Tokyo Was Closed. A Hyperliquid Perpetual Had Already Priced the Nikkei Gap.
A perpetual priced the 31 July move before Tokyo opened. The Nikkei index opening print did not, and the ETF and the shares did.
Posts / 91 records
Searchable archive. For the latest feed with pagination, start on the home page.
A chronological index of field notes on research, engineering, markets, and protocols.
A perpetual priced the 31 July move before Tokyo opened. The Nikkei index opening print did not, and the ETF and the shares did.
Evidence from Hyperliquid, GMX, and OKX
How staking separates capital from operation—and creates new routes to liquidity and credit.
Position size sets dollar PnL. Market structure sets the execution price. Loss-absorbing capital sets which losses can be paid.
Ethereum settles execution against gas used; Solana legacy/v0 calculates native priority fees from requested compute units.
What a fixed-block quote reveals about pool activity, price impact, fees, and slippage tolerance.
Across 23 large USD/JPY moves, U.S. yields, BTC/ETH, and a tokenized-Treasury allocation proxy moved—but not in a stable, repeatable sequence.
Solver competition and endogenous participation shape private execution gains and aggregate welfare across batch durations in a simplified intent-auction game.
TVL is a measure of capital. The marginal price is the current quote. Liquidity tells us how much trading it takes to move that quote.
A settlement workload calibrated to KRX's daily trading value fits within Ethereum's theoretical block-gas ceiling at the baseline. What changes under Glamsterdam is the cost of updating balances and creating new onchain state.
Testing Provider Changes Against Consumer-Declared Requirements
A collateral boundary still exists onchain. The difference is who may submit the liquidation once it is reached.
Korea is moving past deposit-token pilots. What changes when tokenization reaches the scale of a capital market.
When discovery stops being scarce, competition becomes an execution race—and the rule that allocates execution determines how much rent participants keep.
CoW batches trades and auctions execution to competing solvers. Its auction data shows what those competing plans look like.
Thirty externally selected Solana price shocks, one Orca pool: sometimes price and landed execution return to normal together. Sometimes they don't.
Two compiled Solana implementations accepted the same call. Both transactions succeeded; only one preserved the caller’s fixed postcondition.
A minimal benchmark that separates onchain participation from independent information: what actually moves the price when traders are added.
Four synthetic markets priced companies before their shares could trade.
A bounded event study separates observed trading activity from unproven capital migration after a SpaceX-linked token shock.
When the yen broke, what moved onchain, and in what order? Tracing risk transfer through Hyperliquid and DeFi markets.
On 8 July 2026, spot gold fell while WTI moved the other way. A Rust CLI measures how that commodity shock showed up in tokenized markets.
Five Solana wrappers on one company, one issuer-designated conversion route, and nine weeks of swap-level flow through it.
Execution provenance reconstructed for all 6,695 landed transactions in a frozen SPYx x OKX router set, and what that evidence can actually prove.
An empirical note on how multiple Uniswap v3 pools behave like one under an impact-constrained capacity objective.
A transaction-level audit shows why a quoted slippage threshold can disappear before a Solana swap lands.
A transaction-level study traces router, provider, and execution-source claims in Solana tokenized-equity trades.
Part II of the DFlow series: what a landed Solana transaction can and cannot prove about which router executed it.
As execution becomes an embedded wholesale service, synchronized JTX quotes show that app-level fees still shape the price offered before signing.
Why the execution benchmark had to be frozen, with data, configuration, and evaluation fixed, before any result could count as evidence.
A large DEX order pays for a path, not a single price. Framing execution as sequential control over pool state, fees, and liquidity response.
How frozen AMM logs became a matched difference-in-differences pipeline in R, behind the finding of no large short-run liquidity exit after UNIfication.
After the Uniswap fee switch turned on, did liquidity providers leave? A causal look at the short-run liquidity response.
After Uniswap v4 went live, the pool stopped being the whole market. Mapping where trading and liquidity actually sit now.
Hyperliquid is more than a price story. What an outsider can reconstruct from an onchain order book that keeps moving.
When does a dynamic AMM fee actually help? Simulations of which pool-state signals let a fee price toxic flow.
When no AMM pool forms, where does the market go? Hyperliquid shows what price formation looks like without one.
Before studying MEV, a Rust AMM lab that measures pool state directly, so later execution questions start from observable pool mechanics.
Canonical deployment tells you which contract is real, not where the market forms. The same token appeared twice; an AMM market formed once.
The hard question is not whether an AI agent can write to a chain, but who defines its authority and where that boundary is enforced.
A DEX receipt does not explain everything, but it gives outsiders a canonical place to start.
A note on turning a Uniswap quote into API-native calldata, then replaying that transaction against pinned mainnet state.
Exit risk shows up in the quote before any trade fails: route fragmentation, output deterioration, and shifting summary fields.
How separate RWA evidence collectors became one Rust CLI: three contracts, versioned evidence bundles, and reproducible audits.
When the venue becomes part of the product, which record is authoritative?
An interactive P2P attack gym in Rust: slot pressure, fake state, and poisoned discovery, made visible on 127.0.0.1.
When RWA trading becomes a clean service, which parts of the path stop being visible?
Sybil, Eclipse, and network-partition attacks, traced from real incidents and rebuilt as local simulations in a Rust lab.
When real-world assets move onchain, the token can move faster than the evidence. What becomes visible, and what an audit should check.
A Rust-based pipeline that turns stablecoin transfer checks into artifacts, manifests, and an inspectable evidence console.
Stablecoins are becoming crypto's cash layer. The question is not only what they are pegged to, but what they depend on.
SpaceX stock is not public yet, but SPCXUSDT perps already show volume, open interest, funding, and venue concentration.
Auditing onchain liquidity for the local-currency stablecoins XSGD and EURC, and how much of it still runs on dollar rails.
Changepoint detection on Ethereum base fees under EIP-1559: useful as early-warning signals for fee-risk stress tests, not as regime labels.
USDC behaves differently across chains, so stablecoin risk can't be read from a single signal.
Tokenization puts the claim on-chain, but custody, compliance, and settlement still depend on surrounding infrastructure.
Funding links are not identity. A conservative onchain entity-linking method, run on a 500+500 Arbitrum governance cohort.
A mini ZK-rollup verifier in Rust, built from the basics most explanations skip: Pedersen commitments, Sigma proofs, and Merkle state.
Not all state behaves the same. The assumptions and guardrails that bound ZK State Prune's rollup cost models.
Why ZK rollups need smarter state management: survival models estimate slot reuse to drive cost-aware hot/cold placement.
The QA pipeline behind a minimal token: testing the contract, off-chain state reconstruction, and the frontend end to end.
Testing Metering Chain against the guarantees defined in Phases 1-4, with QA cases that follow business logic rather than code paths.
Why digital finance still leans on centralized operators, what blockchains change about trust and state, and why Rust fits the systems underneath.
From accounting to settlement: how Metering Chain defines when a bill becomes economically final.
In DePIN, owners can't sign every event. Capability-driven delegation lets operators submit usage without opening a trust hole.
Deterministic authorization for multi-operator DePIN: deciding who may submit usage, so replayed history still yields the same bill.
Can a token's entire state be reconstructed from events alone, without reading storage? A minimal token built to answer that question.
Replaying usage into deterministic balances (same history, same bill) after one too many "we have the logs, but the bill feels wrong" incidents.
Five consensus algorithms implemented conceptually in Rust to compare PBFT, Gossip, Eventual Consistency, Quorum-less, and Flexible Paxos.
A Private Gallery demo that makes the pay-as-you-go article registries and rental modules visible end to end.
Composing multiple providers into one onchain checkout: a pool layer that turns individual services into products on a pay-as-you-go protocol.
Barlow Twins looks statistical on paper but doesn't behave like PCA in practice. A statistical view of redundancy reduction explains why.
Designing rental services as modules on a pay-as-you-go protocol: the contract design, not a dApp tutorial.
Designing the onchain payment and settlement layer for article subscriptions in Solidity.
A practice project while learning Solidity: a simple pay-as-you-go service contract.
A hands-on tutorial for kb-bridge, the MCP server behind the context-aware retriever, for knowledge-base search.
Part 3: evaluating whether the context-aware retriever actually improves RAG answer quality.
Part 2: implementing the context-aware retriever, from design to working retrieval for knowledge-base question answering.
Part 1: why and how I designed a context-aware retriever to improve RAG answers for knowledge-base question answering.
Training an agent to play Hangman with DQN and Transformers.
A deep-learning proof of concept for GMV forecasting with xDeepFM, with an emphasis on clean, reproducible code.
A summary of LoRA: how low-rank adaptation fine-tunes large language models by training small update matrices instead of the full weights.
A hands-on comparison of sequence-to-sequence decoders with and without attention, going beyond a standard code walkthrough.
A derivation of the Normal Bayesian test for hypothesis testing.
Using SpatPCA in R to capture the most dominant spatial pattern from data on two-dimensional locations, step by step.
A SpatPCA tutorial on recovering the dominant spatial pattern from one-dimensional locations, and how it holds up as the signal-to-noise ratio changes.
Potential pitfalls of Empirical Orthogonal Function (EOF) patterns when analyzing a single spatial variable.
A formal look at exploring the dominant spatial patterns of a single variable, with climate research as the motivating context.
Step by step: loading and working with sea surface temperature (SST) data in R.
Notes on Stone's 1985 Annals paper, Additive Regression and Other Nonparametric Models, and the three aspects it uses to frame a statistical model.