ANT Spread Convergence · Dual-Timeframe Dynamic Grid

Risk warning: Live trading is enabled by default, and the strategy sends real orders. The assumed share count does not guarantee contract equivalence, rebasing or convergence. Cross-exchange hedging does not eliminate spread divergence, single-leg liquidation, funding, liquidity or execution mismatch risk. P&L uses configured fees and funding budgets rather than final exchange settlements. Verify contract units, permissions, fees and margin before running.

Strategy Concept

The strategy studies spreads between Anthropic pre-market contracts. Its fixed direction is long Hyperliquid xyz:ANT and short Binance or OKX ANTHROPIC. Quotes and quantities are normalized using the selected share-count assumption and contract specifications.

Two timeframes serve different purposes:

• The long timeframe sizes the initial position. Recent spread location, available balances, reserve capacity and estimated downside room determine the initial number of paired lots.
• The short timeframe controls grid spacing. The strategy reduces one paired lot when the spread narrows sufficiently and conditionally refills or adds one when it widens.
• A fully flat position ends the round. Once both legs are flat and no request is unresolved, the round is settled and a new plan uses current prices and balances.

The strategy does not automatically reverse its hedge direction. Statistical bands are recent activity ranges, not guaranteed price boundaries.

WS Interface and Execution Architecture

The strategy uses FMZ's WebSocket event interface. ctx.subscribe provides depth and private account streams, while ctx.poll processes quotes, orders, fills and connection events.

1. Executable spreads: five depth levels are subscribed on xyz and twenty on the CEX. Order-size-weighted execution prices produce separate entry and exit spreads. Historical closes support statistics; current depth drives trading.
2. Historical preparation: startup pairs closed candles from the same minute on both venues. WS samples subsequently update the history. Insufficient, sparse, stale or discontinuous history blocks new rounds.
3. Sequential execution: the xyz leg executes first, followed by the CEX hedge sized from confirmed fills. Normal WS execution uses IOC limit orders with price tolerance. Simultaneous or complete fills are not guaranteed.
4. Account and order routing: available balances primarily come from WS, with REST queries when required fields are missing. Detected Binance portfolio-margin accounts use REST limit orders, with WS reports still required to confirm outcomes.
5. Unresolved requests: unknown terminal states remain recorded and block blind resubmission and new exposure. Reconnection alone does not establish that orders or positions have been reconciled.

Use a node supporting the required depth and private streams and ctx.subscribe, ctx.poll and ctx.submit. Support depends on the venue and account type. REST routing does not replace the required WS data and account reports.

Trading Workflow

            text
            
            
Add Hyperliquid futures + Binance or OKX futures
        ↓
Verify conversion assumptions, contracts, positions and open orders
        ↓
Prepare paired history → wait for valid quotes and account snapshots
        ↓
Long-timeframe planning: location, capacity and initial paired lots
        ↓
Confirm plan → execute both legs → split confirmed fills into lots
        ↓
Short-timeframe grid: reduce / refill / add conditionally
        ↓
Both legs flat and no unresolved request → settle the round
        ↓
Refresh balances → plan the next round

Initial Position Planning

Defaults use a 24-hour window and its 10th–90th percentile spread band, with at least two hours of valid history.

Initial size is constrained by funding capacity, band location, reserve requirements and available profit room. Approximately 25% of capacity is reserved where capacity permits; a one-lot account cannot reserve an additional lot. If available room cannot support one grid step, the strategy waits.

The round's capacity and initial planning snapshot remain fixed after entry. Long-timeframe changes do not repeatedly rebuild the initial position during that round.

Dynamic Grid Execution

The default short window is 30 minutes. Reduction and addition spacing are calculated separately, both defaulting to approximately one third of the robust spread range, subject to the cost-and-profit floor.

Confirmed trades update the reference spread. After a reduction, the next refill uses the current reduction spacing.

Normal reductions check the reference level, the lot's own entry cost and estimated net proceeds using actual entry prices, executable exit prices, configured fees and the lot's accrued funding budget. Reaching a spread trigger alone is insufficient.

Additions also require available balances, acceptable effective leverage, round capacity, remaining profit room, sufficient depth and the current grid upper bound. A jump beyond that bound waits for updated statistics rather than being chased.

Round Completion and Risk Handling

A round ends when the final lot is closed, both legs are flat and no request remains unresolved. No permanent base position is retained for eventual convergence.

There is no additional inter-round delay by default, but a new round still requires planning and confirmation. The optional spread protection level is disabled by default. Deleveraging, exception handling and manual closes may realize losses.

Fees and Cost Conditions

A paired entry and exit involve four fills. With similar leg notionals:

Round-trip fees = 2 × (xyz taker fee + CEX taker fee)

Grid spacing also includes slippage allowance and a minimum net-profit requirement:

Cost-and-profit floor = round-trip fees + 2 × per-leg slippage budget + minimum net profit

Defaults are:

• xyz taker fee: 1 bp;
• CEX taker fee: 4 bps;
• per-leg slippage budget: 5 bps;
• minimum net profit: 5 bps.

The resulting default spacing floor is 25 bps. This is a configured budget and trading threshold, not a fixed realized cost or a profit guarantee.

Enter actual account and market fees. Funding data is read from the exchanges where available; otherwise the configured budget is used. Individual grid net estimates do not allocate funding, while round settlement deducts the accumulated round funding budget. Reconcile final results against both exchanges' statements.

Live Configuration

Configure exactly two exchanges:

• Hyperliquid futures: xyz:ANT_USD.swap;
• Binance USDT-margined futures or OKX futures: ANTHROPIC_USDT.swap.

The Hyperliquid leg uses available USDC and the CEX leg available USDT. Use independent accounts or avoid external positions and orders in the target contracts.

The default pre-market share-count assumption is 12.7 billion. Binance quotes are normalized by 127 ÷ 10, and OKX by 127 ÷ 100. Quantities additionally account for contract multipliers, precision and minimum order requirements. The converted stage uses one share per unit in the code; official contract rules must still be verified.

The default planning leverage is 2. The strategy does not configure exchange leverage or margin mode. Set these separately.

Startup checks positions and open orders without taking over external exposure. Ledger mismatches, external orders or unresolved requests block startup. Exchanges, conversion settings and live/paper mode cannot be changed while positions or unresolved requests remain.

Main Live Parameters

These are the main settings; the parameter panel contains all 33. Quote age, timing differences, hedge tolerances, price jumps and contract-rule checks also constrain execution.

Operating Notes

1. Share-count assumptions and contract relationships require ongoing verification. The strategy remains long xyz and short the CEX; it does not automatically reverse after a rule change.
2. Margin is separate on each venue. Deleveraging depends on valid quotes, account estimates and available execution channels and cannot guarantee protection from liquidation.
3. IOC orders may fail or partially fill. Sequential execution creates temporary exposure, and depth changes, rejections or hedge failures may produce losses.
4. The cumulative loss limit uses completed-round net estimates. It is not an account-equity stop and does not include current floating losses.
5. Data or account-stream problems suspend affected decisions. Unavailable channels can prevent reductions. Pending requests and persistent halt states require reconciliation.
6. With flatten-on-stop enabled, normal shutdown attempts REST market closes. Pending requests preserve the request and ledger instead of triggering additional close orders. Check actual positions after crashes or exchange failures.
7. Fees, funding and P&L include estimation assumptions. Displayed results are not final settled returns, and small-account results do not establish suitability for larger capital.
8. Larger allocations require further work on market impact, order splitting, hedge mismatch and execution monitoring. AI can assist with these improvements, followed by testing against actual account conditions and capital size.