The Probability Machine and the 90-Pull Ceiling: Dissecting the Revenue Architecture Behind the 7.1 Banner Cycle

The Probability Machine and the 90-Pull Ceiling: Dissecting the Revenue Architecture Behind the 7.1 Banner Cycle By Do Tri For many consecutive weeks, while reconstructing the timeline of an online content release cycle for analytical work, I kept stopping at the same data point: the 90-pull pity ceiling. It is not a score, not a win rate, not a performance metric of any athlete. It is a promise encoded as a number. After at most 90 attempts, the paying user is guaranteed to receive the five-star character they are chasing. As phase two of version 7.0 closes to make way for 7.1, the centre of discussion no longer revolves around the strength of the new character. It revolves around another, less-discussed story: which architecture operates behind that probabilistic door, and how it shapes player behaviour. My profession is reading decisive moments back in slow motion. My daily work is breaking a situation down into milliseconds, cross-checking it against the original rulebook, and pointing out where the crack lies. When I turned that eye toward a gacha system, where players spend real money for a chance at randomised rewards, the first thing I noticed was a structure suspiciously familiar. There is a published rulebook, a pity mechanism, and a single entity that writes the rules, operates the system, and collects the revenue. This concentration of power is not rare in the gaming industry, but rarely does it reveal itself so clearly through a mere release schedule. Most coverage of this topic stops at service information. It tells you when a banner opens, which character appears, whether to save or to pull. It answers questions about timing and choice, but leaves the harder question blank: what is this system designed to do with players' money and trust. This article does not aim to tell anyone how much to spend. It aims to reconstruct the architecture of a revenue machine, compare it with how the esports industry handles money, and point out a familiar crack. When the rule-writer is also the ticket-seller, transparency becomes an adjustable variable, not a guaranteed constant. To read this machine, one must place it in its operating context. Genshin Impact is an open-world action role-playing game published by HoYoverse, operating on a gacha model. In this model, players use premium in-game currency to pull for randomised rewards, the most valuable being a five-star character or weapon. The content cycle is divided into versions. Each version lasts about six weeks, split into two phases of roughly three weeks each. Each phase opens one or more banners, i.e. pools, with its own promoted character. The two-phase, roughly three-week structure is the machine's heartbeat. It creates a regular repeating cycle, turning spending into a scheduled habit rather than an impulsive decision. According to data recorded during the analysis period, version 7.0 entered its second phase with rerun banners, i.e. bringing back previously released characters. Shortly after, phase one of version 7.1 opened with two new characters simultaneously, while phase two of 7.1 continued with reruns. This rhythm is not accidental. It concentrates spending pressure precisely on the opening phase of a new version, when two new characters appear at once. I recall how I once built tracking sheets for VAR situations. Every controversial decision has a window, the period in which the referee must issue a ruling. If that window is too narrow, mistakes happen; if too wide, the match loses rhythm. Revenue machines operate on a similar logic, except that the window is deliberately designed. A three-week banner is a time-limited purchase window. Players cannot postpone indefinitely; when the banner closes, the chance to pull that character disappears, at least until the next rerun. That deadline is the first psychological lever, and it exists in the rules before the player makes any decision. Notably, the source text provides no data on the strength of the new characters. No tier list, no kit analysis, no stat comparison. It only provides a schedule. This absence is systematic, because in a gacha machine, a character's value is not measured by competitive performance but by scarcity and expectation. Players are not invited to judge whether a character is stronger or weaker. They are invited to consider whether it is worth opening their wallet. That is a crucial shift: from evaluating capability to evaluating desire. In my VAR work, I have seen debates stretch for hours merely to determine whether a player was offside. The final conclusion did not depend on who wanted what, but on a line drawn on the grass. Here, that line does not exist. There is no objective yardstick for how much a character is worth. Value is created by the conversation around it, by waves of anticipation nurtured through each news item, analysis video, and forum comment. The machine does not sell characters. It sells expectation. Three parameters shape the entire spending experience within this machine: the pity ceiling, the 50/50 mechanic, and shared pity across banner categories. These are not trivial technical details. They are the pricing framework players rarely see in full. The 90-pull pity ceiling is a safety floor. It guarantees that however bad luck is, a player will receive a five-star character within 90 attempts. This is a clever design, creating a sense of safety while establishing a predictable spending cap. Knowing the worst case is 90 pulls, players can budget. But precisely because they can budget, they are easily guided into planned spending, which is more dangerous than impulsive spending, because it is rationalised with numbers. The 50/50 mechanic is more complex. On a promotional banner, the first five-star has a 50% chance of being the featured character and a 50% chance of being a standard-pool character. If that pull yields a standard character, i.e. losing the 50/50, the next five-star is guaranteed to be the featured one. This structure creates a probabilistic game with insurance, where players always feel they are near the winning threshold, even when they are far from it. I have a habit of measuring decision latency. In VAR, the time between the moment a foul occurs and the moment the referee rules heavily affects ruling quality. In gacha, the time between the first pull and the ninetieth operates similarly. It is not merely duration but an emotional space where the player convinces himself that the next pull will be different. The number never sits still in the wallet; it always whispers that the threshold is a little closer. Shared pity across same-category banners is the third parameter, and the least noticed. When your pull count on one banner carries over to another of the same type, the marginal cost of switching from a new banner to a rerun, or vice versa, drops. This is a revenue-smoothing design. It keeps the cash flow unbroken between phases, because players lose no progress when switching targets. This is not generosity; it is a technique for retaining capital flow. Having understood the pricing, one must look at the supply policy. Characters cannot always be pulled. After their banner closes, they leave the pool and can only return via rerun. The key point: there is no fixed rerun schedule. Some characters are absent for over a year, while others return after only a few versions. This uncertainty is not an operational flaw. It is a tool. In behavioural economics, when consumers do not know exactly when a scarce item will return, they tend to value it higher and act faster. This is why limited-version events are designed with vague deadlines. You do not know when the next one is, so this one becomes the only one within reach. For a character absent over a year, the pressure is even greater. Players cannot be sure of a second chance within a reasonable time. Alongside this is the Chronicled Wish mechanic, a separate banner type operating under its own rules, usually for older characters. Its existence reveals a secondary revenue lane. Characters seemingly dormant can still be re-monetised without occupying space on primary banners. This is a technique of re-monetising old assets, much like a football club selling the shirt of a retired legend. Assets do not depreciate; they simply wait for the right moment to be repackaged. From a data analyst's perspective, this is an interesting intersection between two seemingly different monetisation models. Esports earns through sponsorship, broadcast rights, in-game item revenue sharing, and prize pools. Gacha earns through direct, recurring spending voluntarily made by players. Both rely on nurturing a committed community. But gacha depends on no tournament, no schedule, no sporting result. It is a self-running machine, as long as players keep paying for expectation. Here a paradox familiar to anyone who has analysed sports rules appears. The gacha rulebook is published publicly, full of numbers: 90 pulls, 50%, guarantee. Reading it, everything seems transparent. But the rulebook does not record the most important part: the reader's emotions. It does not record that tens of thousands of players will stay up until midnight to pull in the first seconds a banner opens. It does not record that a character absent fourteen months will produce a wave of pent-up spending upon return. The noise of the crowd is not written in the law, yet it carries legal weight. And this is where I want to pause longer, because it connects directly to my work. In VAR, we once believed that with enough camera angles and enough time, the truth would reveal itself. We were systematically wrong. Some decisions cannot be agreed upon even after ten reviews from ten angles, because the definition of the foul itself is vague. The same happens with gacha. You can have enough numbers to calculate probability, but you cannot calculate the emotional value a player assigns to a character. That gap is precisely where the revenue machine works most effectively. There is a counter-intuitive view worth considering. Many believe the 90-pull pity system is an expression of kindness, that the publisher has limited its own profit to protect players. I am not sure that is the right reading. A predictable spending ceiling does not protect players from spending; it merely makes spending easier to plan. This distinction matters. In the gambling industry, this is called the illusion of control, the belief that because you understand the rules, you control the game. But the rule-writer, the operator and the beneficiary are the same entity. When one party sets the limit, prices the product, and sells the ticket, the question of whom the limit serves cannot be ignored. Here I must acknowledge a blind spot in the data. The probability figures are published by the publisher itself. No independent body verifies that the actual server rates match the published rates. This is not an accusation; it is a structural gap, like a match with no fourth official overseeing the referee's decisions. As an analyst, I record that gap rather than fill it with speculation. My personal experience makes me cautious about anything that sounds certain. In 2026, my data model flagged a defender as high card-risk and advised against signing him. The club signed him anyway, and he became a pillar who helped the team win the title. My model had ignored teammates' covering ability and differences in how referees in different national leagues interpret the rules. The lesson remains valid: numbers do not tell the whole story. In South Korea's esports industry, where I follow competitions weekly, direct-spending monetisation models are creeping into every corner. Tickets, virtual items, event rewards limited to a short window. An esports player may retire at twenty-five, but the commercial ecosystem around him does not stop. Teams are learning to convert fan attachment into recurring cash flow, much as a gacha machine turns expectation into steady per-version revenue. The analogy is not perfect, and I offer only one comparison to clarify structure. But it is enough to reveal a larger trend: every trust-based entertainment ecosystem is shifting toward direct, recurring, hard-to-verify monetisation. The concern is not that such models exist, but that consumers increasingly cannot distinguish a genuine guarantee from a revenue design presented as a guarantee. For years, the gaming industry has learned to turn uncertainty into an asset. Gacha is one of the most complete examples: turning probability into a price list, scarcity into a schedule, expectation into cash flow. The esports industry takes a different road but is increasingly learning the same logic. Time-limited in-game items, limited events, rewards obtainable only within a short window, all are the same technique applied to a different audience. What deserves thought is not whether the technique is good or bad. It is that players, like sports fans, have fewer and fewer chances to see the mechanism behind numbers presented as self-evident truth. When the 90-pull ceiling is announced as a guarantee, it is no longer read as a revenue design; it is read as an entitlement. That transformation happens quietly, as a controversial refereeing decision gradually becomes part of the match's history, until no one remembers a crack was ever there. For the esports industry, this is not the story of an RPG. It is the story of how every trust-based entertainment ecosystem is shifting toward direct, recurring, hard-to-verify monetisation. An esports player may retire at twenty-five, but the revenue machine he served will keep running, without him, as long as people keep paying for expectation. In every such system, the natural position of the observer is neither fan nor whistle-blower. It is someone standing far enough to see structure, and close enough to see people. The open question, and perhaps one that will stay open a long time, is whether a system can be transparent about numbers while opaque about motives. For if the answer is yes, then every publicly disclosed probability table is merely a mirror reflecting the reader's belief, not a map leading to the truth. Every VAR error is a crack in the mirror that reflects the rules, and so is every published probability table. We search the pitch not for justice, but for an excuse to stop arguing. With revenue machines, we are searching for the same thing, except that the excuse costs real money.