Paired execution
Durable paired instructions precede concurrent, immediate-or-cancel limit orders. The system reconciles uncertain fills before replacement and chases only confirmed residual quantities.
Arbitrage Framework
Cross-venue perpetual arbitrage, built around price convergence and funding carry.
A working system has completed bounded paired execution on Hyperliquid and Lighter testnets. A separate, read-only collection run now studies production market data.
Progress reported as at .
What the execution evidence shows
Four paired ETH round trips completed, with confirmed fills, reconciliation and independently checked return to flat. The demonstrated direction is long Hyperliquid / short Lighter; authenticated execution in the reverse direction remains unproven.
The first cycle opened 0.007 ETH per leg and closed normally, at a recorded cost of 0.027489 USDC. The subsequent repeatability batch obtained three further paired successes in nine attempts:
- Paired round trips
- 3
- Each needed bounded closing chase.
- One-sided fills
- 3
- Bounded recovery returned both accounts to flat without operator closing orders.
- Connection failures
- 3
- Failed before activation; no orders were sent.
The batch cost 0.128436 USDC including recovery attempts. These were deliberately uneconomic testnet execution tests: four successful round trips, not four consecutive successes or evidence of a profitable strategy.
What is implemented
Separate foundation, persistence, engine and venue-adapter layers connect economic checks to durable execution and replay. Venue credentials and external actions belong to the adapters, not the engine.
Economics after costs
Both directions are evaluated for basis convergence and funding carry, after entry and exit fees, executable-depth slippage, capital costs, recovery reserves and uncertainty allowances. Ranking uses the capital reserved on both venues.
Risk at each exchange
Freshness, liquidity, margin, capability and delta checks govern admission. Collateral and liquidation risk remain local to each venue. Exits respond to convergence or reversal, subject to a maximum holding horizon.
Bounded recovery
Price limits, deadlines and attempt budgets constrain recovery. Reduce-only orders address residual exposure; the system halts when completion cannot be established. Restart reconstruction and account isolation are implemented.
Venue coverage
Hyperliquid
Authenticated testnet execution
Permanent adapter implemented; paired ETH execution demonstrated with Lighter.
Lighter
Authenticated testnet execution
Permanent adapter implemented; paired ETH execution demonstrated with Hyperliquid. The v1 zero-taker-fee account requirement is enforced.
Binance USDⓈ-M
Implemented and locally tested
Authenticated paired execution has not been demonstrated.
Bybit V5 Linear
Implemented and locally tested
Authenticated paired execution has not been demonstrated.
Aster
Adapter implemented
Testnet access restrictions prevented the intended paired test.
Production-data research, without trading
The deployed SHADOW run collects public Hyperliquid and Lighter ETH data with no trading credentials, signers, account access or order submission. Bounded, hash-linked journals support offline research in both directions over 60-second, five-minute and one-hour horizons, recording missing, stale, incomplete and censored observations.
At the reported 12:27 BST check on 3 October, it had nine valid observations from each venue and no recorded acquisition issues. This was an early health check, not a completed 24-hour result. The authorised collection window ends on 4 October 2026 at 12:18 BST, with a server-enforced deadline and bounded shutdown period.
No profitability scenario has been approved and fee/FX assumptions remain incomplete. This collection cannot establish profitable hypothetical trades.
Verification and remaining work
The committed SHADOW build reports 7,097 passing tests across foundation, unit, transport, architecture and SHADOW suites; strict typing passed for 218 source files. Installation checks passed 42 tests, and all 463 deployed source files matched the committed archive. These scoped checks are not a fresh full-database aggregate or independent certification of the entire release.
Independent review has driven corrections to execution authority and concurrent claim handling. Repeatability testing also exposed connection failures and one-sided fills; those failures remain part of the evidence.
- Verify the completed SHADOW journals and assess freshness, gaps, executable depth and observed basis behaviour.
- Calibrate convergence, funding, fees and risk assumptions against that evidence.
- Improve measured execution reliability while retaining admission and recovery controls.
- Extend evidence beyond one asset and one authenticated direction, and reconcile release documentation and delivery baselines.
Profitability, sustained operational reliability and mainnet readiness remain unproven. Mainnet trading, unattended operation and multi-pair portfolio behaviour require separate acceptance.
FAQ
Arbitrage Framework FAQ
Scope, execution evidence and current research status
What is the intended strategy?
Matched long and short perpetual positions across two venues, seeking price convergence and funding carry after costs and risk allowances. Both directions are evaluated. Funding can change, price gaps can widen and each venue retains its own margin and liquidation constraints.
What has execution testing demonstrated?
As reported on 3 October 2026: Four completed paired ETH round trips on Hyperliquid and Lighter testnets, with fills reconciled and return to flat independently checked. These were long Hyperliquid and short Lighter, not four consecutive successes. Reverse-direction authenticated execution is not yet established; the tests were deliberately uneconomic and do not establish profitability.
Is the production-data collection trading?
As reported on 3 October 2026, the authorised run observes public Hyperliquid and Lighter ETH data without trading credentials, account access or order submission. Its collection window ends on 4 October at 12:18 BST. The initial health check is not a completed 24-hour result; no profitability conclusion has been established.
Is the system available for live trading?
The system remains under build at the research-validation stage. Mainnet trading, unattended operation, multi-pair behaviour and economic viability require separate acceptance. No launch date is announced.