CHEETAH v5.1.1 APEX — Multi-signal confluence sniper with leverage safety. Purpose-built to win the Senpi Arena by taking ~5 high-conviction trades per week at 80% margin and score-scaled leverage. v5.1.1 adds get_safe_leverage() (clamps requested leverage to the asset's Hyperliquid max via strategy_get_asset_trading_limits) and inner-order success validation (rejects phantom ENTRY logs when the outer envelope lies).
get_safe_leverage() which queries
strategy_get_asset_trading_limits and clamps requested leverage to the
asset's Hyperliquid max before submission.create_position returns,
inspect data.orders[0].success and surface INNER_FAILURE when the
per-order status is false even if the outer envelope claims success.
Prevents phantom ENTRY logs from corrupting the daily counter.SM commits. Quality traders commit. Price confirms. Volume commits. All at once. Cheetah pounces once.
Cheetah iterated through four theses, all on HYPE, all on Wolverine's turf:
| Version | Thesis | Result |
|---|---|---|
| v2.0 | HYPE SM consensus momentum | +7.6% (top of fleet briefly) |
| v2.1 | HYPE momentum (hardened) | 33% win rate, -$175 gross |
| v3.0 | HYPE contrarian SM fade | 40% win rate, -$39 in 20h |
| v4.0 | HYPE funding rate extremes | -35% drawdown → HARD STOP |
The pattern: Cheetah kept fighting for Wolverine's HYPE territory. Fleet analysis showed Wolverine (HYPE momentum), Pangolin (funding fader), and Owl (crowding fader) already cover every HYPE thesis Cheetah was trying. Cheetah was redundant.
v5.0 does something no predator in the fleet does: it refuses to trade unless every major signal aligns simultaneously. MIN_SCORE is 14 out of 15 — the highest gate in the fleet by a wide margin.
The Senpi Arena ranks by ROE %, not absolute PnL, with a $25k weekly volume minimum. Looking at the current week's leaderboard top 5:
Two winning shapes exist. The scalper path is fee-drag death for a fleet agent — our fleet is 37 red / 2 green precisely because Hyperliquid fees eat high-frequency strategies. The sniper path (pr0br000) works because each trade has maximum edge, and 19 trades × ~3.5% ROE per trade = 67% ROE total.
APEX is purpose-built to replicate the sniper shape.
Every scan iterates through all non-XYZ assets from leaderboard_get_markets(limit=100) and scores each one. An entry fires only when score ≥ 14 AND all hard gates pass.
pct_of_top_traders_gain ≥ 10% AND trader_count ≥ 25contribution_pct_change_15m ≥ 1.0 OR contribution_pct_change_1h ≥ 3.0| Signal | Points |
|---|---|
| SM_STRONG (pct ≥10%, traders ≥25) | 4 |
| Velocity gate passed (one of 15m/1h above minimum) | 2 |
| Accelerating (15m > 1h > 0, inflow building) | 2 |
| Dual price confirmation (4h ≥2% + 1h agrees) | 2 |
| Volume spike (≥2× 6h average) | 1 |
| Quality trader alignment (≥1 ELITE/RELIABLE in same direction) | 3 |
| Rank climbed ≥5 positions in last 2 scans | 1 |
Score = 14 requires all signals except either the rank climb OR dropping one of the velocity subcomponents. Score = 15 requires everything. Partial signal stacks (score ≤13) never fire.
| Parameter | Value | Rationale |
|---|---|---|
| Starting budget | $648 | Cheetah's current equity post-drawdown (not rebased to $1000) |
| Margin % per trade | 80% | Maximum conviction commits maximum capital |
| Max leverage | 10x | Fleet cap per H12 audit hypothesis |
| Max positions | 1 | Concentration, no parallel bets |
| Max daily entries | 5 (from dynamic cap) | Matches Arena $25k/week volume floor |
| Notional per trade | ~$5,184 | 80% × $648 × 10x |
| Target weekly volume | ~$25,920 | 5 trades × $5,184 = clears $25k Arena floor |
if pnl_pct >= 5: return 8 # Hot hand — more shots
elif pnl_pct >= 0: return 5 # Target rate for Arena volume floor
elif pnl_pct >= -5: return 3 # Careful
elif pnl_pct >= -15: return 2 # Defensive
elif pnl_pct >= -25: return 1 # Preserve