Start with the assumption almost every dating technology makes without saying it out loud: that chemistry can be predicted before two people are ever in the same room. A profile goes in. An algorithm scores it against other profiles. A match comes out, before either person has said a word to the other, laughed at a bad joke, or noticed the way someone's whole face changes when they talk about something they actually love.

We think that is working from the wrong starting point. And after 19 years in business, with 26,000+ of our verified events run in just the last 10 years alone, we finally have enough data, and enough machine learning built on top of it, to explain exactly why.

Here is the thing nobody in dating tech likes to say plainly: a profile is not a person. It is a person's highlight reel, edited for an audience of strangers who will judge it in under two seconds. The best photo. The bio line workshopped until it sounds effortless. The five interests picked because they sound interesting rather than because they are the ones that actually come up on a Tuesday night.

Train an algorithm on that, and you do not get a system that understands attraction. You get a system that is extremely good at predicting who performs well on paper together. Two people who would talk for hours in person can score badly because neither one's bio used the right words. Two people who would never make it past a swipe can look perfect on paper, because a good photo is not the same thing as a good conversation.

This is the entire premise the Smart-Card was built to reject.

Registration for a MyCheekyDate event asks for one thing beyond the basics: your name and email address. No profile to optimize. No photo to agonize over. No list of interests curated for a stranger's approval.

The bio comes later. At the event itself, before the conversations begin, attendees enter a short bio directly into the Smart-Card. A few lines, written in the room, on the night, without the hours of refinement that a dating profile at home on a laptop tends to accumulate. No time to workshop it. No opportunity to run it past a friend for feedback.

The front end is deliberately simple. After each four-minute conversation, you privately rate the person you just spoke with across five tiers. A spectrum of genuine interest that captures not just whether you would like to see someone again, but how strongly you felt that. The selection window stays open until midnight, so there is no pressure to decide on the spot, in the room, with the other person still nearby.

That five-tier rating is the first of four distinct data signals the machine learning collects from every event. What is happening underneath is where the intelligence lives.

This is the part that separates the Smart-Card from anything else in the dating technology space. Not just what it collects, but how many simultaneous signals it cross-references to build a genuinely accurate picture of real-world attraction.

Your five-tier ratings for each person you met reveal something no stated preference ever could: who you were actually drawn to, in person, after a real conversation. Not who you thought you would like based on a photo. Not who fit your stated criteria. Who actually held your attention across a table for four minutes and made you want more time.

This is the signal most dating technology ignores entirely, and it may be the most revealing one. If someone chose you and you did not choose them back, that selection still tells the machine learning something important: what was it about you, your bio, your conversation style, your presence, that attracted that particular person? Every one-sided selection is a data point about what you project, not just what you prefer.

When two people both selected each other, the system examines why. What did their bios share? What attributes connected them? Repeated across 26,000+ events, the system has built a genuine understanding of what predicts mutual attraction in real-world settings. Not compatibility scores assigned before anyone has spoken, but patterns extracted from conversations that already happened.

Perhaps the most powerful signal of all. At the event, you wrote a few lines about yourself and implicitly signaled what you were looking for. After the event, your selections showed what you actually responded to. The machine learning holds both of those things at once and looks at the gap between them.

All four signals depend on one thing: honesty. When selections are visible, even partially, people stop being honest. They start managing how a "no" will land, softening it, hedging it. A dataset built on strategic, self-conscious answers teaches a machine learning system to model strategy. Not attraction.

Private selections remove that filter entirely. Nobody sees your ratings. Not the host. Not the staff. Not MyCheekyDate internally. The only output that ever surfaces to another person is a mutual introduction, when both people independently and privately chose each other. One-sided interest produces nothing visible. No notification. No nudge. Just silence, unless it is mutual.

The revealed preference gap is consistent and significant. Across the full dataset, there is a reliable difference between the attributes people describe at the event and the attributes that actually predict who they select after a real conversation. This gap is not random. It follows recognizable patterns the machine learning has become increasingly accurate at identifying.

Behavioral signals outperform demographic ones. How strongly someone rates another person, whether they return for a second event, how their selections shift across an evening: these signals predict genuine long-term interest more reliably than the static attributes a profile-first system would weight most heavily.

Accuracy compounds over time. Attendees who return for a second event see a 77% improvement in match rate over their first. With 26,000+ events in the dataset, that compounding has produced significant accuracy gains across the full system. Nationally, the four signals combine to produce an 86% mutual match rate, averaging 2.3 mutual matches per event. Numbers that reflect conversations that already happened, not compatibility scores for conversations that might.

Honest caveat, the way we treat every number we publish: this is observational data drawn from real event outcomes, not a controlled experiment. Strong compass, not a script.

Most dating companies offer one product. A speed dating night. A swipe feed. A single format, repeated. The Smart-Card is different. It is the connective intelligence underneath an entire ecosystem, and the speed dating event is simply the front door. Every product beyond that door gets smarter because of what the machine learning has already observed in the room.

Rather than filtering future introductions based on what you told us at registration, we filter them based on what your behavior across a real evening actually revealed. Who you chose. Who chose you. How strongly. Where the patterns overlap. What the gap between your bio and your actual selections tells us about who you are genuinely drawn to.

A swipe dataset, however large, is built from static images and short bios. A few seconds of judgment, repeated millions of times. Wide, but shallow. It has never watched two people's body language shift mid-conversation. Never seen a room's match rate move because the energy changed after 9pm. Never captured the gap between someone who says they want a good listener and what their actual selections, event after event, say they are genuinely drawn to.

26,000+ verified events across 65+ cities is a different kind of dataset. Not wider, but deeper. It is 10 years of real interactions, each one producing a signal that only exists because the interaction actually happened. That is not something an app can shortcut its way into. It has to be lived, one real conversation at a time.

Every dating app you have ever used has, at some point, asked you to describe yourself in a way that sounds appealing to a stranger. The Smart-Card asks you to do the same thing, but in the room, on the night, before you have met anyone, with no time to overthink it. And then it watches what happens when the conversations start.

That gap, between the bio you wrote at the event and who you actually chose by midnight, is where the real learning lives.

Curious what the Smart-Card actually looks like in your hand at an event? Here is the full breakdown. Want to understand exactly how your data stays private? Read the companion piece: Your Selections Are Private. Here Is Exactly What That Means.