opportunity-radar

Screens all Hyperliquid perpetual markets through a 4-stage funnel to surface the highest-conviction trading setups.

Architecture

Stage 0: BTC Macro Context (EMA 5/13 on 4h, 1h momentum)
Stage 1: Bulk Screen (~500 assets → ~70 by volume)
Stage 2: Top-N Selection (by composite liquidity score)
Stage 3: Deep Dive (multi-TF technicals + 3-pillar scoring)
Stage 4: Momentum + Final Ranking

3-Pillar Scoring (0-400 scale)

Pillar	Weight	Signals
Market Structure	35%	Volume tiers, surge, OI, OI/Vol health
Technicals	40%	4h trend, hourly trend, RSI, patterns, volume
Funding	25%	Neutral (+40), favorable, unfavorable penalties

Hard Disqualifiers

1. Counter-trend on hourly structure
2. Extreme RSI (>80 for LONG, <20 for SHORT)
3. Strong 4h counter-trend (strength > 50)
4. Volume dying on both timeframes
5. Heavy unfavorable funding (>50% annualized)
6. BTC macro headwind (modifier < -30)

Usage

CLI

hl radar once              # Single scan
hl radar run --tick 900    # Continuous (15 min intervals)
hl radar once --json       # JSON output
hl radar once --mock       # With mock data (no HL connection)
hl radar status            # Show last scan results
hl radar presets            # List presets

Standalone

from skills.radar.scripts.standalone_runner import RadarRunner
from cli.hl_adapter import DirectHLProxy

runner = RadarRunner(hl=hl, tick_interval=900)
runner.run()

Configuration

Via YAML config or CLI flags:

• --min-volume: Minimum 24h volume to qualify (default: $500K)
• --top-n: Assets to deep dive (default: 20)
• --preset: "default" or "aggressive"
• --score-threshold: Minimum final score (default: 150)

Agent Mandate

You are the opportunity radar. Your job is to screen the entire Hyperliquid perps universe and rank assets by trading conviction. You do NOT place trades — you surface setups for APEX or the human operator.

RULES:

• ALWAYS check BTC macro context first — if headwind modifier < -30, all scores are suppressed
• NEVER recommend a disqualified asset — hard disqualifiers are absolute
• Present results sorted by score, highest first
• Include direction (LONG/SHORT) and risk factors for each candidate
• Run at minimum every 15 minutes during active trading

Decision Rules

Score Range	Interpretation	Action
250-400	Elite setup — rare, strong multi-pillar confluence	Immediate entry candidate for APEX
170-250	Good setup — solid edge	Standard entry if Pulse confirms
140-170	Marginal — needs confirmation	Queue only, wait for Pulse signal
100-140	Weak — one pillar carrying	Skip — insufficient edge
0-100	No edge	Ignore completely

BTC Macro	Effect	Action
Strong uptrend (mod > +20)	Tailwind for longs	Score longs normally, penalize shorts
Neutral (mod -10 to +10)	No macro effect	Score normally
Downtrend (mod < -20)	Headwind	Raise entry threshold to 200+
Crash (mod < -40)	Major headwind	Skip all entries, wait for stabilization

Anti-Patterns

• Chasing yesterday's winners: An asset scored 300 yesterday but 120 today → do not enter. Scores are point-in-time.
• Ignoring volume decay: High score but volume dying on both timeframes → disqualified. The setup is stale.
• Entering on RSI extremes: RSI > 80 (long) or < 20 (short) means the move already happened. Radar correctly disqualifies these.
• Running in aggressive mode by default: Aggressive lowers thresholds → more candidates but lower quality. Use default unless you have excess budget.

Error Recovery

Error	Cause	Fix
0 candidates after scan	Low-vol period	Normal — no action needed. Don't force trades.
Candle fetch timeout	HL API rate limit	Reduce --top-n or increase tick interval
BTC candle unavailable	API issue	Radar defaults to neutral macro — safe fallback
Score calculation error	Missing data for asset	Asset auto-skipped — check logs for pattern

Composition

Radar is a sub-component of APEX (runs every 15 ticks). Can also be used standalone for manual trade selection. Pairs with Pulse for confirmation — Radar finds setups, Pulse detects timing.

Cron Template

# Standalone radar every 15 min during trading hours
*/15 8-20 * * 1-5 cd ~/agent-cli && hl radar once --json >> data/radar/scans.jsonl 2>&1