The 21-Day Rhythm and the Revenue Machine: Reading the Banner Architecture of Genshin Impact

I opened my stopwatch and set it beside my notebook, a habit I carried from the Mỹ Đình athletics stands down to my desk in Hanoi. This time there was no athlete to time. My subject was a machine without legs: the banner system of Genshin Impact, the open-world action RPG published by HoYoverse. I recorded one number and stared at it for weeks: twenty-one days. That is the length of one phase in each version. Twenty-one days for phase one, twenty-one days for phase two, repeating without a mid-season break, without a transfer window, without qualifiers. When I drew that line on paper, it did not look like a holiday calendar. It looked like a competition schedule. Once I saw it, I could not stop counting.

Outsiders read Genshin Impact as simple entertainment. From the angle I am used to, the angle of someone who watches how sports organizations run money and calendars, it is a revenue machine engineered down to the last mesh. One thing must be said plainly about its nature: this title operates as a role-playing product, outside the professional competitive system. It has no club-tier league, no world championship, no player-transfer market, no balance patch serving head-to-head play. Its versions are PvE content drops running on a release cycle. That means there is no competitive meta to dissect the way I still do with athletics or football.

Yet it owns something I find no less interesting: a randomized-sales system running on rhythm. That machine has a few clear components. Each version splits into two phases, each lasting about three weeks. Each phase sells one or several characters on a limited banner. Players spend premium currency to pull, gambling on obtaining the character they want. At the center sits the pity mechanic, a guarantee threshold. According to the data I hold, a player is guaranteed a five-star character within ninety pulls. On an event banner, the first five-star has a fifty percent chance of being the limited character and a fifty percent chance of being a standard character; if a standard one appears, the next five-star is guaranteed to be limited. This is a pricing mechanism, not a contract clause.

Two more details belong in the notebook. First, the rerun schedule is not fixed: some characters are absent for more than a year, others return within a few versions. Second, a separate banner type exists, called Chronicled Wish, for older characters. And the key point: pity is shared across banners of the same type. I read that much, wrote it down, circled what needed verification, then isolated the structure to analyze it. My principle does not change: anchor on data, never use the crowd's feeling as a compass.

The first thing I want to dissect is the twenty-one-day rhythm. In sports, a schedule is designed to optimize recovery time between matches. Here, the twenty-one-day rhythm does not serve recovery. It serves spending. Three weeks is short enough to create scarcity, long enough for most players to accumulate just enough free currency for one attempt. Each time a phase closes, a new purchase window opens, and the player again faces an allocation decision. That structure repeats in a pattern I can draw in advance: phase one usually carries new characters, phase two usually carries reruns. The analysis I hold states plainly that phase two of version 7.0 was rerun banners, while phase one of 7.1 introduced two new characters at once.

When an organization repeats the same plan across many versions, it is not praying to luck, it is carving a habit into the player. That habit has a name: saving for the next version. I found that exact line in the analysis, advice telling players to prepare for 7.1. In wording, it is the buyer's statement. In structure, it is proof the machine has taught players to count on their own. The most successful sales system is one that makes buyers plan their spending before the seller has to speak.

Phase one of version 7.1, per the data in the analysis, releases two new characters at once. It is a small detail worth weighing. Two new targets in one window means currency-allocation pressure peaks right at phase one, rather than spreading evenly across the version. The player is forced to choose, and every forced choice is a decision the machine collects. Phase two, with its rerun banners, plays the closing role: it creates no new pressure but absorbs whatever currency remains after phase one is spent. That structure mirrors exactly how a tournament places its showcase match in the middle, then lets the undercard fill the rest of the schedule.

Then the pity mechanic. I see it as a scorecard with an uncertainty band. The ninety-pull threshold is a soft floor: it reassures players they will never fall into infinity. But the threshold does not erase variance, it only frames it. Between floor and ceiling lies a wide probability band. A player may hit early in the first few pulls, or hit the floor. That probability band generates the revenue. Put simply: this design optimizes two things at once, the sense of access when everyone has a chance, and revenue variance when big spenders and small spenders coexist in one market. Years ago, when I sat recounting the record tables of forty Vietnamese track-and-field athletes during the shutdown, I learned something: most value lies in the middle band, not at either extreme. Pity works the same way. Floor and ceiling are just the two ends; the money sits in between.

The fifty-fifty mechanic is even subtler. It turns the first five-star acquisition into a game with clear rules: half a chance of the target, half a chance of something else. But it does not let the player leave with lasting frustration; lose now, and the next one is guaranteed. This is retention design: accept a player losing one round, as long as the player stays for the next. In my sporting language, it is stretching a series to raise total minutes played, rather than going for a quick win. In football, people call a draw a disappointment; I call it an evening full of intent. Here too: a miss is designed to become a springboard, not a full stop.

The shared-pity detail across same-type banners is the piece I find most notable, and the easiest to overlook. It means pulls a player has accumulated on one banner are not lost when they switch to another banner of the same type. In experience, this is kindness. In business, it is a revenue-smoothing mechanism. It lowers the marginal cost of switching: a player hesitating between a new character and a rerun suddenly finds spending more across both windows less painful. When marginal cost falls, spending frequency tends to rise. That is not gut speculation; it is the logic of every points-based promotion system.

The unfixed rerun schedule plays a parallel role. If every character returned regularly, players would wait at ease. But when some are absent for more than a year, players cannot know for sure when the next chance is. That uncertainty is a form of time pressure, exactly like a sports organizer releasing a limited number of tickets to push immediate decisions. Combined with Chronicled Wish for older characters and shared pity, the machine gains a secondary income lane: characters thought to be dormant can be monetized again without breaking the main banner rhythm.

Finally, the power structure. In most sports ecosystems, power is dispersed: organizers, clubs, broadcasters, sponsors, each holding a share. Here, the publisher is at once the game's operator, the rule-setter of the pull, and the announcer of the banner schedule. The only official source in the analysis I hold is the publisher's own announcement. The rule-setter is also the rule-announcer, and also the beneficiary. Every number about pity, rates, and schedule is published by one side, a point to remember before entering any calculation. I do not say this to accuse. I say it to set the right confidence level for each number, as I always do when cross-checking official statistics against my own hand-counted tables.

The counterintuitive point lies here. Most analysis of this title stops at the question of which character is strong. The analysis I hold provides not a single line on power or kit. It speaks only of schedule. That makes me think: what is sold here is not power, but rhythm. The player buys a window of time, not an index. When time becomes merchandise, every buyer's calculation revolves around now or later, not strong or weak.

And here is where I want to push back on the habit of analysts themselves. We tend to compare the revenue model of a randomized-sale game with the esports model, then conclude the latter is harder to sustain because it depends on sponsorship, broadcast rights, and prize value. Look closely, and the two systems carry opposite risks. The esports model depends on an external value chain of sponsors, broadcasters, audiences, and event calendars, so it is vulnerable to calendar shocks. The in-game randomized-sale model is closed: the publisher controls both supply and information, so it depends less on outside events. In return, it exposes itself to a different kind of risk. When rules on rate transparency and minor protection change, the whole machine may have to be redesigned from the root. Higher autonomy comes with higher legal exposure, not absolute safety. This trade-off is rarely stated plainly, because it does not sit in the revenue table everyone wants to show off.

There is one more blind spot. The analysis itself admits the exact banner schedule is still awaiting confirmation, and several character names and version numbers in it cannot be cross-checked against the game state I know. For someone who counts for a living, that is a red signal on reliability. My discipline is this: a fact that repeats many times carries weight; a fact appearing once, without a source, is only a hypothesis to be parked. Memory does not yield room to error, and a number without a source should not be granted room in a conclusion.

I closed the notebook and turned off the stopwatch. No finish line was crossed here, and perhaps none was needed. What I carry back is not a prediction of which character will sell best, but a way of reading: every profit-generating machine has a rhythm. The rhythm of a stadium is the whistle; the rhythm of the banner machine is the number twenty-one days flickering on a screen. Whoever takes the trouble to count sees structure before emotion. Every match is a countable bet; all it takes is patience to observe. For a piece built on many unverified facts, the most memorable thing is perhaps not what it asserts, but that it reminds us to count for ourselves, because the number published by the seller and the number recorded by the buyer will forever be two different numbers.