The 90-Pull Machine: HoYoverse and the Lesson Esports Can Never Copy

That night, in the chat of a livestream in Guangzhou, more than two hundred thousand people held their breath for about three seconds. The screen turned gold. A five-star character appeared, but not the one the entire room was waiting for. A collective sigh, then laughter, then the familiar line: Missed the coin flip, back to grinding.

That moment, which the community calls losing the 50/50, lasted under four seconds, yet it carried more emotional swing than extra time in a final. I have sat through hundreds of such moments over the years, and what has always caught my attention was never the character itself. It was the structure behind the moment: a revenue machine engineered down to the smallest detail, running on a cycle, and dependent on no third party whatsoever.

When this analysis document was handed to me under the esports label, my first reaction was to stop. That label is a misclassification that must be corrected before it corrupts anything downstream. But the very reason this subject does not belong to esports is the reason the esports industry should study it.

Let us be clear from the outset: Genshin Impact, a HoYoverse product, has no professional tournament circuit. No Worlds, no TI, no Major, no VCT. No franchised league, no club ecosystem, no player transfer market in the esports sense. What it has is a version release cycle: each version splits into two phases, each phase runs roughly 21 days, and each phase carries one or more banners, that is, limited-character wish pools.

Right now the player community is crossing the transition between version 7.0 and 7.1. Phase two of 7.0 is a rerun cycle, bringing back older characters such as Flins and Ineffa. Phase one of 7.1 is said to debut two new characters at once: Vesna and Vodyanitsa. Phase two of 7.1 continues with reruns, carrying names such as Skirk and Escoffier. Alongside them, characters including Odette, Aino, Iansan and Lan Yan surface sporadically in community discussion.

The language the community uses to talk about all this is telling. They do not say this character is strong or weak. They say save for 7.1. They say should I dump into this banner. They say have I accumulated enough. This is the language of an investment decision, not the language of a competitive decision. That is the first and only trustworthy point of contact between this material and esports industry analysis.

I must be blunt about source reliability. Of the 28 information points I hold, twenty carry no source at all. Only one cites an official announcement from HoYoverse's own channel. Three are the original author's subjective opinion. Several named entities, including Odette, Flins, Ineffa, Vesna, Vodyanitsa, along with version numbers 7.0 and 7.1, cannot be cross-verified against known game state. The document itself concedes that the exact banner schedule is still to be confirmed.

This matters, because everything that follows must be read as an analysis of structure, not of schedule. Structure can be verified. Schedule cannot.

That structure has four layers, and every layer is deliberate.

The first layer is the soft pity floor at 90 pulls. Players are guaranteed a five-star character within a maximum of 90 pulls on the same banner. This mechanism reconciles two apparently opposite feelings: randomness and fairness. Players still believe they are lucky or unlucky, but the system guarantees nobody falls into infinity. It is a familiar psychological equation: give players a floor so they do not leave, but never give them a ceiling so they do not stop.

The second layer is the 50/50 mechanic. On an event banner, the first five-star has a 50 percent chance of being the featured character and a 50 percent chance of being a standard character. If it lands on the standard branch, the so-called loss, then the next five-star is guaranteed to be the featured one. This is the heart of the entire system. It turns a spending decision into a gamble with memory: if you lose, you are credited a reward; if you win, you walk away feeling fortunate. Both outcomes bring the player back next time.

The third layer is shared pity across banners of the same category. Accumulated progress does not vanish when a player switches to another banner in the same group. Technically, this lowers the marginal cost of switching between banners. Behaviourally, it raises spending frequency: players no longer have a reason to freeze their accumulation waiting for a worthier banner.

The fourth layer is the absence of a fixed rerun schedule. There is no long-term published calendar. Some characters are absent for over a year. Others return within a few versions. This is controlled scarcity, the equivalent of limited-time event design in game-development language. And finally there is a secondary monetization lane called Chronicled Wish, dedicated to older characters, allowing the publisher to re-monetize dormant names without disrupting the cadence of primary banners.

Place these four layers side by side and you see a machine running on a 21-day cycle, repeating, predictable in rhythm but unpredictable in content. That is a revenue rhythm, not a competitive rhythm.

The algorithm does not tire, but the fan's heart does. And precisely because it tires, the fan is the only variable this machine must keep feeding.

This is where the comparison with esports becomes interesting.

The esports ecosystem runs on a multi-party value chain: publisher, league and clubs, audience and sponsors. Revenue comes from broadcast rights, sponsorship, in-game item revenue sharing, and prize pools. Every dollar must pass through at least one intermediary. Each intermediary carries cost, risk, and, most importantly, an external calendar the publisher does not fully control. A postponed tournament, a withdrawing team, a region losing broadcast rights, all create revenue holes.

The gacha model is different in kind. It is a closed two-party loop: publisher and player. No clubs, no tournaments, no broadcasters, no sponsors. The publisher is simultaneously the operator, the rule-maker, and the sole information authority. The concentration of power in this model is higher than in any esports ecosystem I have ever analysed.

First consequence: far greater resilience to calendar shocks. When world sport froze in 2026, leagues lost ticket and broadcast revenue; a gacha machine only needed players to keep opening the app. Second: higher margins, because there is no intermediary to share with. Third: near immunity to form risk, because characters do not get injured, do not decline, do not demand raises, do not lose the dressing room.

But everything has a price. The one genuinely material weakness of the gacha model is its exposure to regulation. Because the publisher controls supply, rules and information channels alike, when regulators change the framework, whether on probability disclosure, minor protection, or loot-box debate, the whole chain has no buffer to absorb it. In esports, a new rule can hit a tournament, a team, or a region. Here, it hits the engine directly.

One more point deserves attention: gacha depends on no external cultural or sporting event to sustain attention. It manufactures its own content on a cycle. Esports, by contrast, must constantly cling to fixtures, to team narratives, to player form to hold its audience. That is a structural advantage, but also a structural isolation: the machine has no ally but itself.

The 21-day rhythm also deserves to be read as schedule design, not merely an operating figure. A football league creates 38 matchdays across nine months, and its value comes from the uncertainty of results. A gacha version creates two spending windows every six weeks, and its value comes from predictable supply combined with unpredictable outcome. In sport, uncertainty sits in the result. In gacha, uncertainty is industrially manufactured at exactly one point: the moment the screen turns gold. That is the entire difference between a sporting product and a cyclical consumer product.

Now comes the part where I must challenge myself.

First, source reliability is far too low to support any firm conclusion about schedule. Twenty of twenty-eight points carry no source. It is possible part of this content is aggregated or rumour-based rather than verified reporting. The document's own admission that the banner schedule is unconfirmed is a positive signal of integrity, but it simultaneously confirms that what I am analysing is structure, not event. If the official schedule differs, the structural section stands and the specific forecasts fall.

Second, the gacha-versus-esports comparison risks becoming a false equivalence. These are two different products, serving two different audiences, under two different legal frameworks. Gacha players do not watch their product the way an audience watches a match; they own it. Sports fans buy a collective experience; gacha players buy a digital asset. I deliberately do not call characters players, do not call banner phases tournaments, do not call reruns transfer windows, because doing so would fabricate equivalence to fill a template.

Third, and this I consider most important: people systematically underweight regulatory risk and overweight public-opinion risk. A community furious over an unfair rerun schedule will speak up for a few weeks and cool down. But a new legal framework on probability disclosure or spending limits permanently alters revenue structure. Under most current frameworks gacha is not classified as gambling, but it sits right against the line. That proximity, not the online storm, is what deserves watching.

Fourth, I admit a professional temptation. As a content maker, I know a headline with a shock factor travels further than a dry table of numbers. But if I turn an unconfirmed schedule into a confident prophecy, I am no longer the person supplying evidence; I am only the person supplying noise. Between those two lies the entire difference between an analyst and a loudspeaker.

And there is one more risk I must name: the possibility that the source content was machine-generated or rumour-based. The appearance of unverifiable names and version numbers is a warning sign. Had I copied them without noting it, I would have staked my reputation on a source I cannot verify.

There is one more analytical layer worth discussing: the media narrative around banner cycles.

The current narrative is clear: a new version means a new hype cycle; save now, pull later. This is a schedule-driven narrative, not a performance-driven one. It tells you when, not why. And the gap between market expectation and objective reality sits exactly there: a confirmed schedule is not a confirmed value.

The source document provides no data whatsoever on character strength, kit design, or lineup value. It provides only a calendar. That means this is a schedule explainer, not a decision-support piece. Readers get an answer to when, not to should I.

The ratio of media heat to underlying fundamentals here is high: thick hype, thin verifiable data. This is the content model I call the traffic filter, made of promotional tone plus thin sourcing. It is not ethically wrong, but it is cognitively dangerous, because readers easily mistake the liveliness of discussion for the certainty of information.

A stadium can be empty of spectators, but history never lacks a chronicler. And what is chronicled here, regrettably, is mostly a calendar, not evidence.

So what is worth keeping?

What is worth keeping is not the character list, and not the release date. It is a monetization structure engineered to turn repeated spending decisions into part of a habit, rather than a single event. That is the core difference between a gacha machine and a sports competition: a competition lives on emotional peaks spaced months apart, while the machine lives on small oscillations repeating every three weeks.

What can the esports industry learn from this? Less than people think. It cannot copy a model that requires monopoly control of both supply and information channels. But it can learn how to read it: look at rhythm rather than event, at architecture rather than name, and at regulatory risk rather than public-opinion storms.

Data needs no loudspeaker, but it shakes an empire.

I do not oppose tradition; I am only handing tradition a new piece of evidence. In this case, the new evidence is a structure most observers overlook because they are busy looking at characters.

And if you are wondering whether to pour your entire accumulation into phase one of version 7.1, my answer is simple: do not ask a structural analyst about an emotional decision. Ask yourself whether you are buying a character, or buying the feeling of not missing out.