100 Percent Versus 9 Percent: KRAFTON's Four Fronts and a Trust Deficit Born From a Drop Rate

The fifth Premium Bundle came and went. The blueprint set never dropped. The tooltip on the in-game store still said that four consecutive misses guaranteed the item on the fifth attempt. He bought a sixth. Then a seventh. On the forum he found thousands of players waiting in the same queue. The displayed number was 100. The number buried in the code, a Korean regulator later confirmed, was 9.

That single gap opened four separate battlefields for KRAFTON over the past two years. One involved a regulator's fine. One involved minors' faces on a skin. One involved service fees from China. One involved four point one million signatures in Vietnam.

They look like four unrelated incidents. They share one structure, and it has almost nothing to do with drop rates themselves.

Here is the verdict first. A game company is never punished for the size of its lie. It is punished for the size of the price tag attached to that lie. A lie worth nothing produces no lawsuits. A lie worth a few thousand won per user becomes a business model.

I work as a junior betting analyst at a New York sportsbook with a heavy esports book. I built my xG model before I understood the market, and that taught me something most match-preview writers never learn: the reliability of a number is set by the transparency of its source, not the beauty of its presentation. A tooltip is marketing until there is code behind it.

My central argument is this: KRAFTON's four incidents are not four separate mistakes. They are four surfaces of one compromise. Wherever transparency collided with monetisation, the company repeatedly chose short-term revenue.

The spreadsheet said one thing. The stadium said another. The wider that gap, the larger the regulatory risk.

In 2026, KRAFTON ran its most discussed collaboration of the year with the Korean girl group NewJeans. Two paid products entered the in-game store: the PUBG X NewJeans Loot Pack and the PUBG X NewJeans Premium Bundle. Players opened them for collectible items and character customisation inspired by the group.

The problem began with the published odds. KRAFTON had announced that failing to receive the blueprint set four times in a row guaranteed it on the fifth attempt at 100 percent. Players who bought five, six, seven bundles did not receive it. After community pressure, the PUBG team confirmed the Premium Bundle did not actually fall under the bad-luck prevention mechanism, and that some in-game information was inaccurate. KRAFTON corrected the information, apologised, and issued compensation. Some high-spending players demanded cash refunds.

One distinction matters here, because much of the community blurred it. A refund and a compensation package are not the same thing, and a company's word for correction is not accuracy.

I do not trust a signal until it survives a cold Tuesday in February. That Tuesday arrived in June 2026. According to Yonhap, the Korea Fair Trade Commission fined KRAFTON over misleading drop-rate information in PUBG: Battlegrounds. The reported fine was two point five million won. The FTC found the company had told players a guaranteed item would arrive after four failed attempts, while the actual rate was set at 9 percent, and that other items' odds had also been misstated.

Then comes the number this entire piece turns on. KRAFTON refunded roughly one point one billion won to about three hundred eighty thousand buyers, and added in-game compensation worth approximately nine point eight billion won. Because the company corrected the information and paid compensation on its own initiative, the regulatory action was limited to a financial penalty.

Consider the scale. One point one billion won refunded. Nine point eight billion won in in-game compensation. A fine of two point five million won. The fine is a rounding error against the total remediation.

The newsletter began as a way to argue with my own numbers. Here I am arguing with them again: the fine is too small to function as a deterrent, and the remediation is so large it serves as a four-column billboard for the industry.

KRAFTON said the bad-luck mechanism did not apply to the Premium Bundle. The player's screen said it did. That distance can be measured.

Assume the mechanism worked as displayed. Then the effective cost of a blueprint set capped at five packs. Now assume a real rate of 9 percent with no guarantee. Under a geometric distribution the expected number of packs is about eleven point one. The displayed mechanism priced the item at roughly half its real cost.

The refunded one point one billion won divided by three hundred eighty thousand buyers is about two thousand eight hundred ninety-five won per buyer. That is only one slice of the cost. The rest is invisible.

The invisible part is the real damage. Money spent by users who were misled but never qualified for a refund appears in no ledger. Only what was returned appears.

An old habit applies. In betting markets I evaluate any offer by its implied probability, not its banner art. A transfer fee is a story the market tells before the player speaks, and loot packs work the same way. A time-limited collaboration pack removes the player's opportunity to verify a rate before purchase, which is exactly when verification matters most.

The NewJeans collaboration carried a second, different problem. Korean media reported that some users combined NewJeans face skins with revealing outfits to create and share sexualised, harassing images and videos. The case drew extra attention because Haerin and Hyein were minors at the time.

KRAFTON and ADOR, the group's management company, said they would act against uses of items outside the collaboration's original purpose. KRAFTON then restricted combining NewJeans face skins with certain in-game outfits.

The restriction reignited debate. One camp blamed user behaviour and wanted users punished. Another blamed PUBG's character customisation design, since the system is built to allow near-infinite combinations.

Both camps are partly right and both avoid one arithmetic fact. With ten face skins and two hundred outfits there are more than two thousand states. No single rule governs two thousand states, because moderation would require writing rules pair by pair while the number of states multiplies with every new cosmetic.

The fix that arrived was a configuration patch, not a principle. Blocking a face skin against a handful of outfits reduces the count of abusive states; it does not reach zero.

My view is that the real fix is a design-level limit: items that should never exist in combination with a given character age profile. That is a product decision, not a moderation decision. Moderation cannot repay design debt.

Working on closed-door match data taught me that a crowd and a behaviour are different variables. In PUBG's case, underage viewers, anonymous accounts and instant sharing raise the risk of any face-based content. A studio cannot solve the social problem, but it can decide how much of that risk sits on its own balance sheet through its design boundaries.

The China question looks unrelated and shares the same skeleton. In 2026, ZDNet Korea reported that, ahead of its IPO, KRAFTON disclosed for the first time in its registration filing that it provides technical services to Peacekeeper Elite, the Tencent-operated game in China, and receives service fees.

PUBG Mobile in China shut down in May 2026 after licensing problems. Tencent moved players to Peacekeeper Elite, a game with strong similarities to PUBG Mobile but altered gameplay, visuals and content. KRAFTON has consistently said the two are separate products. Some analysts argue Peacekeeper Elite was changed enough to meet licensing requirements. Others ask whether it was a route to keep PUBG alive in China.

Separate the analysis from the moral verdict. Technical service fees are real revenue. A company can call a product separate while sharing codebase, pipelines or service stacks at the engineering layer. The question then is not ethics but disclosure of revenue concentration and licensing risk.

As a betting analyst I am rarely satisfied by corporate language. Filings seldom detail what share of revenue such services contribute. But the question is now on the record, and every subsequent headline reactivates the old suspicion. That is the real cost, and nobody can amortise it.

The final incident produced the largest reaction, in Vietnam. At PUBG Asia Stars 2026, two Vietnamese players were implicated: Himass of Anyone's Legend and TanVuu of The Expendables. The case opened when Korean streamer Soopi of Gen.G accused the two of watching opponents' livestreams.

On September 23, KRAFTON published its investigation, finding that Himass and TanVuu used outside information, including other players' livestreams, to make judgments and build tactics during matches. KRAFTON classified this as stream sniping, violating PUBG: Battlegrounds operating policy and the conduct standards for professional players.

Both players were permanently banned and barred from official PUBG Esports events organised or approved by KRAFTON, including PGC, PGS and PNC. Previously, organisers removed them from the rest of the event, adjusted points, increased stream delay, cancelled the third day of play and split the prize money.

KRAFTON said the investigation found no further violations. Reaction in the Vietnamese PUBG community was immediate. Do Mixi, PewPew, Rambo, DjChip and Ngan Sat Thu protested or stepped away from PUBG. GAM x The Expendables demanded a clear explanation, while Anyone's Legend defended its player's right to be heard. Debate shifted from whether the two violated rules to the severity of the sanction, the organiser's handling, and whether standards were applied equally. More than four point one million signatures were collected.

Now the uncomfortable part the community would rather skip. Stream sniping is detected by inference, not observation. Nobody sees what is on a player's screen without a direct feed or system-level logs. What exists is pattern: when a player turned, when a check was anticipated. These are strong indicators, not proof.

The rule I use in modelling applies here. Write the kill criteria before the verdict. Which data proves the allegation, and which data kills it. Written after the fact, there is no difference between an allegation and a defence.

Second, there is a proportionality question. A permanent ban reaching PGC-level events was issued from a showmatch incident. That does not make it wrong, but it sets a precedent that the same standard applies everywhere. If so, the standard should be published in advance, not only in a rulebook but in the testing procedure.

Third, corporate fault and player fault are not the same thing. Organisers removed players first, investigated second, and disclosed results third. The correct order runs the other way. When results arrive late, users assume the outcome was tailored rather than drawn from evidence.

In 2026 I built a logistic regression model for a small betting syndicate on closed-door Bundesliga matches. It returned eight point four percent over twelve weeks. The most valuable lesson that year was different: when a league withholds its reasoning, people manufacture one, and manufactured reasons are usually more sinister than reality.

Four point one million signatures are not proof of innocence, and nobody claims they are. They are proof of a legitimacy deficit, and that deficit is a number, not a mood.

My final position on inference-based sanctions is this. Stream sniping is a real problem that strikes at competitive integrity, and organisers have the moral authority to act. But the burden of showing that the action is evidence-based sits with the organiser. That means publishing the method in advance, applying it equally, and scaling the penalty to the strength of the evidence. The strongest anti-sniping tool in this incident was never a ban. It was stream delay management and schedule changes, both of which are information-security decisions, and both of which cost broadcast value.

A final note on the shape of this argument. I am not defending the company or the community. I am reading the market. My job is not to amplify noise but to name its structure. On one point I will take a side without hesitation: the refund ledger. If KRAFTON could return one point one billion won to three hundred eighty thousand buyers, it holds the list of everyone who bought more than five packs but did not file on time. Only those who claimed correctly appear in the ledger. Escrow inquiries never happen.

Watch three numbers next. First, app store policy: if Apple and Google mandate verifiable odds APIs for paid random items, the 100-versus-9 gap becomes mathematically impossible to hide, and that will matter more than any fine. Second, the PGC 2026 rules annex: if it does not define what counts as external information and what evidence is admissible, the Himass and TanVuu case becomes a precedent rather than an incident. Third, whether Vietnamese organisations enter PGC 2026 qualifying at full strength. If they do, the community was damaged, not broken. If they do not, every future Vietnam-related decision carries a permanent legacy risk that no fine settles.

One name and one number to close. The names are Himass and TanVuu. The number is 9 percent. A company that runs 9 percent in its code and shows 100 percent on the screen does not need accountability as much as it needs a culture of proof, and that cannot be bought with a fine. It can only be printed. For the Vietnamese community the question is simpler still: four point one million signatures were collected against a sanction, and twelve months later nobody knows who will be in the PGC lineup.