The five rules
| Rule | What it does | Works with | Example |
|---|---|---|---|
| Ask only | Collects an answer without using it to score fit. | Any type | The second half of a needs and offerings pair, or a fun warm-up. |
| Shared answers | Prefers people who gave the same answer, or overlapping answers. | Single and multiple choice | “Which first-date activities would you enjoy?” |
| Similar answers | Prefers people who are close on a scale. | Scale of 1 to 5 | Quiet evening (1) to lively evening (5). |
| Needs and offerings | Compares what one person needs with what the other can offer, in both directions. | Two choice questions with identical choices | “What would you like help with?” paired with “What could you help someone with?” |
| Conversation context | Lets the ranking use the answer as soft background. It’s never shown to anyone. | Short text | “A conversation I would love to have…” |
Importance
Questions that score fit can be Normal or High. High means an answer counts for more in the ranking. It’s still a preference. If you want one thing to dominate, set only that question to High.
What always applies
These are checked by code before any ranking happens. They hold for every event, and there’s no setting to loosen them:
- Everyone has confirmed they’re 18 or older and has finished the required questions.
- Both people chose the same context at this event.
- At dating events, each person is someone the other wants to meet. It works both ways or not at all.
- Dealbreakers a guest has set stay in force.
- Neither has blocked the other, and they haven’t already met or cleared each other.
- Both are available, and neither is resting after an earlier introduction.
You can’t currently write your own hard requirements. A rule like “only introduce investors to founders” isn’t possible. Needs and offerings will prefer that pairing, but it won’t forbid others.
How a match gets made
- Check. The rules above remove anyone who can’t be introduced.
- Shortlist. Your questions are scored by code, and each person’s best candidates go forward.
- Rank. An AI model (Cloudflare’s Clef) ranks those shortlists using answers alone, with no names or IDs.
- Choose. Flare picks the best pairs across the whole room, so nobody is offered to two people at once. Each pair has to reach the higher of the two guests’ match levels.
- Confirm. A final check can stop a pair that clashes with a dealbreaker a person wrote, and picks something they have in common.
If the AI model isn’t available, Flare carries on using your rules alone. Introductions keep happening, though the ranking is simpler.
Guests choose a match level
While their light searches, each guest can choose Okay, Good or Strong beside Light up for. Every event starts at Okay, and you can’t set it for them. A pair has to reach the higher of its two levels, so lowering yours never gives someone else a weaker match than they asked for. Nobody sees anyone else’s level. Good and Strong mean fewer introductions, and Strong may be rare.
What Spark, Glow and Flare mean in your results
Your results count introductions by strength: Spark is compatible, Glow is strong and Flare is rare. Guests never see a strength on a match. These are estimates of fit from answers. They don’t predict how a conversation will go, so don’t promote your event as guaranteeing either.