The Deal-Breaker Audit: Why Your List of Non-Negotiables Is Probably Wrong

Everyone carries a list. It lives somewhere behind the eyes and it comes out at brunch, in bullet points: no smokers, must want kids, taller than me, no one who says "per my last email" unironically, nothing under a graduate degree, definitely not someone who votes like that.

The list feels like self-knowledge. It feels like the hard-won residue of every relationship that went wrong. And some of it is. But a surprising portion of the average deal-breaker list is doing something else entirely: it's shrinking the pool without improving the outcome. It's a filter dressed as a principle.

This piece is a way to tell the two apart.

The uncomfortable evidence

Start with the most-replicated finding in modern attraction research, which is also the least flattering: what people say they want in a partner does a poor job of predicting who they actually want once a real human is in the room.

Speed-dating studies let researchers test this directly. Participants state their ideal preferences in advance, then meet a dozen people for four minutes each, then report who they liked. In one of the foundational studies in this literature, stated preferences were essentially uncorrelated with the preferences participants revealed in person — suggesting that what we report about hypothetical partners may not predict whom we desire after actually meeting them.[1] The researchers put it bluntly: people appear to lack accurate introspective access to the preferences they'll display when confronted with an actual candidate.[1] The gap between stated and revealed preferences has since become its own research programme, with the discrepancy replicated across attractiveness, ingroup preference, and beyond.

Then it gets worse. In 2017, Samantha Joel, Paul Eastwick and Eli Finkel threw machine learning at the problem. Unattached participants completed more than 100 self-report measures — traits, preferences, everything prior research had flagged as relevant to mate selection — and then went on four-minute dates with everyone in the room. Random-forest models did okay at predicting general tendencies: 4% to 18% of the variance in how romantically desirous a given person tended to be, and 7% to 27% of how desired they tended to be. But the thing everyone actually cares about — whether these two specific people would uniquely click — was unpredictable from any combination of pre-date traits and preferences. Eastwick's comment on the result was appropriately humble: "It may be that we never figure it out."[2]

Read that carefully, because it's easy to over-read. The finding isn't that compatibility is random. It's that spark is not forecastable from a questionnaire — which is a very different claim from "nothing about a person predicts anything." Traits predict who tends to be liked. They don't predict the chemistry between a specific pair. Any honest matching system, ours included, has to hold both facts at once.

Why your predictions about yourself are systematically off

There's a second body of research that explains why the list misfires, and it has nothing to do with dating. It's called affective forecasting: our ability to predict how future events will make us feel. The evidence here is consistent — people tend to overestimate the emotional impact of future events.[3]

The mechanism has two familiar parts. Focalism: when you imagine the smoker, you imagine the cigarette, in close-up, forever. You don't imagine the Tuesday evening where he's making dinner and being funny and the cigarette is a fifteen-minute event on a balcony you'll barely register. Adaptation neglect: we underweight how fast the nervous system normalises things. The two-inch height difference that feels catastrophic in a photograph becomes invisible by date four, because you've stopped looking at it and started looking at a face.

Adaptation cuts both ways, and this is the part people skip. It means the cosmetic annoyance you swore you couldn't live with fades. It also means the charming thing that made you swoon fades — which is exactly why "he's so ambitious" has a habit of becoming "he's never home." A good deal-breaker audit corrects in both directions: it demotes the things you'll adapt to, and it promotes the things you'll adapt into misery.

What actually breaks deals

Researchers have studied the negative side of mate preference directly. Across six studies with more than 6,500 participants, Jonason and colleagues found that deal-breakers cluster around undesirable personality traits, unhealthy lifestyles, and divergent mating strategies — and that these operate differently depending on whether you're evaluating someone as a friend, a hookup, or a long-term partner. Later work sorting the items produced categories including unattractiveness, unhealthy lifestyle, undesirable personality traits, differing religious beliefs, limited social status, divergent mating psychologies, and differing relationship goals.

Notice what's doing the heavy lifting in those taxonomies. Personality, lifestyle, relationship goals — these are structural. Unattractiveness and social status are also on the list, and they're precisely the dimensions where stated preferences have the worst track record of predicting real-world behaviour.

So here's the working taxonomy. Every item on your list belongs in one of three buckets.

1. Structural incompatibilities. Things where two reasonable people simply cannot both get what they need from the same relationship. Children: want them, don't want them, already have them. Geography and mobility. Monogamy versus non-monogamy, and the specific definition of fidelity each person is operating under. Religious practice — not identity, practice: whether Saturday is negotiable, how children are raised. Money scripts: a saver married to a spender is a structural problem, not a stylistic one. Substance use at the level where it organises a life. These are real. They don't soften with exposure. Encode them.

2. Behavioural patterns. Contempt in conflict. Chronic dishonesty. A partner who withdraws for three days after a disagreement. Gottman's research on what predicts relationship dissolution points at how couples fight, not what they fight about — and these patterns are the strongest candidates for genuine non-negotiables, with one catch: you can't observe them in a profile. Nobody writes "I stonewall." We've written about testing conflict style before it costs you three months, because this bucket is where the predictive value hides.

3. Cosmetic filters. Height, hair, job title, alma mater, the specific flavour of taste in music, whether they text with punctuation. Some of these correlate loosely with things you care about. Most are proxies you've never audited. Every one of them costs you candidates.

The audit

Twenty minutes, a piece of paper, and a bit of ruthlessness.

Step 1 — Dump the list. Write every non-negotiable you can name, including the embarrassing ones. Especially the embarrassing ones. Aim for at least fifteen items; the real ones surface after the polite ones are out of the way.

Step 2 — Run the ex-test on each item. For every entry, ask: has this actually ended a relationship of mine, or ended someone else's that I watched closely? Not "could it" — did it. Most cosmetic filters have zero case history behind them. Most structural ones have a body count. If an item has never caused you a single hour of real distress, it isn't a deal-breaker; it's a preference wearing a costume.

Step 3 — Run the Tuesday test. Don't imagine the trait in the abstract. Imagine an ordinary Tuesday, eleven months in. Where exactly does this trait show up in the day? How many minutes does it occupy? If you can't locate it in a concrete hour of an ordinary week, focalism is doing the talking. Height doesn't show up on Tuesday. "Never initiates repair after a fight" shows up on every Tuesday for the rest of your life.

Step 4 — Name the real want underneath. Every cosmetic filter is a proxy for something legitimate. "Must have a master's degree" is usually reaching for intellectual curiosity. "Must be over six feet" is often reaching for feeling physically at ease, or for a status signal you might want to examine in private. "Must love hiking" usually means I don't want to be the only one who plans things. Rewrite the proxy as the underlying want. The rewritten version is almost always broader, more honest, and more findable.

Step 5 — Cap it. Keep five hard non-negotiables. Five. Everything else moves to a "strong preference" column, where it can influence a decision without vetoing one. The discipline of the cap is the entire exercise: it forces you to rank, and ranking is where you learn what you actually believe.

What a matching system should encode — and what it shouldn't

This is the part most dating products get wrong, and it follows directly from the research above.

Filters are for structural items only. Kids, geography, monogamy norms, religious practice, timeline. These are the things where a mismatch is a genuine dead end, and where filtering saves both people time rather than wasting an opportunity. A filter is a blunt instrument; use it only where bluntness is correct.

Everything else should be a weight, not a gate. Preferences should tilt the ordering of who you meet, not amputate the population. The stated-versus-revealed literature is a standing warning that your self-reported preferences are a noisy signal about your future behaviour — so a system that treats them as absolute law is amplifying your own measurement error.[1]

Behavioural patterns should be probed, not asked about. You cannot self-report your conflict style accurately, partly because you don't see it and partly because you'd rather look good. This is where an avatar earns its keep: a conversation between two digital twins can surface how someone handles disagreement, repair, and pressure, without either human having to spend three weeks finding out. That's the core of how matching works here — and if you'd like to see what your own twin says when you're not in the room, you can create your avatar.

And the spark stays with the humans. The 2017 machine-learning result is a boundary condition on the entire category: relationship-specific chemistry was not predictable from pre-date self-reports, using any combination of the variables tested. The honest job of a matching system is not to manufacture chemistry. It's to make sure that when chemistry does happen, it isn't happening with someone who wants a different life than you do.

Your list isn't wrong because you're shallow. It's wrong because it was assembled by a version of you who was imagining, not living — and imagination systematically overweights the vivid and underweights the daily.[3] Cut it to five. Make those five structural. Let the rest be surprises.

Sources

1. Speed-Dating (Current Directions in Psychological Science) — Eli J. Finkel & Paul W. Eastwick, Northwestern University
2. Dating? A magic formula to predict attraction is more elusive than ever — ScienceDaily
3. A pilot study investigating affective forecasting biases with a novel virtual reality-based paradigm — Scientific Reports (PMC)