The Elo score problem

What the Elo score actually is

The Elo rating system was invented by Hungarian-American physics professor Arpad Elo in 1960 to rank chess players. The system is elegant: when two players compete, the winner gains rating points and the loser loses them. The number of points transferred depends on the expected outcome - beating a higher-rated player earns more points than beating a lower-rated one.

The key properties of Elo are:

• Zero-sum. Every point one player gains, another player loses. There is a fixed pool of "desirability" being redistributed.
• Transitive. If A beats B and B beats C, the system assumes A would beat C. Ratings form a linear hierarchy.
• Single-dimensional. There is one number. You are either good at chess or you are not. The system does not account for different styles, strengths in different openings, or matchup-specific advantages.
• Competitive. The system assumes the core interaction is a contest with a winner and a loser.

For chess, these properties make sense. Chess is a zero-sum game. There is one winner. Skill is largely transitive. A single rating captures meaningful information about playing strength.

For human attraction, none of these properties apply.

How dating apps adopted it

In 2016, Fast Company revealed that Tinder used an internal Elo-style desirability rating for every user. The system worked like this: when you swiped right on someone, their "desirability" score was factored into yours. If a high-rated user swiped right on you, your score went up. If they swiped left, it went down. Your score determined who you were shown to and who was shown to you.

As WIRED reported, Tinder was essentially creating an internal "desirability" ranking and using it to pair people of similar "attractiveness tiers." High-rated users were shown other high-rated users. Low-rated users were shown other low-rated users. The system created invisible hierarchies that users could not see, challenge, or opt out of.

The chess analogy was explicit. A "win" (right swipe) from a "strong player" (high-rated user) was worth more than a win from a "weak player." Your dating life was being governed by a competitive ranking system designed for a board game.

Why it is the wrong model for attraction

The Elo system makes assumptions about attraction that are not supported by research on how relationships actually work.

Attraction is not zero-sum

In chess, if I beat you, you lose. In attraction, if I find someone attractive, that does not make you less attractive to someone else. Attraction is not a competition for a fixed resource - it is a subjective, individual assessment that varies enormously between people. The Elo framework assumes a universal hierarchy of desirability. Reality is much messier and much more interesting.

Attraction is not transitive

If Person A finds Person B attractive, and Person B finds Person C attractive, it does not follow that Person A will find Person C attractive. Taste is personal, idiosyncratic, and often contradictory. The person who loves your sense of humour might not overlap at all with the person who loves your music taste. Elo assumes transitivity. Attraction refuses it.

Attraction is not single-dimensional

A single number cannot capture what makes someone appealing to a specific other person. Compatibility operates across dozens of dimensions: shared interests, complementary personalities, physical preferences (which vary enormously between individuals), life goals, communication styles, values, humour, and countless other factors. Reducing all of this to one number is not just reductive - it is architecturally wrong.

Research on face perception in dating apps has found that even rapid swiping decisions are based on more than just physical attractiveness - users also make judgements about moral character, social desirability, and perceived personality. A single desirability score cannot capture this multidimensional assessment.

The matching problem is not competitive

Chess has winners and losers. Dating (ideally) has two people who both benefit from finding each other. The fundamental interaction is cooperative, not competitive. Using a competitive ranking system to facilitate cooperative matching is a category error - like using a war simulation to plan a dinner party.

The psychological damage

The Elo system does not just produce bad matches. It produces measurable psychological harm.

A 2016 study presented at the American Psychological Association found that Tinder users reported lower self-esteem and more negative body image than non-users. While the study could not prove causation, the researchers noted that the swipe-based evaluation mechanism - being reduced to a binary yes/no judgement by strangers - was a plausible contributing factor.

The Elo system compounds this. It is not just that other users are evaluating you - it is that a hidden algorithmic system is aggregating those evaluations into a single "desirability" number and using it to determine your entire experience on the platform. Users who receive fewer right swipes are progressively shown less desirable profiles, creating a negative feedback loop that reinforces the system's initial assessment.

As ethicist Neil McArthur has argued, the apps "compound existing inequalities in the dating market" by ensuring that "the most popular users, measured by the total number of swipes they receive, are shown to the greatest number of potential partners." This creates a winner-takes-all dynamic where a small number of highly-rated users receive a disproportionate share of attention, while the majority are left competing for scraps of visibility.

"It's hard not to perceive the rating as a definitive scoring of our attractiveness, a supercharged Hot or Not-style algorithm culled from thousands and thousands of signals. Should Tinder make your score available to you? And if the company did, would you even want to know it?"

- Austin Carr, Fast Company, after discovering his own Tinder Elo score

"We do not use Elo anymore" (but do they?)

In 2019, Tinder announced that it had deprecated the Elo score. Their official blog post stated: "Elo is old news at Tinder. It's an outdated measure and our cutting-edge technology no longer relies on it."

But the replacement was described in conspicuously vague terms: "a dynamic system that continuously factors in how you're engaging with others on Tinder through Likes, Nopes, and what's on users' profiles." This sounds suspiciously like... a more sophisticated version of the same thing. A system that still tracks who likes whom, still factors in the "quality" of likes received, and still uses this data to determine visibility and matching.

As the Feminist Science newsletter noted, the fundamental problem was never the specific Elo algorithm - it was the underlying assumption that users can be ranked on a single dimension of desirability. Whether that ranking uses Elo, TrueSkill, or a neural network, the conceptual framework remains the same: some people are more desirable than others, and the system's job is to figure out who.

What matching should measure instead

The Elo model asks: "How desirable is this person?" This is the wrong question. The right question is: "How compatible are these two specific people?"

Desirability is universal, hierarchical, and competitive. Compatibility is specific, multidimensional, and mutual. They are fundamentally different concepts, and building a matching system around the wrong one produces fundamentally different outcomes.

From universal ranking to specific compatibility

Affinity Atlas does not rank users. There is no single "desirability" score. There is no hierarchy. There is no system where one user's gain is another user's loss.

Instead, every match is calculated as a pairwise comparison between two specific users, across multiple dimensions of shared interest. Your Affinity Score with Person A might be 82%. Your score with Person B might be 34%. These scores are independent - they are not competing for position in a league table. They are measuring the specific overlap between two specific people.

From attractiveness to shared depth

The Elo model privileges surface-level attractiveness (because that is what drives swipe behaviour). Affinity Atlas privileges shared depth - the niche interests, obscure passions, and genuine behavioural patterns that indicate real compatibility.

Two people who both love Lingua Ignota, Hollow Knight, and barrel-aged imperial stouts share something genuinely rare and meaningful. That overlap has nothing to do with how conventionally attractive either person is - and everything to do with whether they would actually enjoy spending time together.

From hidden scores to visible reasoning

The Elo score was secret. Users could not see it, understand it, or challenge it. Affinity Atlas's matching is transparent - every match comes with an explanation showing which interests drove the score, how much each category contributed, and what the confidence level is. If the system suggests someone, you can see exactly why.

This is not just a UX improvement. It is a fundamentally different relationship between the user and the system. The Elo model treats users as subjects to be ranked. The Affinity model treats users as participants who deserve to understand the system they are using.

From competition to discovery

The Elo system gamifies dating by turning it into a competition. Your score goes up or down based on outcomes. You "win" by being swiped right on by high-value users. This framing encourages optimisation behaviour - carefully curated photos, strategic swiping, and the entire "Tinder strategy" industry that has grown up around gaming the system.

Affinity Atlas is designed around discovery, not competition. The system is looking for genuine overlaps in taste and behaviour. There is no way to "game" the system except by genuinely engaging with the platforms you have connected - which is just being yourself. The matching quality improves with authenticity, not with optimisation.