Onimusha: Way of the Sword and the Esports Mislabel Inside the Data Pipeline

Last Tuesday night, the second monitor in my small Shenzhen apartment lit up with a spreadsheet 4,117 rows long. I was filtering fresh records for the morning esports bulletin when row 3,842 stopped my finger mid-air. The classification tag read, plainly: esports. Game title: Onimusha: Way of the Sword. Publisher: Capcom. The laptop fan hummed evenly, and I scrolled down.

No teams. No tournaments. No rosters. No qualifiers. No prize pools. Not a single line about a competitive patch cycle or a meta rotation. Just a single-player action game, a system-generated tag, and one record sitting neatly inside a data pipeline it never belonged to.

Ten years covering this industry taught me one thing: when a record is wrong at the labelling stage, every number downstream deserves to be challenged. This time, though, the game itself carried some very concrete figures, and they are worth dissecting properly.

A single-player title walking into the arena of professional teams

Onimusha: Way of the Sword is one of three major titles Capcom plans to release in 2026, alongside Resident Evil Requiem and Pragmata. Players control Musashi, upgrading his sword and clothing with iron and leather gathered from side quests and Ace Archer challenges. That is the entire game structure the source describes: one person, one progression track, one story.

In our pipelines, game content usually splits into two branches: single-player, and esports. The second branch requires teams, tournaments, schedules, pick-and-ban data. When a single-player title carries an esports tag, it takes up a row and drags a whole chain behind it: win-rate prediction models, form indices, team rankings. All of them can be poisoned by an entity that does not exist on any competitive stage.

A mislabelled record does not stay put in its own table. It flows down into every report generated from that table.

That is why I spent the whole morning dissecting what this game actually says about itself.

The numbers that can be verified, and the numbers that cannot

First, length. According to the source, players on the Action difficulty setting need roughly 30 to 40 hours to finish the main story. The main story is separately estimated at about 30 hours. Players unfamiliar with action games may need more than 50 hours. Completionists add another 10 to 15 hours of optional activities.

Here the first problem appears, and it is the kind I run into most often in this job: two figures that do not match sit inside a single source, and neither carries a stated method. The gap between 30 hours and 30-40 hours runs to ten hours — for an action game, that is the length of an entire major chapter.

I am not saying the source is wrong. I am saying it is not yet enough to be called data. The 30-40 hour figure could be the result of a real measurement; it could also be the writer's impression after one playthrough. Without a method, I am not allowed to add or subtract them.

But there is one point I can calculate, and it is far more interesting. The source records 36 boss encounters inside a campaign lasting roughly 30 hours. Divide 30 by 36 and I get about 50 minutes per boss. That is an unusually dense cadence for a single-player action game.

A density of one boss every 50 minutes forces the designers into one of two directions: either boss fights are the heartbeat of the whole game, or they become a repetitive burden the player has to endure.

I lean toward the first, but only at medium confidence. The reason: the source itself admits some bosses are arguably tougher than the rest, meaning there is difficulty stratification between encounters. A game padding its boss count for volume would not need that stratification.

The second point worth dissecting is the upgrade economy. Iron and leather — the two main materials for upgrading Musashi's sword and clothing — come from side quests and Ace Archer challenges. The source calls them essential materials. If that holds, the optional content layer stops being optional.

When core upgrade materials sit behind side content, players who rush the main story will face the late bosses under-equipped.

This is a familiar design pressure, and it never appears in the source's description. It is the gap I have to fill myself by reasoning from structure.

The third point is the achievement structure. To claim the platinum trophy, a player must finish the game at least twice. Add 10 to 15 hours of optional content, and the arithmetic lands above 50 hours — and this time the arithmetic checks out on its own. It is the only figure in this whole article with visible internal consistency.

The fourth point, and the one I trust least: the claim that Onimusha's campaign is significantly longer than Resident Evil Requiem and Pragmata combined. No figures are given for either comparison title. No definition of longer is offered — longer main story, longer total content, or longer maximum completion time.

A comparison without a denominator is not a comparison. It is media positioning.

That is the line I wrote in my notes before closing the spreadsheet.

The problem is not the game

Read only the sections above and you would think my problem is a problem of length. The real problem sits elsewhere.

It sits in the pipeline itself. The esports analytics industry built its models on an assumption that every entity in the table has an opponent, a schedule, a result. When a single-player title slips in, it does not produce an obvious error. It produces an empty cell carrying a valid label. And an empty cell with a valid label is the hardest class of error to catch, because nobody audits a row that looks normal.

I have seen the same thing with football data. A match gets assigned to the wrong competition, and three weeks later the league-wide average defensive index has drifted with nobody knowing why. Whether a stadium has fans or not, the match still needs someone to retell it — but the match has to exist first.

This is where I have to argue against myself. Maybe this labelling error is harmless. Maybe esports is just a broad bucket for all game content, and nobody actually uses it for computation. I have no evidence to rule that out. But in my experience, labels like that rarely stay put. They get recalled, regrouped, pulled into quarterly reports. I do not build tables for matches; I build tables for doubt. And one bad row is fair reason to doubt the whole table.

One more thing deserves saying: a purely single-player title being recognised as esports reflects something broader. The boundary between game content and esports content is blurring in the eyes of automated systems, while it has not blurred at all in operational reality. One side needs a roster. The other needs a person sitting alone in front of a screen until two in the morning.

A thought to carry forward

Data does not live inside the cells; it lives between them. And in my industry, a number does not lie — it simply never tells the whole truth, and sometimes it is spoken by a tag someone else attached to it.

If you are building models on your own esports dataset, do one simple thing this week: filter out every entity that has no opponent. You may find a row like my row 3,842. The remaining question is not whether Onimusha is worth playing, but how many other rows in your table are carrying a name they never deserved.