SoulCode

What SoulCode gets right is the framing. Instead of handing you a four-letter label, it hands you a vector: mbti = [0.72, 0.35, 0.81, 0.44], attachment = [0.18, 0.62, 0.74, 0.08], six values weights and a set of interest tags. That matters because vectors can be compared arithmetically, and SoulCode actually publishes how it compares them — cosine similarity for personality, attachment and values, a coverage-weighted semantic match for interests, then a single resonance score with fixed weights (0.25 / 0.25 / 0.30 / 0.20). Very few matching products tell you their weights. You also get three distinct routes rather than one: link sharing, where you send your code to someone and see the resonance before you see each other; a capped discovery feed of ten candidates a day, which deliberately avoids infinite scroll; and mutual-match chat that stays closed until both sides say yes. The AI explanation layer is the piece that makes the arithmetic legible to someone who does not want to read vectors. Where it is thin: attachment contributes four numbers and values contribute six, but the site does not document how those numbers are elicited, so the input side is a black box even though the output side is open. Interests are free-text tags, which means your match quality on the 20% interest weight depends on how you and a stranger happen to phrase things. The social graph is the real constraint — a resonance score is only as useful as the number of people who have generated a code, and link sharing shifts that burden onto you. Treat it as a structured self-reflection tool with a matching feature attached, not a substitute for a large dating pool. Use the free feed to test whether the resonance scores feel meaningful to you before paying for anything.