CIPHER Trading Playbook

April 17, 2026 · View on GitHub

Author: Head of Trading / Quant Strategist Date: 2026-04-17 Capital: $1000 USD equivalent on Solana (wallet-held SOL + USDC) Horizon: 60 days live trading Jurisdiction: Canada (no US-only brokers)

1. Executive recommendation

We trade Solana-native only: SPL spot via Jupiter for longs, Drift perps for shorts and leverage, Jito-LST for idle cash. We take the top 3 CIPHER signals per day filtered at composite >= 0.65 and confirmation STRONG+, sized at eighth-Kelly (roughly 5-8% notional per trade), on 3-10 day swing holds with ATR-based stops. Target: Sharpe 0.8-1.2, expected 60-day P&L band of -15% to +25% with a hard kill switch at -20% drawdown.

2. Universe + strategy

Universe (ranked by priority)

TierAssetsVenueRationale
CoreSOL, BTC (wrapped), ETH (wrapped), JTO, JUP, RAY, PYTHJupiter spotDeep liquidity (>$5M 24h vol), <20bp slippage on $50 clips
SatelliteBONK, WIF, POPCAT, top-20 memecoins with CIPHER STRONG+Jupiter spotHigher edge, cap at 20% of book
Shorts / leverageSOL-PERP, BTC-PERP, ETH-PERPDrift ProtocolDecentralized, no KYC, Canadian-legal, <5bp funding drift
Yield on idle cashJitoSOL, bSOLJito/BlazeStake7-8% APY on unused SOL while waiting for signals

Explicitly excluded: cross-chain bridges (fragile at $1k scale), CEX (Binance/OKX custody + Canadian tax friction), US-only perps (dYdX geoblocks Canada on some pairs — verify before use, prefer Drift), Alpaca/stocks (no Canadian broker API exists with sub-$1 fees). Stocks stay paper-traded only until we find a Canadian broker with a REST API (Questrade IQ? needs research).

Strategy: filtered top-K swing

  • Signal filter: composite_strength >= 0.65 AND confirmation_count >= 5/10 (STRONG+) AND regime_gate passes.
  • Regime gate: RISK_ON -> longs + shorts both live. RISK_OFF -> shorts only, longs paused. CRISIS -> flat, all positions liquidated to USDC within 4 hours.
  • Top-K: N=3 per day, ranked by composite score. If fewer than 3 pass the filter, take fewer. Never force-fill.
  • Hold period: 3-10 day swings. Exit on (a) ATR stop hit, (b) TP2 scaled out, (c) CIPHER emits an opposing signal on same asset, (d) 10-day max holding time.
  • Entry: Limit order at signal-bar close +/- 10bp, valid 60 min. If unfilled, cancel. No market orders (protects against Jupiter route degradation on memecoins).
  • Exit:
    • Stop loss: entry - 1.5 * ATR(14, daily).
    • TP1: entry + 1 * ATR, scale out 50%.
    • TP2: entry + 2.5 * ATR, scale out remaining 50%.
    • Trail stop on remaining position after TP1 moves to breakeven.

Why not trade all 18-22 signals?

At ~$1000 capital, each trade has a hard floor of ~$75 notional to overcome fees + slippage (Jupiter 20-50bp + priority fee + Jito tip ~$0.10). Trading 20 signals means $50 clips, which is unprofitable after round-trip costs. Top-3 per day with $60-120 per trade is the sweet spot.

3. Position sizing + risk budget

  • Per-trade size: 6% of equity (eighth-Kelly approximation for composite ~= 0.7 signals with assumed 52% hit rate and 1.5R average winner). On $1000: $60 notional nominal, $90 max after winners compound.
  • Max simultaneous positions: 5. Hard cap — if 5 open and a new STRONG signal fires, skip it.
  • Max per asset class:
    • SOL + wrapped majors (SOL/BTC/ETH): 60% of book.
    • Satellite memecoins: 20% of book.
    • Perps (net notional): 30% of book (leverage capped at 2x).
    • USDC / JitoSOL yield bucket: minimum 20% always (dry powder for new signals + drawdown cushion).
  • Correlation adjustment: If 2+ open positions are in the same CIPHER correlation cluster (e.g., three memecoin longs, or SOL-spot + SOL-PERP both long), count them as one "meta-position" and cap total exposure at 10% of equity.
  • Fee budget: $40 / 60 days (4% of capital). If fees exceed this by day 30, reduce trade frequency to top-2/day.

4. Expected P&L + drawdown

Scenario60-day P&LSharpeMax DDNotes
Edge works (base case)+10% to +25%0.8-1.212%What we plan for
Break-even-3% to +3%015%Fees eat the alpha; signal quality = market
Edge broken-15% to -25%negative20% (kill switch trips)Regime change; CIPHER miscalibrated
Black swan-40%+N/A> 20% killExchange exploit, SOL depeg, correlated crash

Volatility envelope: daily return std-dev ~2.0-3.0% on $1000 with 3 concurrent positions. Weekly drawdowns of 6-8% are NORMAL and not cause to intervene. Only the 20% all-time-drawdown kill switch halts the bot.

Realistic prior: retail algos without institutional edge typically net -5% to +15% annualized after costs. We should not expect hedge-fund Sharpes. If we hit +10% in 60 days with Sharpe > 0.8, that validates CIPHER alpha exists and justifies scaling to $5k.

5. Day-in-the-life walkthrough

10:00 UTC — CIPHER pipeline finishes daily run. Emits 20 signals across stocks + crypto.

10:05 UTC — Trade executor (new component, see Section 6) filters:

  1. Drop all stock signals (no Canadian broker yet).
  2. Drop all crypto signals not in our universe (tier 1 + 2).
  3. Apply regime gate — assume RISK_ON today.
  4. Filter composite >= 0.65, confirmation >= 5/10.
  5. 6 signals survive. Rank by composite; take top 3: SOL long (0.78), JTO long (0.71), WIF long (0.67).

10:06 UTC — Pre-trade checks:

  • Wallet balance: 4.8 SOL + 200 USDC = ~$950. OK.
  • Open positions: 2 (RAY long, SOL-PERP short as hedge). Adding 3 more = 5, at cap. OK.
  • Correlation cluster: all 3 new + RAY = 4 crypto longs. Meta-position check: sum notional $240 = 25% of book. Above 10% cluster cap. Reduce: take only SOL + JTO, skip WIF.

10:07 UTC — Place orders:

  • SOL: buy $60 at limit $148.20 (signal close + 10bp), slippage cap 30bp, TTL 60 min.
  • JTO: buy $60 at limit $3.42, slippage cap 50bp (thinner book), TTL 60 min.
  • Send via Jupiter v6 swap API, signed locally by wallet module, priority fee from Helius estimator, Jito tip 5000 lamports.

10:08-11:07 UTC — Fill monitor polls every 30s. Both fill within 2 min. Record entries, compute ATR-14 from yfinance-equivalent (Birdeye daily OHLC), set stops + TPs in local state DB.

Hourly — Exit monitor ticks: pulls current prices via Jupiter price API, checks each open position against stop / TP1 / TP2 / max-hold. Fires exit swap if triggered. Same Jupiter path, slippage cap 50bp on exits (prefer fill over price).

23:45 UTC — H6 NightlyQualityCheck runs, computes realized P&L, updates IC history, rolls closed trades into H8 PerformanceReporter.

Weekly — Manual founder review: check Sharpe, hit rate, fee-to-P&L ratio. If fee ratio > 30%, cut trade count.

6. Critical missing components (ranked)

P0 — blocker for go-live

  1. Wallet signing module (cipher/trading/wallet.py) — solders Keypair load from encrypted keyfile, ed25519 sign, broadcast via Helius RPC. Est 300 LOC + tests.
  2. Jupiter swap client (cipher/trading/jupiter_client.py) — v6 /quote + /swap REST calls, priority fee from Helius getPriorityFeeEstimate, Jito tip injection, slippage enforcement, retry with route refresh on ROUTE_NOT_FOUND. Est 400 LOC.
  3. Order manager (cipher/trading/order_manager.py) — state machine (PENDING -> FILLED / CANCELLED / FAILED), fill confirmation via getTransaction polling, idempotency by client_order_id. Est 500 LOC.
  4. Position state store (new table positions in SQLite) — open qty, entry px, stop, TP1, TP2, ATR, opened_at, meta-position cluster id. Migration 0004.
  5. Exit monitor (scheduled hourly via existing Scheduler) — reuses CIPHER pricing layer, evaluates stop/TP/time, fires exit via order manager.
  6. Pre-trade risk gate — correlation cluster check, per-asset-class cap, fee-budget check. Hard-rejects trades violating limits.
  7. Emergency liquidation path — single CLI command cipher panic-sell that converts ALL SPL + perp positions to USDC within 4 hours using Jupiter TWAP. Triggered by (a) manual invoke, (b) D8 CircuitBreaker 20% DD, (c) regime = CRISIS.

P1 — needed by day 14

  1. Drift perp client — Python SDK (driftpy) for SOL/BTC/ETH-PERP open/close/flip. Deferred until spot legs stable.
  2. PnL attributor — closed-trade realized P&L, unrealized mark-to-market, fee breakdown. Feeds H8 PerformanceReporter.
  3. Slippage post-mortem — log expected vs realized fill for every swap; alert if 7-day average slippage > 40bp (route quality degrading).
  4. JitoSOL sweeper — idle USDC above 20% floor auto-stakes to JitoSOL; unstakes when signal fires. Yields ~7% on dry powder.

P2 — nice to have

  1. Transaction simulator (dry-run Jupiter quotes without signing) for UI preview.
  2. Multi-wallet split (Wallet A = long book, Wallet B = short book) for P&L attribution clarity.
  3. Telegram alerting on fills + kill-switch events.
  4. Canadian broker research (Questrade IQ / IBKR Canada / Wealthsimple API) for stock leg.

7. Go-live gate (must be GREEN)

Before a single real SOL moves, every item below must be checked off:

  • Paper trade 14 consecutive days with the full pipeline — same signals, same sizing, same execution timing, but no wallet signing. Log hypothetical fills at mid-price + realistic slippage.
  • Paper Sharpe >= 0.5 AND paper max DD < 15% over those 14 days.
  • Wallet module passes signing tests on Solana devnet (never touch mainnet keys in CI).
  • Jupiter client tested on mainnet with $5 clips (micro dry-run), >= 20 successful round-trips, < 30bp average slippage.
  • Emergency liquidation cipher panic-sell tested end-to-end on devnet AND once on mainnet with a single $10 position.
  • D8 CircuitBreaker 20% DD trigger verified via fault injection test.
  • Oracle Cloud VM running cipher serve 72 hours with zero restart, zero unhandled exceptions.
  • Separate hot wallet funded with exactly $1000 — never reuse Wallet A (creator) or Wallet B (GMGN rep-building), per memory.
  • Kill switch documented: founder can SSH and run one command that stops bot + liquidates all to USDC.
  • Fee budget tracker shows hypothetical fees during paper trade < 4% of capital over 14 days.

If any gate is red on day 60 of current roadmap, we delay go-live. No exceptions.

End of playbook. Next action: scope P0 components into Sprint 24 plan.