Data Discipline in Esports: When an Analyst Has to Say 'Insufficient Information'

3 a.m. in Seoul. A small apartment in Mapo district, blue light from the screen washing the ceiling. I opened the second-stage analysis file my team had just run, and what struck me was not a heat map but a table of nine sections, each carrying the same repeating status line: 'insufficient information.' Game title: blank. Tournament: blank. Team: blank. Player: blank. The list of information points: empty, literally an empty array. There was no game title to latch onto, and the engine — programmed to be honest — refused to invent anything. The interesting part was the human reaction. The new intern, very sharp, very impatient, messaged through the internal channel: 'We have a deadline tomorrow morning. Should we just grab a few familiar metrics and write it out?' He meant no harm. He was simply doing what most sports writers are taught to do: fill the gap with something that sounds plausible. A bit of xG here, a bit of PPDA there, a few national-team names, and the piece would flow. I understand that temptation, because I once stood inside it. At fifteen, during the World Cup in Russia, I wrote about Germany's loss to Mexico and was scolded by a male reader: 'Girls shouldn't speak about tactics.' Back then I could have answered with emotion. Instead I reopened the spreadsheet, calculated xG for both teams — Mexico created 1.8, Germany only 0.9 — and published a new piece with a chart. The numbers convinced more than the noise. But from that moment I also recognized a trap more dangerous than prejudice: the trap of using numbers to fill a void, turning data into a thin coat of paint over the truth that you actually have nothing at all. There are matches the naked eye cannot see; you have to let the spreadsheet tell the story. But when the spreadsheet has nothing to tell, the writer's first task is not to compose — it is to confess. The esports data-journalism trade sits exactly at the intersection of a decade of explosive growth and a crisis of trust. Over the past ten years, esports has gone from a playground of cramped PC rooms to an industry with billions of dollars in revenue, with tournaments streamed live to tens of millions of viewers at once. Along with it, data poured in. Before 2026, almost nobody discussed a League of Legends player's metrics beyond KDA. By 2026, a top-tier match could generate hundreds of data columns, from gold difference at minute 15 to objective-control rate, from vision score to the number of solo deaths in mid lane. This explosion brought two opposing consequences. On one hand, it gave analysts a toolkit never seen before to see what the naked eye misses — silent matches where a team wins without holding the ball, or loses despite creating more chances. On the other hand, it produced a generation of writers who believe that if there is more data, there is automatically more truth. That belief is wrong, and it is degrading the quality of most esports analysis content online. Before diving into the specific traps, I want to rebuild the full foundation. Without a foundation, every judgment is just sentiment dressed in numbers. And in this trade, sentiment dressed in numbers is the gravest sin. A data journalist's vocabulary in esports differs by title, because each game measures what it is designed to reward. This is a point newcomers often miss: there is no universal metric set. A good metric in League of Legends can be meaningless in Dota 2, and vice versa. Let us start with League of Legends, the title with the most developed public data ecosystem. The most basic metric is KDA — the average of total kills plus assists divided by deaths. But KDA is a lazy metric, because it rewards caution. A mid-laner who stands far from teamfights can keep a pretty KDA while contributing nothing to victory. For this reason, serious analysts move to the time-stamped group, most importantly GD@15 — gold difference at minute fifteen — and CSD@15 — creep-score difference at minute fifteen. GD@15 measures a player's economic strength during the laning phase, before large teamfights begin. A top-laner who wins lane will have positive GD@15; one who is pressured under tower will have negative GD@15. But reading GD@15 without context is a classic error. If a team deliberately sacrifices the top lane to funnel resources to the bottom lane, the top laner losing gold at minute 15 is not a failure — it is a strategy. That is why I always require my team to cross-check at least two independent data sources and state the sample size before concluding anything. The second notable metric is DPM — damage per minute. This measures involvement in combat. A high-DPM player applies constant pressure. But DPM can also be inflated: a marksman on a weak team often has to deal more damage because they are constantly pressured, so high DPM is not necessarily a sign of quality. I have seen analyses praise a player simply because his DPM topped the chart, without checking whether his team won or lost, or whether that damage converted into objectives. The third metric is KP — kill participation, the percentage of the team's kills in which a player was present. High KP in a mid-laner or support usually signals a coordinating role. Low KP in a marksman can be normal if the team plays around the top lane, but it can also signal isolation. And finally, the group least mentioned but often most important: vision metrics and objective-control rate. In League of Legends, information is a resource. A team that controls vision around the river and dragon pit forces the opponent to guess. Objective-control rate — the percentage of dragons, heralds, towers, and Barons a team secures — often predicts match outcomes more accurately than KDA. In a recent season, I tracked a team whose average KDA was only mid-tier but whose objective-control rate was among the leaders, and they finished the group stage far above many teams with more impressive kill numbers. That is an example of the principle I always repeat: when I predict, I do not look at emotion; I look at operating structure. In football, that structure is PPDA; in esports, it is the combination of vision, objective control, and tempo. Moving to Dota 2, the metric set changes entirely because the game's nature differs. Dota 2 has no clear notion of 'winning lane' as League does; instead it has a far more complex economy. Three foundational metrics are GPM — gold per minute — XPM — experience per minute — and net worth, a player's total asset value including held items. Net worth is the metric I trust most in Dota 2, because it directly reflects a player's immediate power at a given moment. A carry with two thousand gold more than his opponent at minute thirty is a threat that can convert into real power in fights. But high net worth does not automatically mean victory, because Dota 2 is a game of comebacks. There are matches where a team leads by tens of thousands of gold and still loses because of one poorly positioned teamfight. This is why I never conclude from a single number, and always remind myself that correlation is not causation. Hero damage — damage to heroes — and tower damage split damage into two tactically different categories. A team dealing high tower damage plays a push-tower tempo, while a team dealing high hero damage plays a fight tempo. Recognizing which tempo a team is playing is the first step to predicting how they will win. In Counter-Strike 2, the metric world is rewritten once more. Here the unit is the round, not the minute. The most important metric is ADR — average damage per round. A player with ADR above one hundred is usually a constant threat. The second metric is KAST — the percentage of rounds in which a player has at least a kill, an assist, a survival, or a trade. KAST measures consistency, and it often distinguishes a true star from someone who only occasionally shines. The third metric, and the most debated, is the rating from an independent statistics system the community uses as its standard measure. Rating aggregates many factors into a single number, and precisely for that reason it is both convenient and dangerous. Convenient because it allows quick comparison; dangerous because it hides the method behind it. I always advise young journalists to read the metric breakdown rather than only the final rating number. Finally, Valorant, the youngest title in the group but already with its own data ecosystem. The leading metric is ACS — average combat score — combining damage, kills, and assists into one measure of contribution. Alongside it are KAST and the number of first bloods — the first kill of each round. First blood is especially important because it creates a numbers advantage from the start of a round, and in a game where a round lasts barely two minutes, that advantage is often decisive. The common thread across all four titles is this: every metric is a perspective, not the full picture. A poor writer picks one metric and builds the whole story around it. A good writer builds a network of metrics that support one another, and only then concludes. The spreadsheet does not lie; the reader must learn how to listen — but the writer must also learn how to ask the right questions. To see how a network of metrics operates, look at one of the most analyzed stories in esports history: T1's championship run at recent League of Legends World Championships. In 2026, T1 defeated Weibo Gaming 3-0 in the final. In 2026, they won again, overcoming Bilibili Gaming 3-2. With those titles, Faker — Lee Sang-hyeok — raised his total world championships to five, a record unmatched in the discipline's history. The emotional story is well known. But looking at the data, we see something else. In the decisive matches, T1 was not the team with the most dominant kill numbers, nor the one dealing the most damage. What they dominated was the ability to convert small advantages into big objectives. Their objective-control rate in the mid-game was usually higher than their opponents', and their count of wasted Barons or dragons was very low. In other words, they won through structure, not through moments. I had the chance to follow some T1 matches in the knockout stage directly in my role as a data reporter. What I recorded in my notebook was not highlight plays but tempo. T1 typically forced fights around major objectives once they had a vision advantage, and they were willing to retreat when the advantage was not ripe. It was a repeating pattern, and that repetition — not luck — is what data can measure. Compare this with another story, this time in Dota 2: Team OG's back-to-back championships at The International in 2026 and 2026. OG became the first team to win The International twice in a row, and the way they did it is a lesson in how data cannot capture everything. Before the tournament, prediction models based on individual metrics did not rate OG highly, because their roster was assembled late and they had no outstanding GPM or XPM in the group stage. But looking at situational combat metrics — the efficiency of coordinated plays at decisive moments — OG stood out. This is a lesson every analyst must burn into memory: there are dimensions of the game that current data cannot yet measure, and the silence of data does not mean the absence of value. A stray number can be a truth hiding where nobody expected — but a gap in data can also be a limitation of the tool, not a statement about reality. At this point, I want to pause on the least-discussed aspect of the trade: the data pipeline. Most viewers only see the final product — an analysis, a chart, a prediction. They do not see that behind every number is a chain of fragile steps: collecting raw data from publisher APIs or statistics sites, cleaning it to remove duplicates and errors, standardizing fields so they can be compared, then extracting meaningful information points. In the process my team uses, the first step — source deconstruction — turns a raw article or match into a structured set of information points: who, did what, when, with what number. The second step — deep analysis — takes that set and examines it across many dimensions: patch and meta, tournament format, rosters and players, regional landscape, club finance, rules and governance, risk profile, public narrative, and the industry's transmission chain. The important thing is this: the second step depends entirely on the first. If the first step returns an empty set, the second has nothing to analyze. And that is exactly what happened in the file I opened at 3 a.m. All nine analytical dimensions were fully framed, but every cell carried the line 'insufficient information.' No game title was identified — not League of Legends, not Dota 2, not Counter-Strike 2, not Valorant. No team, no player, no event or transaction to examine. For an ordinary database, this would be a disaster. For a data journalist with discipline, it is a moment that tests professional dignity. And this is where I want to speak directly to something many in the industry avoid: the pressure to produce content is making writers fabricate data without realizing they are fabricating. Nobody sits down and decides 'today I will lie.' Instead, the fabrication happens in small steps: taking a familiar metric from another match, assigning it to an unverified team, adding an adjective for weight, and finally a conclusion that sounds very expert. The whole piece flows, has numbers, has charts — and is wrong from the root. Even more dangerous is the sophisticated kind of fabrication: using real data in the wrong context. A metric correct in one season is applied to another season with a different meta. A figure from one region is used to conclude about another. This is the hardest error to detect, because every number is real. Only the connection between them is fabricated. I once watched a colleague heavily criticized for predicting a team would win based on their group-stage metrics, without checking that the knockout stage was played on a different game version than the one they played in groups. The patch changed, the old metrics lost value, but the spreadsheet in the article still looked pristine. The piece drew high readership, and readers' trust eroded a little more. That is why I built a hard rule for my team: before concluding, cross-check at least two independent data sources, state the sample size, state the assumptions, and leave one cell empty for the possibility of being wrong. A model with no empty cell is a model fooling itself. Alongside this are the three traps a data writer most easily falls into, and how to escape them. The first trap is metric intoxication — believing in a single metric as if it were truth. For someone whose identity is numbers, this temptation is especially strong, because a beautiful metric gives a sense of control. But every metric has blind spots. High ADR does not tell you whether a player knows how to hold position in a low-numbers situation. Positive GD@15 does not tell you whether a player knows how to convert advantage into objectives. The escape: always cross-check, and always ask 'what does this metric fail to measure?' The second trap is writing for insiders. After years of following the scene, technical language becomes a reflex, and the writer unwittingly builds a wall that new readers cannot climb. A piece meant only for a small group of experts is a piece that has abandoned most of the public. The escape: re-read the piece with a newcomer's eyes, and add one line of explanation for every term that appears for the first time. Precision does not require obscurity. The third trap — and the one I nearly fell into myself — is contempt for the naked eye. Because the nature of the work is to find the counter-intuitive, a data writer easily slides into the mentality that the naked eye is always wrong. But the naked eye is usually right at the level of phenomena. A team playing well looks good; a beautiful play is beautiful. The escape: start by acknowledging what the naked eye sees, then use data to explain why it happened or why it deceives us. Stand above the naked eye, not outside it. At this point, I must return to the blank spreadsheet at 3 a.m., because it is the hardest test of every principle just stated. When the input source is empty, there is no patch to examine, no format to analyze, no roster to assess, no regional picture to compare, no financial event to weigh, no rule to check, no risk profile to build, no public narrative to dissect, and no industry transmission chain to trace. All nine analytical dimensions stand still. A writer without discipline would treat this as an opportunity. They would fill it with the familiar: a few national-team names, a few football metrics, a few observations that sound profound. Readers would not notice, because the piece flows. But that is precisely the moment the data-journalism trade betrays itself. Because a piece with no source is not a bad piece — it is a toxic one, planting 'truths' that do not exist into readers' minds. The right response is not silence but transparency. State clearly that the input is empty. State clearly that this is a failure at the data-collection layer, not an analysis task. State clearly that the corrective action is to re-run the source-extraction step, not to invent an analysis out of thin air. And most importantly: propose a gate so that an empty set never slips through. That is not weakness. It is the kind of courage this trade needs more than any bold prediction. Saying 'I do not know' is harder than saying 'I predict.' But precisely because it is hard, it is worth it. Looking more broadly, this incident is not the private business of one analysis team. It is a symptom of how the esports industry operates its data. When speed is placed above accuracy, when the number of articles is treated as the measure of success, then empty sets — whether empty due to technical failure or empty because the subject has nothing new — will always be forced to generate content. And content generated from nothing is destructive content. There is a lesson I learned very early, when I was a fourteen-year-old girl sitting on the sideline with a notebook, recording every pass of a young midfielder. That day I noticed he had a ninety-two percent pass-accuracy rate but had made only three forward passes. I wrote in my report that his midfield control was 'soulless' because it lacked through balls. After the match, the coach confirmed the observation and used it to adjust tactics. It was the first time I saw data speak a truth the naked eye had missed. But the bigger lesson came from that very number, ninety-two percent. If I had looked only at the number, I would have praised him. It was because I placed the number in context — counting forward passes — that I saw the problem. A number detached from context is a number lying innocently. And a data journalist, for life, must fight that dangerous innocence. At fourteen I sat on the sideline with a notebook; football did not look at me, but the numbers did. That line holds for football, and holds many times over for esports — a younger, faster industry with fewer gatekeepers. So which signals should be tracked in the next cycle? I propose three directions. First is the health of the data pipelines themselves. As more and more analysis is generated by automated systems, the quality of the writing depends on the quality of the source-extraction step. A system without a gate for empty input will silently produce pieces that are formally perfect but substantively hollow. This is a slowly spreading risk, and it is more dangerous than an obvious error, because it makes no noise. Second is a culture of stating confidence levels. A mature analysis must be able to say what it believes, how strongly, based on how many samples, and which assumptions could be wrong. This is the standard of other serious analytical fields, and esports needs to adopt it. Readers are not afraid of the truth that a prediction carries uncertainty; they are afraid of being deceived by a false certainty. Third is training the next generation of writers to say 'insufficient information.' In an industry that rewards confidence, teaching young people that sometimes the right answer is an admission of one's limits is a difficult and rarely done thing. But if we do not, we will keep building a house on sand — the higher we build, the faster it falls. I do not believe in luck. I believe in the number of blocked shots and forgotten gaps. And in this specific case, I believe in something even humbler: when there is nothing to count, do not count. Reopen the pipeline, fix the failure at the collection layer, and run it again from the start. That is the only way the next number can mean anything. That night, I sent the intern a short message. I wrote: you were right to want to submit on time, but our trade is not a trade of submitting on time. Our trade is a trade of telling the truth, even when that truth is a void. He was silent for a long while, then replied: 'So can we write a piece about this very void, noona?' And that was the beginning of the piece you are reading. A piece about data discipline, born from a spreadsheet that had nothing. Perhaps in the future, when analysis engines become smarter, people will reconstruct an entire match from scattered fragments. But until then — and perhaps forever — the dignity of data journalism lies here: knowing enough to speak, and knowing when to stop because there is not yet enough to speak. They told girls not to speak about tactics; I drew charts instead of answering, and sometimes a blank chart is the most honest answer I can give.