AI Rebuilt RuneScape in Weeks: Vietnamese Football Data Is Next in Line

A website ran RuneScape inside a browser. In its first 24 hours it recorded roughly 3,000 registered accounts and more than 1,000 hours played. The person who built it was Rohan Varma, an independent game developer. He asked GPT-6 Astra in Codex to “build RuneScape” and received something complete enough to name: Elderwood Realms, with maps, a skilling system and combat mechanics inspired by Jagex's long-running MMORPG.

Then a letter arrived. Jagex asked for the website to be taken down. Varma confirmed to the press: “We received a notice from Jagex asking us to take down the website,” and “following that notice, we are discontinuing Elderwood's operation.” The site survives only in a limited capacity, preserving players' high scores. Before that, on the Old School RuneScape forum, players had flagged the similarities and called on Jagex to act. The community spoke first; the law followed.

I read the report in Express Tribune. What made me stop was the timing between three stages: copying, spreading, being removed. A few weeks, three thousand accounts, one letter. Football has all the raw material for the same sequence, only in a different order.

RuneScape launched in 2026, developed by Jagex, and is one of the longest-running MMORPGs. Old School RuneScape, a rebuild in the 2026 style, arrived in 2026 and still carries a busy community. That community spotted Elderwood Realms before Jagex responded. Varma used licensed AI tools, wrote his own code and published openly. The one thing he lacked was rights.

Replace the word “game” with the word “data” and the story changes subject but keeps its structure. Football data sits in four layers: event data (every pass, every shot with coordinates), tracking data (player positions over time), transfer data (fees, clauses, agent commissions) and scouting data (reports, potential ratings). Each layer has an owner, a contract and a price. Each layer can be regenerated.

In V.League 1, most match data is collected by the league organiser and its partners, then redistributed. Clubs keep internal training data. International platforms buy exploitation rights. Those three tiers work as long as regeneration stays expensive. When a language model can read thousands of reports and hundreds of matches on video, then generate a plausible-looking dataset, the cost of regeneration collapses. The line between copying and synthesis disappears at the same moment.

After the takedown, Elderwood was reduced to a page holding high scores. But three thousand accounts in 24 hours says one thing: the user's “good enough” threshold sits far below the legal threshold. By the same logic, a free football dataset that is smooth and detailed enough will be used before anyone asks how it was collected. In any data market, the fastest-spreading product is always the cheapest copy.

I have followed football since 2026, starting on local radio stations, and learned early that data does not speak for itself. In 2026, still a journalism student, I built a World Cup prediction model from xG and xA across five European leagues over three consecutive seasons. The model gave Germany a 78% chance of reaching the semi-finals. Germany lost 0-2 to South Korea in their final Group F match and went out in the group stage. The model got 12 of 16 knockout qualifiers right and was wrong about the one team I trusted most. The variables I ignored — internal conflict, complacency, declining fitness — all sat outside the spreadsheet.

When the model is wrong, the data starts telling the truth.

In 2026, when the Bundesliga returned to empty stands, I gathered data from nine rounds. The home win rate fell from 44.2% in 2026-19 to 36.7%; average goals per match dropped from 3.1 to 2.8. The home advantage every model treated as a constant turned out to be a variable dependent on the crowd. Home is not sacred ground, only a variable frozen in place.

In 2026, ahead of the Euro quarter-final between Italy and Belgium, I combined PPDA with distance covered. Italy pressed with an average PPDA of 8.2, meaning opponents completed only 8.2 passes before an intervention; Belgium ran 17% less than in previous matches. Italy won 2-1. PPDA is the signature; distance covered is the confession.

In 2026, I tracked Enzo Fernández's move from Benfica to Chelsea at 121 million euros. My valuation report leaned on World Cup data: 82% pass accuracy and 14 successful tackles. But the deal also depended on intermediaries, payment terms and the buyer's urgency. Data does not capture those things. Transfers do not pick the best player; they pick the one you mis-measure least.

Those four stories lead to a conclusion that applies to Elderwood Realms too: the most expensive part of a data system is not the ability to generate it, but the ability to prove it correct. Varma built a game in a few weeks. He could not build ownership of it. An AI model can generate thousands of rows about a V.League match in minutes. It cannot generate the provenance of a single row.

Verification is expensive because it demands three things a model does not have: a person present at the stadium, a recording process consistent across seasons, and lawful access to the source. A single V.League match can produce more than a thousand events. Checking each event takes people, time and contracts. Generating takes nothing. That is the core asymmetry of this decade.

In Vietnam, data contracts are usually signed at league level, not club level. A club knows how many metres its players ran, but does not hold distribution rights to that number. When data is regenerated from outside, the first loser is the club that invested in producing it. The second loser is the league itself, because a league's asset value rests on the exclusivity of its source.

There is a paradox here that football usually reads backwards. The common fear is that AI will replace the analyst, the match reporter, the scout who builds the spreadsheet. Elderwood Realms shows the real barrier sits at the verification layer and the rights layer. The product was blocked on legal grounds, not technical ones. It ran well. It simply had no right to exist.

In football, the verification layer is the most fragile part. An AI-assembled dataset can pass through several media layers before anyone checks it. Nguyễn Quang Hải moved to Pau FC in Ligue 2 in 2026; Đoàn Văn Hậu joined SC Heerenveen in 2026; each move generates hundreds of articles with numbers attached, and not every article traces its source. A wrong number about a player does not kill anyone. It erodes the very thing that gives data its value: trust in the source.

Data feels nothing, but it remembers everything the press forgets.

There is one more layer I always want stated clearly when writing about the digitisation of sport: live data supplied to betting companies is the darkest side effect of the whole process. A data system unverifiable at its root does not stop at a statistics page. It flows straight into odds, into derivative products, into decisions audiences never see.

One detail deserves keeping. The Old School RuneScape community spotted the duplication and applied pressure before Jagex acted. In football, the equivalent enforcement layer is the fans. They are the first to notice an absurd table or an index that does not match the match they just watched. The problem is that Vietnamese fans have no tools to check, and no habit of demanding sources. A wrong table still gets shared thousands of times if it looks good enough.

The next cycle will not begin with a big lawsuit. It will begin with a small platform: fast, plausible-looking, free, and with nobody asking where the data came from. When “where did this number come from” becomes the first question instead of the last, Vietnamese football data will finally have ground to stand on. Until then, every model can be right on average and wrong at the one point that matters — and no letter will arrive to take it down.