🐕 DOG v3.0.0 — The Contrarian Pup

Smart Money goes one way. Dog goes the other. The crowd is already in — let them eat the unwind.

v3.0.0 (2026-05-12) — plumbing-only migration

NO thesis change. v2.5 scoring tables, contrarian flip, exhaustion gate, regime confirmation, persistence bonus, hard gates all preserved verbatim. Six-layer plumbing flip:

1. MCP calls: cfg.mcporter_call(...) → senpi_runtime_helpers.SenpiClient.mcp_call() — direct HTTPS, kills mcporter subprocess cold-start (~280 ms vs 2.5–5 s per call).
2. Signal emit: create_position call site removed; producer emits via cfg._wrapper_client.push_signal(...) to runtime /signals. Runtime opens the position via the LLM-gated dog_entry action.
3. Reentrancy: hand-rolled fcntl lockfile dropped. producer_daemon owns the per-tick scanner_lock with stale-PID auto-recovery.
4. Tick scheduling: bash while true; sleep 180 loop replaced by producer_daemon(fn=main, interval_seconds=180, ...). Long-lived Python process, no per-tick cold-start cost.
5. Risk gates: Python MAX_DAILY_ENTRIES, COOLDOWN_MINUTES, dynamic daily cap circuit breaker → declarative risk.guard_rails block in runtime.yaml. No state files to crash.
6. Exit fee: DSL exits switched from MARKET (taker, 0.045%) to FEE_OPTIMIZED_LIMIT (maker-first, 0.015%, 60 s ALO timeout, taker fallback). Saves ~0.020-0.030% per maker-filled close.

Why the contrarian thesis works (preserved from v2.0)

Hyperliquid is dominated by leverage traders chasing momentum. When a coin moves 3%+ in 4 hours and SM consensus piles in 15%+, the crowd is already maximum exposed. The unwind happens for two reasons: