Perpetual futures on Solana
Know the price. Not the size.
A size-blind CLOB for perpetuals on Solana. Your exact size hides behind a SHA-256 commitment until settlement, while side, price, and size band stay public. The privacy is not free: it costs about 0.2 bps in a wider quote, a number we measured instead of assuming.
What sealing size buys and costs
Three properties we actually ship, with the evidence attached.
Slicing an order advertises it
Reinforcement-learning market makers detect and adversely select sliced meta-orders (arXiv 2510.27334). Splitting a whale order does not hide it, it advertises it. Here an order carries a band, not a quantity, so the size you are working never reaches the tape until the fill settles.
Dark Commit, at ER latency
commitment = SHA-256(salt || size_lots || leverage_bps || trader), published at placement and verified on-chain at settlement. The commit-reveal shape is the one the literature proves stops severe front-running (arXiv 2301.13785), and the two-phase delay costs nothing at Ephemeral Rollup speed.
The cost is measured, not assumed
56 simulated configurations in research/band-signal price what size blindness does to a quote. On the live 5-band schedule the maker's break-even half-spread is 1.991 bps against 1.790 bps on a transparent CLOB, and the uninformed taker pays 1.771 bps against 1.532 bps. Hiding size costs the taker about 0.2 bps, and we publish that because we ran it.
What is not claimed: flow privacy, or a better deal for the maker. The fill tape after settlement leaks magnitude, and under size blindness the maker quotes about 11% wider, so the taker carries the cost. For a risk-averse maker the gap widens to 1.351 bps and BLOCK-band volume falls from 14.7% to 2.8% of flow. DarkBook hides pre-trade size, nothing more.
docs/PRIVACY_MODEL.md, research/band-signal
How DarkBook works
From a committed order to atomic settlement. Exact size exists on-chain only as a SHA-256 digest until claim_fill.
Exact size committed. Band public.
Book shows side, price, and size band. Exact lot size stays behind SHA-256 until settlement.
DarkBook vs the field
Graded honestly. Partial marks are partial: residual trust in the ER validator and settler is documented, not hidden.
| Protocol | Privacy | Latency | Decentralized | Max Leverage | Settlement | MEV Resistant |
|---|---|---|---|---|---|---|
DarkBook This protocol | <50ms | 20x | Solana devnet | |||
| Hyperliquid | ~100ms | 50x | HyperEVM | |||
| Drift Protocol | 400ms+ | 20x | Solana | |||
| Dexlab | 500ms+ | 10x | Solana |
* Latency measured as order submission to match confirmation. Decentralization based on validator set / admin key control.
The stack
Every layer was picked for execution latency, not for a logo wall.
Read the mechanism, not the claims
Every claim on this page is explained in full, including what it costs you and what is not built yet.
The full order lifecycle: what you publish at placement, what the matching engine sees, and what the contract verifies before a position opens.
Field by field, what is public and what is hidden, when each becomes public, and the residual trust in the ER validator and the settler.
Size blindness is priced, not free. 1.991 bps of break-even half-spread against 1.790 on a transparent book, measured across 56 configurations.
Devnet only, no mainnet deployment, no audit, no external desk quoting yet. What is actually running, and what is not.