About Sealed

Sealed lets you commit to claims with a SHA-256 digest, then reveal the original text on a schedule.

Making claims is free. Servers and inference are not. If Sealed is useful to you, you can support it here. Every contribution goes directly to keeping it running.

How it works

You write a prediction or claim. The server immediately generates a SHA-256 hash of your text — a one-way cryptographic fingerprint — and publishes it. At the same time, the original text is encrypted with AES-256-GCM and stored. The decryption key is held in a separate key management service. Once posted, your claim is scheduled for the reveal date and time.

To prevent anyone from guessing your claim, a random nonce is mixed in:

hash = SHA-256("sealed:{nonce}:{your claim text}")

On your chosen reveal date, the system decrypts the text, verifies it, and publishes it. If even a single character doesn't match, the claim is rejected. Once revealed, anyone can independently re-hash the text and nonce to check they match the digest stored before reveal.

Verify any claim yourself:

printf 'sealed:<nonce>:<claim_text>' | shasum -a 256

After reveal, three independent AI models search the web and evaluate the claim against real-world evidence. Each returns a verdict — supported, contradicted, partially supported, inconclusive, or not yet verifiable — and they're combined into a consensus score. Every judgment includes its sources so you can check for yourself.

Why this exists

Prediction markets let you bet on outcomes. Social media lets you say “I called it” after the fact. Neither gives you a simple way to commit to a statement before the outcome without money or hindsight bias. Sealed gives you a way to put a stake in the ground, and have it checked later.

Who built this

I'm George, a developer in the UK. I built Sealed because I'm interested in the time value of information, and how to generate authentic signals for high quality forecasters. If you find it interesting or have thoughts, I'd love to hear from you.

Trust and transparency

The digest is stored before reveal and is independently verifiable after reveal. This proves content consistency against that stored digest; it does not prove identity, truthfulness, or intent. Before reveal, Sealed keeps the claim text encrypted. When the reveal time arrives, the app decrypts it, recomputes the SHA-256 digest, and publishes the text only if it matches.