About This Project

The Big Kink Survey was created and administered by Aella, collecting responses from ~970,000 participants. The survey is still open — you can take it yourself. Topics span sexual interests and kinks, personality traits (OCEAN model), demographics, political orientation, relationship structures, and psychological characteristics.

The publicly released dataset powering this explorer is an anonymized subset of the original responses. To protect participant privacy, the published data is limited to respondents aged 18–32 from Western countries (US, Canada, and Europe), and has been processed with aggressive binning, demographic column removal, and noise injection.

These privacy measures attenuate correlations by roughly 15–30% compared to the original, depending on the variable. The dataset supports directional exploration and pattern discovery, not precise point estimates.

The columns span five broad categories: demographic variables (age, gender, orientation, politics, religion), OCEAN personality traits and sub-scales, fetish interest ratings across dozens of categories, derived composite scores, and other variables covering relationship style, mental health indicators, and behavioral patterns.

Not every respondent answered every question. Many columns have significant missingness: some questions were conditionally shown (gated), some were added partway through the survey (late-added), and some are missing for reasons we can only label unknown. This explorer surfaces missingness metadata throughout so you can judge reliability yourself.

This tool loads the full parquet dataset into DuckDB-WASM directly in your browser. Every query runs locally. You can explore the full dataset without setting up a local analysis environment.

• Self-selected sample. Respondents opted in — this is not a probability sample of any population. Patterns describe this group of respondents, not humanity at large.
• Attenuated correlations. The anonymization process (noise injection, binning) weakens real relationships. If you see a moderate association, the true association in the original data was likely stronger.
• Missingness is informative. A column with 60% null values tells you something. The pattern of who answered may itself be meaningful. Check the missingness badges throughout the explorer.

This explorer exposes both a REST API and an MCP server so AI clients can inspect schema metadata, retrieve summaries, and run bounded read-only queries against the same dataset that powers the UI.

{
  "mcpServers": {
    "bks": {
      "type": "streamable-http",
      "url": "https://bks-mcp-server-production.up.railway.app/mcp"
    }
  }
}

Machine-readable endpoint for agent discovery: /llms.txt