Technical Architecture

AIP-120. The Foundation Layer

AIP-120 is the infrastructure upgrade that made Decibel possible. Built by Sital Kedia while still at Aptos Labs.

What AIP-120 specifically changed:

• ›Replaced Econia's AVL queue with BigOrderedMap (B+ tree)
• ›AVL rebalance had unpredictable gas. O(log n) with variable cost depending on tree state. Transactions could cost wildly different amounts of gas, making HFT market-making economically unpredictable.
• ›B+ tree leaf split → O(1) with flat gas. Each write costs a predictable, constant amount regardless of tree state. This is what makes HFT maker economics viable on Aptos.
• ›57% reduction in bytecode size
• ›Added: bulk orders, TWAP, Dead Man's Switch, Pre-cancels

Everything in the Whop exchange stack runs on top of AIP-120.

Block-STM v2

Aptos's optimistic concurrent execution engine. Runs all transactions speculatively in parallel, then resolves conflicts after execution via a validation pass. For a CLOB, this means multiple makers updating the same market's orderbook in the same block can execute concurrently. Conflicts get caught and re-executed, not pre-prevented.

Block-STM v2 scales to 256 cores. 8x parallel execution capacity over the previous generation.

Why this matters vs Solana: Solana uses Sealevel (static parallelism). Transactions declare all accounts they'll touch upfront. Any two transactions that declare the same account as writable are forced to serialize. For a CLOB where every order writes to the same market account, every order serializes regardless of Sealevel. Dropset compensates for this. Brilliant engineering that minimizes the cost of unavoidable serialization. Aptos Block-STM removes the serialization constraint entirely.

AIP-125. Scheduled Transactions

Enables time/signal/price-triggered transactions. Used by the Indicator Marketplace for automated bot execution.

When an onchain indicator fires a signal, AIP-125 auto-executes the Decibel perps order without any human intervention. The creator sets the trigger conditions (crossover of SMA, RSI threshold, price level) and the keeper evaluates them on a 60-second cycle. When conditions are met, the scheduled transaction executes.

Gas deposit bounty: third-party keepers can be paid to execute scheduled transactions, keeping the system decentralized. The keeper service is already built in the Indicator Marketplace codebase.

AIP-143. Confidential Assets

Pending governance. Twisted ElGamal homomorphic encryption for token balances and transfer amounts. Sender and receiver wallet addresses remain publicly visible, only amounts are hidden.

The homomorphic property: you can add two Twisted ElGamal ciphertexts without decrypting them. This is what enables the balance update math. The new balance is computed as encrypted_balance + encrypted_transfer_amount without any party seeing the plaintext values.

ZK proofs (Bulletproofs or range proofs) verify that the encrypted balance remains non-negative after a transfer. Preventing someone from sending more than they have without revealing the balance amount.

TypeScript SDK live at aptos.dev. Two public test workers available.

RAPT BFT Consensus and Zaptos

rAPT BFT. Aptos's consensus mechanism, designed for high TPS and resilience under load. Zaptos is the parallel pipelined architecture built on rAPT BFT for sub-second finality and low latency.

Archon. Speed Layer

30ms inclusion confirmations. ~10ms block times. CEX-level responsiveness for traders, payment flows, and RWA platforms. Multi-Leader Primary-Proxy architecture for deterministic confirmation at the speed modern markets expect.

Event-driven Transactions. Native automation layer for real-time trading logic optimized for High Performance trading. This is the full description of what AIP-125 enables at the protocol level: not just "scheduled transactions" but a native event-driven automation layer where onchain events (indicator crossover, price threshold, Pyth update) trigger transactions automatically. This is the execution model for the entire Indicator Marketplace bot infrastructure.

Archon is about speed, not fair ordering. MEV protection comes from the encrypted mempool, not Archon.

Encrypted Mempool (proposed October 2025, Pending Governance)

Hides transaction contents before block inclusion. Nobody can see your order before it fills. Front-running via mempool observation becomes impossible. If approved, Aptos would be the first L1 to offer users the option to send transactions via a native encrypted mempool.

The encrypted mempool and Confidential Assets work at different phases:

• ›Encrypted mempool: invisible before execution (intent hidden)
• ›Confidential Assets: invisible after execution (amounts hidden)

Together: completely invisible before and after.

Move v2

The Move language upgrade that enables the UTT implementation. Key capabilities:

• ›Zero-copy frames: Reduces memory allocation during complex operations
• ›Inline functions: Eliminates function call overhead on hot paths
• ›Struct variants: Enables enum-like patterns for coin types
• ›Public(package) visibility: Finer-grained access control for module composition

The UTT Move v2 implementation uses:

• ›aptos_std::ristretto255::CompressedRistretto. for Pedersen commitments (coin envelopes)
• ›aptos_std::groth16_bn254. for ZK proof verification
• ›aptos_std::bls12381. for PS (Pointcheval-Sanders) mint signatures
• ›aptos_std::table::{Self, Table}. for the nullifier set (O(1) lookup, append-only)

UTT Deployment Requirements

One governance vote for:

• ›Groth16 verifying key (VK) for the UTT circuit. Set once at init, then immutable
• ›IBE master key parameters for the send-to-identity mechanism

No consensus-layer changes required. UTT deploys as a standard Move module using native functions already in the Move VM. This is Tier 3 (Move contract) not Tier 1 (consensus) or Tier 2 (Move VM native functions).

UTT global state serialization bottleneck: Block-STM must serialize on UTTState (frontier/root/next_index) because all transactions write to the same global resource. Estimated throughput ceiling: ~200-500 tx/s. Acceptable for a privacy layer at Whop's launch scale.

Decibel. Onchain Risk Engine

Fully onchain: auto-deleveraging logic, mark price computation, and market parameterization are always verifiable. No discretionary liquidations. No off-chain risk engine that can be manipulated.

Auto-Deleveraging (ADL): The mechanism for forced position reduction when the insurance fund is exhausted. Decibel's ADL logic has a known regression. The insurance fund calculation has an off-by-one error in specific edge cases. This doesn't affect normal operations but surfaces in adversarial conditions. OtterSec audit found 91 total findings. This regression was not in the original critical findings but was identified separately.

Pyth Oracle Integration

The Indicator Marketplace uses Pyth for:

• ›Real-time price feeds (5s polling via /price-tick route)
• ›2yr historical OHLCV data for Monte Carlo backtesting
• ›Onchain price pushes by the keeper every 60s for indicator evaluation

Pyth provides sub-second price updates across 500+ price feeds. The keeper service pushes Pyth prices onchain before evaluating scheduled indicator jobs. Ensuring all onchain indicator computations use current market prices.

Fixed-point Math Library

Onchain indicators require precise decimal arithmetic without floating-point. The Indicator Marketplace fixed-point library implements:

• ›exp. via Taylor series at 1e8 scale
• ›ln. via Taylor series at 1e8 scale
• ›sqrt. integer Newton's method
• ›tanh. via Taylor series at 1e8 scale

All at 1e8 (8 decimal places) fixed-point precision. This is sufficient for TA indicators. RSI, MACD, Bollinger Bands, etc.. where 0.000001% precision differences are irrelevant.

Econia. Historical Context

Alex Kahn's original CLOB on Aptos. One of only three crankless, atomic onchain CLOBs ever built (the other two being serum on Solana and one other). Dragonfly Capital invested. Kadin led the investment.

Econia's matching engine was atomic: the order is matched, the book is updated, and the assets are routed in one single transaction. No separate crank transaction. This was the core technical achievement, not the orderbook itself but the crankless atomic settlement.

Econia shut down. Decibel (built by Sital Kedia on Econia's architectural foundation) replaced it. Alex went to Solana and built Dropset.

Econia was crankless. The matching engine was fully atomic, settling in a single transaction with no external crank operation.

The 22-hour Timeline

• ›March 31, 2026: Alex Kahn reveals Dropset. The first formally verified onchain CLOB with assembly-level optimizations and HFT maker economics on Solana
• ›April 1, 2026: Sital Kedia ships Bulk Orders on Decibel. Atomic cancel+replace, the identical primitive, on Aptos. Less than 22 hours later

No direct mention of Dropset in Sital's announcement. 95% conceptual overlap. Decibel is reacting to Alex. The two teams independently arrived at the same solution on different chains, but Alex got there first and Sital's response confirms the insight.

What this proves: Alex is the technical leader in this design space. Decibel validated the insight by reacting within 22 hours. The fastest possible acknowledgment that Alex had identified the right problem and the right solution.

Dropset Technical Details

The drop-and-set primitive:

• ›One atomic transaction cancels the full order ladder
• ›Replaces it with a new one
• ›Silent return if the drop fails. Race condition protection built in
• ›No cancel transaction, no waiting, no race between cancel and replace

Engineering tricks that make HFT viable on Solana despite serialization:

• ›SBPF assembly with RB-tree data structures (directly against Solana's runtime)
• ›Single canonical PDA with one ATA vault
• ›Absolute pointer addressing with pinned empty signer
• ›Chunked 64-bit pubkey compares
• ›CLRS spec compiled to Lean formal proofs. First formally verified CLOB

All 100% open source. Pre-mainnet beta. Multi-assembly refactor in progress.

Dropset vs Decibel Bulk Orders:

• ›Same core insight: atomic batch ladder replace reduces serialization writes and eliminates race conditions
• ›Different runtime: Dropset fights Solana's hostile serialization constraints via assembly, while Decibel runs high-level Move 2 on Aptos's more accommodating runtime
• ›Dropset verdict: more impressive engineering. Fought a harder problem
• ›Decibel verdict: cleaner implementation. Expected result on a better runtime