🔗
CommitmentOracle
🧭 How to use this page start here ▶

This page proves you hold data without revealing it. Everything is verifiable by recomputation — no trust in us required.

1
Bind — paste any data, commit its SHA-256 on-chain. Only the 32-byte hash touches Stellar; your data stays local.
2
Prove — paste the same data + the hash from Step 1. The contract recomputes and matches. Change one character → fails.
3
Reveal — for big objects (10k-dim vectors), split into 32 Merkle chunks; commit only the root, prove one chunk at a time.
4
Verify / Compare — check a document against the chain head, or anchor a semantic distance between two documents.

💡 Every form has a Load example chip — click one, then hit the submit button. No data to invent.

SVEC · Layer-Commitment Oracle

Proof of reserve for AI, without spilling the magic.

Bind any data object — a 10k-dim VNA vector, a document, a model snapshot — to a SHA-256 commitment on Stellar testnet. Only H(v) (32 bytes) + provenance touch the chain.
Verification is recomputation, not trust.

Commitments
235
Contract
CCXPK442UI5SNCLDPKJ3J5H33OEPVXU7UCN3TZG6XFCVPYK5I2VMFICN
Chain head
a9d511f2a765f563cb8cc1b28f4f037b9b9c6f0ccffe8ed30e17cea6ab185050
Walked
1 links
🧪 Example data one click fills Step 1 — then just hit Commit
Scenario A — tamper proof: load Deal bundle → Commit (Step 1) → copy the hash → Verify (Step 2) with the same data ✓ → then change one digit and Verify again ✗.
Scenario B — private reveal: commit the Merkle root (Step 3, example vector pre-filled) → Prove leaf #7 → the 5-hash path ties one chunk to the root while 31 stay hidden.
Scenario C — semantic audit: in Step 3c paste Doc A + Doc B (same cohort) → high similarity; then Doc A + Doc C (fleet telemetry) → near zero. The distance is anchored on-chain.
1 Bind — commit H(data) on-chain
write → contract
2 Prove — verify by recomputation
read → contract
Tip: change one character in the data → verify fails. That's the no-trust proof.
3 Selective Reveal — prove one chunk, hide the rest
root on-chain · chunks local
Split the vector into 32 chunks → each becomes a Merkle leaf → commit ONLY the root (32 bytes). Then prove ONE chunk matches the root with a 5-hash proof path. The other 31 chunks never leave your machine.
Commit root
Prove one leaf
3b Neural Bundle — document → VNA vector → Merkle root
weights stay off-chain · H(v) on-chain
Paste a document (deal bundle, policy, VNA output). The Actuary.Neural module vectorizes it deterministically → 10k-dim sparse vector → 32 chunks → Merkle root. Only the root goes on-chain. Reveal one chunk as proof — the other 31 stay private.
Verify doc vs root
Reveal one leaf
3c Semantic Compare — two bundles, one distance, on-chain
public cosine · result anchored
Vectorize two documents, compute cosine similarity over their sparse features, and record the comparison on-chain through the registered comparator. Anyone can later prove what was compared and how close it was — without seeing either vector.
examples:
4 Verify Document — against the latest chain head
read → contract
📜 The Chain 1 walked
newest first
a9d511f2a765f563cb8cc1b28f4f037b9b9c6f0ccffe8ed30e17cea6ab185050
svec_vna cnn-splade-v2
ledger 4825591 · prev 26815494b777d2802b2dfaa83f06642fdb79cf98bb98144f720baaaed267291b
🕐 Session log
Nothing yet.
Layer-commitment oracle · Stellar testnet · contract CCXPK442UI5SNCLDPKJ3J5H33OEPVXU7UCN3TZG6XFCVPYK5I2VMFICN · verify by recomputation, no trust