OWL v6.1 — Pure Contrarian (universe expansion)

Wait for the crowd to overcommit. Wait for them to exhaust. Then eat their liquidations.

One scanner. One thesis. Every other skill in the zoo enters WITH momentum, WITH the trend, WITH smart money. OWL is the only skill that enters AGAINST the crowd. The edge: crowded trades unwind violently and predictably.

v5 is a complete rebuild. v1-v4 had 3 scanners (contrarian + momentum + correlation). The momentum and correlation scanners caused the agent to enter WITH the crowd while thinking it was contrarian. v5 has one scanner that does one thing: find crowded assets that are exhausting.

v5.3 — the 30-day-drought fix

Owl v5.2 took zero trades between 2026-03-15 and 2026-04-14 despite running continuously. Two root causes, both fixed in v5.3:

1. Self-executing scanner (the main bug). The v5.2 scanner emitted a signal JSON via cfg.output() and assumed an external "action layer" would read it and call create_position. Owl has no runtime.yaml action layer wired up, so signals died at stdout. Even perfect signals never became trades. v5.3 calls create_position directly via mcporter, matching the Wolverine/Phoenix self-executing pattern. When a score ≥ minScore (14) candidate is found, the scanner executes the entry order itself — no external action layer required.

2. Persistence tolerance. The v5.2 check_persistence/clear_persistence pair cleared the 4-hour persistence timer on any single scan that fell below minCrowdingScore. At 15-minute cadence, a single noisy scan reset the clock to zero. Over 30 days, no coin ever accumulated 4 uninterrupted hours of high crowding. v5.3 adds a mark_below_threshold tolerance counter — persistence is only cleared after 2 consecutive below-threshold scans (30 min at 15-min cadence). A single noisy dip no longer destroys the 4h window.

Fleet-standard guardrails layered on top:

Drawdown circuit breaker. Below -5% PnL from starting budget, the effective entry cap is hard-clamped (defensive). Below -25%, HARD STOP (circuit breaker). Preserves capital in drawdowns while still allowing Owl's dynamicSlots earning-forward logic to unlock more entries when winning.
Auto-cancel stale resting orders. has_resting_orders() cancels any non-reduceOnly maker order older than 10 minutes so the scanner is never locked out indefinitely by a FEE_OPTIMIZED_LIMIT order that didn't fill (fleet-standard pattern from PR #177).
runtime.yaml added. Owl now follows the standard Predators deployment pattern with a position_tracker scanner + DSL High Water exit engine preset.

v5.2 fixes (preserved in v5.3):

Funding floor lowered from 20% → 12% so the five-factor scoring model actually runs. Assets below the funding floor score 0 on funding but continue through SM/OI checks instead of early-returning. minCrowdingScore of 8 remains the real quality gate.
Observability logging: every scan cycle logs the top 3 crowding scores and active persistence timers to stderr (internal log only, not notifications). Lets us answer "is OWL seeing anything?" without changing config or asking the agent.

MANDATORY: DSL High Water Mode

OWL MUST use DSL High Water Mode. This is not optional. Do not substitute standard DSL tiers.

Spec: https://github.com/Senpi-ai/senpi-skills/blob/main/dsl-dynamic-stop-loss/dsl-high-water-spec%201.0.md

When creating DSL state files for any OWL position, you MUST include:

{
  "lockMode": "pct_of_high_water",
  "phase2TriggerRoe": 10,
  "tiers": [
    {"triggerPct": 10, "lockHwPct": 20, "consecutiveBreachesRequired": 3},
    {"triggerPct": 20, "lockHwPct": 40, "consecutiveBreachesRequired": 3},
    {"triggerPct": 35, "lockHwPct": 60, "consecutiveBreachesRequired": 2},
    {"triggerPct": 50, "lockHwPct": 75, "consecutiveBreachesRequired": 1},
    {"triggerPct": 75, "lockHwPct": 85, "consecutiveBreachesRequired": 1}
  ]
}

FALLBACK: Use tiersLegacyFallback from config until engine supports pct_of_high_water.

How OWL v5 Works

The Three Phases (all must pass before entry)

Phase 1 — CROWDING (score ≥ 8) Scan top 30 assets every 15 minutes. Score how one-sided each asset's positioning is:

Signal	Max Points
Funding extremity (annualized rate)	4
SM concentration (leaderboard tilt)	4
OI concentration (USD-weighted)	2
SM confirms funding direction	1

Most assets score 0-3 (not crowded). Only assets scoring 8+ advance.

Phase 2 — PERSISTENCE (4+ hours) Crowding must persist for at least 4 hours. A brief funding spike that resolves in 30 minutes is noise. True crowding builds over hours — funding stays extreme, OI keeps growing, SM stays tilted. The longer the crowding persists, the more violent the eventual unwind.

Phase 3 — EXHAUSTION (score ≥ 5) The crowd is positioned, and they've been positioned for hours. Now: are they exhausting? Four signals:

Signal	Points	What It Means
Volume declining (recent vs 6h avg)	3	Conviction leaving — nobody new is entering
Price stalling (crowd long but price flat)	3	The trade stopped working — crowd is trapped
Volume spike without follow-through	2	Capitulation wick — someone tried to push, failed
4h RSI divergence	2	Momentum dying despite positioning

Entry

Total score (crowding + exhaustion) must be ≥ 14. Entry direction is OPPOSITE to the crowd. If the crowd is long, OWL goes short. If the crowd is short, OWL goes long.

This means OWL enters 1-2 times per day at most. Often zero. That's by design.

Hold

Every 15-minute scan re-evaluates held positions FIRST. The position holds as long as the crowd doesn't come back.

Re-Crowding Exit (unique to OWL)

If the crowd rebuilds in their original direction (re-crowding score ≥ 6), the unwind thesis is dead and the position exits immediately. This is OWL's equivalent of SCORPION's "sting" — an instant, thesis-based exit that overrides DSL.

DSL: Widest in the Zoo

Contrarian entries retrace hard before working. The crowd doesn't unwind smoothly — they fight back first. OWL needs the widest DSL bands of any skill.

Setting	Value	Compare to FOX
Phase 1 floor	4% notional (~40% ROE at 10x)	FOX: 1.5%
Phase 2 trigger	+10% ROE	FOX: +7%
T1 lock	20% of HW	FOX: 40%
85% trail at	+75% ROE	FOX: +20%
Stagnation TP	15% ROE, 120min	FOX: 10%, 45min
Time exits	All disabled	FOX: 30min hard

The tradeoff: OWL loses bigger on losers (-35 to -40% ROE) but catches crowding unwinds that produce 50-200%+ ROE when the cascade hits.

Risk Management

Rule	Value	Notes
Max positions	2	Rare, concentrated bets
Max entries/day	2 base, up to 4 on profitable days
Phase 1 floor	4% notional	Widest in the zoo
G5 per-position cap	10% of account	Wider than most — contrarian needs room
Drawdown halt	25% from peak
Max consecutive losses	2 → 180 min cooldown	Long cooldown — if the contrarian thesis failed twice, something changed
Re-crowding exit	Immediate	If the crowd comes back, thesis is dead
Loss cooldown per asset	6 hours	Don't re-enter same contrarian trade too soon

Cron Architecture

Cron	Interval	Session	Purpose
Scanner	15 min	isolated	Crowding scan + exhaustion detection + re-crowding check
DSL v5	3 min	isolated	High Water Mode trailing stops

15-minute scanner interval is intentional. Crowding builds over hours, not minutes. Scanning every 3 minutes would waste tokens on data that hasn't changed. The DSL cron still runs every 3 minutes for trailing stop protection.

Notification Policy

ONLY alert: Position OPENED (asset, direction, crowding score, exhaustion signals, how long crowded), position CLOSED (DSL or re-crowding exit with reason), risk guardian triggered, critical error. NEVER alert: Scanner found no crowding, scanner found crowding but no exhaustion, persistence tracking updates, DSL routine check, any reasoning. All crons isolated. NO_REPLY for idle cycles.

Bootstrap Gate

Check config/bootstrap-complete.json every session. If missing:

Verify Senpi MCP
Create scanner cron (15 min, isolated) and DSL cron (3 min, isolated)
Write config/bootstrap-complete.json
Send: "🦉 OWL v5 is online. Pure contrarian. Scanning for crowded exhaustion. Silence = no opportunity."

Expected Behavior

Metric	Expected
Trades/day	0-2 (crowding unwinds are rare)
Avg hold time	4-24 hours
Win rate	~45-55% (wider stops, contrarian timing is hard)
Avg winner	40-150%+ ROE (crowding unwinds are violent)
Avg loser	-25 to -40% ROE (wide floors, structural invalidation)
Fee drag/day	$2-8 (very few trades, all maker entries)
Profit factor	Target 1.3-2.0 (big winners compensate for wider losers)

Files

File	Purpose
`scripts/owl-scanner.py`	Crowding + exhaustion + re-crowding — the only scanner
`scripts/owl_config.py`	Shared config, MCP helpers
`config/owl-config.json`	All variables with DSL High Water tiers + legacy fallback

License

Apache-2.0 — Built by Senpi (https://senpi.ai). Attribution required for derivative works. Source: https://github.com/Senpi-ai/senpi-skills

Skill Attribution

When creating a strategy, include skill_name and skill_version in the call. See references/skill-attribution.md for details.

owl-strategy