Privacy policy

Submind is a personal research project that mints publicly auditable predictions in response to questions visitors submit. Questions you submit, and the predictions Submind produces from them, are public. They are rendered at /p/<slug> URLs, listed on the homepage feed, and aggregated on the scores page. Don't submit anything you would not put in a public blog post under your own name.

Your IP address is used briefly for rate-limiting and is stored server-side in short windows (about an hour for the burst counter, up to a day for the daily quota counter). Beyond that, Submind keeps no account, no profile, and no cross-session identifier tied to you.

• The question text. Sanitized and truncated to 1000 characters, then sent through the prediction pipeline.
• Your IP address. Read from the x-forwarded-for header (Vercel proxy). Used as the identifier for two rate-limit layers (see “What is stored server-side” below).
• No cookies, no account, no email, no browser fingerprint. Submind does not require sign-in. There is no analytics package on the site that captures clicks, scroll depth, URL parameters, or referrer.

• The question, the sub-claims Submind decomposed it into, the evidence URLs the engine retrieved, and the resulting prediction. These are persisted to the project's Supabase database in the submind_predictions, submind_sub_claims, and submind_receipts tables, and rendered publicly at /p/<slug>. This is the core product, not a side effect. Submind is a public predictions ledger. There is no “private question” mode.
• Your IP address, as a rate-limit key.Stored in two places:
  • An Upstash Redis sliding-window counter (rl:predict:<ip>) with a 60-second TTL. The key auto-expires; it is not retained beyond the burst window.
  • A Supabase submind_rate_limits row for the hourly and daily quota buckets. The IP is the literal key column for anonymous callers. Stale rows are pruned opportunistically (one in every hundred requests sweeps the table). In practice, hourly rows are gone within a couple of hours; daily rows within a day or two.
• Receipts. Once a prediction is minted, its receipt is content-hashed and persisted indefinitely so the cryptographic verifier can confirm it later. The receipt contains the question, sub-claims, evidence metadata, and the synthesis output. There is no expiration on receipts. Removing a question means removing its receipt, which breaks any external link that previously cited it.
• Server logs. Vercel retains standard request logs for a short window (typically 7–30 days depending on plan). These include the IP, method, path, and timestamps. Submind's operator does not process these logs for any purpose beyond debugging failures.

The /historypage lists the questions you have asked from this browser. That list is kept in your browser's localStorage at the key verdict_history. It is never sent back to the server. Clearing your browser data, or clicking “Clear all” on /history, removes the list from your machine.

This is the part the prior copy got wrong. Clearing your local history removes the list of questions you asked from this device. It does not delete the corresponding submind_predictions, submind_sub_claims, or submind_receipts rows on the server, and it does not remove the public /p/<slug> page. If you need a question (and its prediction) deleted from the server, email the operator at the address in the footer with the link to the /p/<slug> page.

Submind uses a cascade of large-language-model providers to decompose, grade, and synthesize each prediction. When you submit a question, its text (and the sub-claims extracted from it, and the evidence snippets retrieved from the public web) are sent to one or more of the following providers' APIs:

• Cerebras (cloud inference on free-tier models)
• Cloudflare Workers AI (free-tier inference)
• Groq (free-tier inference)
• Google AI Studio — Gemini (free-tier)
• OpenRouter (free-tier routing)
• Hugging Face Inference (free-tier)
• Anthropic and OpenAI (only when free providers are cooling down or unavailable — this is a fallback, not the default)

Each provider has its own privacy policy. Submind does not control how these third parties retain or process the data they receive. If your question contains anything you would not want a cloud LLM provider to see, do not submit it.

• No advertising, no ad networks, no tracking pixels.
• No third-party analytics that capture page views or URL parameters.
• No reading or transmission of browser history, bookmarks, or saved passwords.
• No sale or sharing of question text with anyone outside the LLM-provider chain listed above.
• No facial-image collection. No biometric data.
• No accounts. Submind does not maintain user accounts. Anything you do is anonymous at the application layer.

To request deletion of a specific prediction (and its backing rows + receipt), email the operator with the URL of the /p/<slug> page. The operator commits to acting on legitimate deletion requests within 14 days. There is no automated self-serve deletion endpoint at the moment; this is a stated limitation, not a refusal.

For access or correction requests, the same channel applies. Submind does not keep an account record to attach a request to, so requests must identify the prediction by URL.

For visitors in the European Economic Area or the United Kingdom: Submind processes question text and IP addresses on the lawful basis of legitimate interest — specifically, operating the public predictions ledger that this site exists to be. Rate-limit IP storage exists to protect the service from abuse and is processed on the same basis.

You have the right to object to this processing under Article 21. To exercise that right, email the operator at the address in the footer; the practical result will be deletion of any matching prediction(s).

Submind does not sell or share personal information for cross-context behavioral advertising purposes. The only categories of information Submind collects are described above (question text, IP, server logs). You have the right to know, delete, and opt out of any sale (there is no sale). Contact the operator at the address in the footer to exercise these rights.

Submind is not directed at children under 13 and the operator does not knowingly collect information from them. If you are a parent or guardian and believe a child has submitted a question that should be removed, email the operator with the URL.

When this policy is materially changed, the “Last reviewed” date at the top will be updated, and the change will be visible in the git history of this file in the public repository. There is no separate mailing list for policy-change notifications because Submind does not maintain user accounts.