How many people does it take to find a match?

01 / Start with the right question

Two people is possible. It is not a guarantee.

If you deliberately recruit one man and one woman who both want to meet, two attendees are enough for one introduction. If you draw randomly from a 50/50 pool, the expected wait is two additional people to find an opposite-gender candidate: three people including you. Two additional draws give a 75% chance; four additional draws give 93.75%.

There is no finite event size that guarantees a compatible stranger under random recruitment. We can instead choose a probability target. Let q be the chance that one candidate meets all requirements and accepts yours; let m be the number of relevant candidates:

In a balanced event, m men and m women means 2m total attendees, including the person seeking a match. Gender has already been handled by counting only the opposite side; do not multiply by 50% again. With unbalanced random arrivals, use all N−1 others and include the desired gender’s share in q.

One match anywhere in the room is a different target. In the homogeneous independent-edge model with q = 10%, five men and five women create 25 possible pairs: a 92.8% chance that at least one pair is eligible. A particular attendee in that same room has only five candidates and a 41.0% chance of an option. Their 90% target instead needs 22 candidates, or 44 total attendees. “At least one pair exists” is therefore much easier than “most people get a match.” Shared-trait bottlenecks also break the independent-edge assumption.

City size matters for whether a candidate pool exists and how much recruitment is needed. For a room of a fixed size drawn from a sufficiently large pool, compatibility rates matter much more than the difference between one million and ten million residents. Population residents are not automatically single, interested, available or willing to attend.

02 / The actual questionnaire

What does each question cost?

We audited all 13 current production profile questions. Gender and age are the baseline below. Every other row adds just one condition to that baseline; it does not include the rows above it. Height is an extra thought experiment, because it is not currently a profile question.

Answer-choice counts alone cannot determine a room size. If you accept several categories, add their observed shares; overlapping identities need the union, not a simple sum. Reciprocal acceptance must then be estimated among those candidates. A seven-option question does not imply a one-in-seven match rate.

Rates measured for ages 15+, 25+, 5+ or all ages are explicitly transferred to the 25–39 example as proxies. They are not measured age-specific joint distributions. ACS recorded sex is a demographic proxy, not a measure of gender identity or sexual orientation. This worked example covers men seeking women and women seeking men; other pairing pools need their own data.

03 / The AND effect

Preferences multiply. Event sizes grow.

For one person seeking a man, the illustrative pass rate is 29.4% for age, then 16% for height, 18.8% for personal income, and 35.9% for a degree. Multiplying these leaves 0.316% of broadly recruited adult men.

That produces 728 men + 728 women for a 90% individual opportunity estimate. At an assumed 70% reciprocal acceptance rate among candidates passing all these conditions, it becomes 2,080 total. Neither result means everyone in that room can find a partner.

Recruiting directly into the desired age band removes the age factor: the same height + income + degree example drops to 426 total at 100% reciprocal acceptance. Recruiting only people who meet every condition can make a much smaller room viable; the difficult part then moves into recruitment.

04 / Explore the assumptions

Design a room around a person.

Gender and an age range are always included (default: ages 25–39). Select extra requirements, then adjust recruitment and reciprocity. The calculator runs entirely in your browser and sends none of these selections.

05 / More people, or more people matched?

An available option is not an exclusive partner.

Imagine ten women all meet the preferences of one man, and all ten would accept him. Each woman has an option, but only one woman can be paired with him in that round. At most 2 of the 11 attendees can be paired at once.

To study this gap, we generated 1,000 synthetic bipartite graphs per row, with a fixed random seed. In this deliberately simple model, each man–woman edge independently exists with probability q. We calculated both the fraction with any eligible edge and the maximum conflict-free pairing. This is a mathematical benchmark, not a simulation of real LA profiles or the production scheduler.

These are averages across synthetic rooms. A 90% chance for each statistically identical attendee is also 90% expected opportunity coverage in that homogeneous model. It is not a 90% chance that everyone has an option, and it does not cover heterogeneous preferences or scarce shared traits.

The absolute gender-balance ceiling for one round is 2 × min(men, women) / total attendees. A 70/30 split cannot pair more than 60% at once, even with perfect compatibility. Multiple no-repeat rounds also need multiple distinct neighbors: one candidate cannot provide three different dates, and individual degree counts alone do not guarantee a workable multi-round schedule.

For q = 10%, an individual 90% chance of at least one candidate needs 44 attendees; at least three candidates needs 104. The calculator handles this binomial tail. Actual rounds still require the shared matcher and event roster.

06 / A practical event plan

Recruit for coverage, not just headcount.

1. Start with a clear age cohort and mutual gender preferences. Balance actual checked-in attendees. Estimate separate show-up rates by side; a balanced RSVP list can become an unbalanced room.
2. Keep genuine requirements distinct from preferences. Every extra strict AND can shrink the pool. Use preferences to prioritize introductions where appropriate.
3. Measure the roster before promising rounds. Use the existing admin potential-matching preview on consenting registrations. Count reciprocal eligible neighbors, simulate the configured no-repeat rounds, and inspect sit-outs and scarce subgroups.
4. Set an operational target. For example, 90% of checked-in attendees get at least three distinct eligible introductions. That is a chosen target to validate on the roster, not something Census data can guarantee.
5. Recruit the bottleneck. If many attendees compete for a few eligible partners, add people who expand that scarce pool or offer a more focused event. An indiscriminately larger room can preserve the same problem.

A useful next analysis would compare anonymized, aggregate roster simulations across cohort sizes and return a distribution of paired coverage by round, including the worst-served groups. This report uses public demographics and synthetic graphs only; no private attendee answers were analyzed.

07 / Sources and reproducibility

What supports these numbers?

• Census QuickFacts: LA County and LA city: July 2025 population context, accessed October 13, 2026.
• 2024 ACS 1-year official table-based summary files: B01001 (sex/age), B15002 (education, 25+), B19325 (personal income, 15+), B12001 (marital status, 15+), B03002 (race/Hispanic origin, all ages), C16001 (language/English ability, 5+). County GEO_ID 0500000US06037. Extracted estimates and 90% margins of error are retained in the research source files; scenario sensitivities are not ACS confidence intervals.
• CDC/NCHS NHANES, August 2021–August 2023, Table 8: measured US heights, age 20+. Interpolated, rounded proxies, not local estimates.
• Pew 2023–24 Religious Landscape Study: California: broad Christian identification, including Catholic. State proxy; overall state survey margin of error ±2.4 percentage points.
• Inner Product public question bank: 13 production profile questions, audited October 13, 2026. Event-specific questions and later edits can differ.

Reproduction: frozen county rows → demographic fractions → individual binomial thresholds → seeded synthetic graphs and maximum bipartite matching. Seed 20261013; 1,000 graphs per scenario. Assumptions about intentions, lifestyle, distance and reciprocal acceptance require event-specific validation. Neither a compatible profile nor a mathematical pairing predicts attraction, consent or an eventual relationship.

A bigger room creates more chances. A thoughtfully recruited room gives those chances somewhere to go.