How matching works.

When you sign up as talent, you describe 2–3 things you have recently built or shipped. Each capability has a name, category, proficiency (1–5), and recency in months. Skills you used three years ago are weighted less than skills you used last quarter.

Companies don't post titles. They describe a problem they're trying to solve this quarter — required skills, budget range, timeline, urgency, and what success looks like. A problem is a target shape; we look for talent that fits that shape.

Each talent–problem pair gets a deterministic fit score:

• · skill_fit     40%
• · recency_fit   25%
• · adjacency_fit 20%
• · constraint_fit 15%

On top of the score, an AI reasoning pass generates a short, specific explanation of why this talent matches this problem — what overlaps, what doesn't, and what adjacent skills compensate.

Talents see only the problems they're unusually well matched for — the company stays anonymous. Companies see only the top 5–10 anonymized candidates per problem. When both sides click "Interested," identities unlock and a direct thread opens. Never before.

No browsable candidate directory. No public profiles. No social feed. No engagement metrics. No DMs before a match unlocks. Every refusal is intentional — they protect signal density and trust.