them in line with how she speaks in the final release would make her personality more consistent.
- CBT3 was released on July 2, 2020, but was compiled 9 days earlier on June 23, three months before the game’s launch. Most of the basic things, such as the UI and Barbara’s voice, have been finalized, but some EN VOs are missing, and there are minor differences.
There are some private betas and dev builds that were never intende for public viewing, however, and these would be the version I’d datamine.
TCRF already did most of the datamining job
Like many other popular games, the TCRF community has already put the effort to dig up the differences between the prerelease builds and final releases. However, I decided to download a few of the builds myself anyway, to try and narrow down the time frame of when she was first introduced.
The information is quite detailed, and reveals that many of the playable characters had undergone a weapon type change, element change, and/or kit change throughout development. The community also notes that the dev builds usually contain more unused assets in the game files that are absent in the CBTs (presumably to prevent people from discovering unreleased assets), which further convinced me to look into these versions.
Nonetheless, Signora was never mentioned anywhere on the wiki, which suggests that she either had no noteworthy changes (unlike the playable characters), or no one had ever thought of checking her assets. With the absense of her info in there, I had to turn to gameplay videos to try and narrow it down.
Tracking down gameplay and cutscene videos from the CBTs
TCRF mentioned a YouTube channel that contains playthroughs of CBT1 and 2. This was a good place to start because the videos in there are several hours long, which means that they must have played through everything the betas had to offer.
The point of interest would be the end of Prologue Act 3, specifically, the cutscene where the MC returns the lyre, as this is followed by Signora’s gnosis heist cutscene.
The first video I checked was this:
This is the last video of the CBT1 playthrough, which means that they must have been approaching the end of the story. The embedded video skips straight to the cutscene, and not only is it very primitive and unfinished, the gnosis heist scene is also not there yet.
As for CBT2, while the channel does have playthroughs of them, unfortunately it appears that they might have not recorded the lyre return cutscene, as the quests skip straight to “meet Katheryne in Liyue”. Therefore, I decided to search for other sources.
This cutscene is closer to the final release, but it’s still somewhat unfinished. Barbara’s voice is notably raspier and deeper, which doesn’t sound appropriate for her youthful build (her scream is hilarious though), and the final part of the cutscene where Venti leaves the church is still a static cutscene (where the player has to manually go to the next line of dialogue) rather than the dynamic one in the final (where it plays out like an FMV, but uses the game engine in real-time). And just like CBT1, the Signora gnosis heist cutscene isn’t there yet.
Finally, there’s the CBT3 cutscene. This one is very close to final, with it being mostly finished, but has missing VOs for the latter half of the lyre cutscene (which is now dynamic), presumably because this was a new addition and no one had thought of moving the appropriate voice lines from CBT2 into this new cutscene. Signora’s gnosis heist scene is finally added in this version, but with the VOs being completely missing, and the Cicin Mages and Pyro Agents having a different, less detailed model than the final release. Additionally, the subtitle “The Fair Lady” underneath “La Signora” is missing from the title card, and some differences in the dialogue’s subtitles can be seen.
For comparison, here’s the cutscene in the final release:
Datamining the prerelease builds
With the information I’ve gathered, this narrows down the time frame of Signora’s conception to April 14, 2020 (after CBT2 ended) and June 23, 2020 (when CBT3 was compiled). I decided to download some of the earlier versions to pinpoint a more exact time frame, namely:
- 0.9.0 (a CN-only pre-CBT3 / 0.9.9 build provided for an even more limited number of players under NDA)
- 0.7.0 (another NDA’d build, but pre-CBT2 / 0.7.1)
- CB1.3.0_a61ca9b661 (a dev build compiled between CBT1.0.0 and CBT2 / 0.7.1)
- CB1.2.0_29221440b3 merged with assets from CB1.2.0_4e105df375. The former is a dev build that was compiled earlier, but with missing assets that had to be transplanted from a later build in the rel (RC) branch.
- CB1.0.0 (CBT1, the very first publicly available preview)
I can’t really tell you where to download these as Hoyo seems to be strict even when it comes to old prerelease versions and took down the original GitHub repo that listed the download links (I had to use a more up-to-date fork), but a quick DuckDuckGo search should point you in the right direction (Google doesn’t seem to index the repo’s contents for some reason). CB1.1.0 builds do exist, but lack any assets, making them useless. There’s a CB1.1.0 build that contains the assets, but some of them are corrupted, which renders it unplayable. It isn’t in the repo though, and while I’ve reached out to the user who leaked it, they have yet to provide me the download link at the time of writing.
I started by looking into the 0.9.0 assets. This was compiled on May 15, 2020, which predates the public CBT3 (and therefore, Signora’s first public appearance) by over a month.
Surprise surprise, she’s present in there. And it’s a “monster” model. Ouch.
This narrows down the dates to April 14 - May 15, 2020. This sounded odd because this could imply that she was hastily added into the game within the span of a month. Or was she?
To confirm this, I looked into 0.7.0’s assets, which was compiled 5 days before the final 0.7.1 CBT2 on March 9, 2020.
No signs of Signora in here. However, I wasn’t convinced that they developed her character model within the span of just a month. Also, remember how TCRF mentioned that the dev builds contain more unused assets than the CBTs? That was exactly why our next target would be CB1.3.0.
This version was compiled on December 13, 2019, over nine months before the final release, and 20 months before Signora’s death in 2.1.
*cinematic, low boom sound* Well, lookie here. She’s absent in CBT2, but present in CB1.3. There’s fewer files in here too compared to CBT3. Again a monster model. This is looking bleak.
So that pretty much confirms that they omitted some of the unfinished assets in the CBTs and NDA betas.
I dug even deeper, this time with CB1.2.0. This is a merge of two different builds: 29221440b3 (256355_256355) and 4e105df375 (265394_265694). The former was a dev build compiled on November 8, 2019, while the latter was a rel build compiled three days later on November 11, 2019 with fewer debug features. The reason why the two build had to be merged is because the former build lacks the assets required to properly run the game, and merging them effectively makes it a debug version of the latter.
Build 265394_265694 predates the game’s release and version 2.1 by 10 and 21 months respectively. It’s the first version to feature a playable but unfinished version of Childe. If Signora is also present in here, then that would mean both characters were developed at around the same time.
*cinematic boom* Turns out she’s also present in here. Still a monster model. This definitely does not bode well for us Signora mains.
In contrast, Childe has the “avatar”, “monster”, and “NPC” models all at once, which suggests that not only was he planned to be playable since the beginning, but his boss had already been in development at least exactly a year prior to version 1.1, when he made his first appearance as both a weekly boss and a playable character.
At this point, I was getting desperate. Were there ANY signs that she was originally meant to be a playable character? I wasn’t gonna wait to get CB1.1.0 assets, so I just go for broke and datamined the version that started it all, CBT1. After all, this version has a ton of unfinished assets, including characters that wouldn’t be playable until many years after it was first compiled in June 2019, specifically, Shenhe (2.5 years) and Yaoyao (3.5 years).
There’s no signs of her assets in here, which unfortunately means that there’s also no signs of her originally being planned for playability, at least not in the versions that were leaked.
I also looked for Childe’s assets and I couldn’t find them in this version either, so neither of them had been conceived yet at the time.
It’s possible that they might have already made some of their concepts outside the game at the time, but we can’t know for sure. That means the earliest possible time frame is actually between June 27, 2019 - November 11, 2019, which was when CBT1 ended and CB1.2.0_265394_265694 was compiled respectively.
Let’s look at Signora’s models and textures from CB1.2 though, because there might be some changes there, compared to the final version (click on the images to expand).
She looks mostly the same as the final version, but her face is glitched as it uses this weird, mostly transparent texture. Let’s have a closer look.
This is caused by the fact that her face has two parts: face and eyes. I don’t know much about Genshin character models, but presumably she originally would’ve worn a mask as the cutouts seem to fit a mask of sorts before they changed it with the crown-and-right eye mask design.
Signora’s textures only underwent some slight changes, unlike some characters. The textures from CB1.2 are on the left while the final textures are on the right:
The million-dollar question: Was Signora originally never meant to be playable since her conception?
Given the “monster” designation throughout all the builds, it’s pretty clear right? This is a strong indicator that, from a development standpoint, she was never intended to be on a player’s team. Conversely, characters intended to be playable have the “avatar” model type, character icons, and in some cases, kit placeholders. Childe had these early on, and in the same version as Signora’s first appearance in the game’s assets no less.
I think Signora was created solely to serve as a powerful “hook”. She needed to be shocking and establish the Fatui as a credible threat right out of the gate. Xiao Luohao, the game’s editor-in-chief, allegedly stated in an interview at Fudan University that the story’s outline is written 1-2 years in advance. Her model being created in late 2019 is well within the timeline for the Inazuma AQ Act 3’s release in September 2021. The plan was likely always “Create this tragic, intimidating character in 2019 to serve as a major plot death in 2021”.
As the developers planned the game’s first year, they mapped out the major story beats. Inazuma needed a powerful, high-stakes antagonist. Signora, having already been established as a ruthless Harbinger who has managed to steal two of the gnoses, was the logical choice to become the region’s main Fatui boss. The team of animators and artists was assigned to start work on her boss fight in around December 2020 or January 2021 during version 1.2, eight months prior to the fateful 2.1 release, according to this VOD. This is significant because it means the decision to make her a major combat antagonist in Inazuma was made relatively early in the game’s live service history.
There’s also the rose bun foreshadowing. The fact that her hairstyle (and the white roses on her outfit) hinted at her real name, “Rosalyne”, from her very first appearance in CB1.2.0 all the way back in November 2019 means her tragic backstory wasn’t an afterthought. The writers knew who “Rosalyne” was and what her fate would be from the moment they designed her model.
All signs point to a deliberate, pre-planned arc: introduce a mysterious Harbinger, foreshadow her hidden tragic identity, and then kill her off to reveal that identity and create a major, shocking story beat.
So, if it was planned all along, why did it backfire so spectacularly? In my opinion, Signora’s execution flopped for two main reasons:
The first one would be the poor narrative execution. The Inazuma AQ is widely criticized for being rushed. Signora’s character received almost no on-screen development. She shows up, acts arrogant, challenges the MC, and then dies. Because the story didn’t give players a reason to be invested in her journey, her death felt abrupt and unearned, creating shock value but no emotional weight.
The development of the Inazuma AQ, especially the final act, was likely affected by the pandemic and plagued by the same remote work challenges that affected many industries. While the decision to have her be defeated by the Traveler and executed by Raiden was made early on, the writing and execution of that plot point were likely rushed and suffered from a lack of polish.
This would explain why it felt so abrupt. The narrative buildup wasn’t there because the focus was on building the complex boss mechanics, not on fleshing out her character in the AQ. As for why no one seemed to question if killing off a major character this early on would make sense? Well, from a high-level planning perspective, it “made sense” to have a major Harbinger defeated to raise the stakes. The devs likely didn’t anticipate the massive emotional connection players would form with her or the backlash that would result from killing her off before she was properly developed in-game. They saw her as a “villain of the arc”, essentially a plot device, not as a future playable character.
The second one is that they set the wrong precedent with Childe, and this is their fatal mistake. By releasing Childe as a playable character in Version 1.1, Hoyo immediately established a powerful pattern in players’ minds, that “the Harbingers are cool villains, and we get to play as them”. This created an expectation among the players where every humanoid character, including the villains, will become playable as long as they have unique and detailed designs. When they broke that expectation with Signora just a year later, it felt like a betrayal. It didn’t matter that she was only the second Harbinger boss; the precedent had been set, and breaking it felt unfair and inconsistent.
Additionally, Hoyo probably didn’t anticipate how popular their game would be, and they presumably thought they could get away with killing a major character early on because not much people really ever complained about it when HI3 was their latest installment. This could explain why Wendy and Himeko in HI3 had a similar fate as Signora where they died early on, but the reception wasn’t as negative as Signora’s. They simply didn’t have as big of a playerbase who would complain about their decision to kill uniquely-designed humanoid characters like them at the time.
Essentially, they tried to have it both ways. They wanted to sell one Harbinger as a playable character while using another as a disposable plot device, and people immediately called them out on the inconsistency. This created the perfect storm of disappointment: a death that felt narratively empty combined with the betrayal of a pattern players thought they could rely on.
Honestly, if Childe simply ran away or died after doing the boss battle without becoming playable, it likely would’ve softened the blow (even if it would probably still lead to a poorly-executed plot nonetheless), as it wouldn’t have led to people becoming too attached to the Harbingers and expect them all to become playable.
Conclusion: She really deserves better
Honestly, the fact that Signora was never intended to be playable, as evidenced by her not having an avatar at any given point in time, really paints a grim picture that they very likely underestimated the potential for her popularity and profitability. This whole incident serves as a reminder that, in gacha games, companies should commit to releasing all uniquely-designed humanoid characters as playable to satisfy everyone without making anyone feel like they’re being left behind.
Alternatively, they should take the Kuro Games approach and make them playable, regardless of whether they died or not, like Phorolova in WuWa who recently died but became playable anyway. At this point, if they wanted to make Signora playable, they would have to put the effort to bring her back in a way that makes sense, which, once again, likely restricts their options to either Mare Jivari or Snezhnaya because of all this mess of vaporizing her in Inazuma.