Inside Riot's Anti-Boost Engine: 296,416 Accounts, a Four-Tier Penalty Ladder and Three Unnamed Blind Spots

In August 2026, three days after starting my job as a transfer market administrator at a sports data analytics firm in Chicago, I was handed a file with a name so tidy it was suspicious: integrity_flags_q3. It contained no player data, no expected goals figures, no contract valuations. It contained a list of accounts flagged for abnormal behaviour on online ranked ladders. Inside was a summary table I copied for my own reference. The number 296,416 sat on the final row, with no annotation, no breakdown by title, no regional split, no baseline against any prior period. A large number. But when I tried to disaggregate it — by VALORANT and League of Legends, by season, by region, by violation type — the file gave me no basis to work with. A total is not a trend. It is a single point. And in my daily work on transfer valuations, I learned a rule I carry into every analysis: a data point without a baseline is a claim, not evidence. An empty stadium does not falsify the numbers, it exposes them. By the same principle, a cumulative figure does not falsify the story about enforcement effectiveness — it merely exposes that the story remains unproven. This is the story I want to tell about Anti-Boost, the anti-boosting enforcement system Riot Games operates across its two largest competitive titles. Not from the perspective of a banned player, not from a press release, but from the perspective of someone who works with data and tries to break a governance system into readable components. Context: why the ranked ladder is not a tournament To understand why Anti-Boost exists and why the machine is more complex than a press release suggests, you have to understand what a ranked ladder is inside a competitive online title. In football, the divisional system rests on hard boundaries. A Premier League club cannot simply appear in League Two without a recorded promotion or relegation event. Boundaries between divisions are defined by associations, validated by seasons, and recognised across the ecosystem. In Riot's ranked ladder, boundaries are far more fluid. A player can drop from Platinum to Gold in a bad week, or climb from Silver to Platinum in a lucky one. The system does not distinguish genuine skill from rented skill. It records match results and adjusts points accordingly. That gap — between the recorded outcome and the origin of that outcome — is what creates the boosting market. Boosting works on a logic so simple it fits in three steps. A high-skill player logs into someone else's account. That person plays ranked matches on behalf of the account owner, accumulating rank points for them. The owner pays for the service and receives a higher rank without doing the climbing. The ranked system, designed to reflect a player's actual skill, is distorted at both ends simultaneously. One account holds a rank above its owner's true level. And one high-skill player is competing at a rank below their true level, influencing the outcomes of matches at that lower tier. For the publisher, the damage is not technical. The game still runs. Servers stay stable. Matches still fire. The damage is to the legitimacy of the ranking system. If players stop believing their rank reflects real skill, the value of climbing — the core motivation for playing a seasonal competitive title — collapses. And when that motivation collapses, daily active players follow. That is why Riot, a company operating both a community ladder and a professional esports ecosystem, cannot let the ladder become unmanaged chaos. The ladder is not merely where players play. It is the intake pipeline for the entire esports ecosystem. Future pros are discovered on the ladder. Academies recruit from the ladder. Teams analyse the ladder to find undervalued talent. If the ladder is distorted, everything downstream is affected. I know this from the other side of the desk. In my transfer market work, part of the job is evaluating young players from smaller leagues. We rely on data to find pricing asymmetries — players whose market value sits below their model-derived value. If the input data is contaminated, the model behind it is meaningless. That principle holds in football, and it holds in esports. Anti-Boost is Riot's answer to that question. It is not a patch. It is not a new agent or a balance change. It is an enforcement system operating at the account and behavioural layer, parallel to the gameplay balance layer. It is an invisible piece of infrastructure most players never see — until they become its subject. Four violation categories: how Riot defines boosting The core of Anti-Boost sits in a set of definitions and a penalty structure. Riot does not merely declare it will tackle boosting. It specifies what a violation is, groups violations into categories, and assigns a penalty tier to each. This is an important system-design move: it turns a vague concept — unhealthy play — into an operable classification at scale. The first category is direct boosting. A high-skill player logs into another person's account and plays ranked matches on their behalf. This is the narrowest and most classic definition. It requires two parties: the booster and the account owner. It requires a transaction, whether cash, transfer, or some other exchange of value. The second is account buying, selling or transferring. This is the act of purchasing, selling or transferring ownership of an account between individuals. Economically, it is the raw-material supply layer for the whole boosting market. Without an account market, boosting struggles to operate at scale, because every boosting transaction would require an owner who wants the service on their own account. Once accounts become tradeable assets, the flows get far more complex. The third is intentional deranking. A player deliberately loses matches to drop to a lower rank. The motive may be to face weaker opponents, or to prepare an account for a boosting transaction. In the second case, deranking is a preparatory step: drop low, boost high, and produce an account with an impressive climb history that does not reflect true skill. The fourth is using a smurf account to assist climbing. This is the murkiest zone of the entire system, because smurfing itself is not a violation. Riot explicitly distinguishes self-created, self-operated alt accounts — normal activity, accepted, even encouraged as a way to test new tactics — from alt accounts used to manipulate rank. That last point is a notable governance boundary. Riot does not ban alt accounts. It bans the intent to manipulate through them. This is a very narrow standard, and it differs meaningfully from how many other publishers handle the issue. That difference matters. A system banning all alt accounts would be easier to run technically, but it would conflate two entirely different populations: players testing new tactics on a separate account, and players manipulating rank. A system that distinguishes them — as Riot is attempting — demands higher detection precision, but protects honest players. This is a sign Riot is not simply trying to ban as many as possible. It is trying to ban the right targets. That is a much harder problem. Four penalty tiers and the accounting logic behind them From the four violation categories, the penalty structure is built as a four-tier escalation. Each tier handles a different severity level, and each reflects a judgment about motive and spillover. Tier one handles routine detections. When Anti-Boost detects manipulation, rank points and rewards earned from manipulated matches are cancelled. The account is returned to its pre-manipulation rank. A temporary suspension is applied. This is the lowest rung, designed both to punish and to give a correction window to players who may have violated without fully realising it. Tier two handles repeat offences. For repeat violations, suspension duration lengthens. Riot does not publish the exact escalation formula, but the logic is simple: each repeat offence brings a heavier penalty than the last. This escalating structure carries an important statistical implication: its existence implies the recidivism rate is non-trivial. If nobody reoffended, escalation rules would be unnecessary. Tier three handles commercially motivated violations and intentional deranking. Account buying and selling, or deliberate deranking, can lead to a permanent ban. This is the heaviest sanction in the system, reserved for the most clearly commercial behaviours. Tier four is joint liability. This is the most contested tier, and I will spend most of the contrarian section on it. Riot does not only action the directly manipulated account. It may also action the booster's main account, and — crucially — teammates who frequently play with the booster. By design, this is a heavily tiered escalation. It separates technical violations — boosting, smurfing — from commercial violations — account trading, deranking. It applies the heaviest sanctions to the behaviours with the clearest monetary motive. This is an accounting logic more than a purely moral one, and it is an interesting design choice. That logic makes sense operationally. If you intend to tackle a problem at the scale of tens of millions of users, you need a structure that prioritises resources. You cannot investigate every case evenly. You cannot guarantee every violation is caught. You must focus on the highest-spillover nodes — and in the boosting market, the highest-spillover nodes are the operators of commercial infrastructure: account traders and large-scale boosting service operators. But any prioritisation structure has a flip side. Here, the flip side sits at tier four. Joint liability: when punishment extends beyond the violator Joint liability in this context means punishment extends beyond the directly offending party to related parties. For Riot, those include two groups: the booster's main account, and players who frequently queue with them. The first group has clear logic. If someone uses their main account to boost, actioning it is a natural, even necessary, consequence. It raises the expected cost of supplying boosting services, because the booster has more to lose than a single manipulated account. The second group is the problem. How do you define frequently plays with. No threshold is published. No clear definition of frequency, duration, or queue pattern. And — the key point — no appeal mechanism is described for players actioned under this clause. Think about that from a false-positive risk perspective. Two players become friends, climbing together for three months. One of them, unaware that their partner is boosting accounts for others during other hours, keeps queuing alongside. Both show a high overlap in queue patterns — entirely normal for a ranked duo climbing together nightly. When Anti-Boost flags the other account for boosting, the friend can also come into scope. That is a real risk, and the documentation does not address it. I understand the motive behind the measure. In many boosting cases, the booster does not operate alone. They may play with a group, and that group may know about the boosting, or even participate in part of it — supplying connections, intermediary accounts, or technical support. The problem is distinguishing aware accomplices from unaware teammates. An automated system based on behavioural patterns can detect queue overlap. It struggles to detect the difference between a teammate who knows about the boosting and one who does not. That is a structural limitation of any automated system, and in this case it can affect honest players. Another angle worth weighing is the effect on social behaviour in the community. If players know that regularly queuing with someone can lead to punishment if that person violates, they may become more cautious about making friends and queuing. That is an unintended effect, and it can shape the game's social experience in ways that cannot be measured directly. The system also has another notable structural feature: it is a reactive-with-rollback system, not a preventive one. The measures described are detect, then cancel and roll back. That implies a lag — a window between when manipulation occurs and when it is detected and actioned. What does that lag mean. During it, matches already happened. Rank points already distributed. Other players already affected by those match outcomes. When Riot rolls back points for the manipulated account, it cannot roll back the impact on players who competed against that account during the preceding window. Structurally, this is a game-balance system continuously skewed by actors outside the model, with correction always arriving afterwards. That is a structural limitation, not a design flaw. But fairness matters here. This is not a system that can be solved perfectly, and nobody should expect it to be. Detecting manipulation at tens-of-millions scale is an extremely hard technical and statistical problem. No algorithm can perfectly distinguish a player performing above normal from a player being boosted. The improvement noted in the documentation — better match-level detection of boosting signs — implies the current system leans mainly on behavioural and telemetry signals rather than direct proof of account ownership. That implies false-positive risk cannot be zero. And false-positive risk, in a system without an independent appeal mechanism, is a trust problem. Because Riot is simultaneously the detector, the adjudicator and the enforcer. No independent third party appears in the described process. This structure — concentrating all governance power in the publisher — is not unusual in the games industry. But it is not unworthy of analysis either. It is a highly centralised governance model, run by an automated system, on self-reported data, with no published independent check. The economics of the boosting market: expected cost and supply-demand structure Now consider boosting as a market, because Riot's penalty structure is not merely a rulebook. It is an intervention in a market, and its effectiveness depends on how the market responds. The boosting market is not formal. It operates in grey space, through unofficial intermediary platforms, private contact groups, and closed channels. But it has the structure of a real market: supply, demand, price, intermediaries, and dispute resolution, however informal. Supply is high-skill players — sometimes including professional or semi-pro players — able to monetise their skill in various ways. Boosting is one, attractive for its low entry barrier, predictable income, and lack of organisational commitment like a team contract. Demand is players who want a higher rank without the climb. Their motives may be social prestige, access to community tournaments, or simply a sense of achievement — even if the achievement is not theirs. When Riot acts against boosting by banning accounts and cancelling points, it hits both sides of the market. But the effect on each side differs in expected cost. For the buyer — the account owner — expected cost is the service price, multiplied by detection probability, multiplied by damage on detection. If detection probability rises, expected cost rises. For some, the rise is enough to deter. For others, not. It depends on how much they value the rank against the risk of losing it. For the seller — the booster — expected cost includes lost income, main-account ban, and loss of future service capacity. Here, actioning the main account bites far harder than actioning only the boosted account. If only the boosted account is actioned, the booster loses a working account — low cost, replaceable. But losing the main account, containing their full history, reputation and ranked skill, is a much higher cost. This is why extending enforcement to the booster's main account has clear economic logic. It raises the expected cost of supplying boosting services, and therefore reduces supply at the current price. One more angle: boosting rarely exists alone. It is part of a larger grey economy including account trading, account-levelling services, and to some degree, betting-adjacent activity. When Riot bans account trading, it strikes at the raw-material supply for that whole ecosystem, not just boosting. That is strategically meaningful. Rather than targeting one service, it targets the infrastructure that allows the service to exist. But — and this matters — there is no data on the real effectiveness of the measure. The documentation gives no recidivism rate, no data on how the boosting market shifted after enforcement waves, no data on service price changes. We know what Riot does. We do not know how well it works. That is the biggest information gap in this whole story. And it is a gap a single cumulative figure like 296,416 cannot fill. Detection: signs, telemetry, and structural lag The question of detection mechanism is the murkiest part of the whole system, and that is not accidental. Riot describes Anti-Boost as an automated detection system. It also mentions improving detection of boosting signs at match level. The word signs matters. It implies the system does not rely on direct proof that an account was logged into by someone else, but on behavioural patterns within matches. Those patterns may include sudden personal performance shifts, changes in play pattern, inconsistency between activity history and current performance, and other telemetry signals. But here is the key point: any behaviour-pattern-based detection system has a false-positive rate greater than zero. And that rate rises as you try to catch increasingly sophisticated behaviour. A concrete example: a genuine player may go through a sudden improvement phase due to practice, psychological change, hardware change, or tactical shift. That phase can produce a behavioural pattern similar to a boosted account. The only difference may be external context — which an automated system cannot see. This is why any automated system needs a human review layer, and why the absence of an independent appeal mechanism in the documentation is concerning. I need to be clear: I am not saying Riot lacks an appeal mechanism. I am saying the documentation does not describe one. In data analysis, absence of information is not evidence of absence of a mechanism. But it does mean we cannot evaluate it. There is another technical dimension. Detecting boosting at match level requires analysing behavioural patterns over time, not in a single match. A player having one good match does not mean they are boosted. A streak of ten matches of sustained outstanding performance, against a prior history of average performance, is a stronger signal. But even such a streak is not conclusive, because there are many legitimate reasons for a sudden improvement. That is why match-level detection is described as an evolving improvement, not a finished solution. This is an arms race between detection and evasion. And in any arms race, the attacker holds a structural advantage: they only need to find one gap, while the defender must close them all. Contrarian view: three unnamed blind spots Here I want to challenge the very premise of the story as Riot and media outlets present it. The premise is that Anti-Boost is a competitive-integrity safeguard. It protects honest players from rank manipulators. It makes the ladder fairer. I do not dispute that premise in principle. But I want to point out three unnamed blind spots in how the system is presented. Blind spot one is concentrated power structure. Consider the whole governance chain. Riot defines the violation. Riot detects the violation. Riot adjudicates the violation. Riot enforces the punishment. Riot publishes the enforcement data. No step in that chain involves an independent party. That means every piece of information we have about the system — including the 296,416 figure — comes from a single source. And that source is the one with a direct interest in presenting the system favourably. This is not an accusation of dishonesty. It is a structural observation. Any organisation publishing data on its own performance has an incentive to present that data favourably. That is a fact of every self-reported measurement system, from firms publishing earnings to governments publishing national statistics. When you have a self-reported system, no independent audit, and no way to verify claims externally, you are not evaluating the system. You are evaluating the press release about the system. In practice, that means we do not know the false-positive rate. We do not know how many cases were wrongly actioned. We do not know how many cases slipped the detection net. We do not know whether the system is improving over time. All we know is that Riot claims it actioned 296,416 accounts over an imprecisely described window, with no breakdown by title, violation type, or region. Blind spot two is the methodological problem of the trend claim. Riot presents the figure as evidence of intensifying enforcement. But there is a fundamental methodological problem: a cumulative number does not establish a trend. To establish a trend you need at least two data points at two different times, ideally a time series. With a single number, you have one point. A point without a baseline cannot say anything about direction. If Riot publishes a figure covering a vaguely described window, and six months later publishes another, only then do we have a trend. For now, the claim of tightening is an inference by the writer, not a finding from data. An outlier can tell a whole season's story. But a number without a baseline only tells its own. Information structure also matters. Riot pools two mechanically very different titles into one number. VALORANT is a tactical shooter, where individual skill depends on reflexes, precision and tactical decisions in very short time windows. League of Legends is a multiplayer online battle arena, where skill depends on map reading, resource management, and team coordination over much longer windows. These two ecosystems have different economic dynamics, different account markets, different climbing pressure, and most importantly, different detection complexity. A boosted VALORANT account may show a skill shift far more clearly than a League of Legends one, where single-match variance is much larger. Pooling them into one number hides those dynamics. That is not technically wrong, but it removes analytical readability. This is a principle I learned in player valuation: data knows the story before we do, we are just late to it. But for data to know the story, it must be disaggregated enough to be read. A pooled number is not readable. And one question the documentation does not answer: of the 296,416 actioned accounts, how many were owners, how many were boosters, how many were teammates hit by joint liability. That breakdown would entirely change how we judge the system. If most are teammates, the system has a targeting problem. If most are owners, it targets correctly. Without data, no conclusion. Only the question. Blind spot three is the transparency-versus-effectiveness trade-off, and how it is handled here. There is legitimate reason for Riot not to publish detailed detection data. Publishing too much about how detection works could help manipulators adjust to evade it. That is a sound consideration in any anti-cheat system. Other publishers do not publish anti-cheat details. Casinos do not publish their fraud-detection details. Exchanges do not publish their manipulation-detection details. This is a fundamental trade-off between transparency and effectiveness. Transparency helps the community trust the system. But transparency also helps manipulators bypass the system. The industry answer is not to publish everything, but to publish enough to build trust. And the way to build trust in a self-reported system is to provide data comparable over time, disaggregated by clear criteria, and where possible, verified by an independent party. With Riot, at this stage, we have one number. And one number is not enough to build trust. That is the crux of blind spot three. This leads to a final governance observation. In traditional sport, competitive integrity is usually managed by bodies independent of the leagues. Ethics committees, arbitration panels, anti-doping agencies — these exist to ensure decisions are not made only by parties with an interest in the outcome. In esports, this model is less common. Because publishers hold intellectual property, they also hold governance of the competitive ecosystem. They make the rules, enforce the rules, and own the platform where the rules apply. This is a uniquely esports governance model, with both upsides and downsides. The upside is consistency and speed: a single publisher can deploy enforcement across its ecosystem quickly. The downside is concentrated power and missing checks and balances. In that context, Anti-Boost is not merely an anti-cheat system. It is a statement about how Riot positions its role in the ecosystem. And how it positions that role will shape not only the ladder, but the relationship between publisher and player community. The transfer market is where emotion is priced in numbers. So is the ranked ladder. Progressive thought: what to track in the next cycles This is where I stop, and this is what I carry. 296,416 accounts is a substantial number. But it does not answer the most important question: whether the system is getting better. To answer that, we need more than a data point. We need a series. And we need a series disaggregated enough to be readable. What I will track in the coming periods is not a new cumulative figure, but structural changes. Whether Riot publishes a specific threshold for the frequent teammate clause. Whether it adds an appeal channel for joint-liability cases. Whether it disaggregates figures by title and violation type. Those structural changes will tell a more accurate story than any cumulative number. Because they will tell us whether Riot is learning from its own system. I will also track a less-noticed signal: how the community reacts to publicly known false-positive cases. In any automated enforcement system, sooner or later a wrongly actioned case goes viral. How Riot handles it will be the real test of the legitimacy of the intent-based standard it operates. And finally, I will track how other publishers respond. If Riot publishes enforcement data in a comparable way, and other publishers begin to do the same, we will have a basis for cross-ecosystem comparison. That is when we can start evaluating not just one system, but a governance model. Football does not lie, we just hear it on the wrong frequency. Governance systems are the same. They do not lie about their effectiveness. They simply publish what they choose to publish. The analyst's job is not to believe what is published, but to understand what is not. That is the gap between the number and the story. And that is where the real work begins.