condor-strategy

🦅 CONDOR v4.0.1 — One Amazing Trade per Day

Top 50 assets. Pure trend continuation. Apex confluence only. One trade a day.

v4.0.1 (2026-05-18) — held-asset dedup race-fix + DSL throttle

Operational reliability patch. NO thesis change. NO scoring change. NO gate change. Direct port of the Bison v3.0.1 patches to address the same race-window + position-state-drift bug class observed on Condor's decision log.

Bug observed. Audit on Condor2 (M193171) 2026-05-14 to 2026-05-17 showed 4 consecutive CONDOR_APEX HYPE LONG ENGINE_FAILUREs, all for HYPE LONG signals fired while HYPE was already held by the runtime executor. The producer's on-chain held_assets check leaks during the race window between push_signal() returning OK and the resulting position appearing in the next-tick clearinghouseState pull. Same bug class Bison v3.0.1 fixed; same fix here.

Fix 1 — recent-signals cache (scripts/condor_config.py). record_signal(coin) writes {coin: epoch_seconds} to state/recent-signals.json after every successful push. main() calls was_recently_signaled(coin) after candidate-best selection and BEFORE push_signal, skipping emission for any coin seen within RECENT_SIGNAL_TTL_SEC (default 240s ≈ 4× the typical ALO open-fill window + next-tick cadence). Skipped coins are reported with DEDUP_SKIP notes in the per-tick output for audit.

Fix 2 — DSL exit interval 30s → 60s (runtime.yaml). Defensive mitigation of the runtime position-state drift bug seen in Bison's audit (CLOSE_FEE_OPTIMIZED_FAILED "Reduce only order would increase position" errors firing 1-2 ticks after a successful close). Condor's close-side audit shows no such failures yet, but the bug is fleet-wide — applying the throttle here proactively. Root cause (runtime-side position-state sync) is escalated to the runtime team — this YAML change is downstream mitigation only.

What this does NOT fix. Phantom orphan positions where DSL state and HL state are durably out of sync. That class of bug needs runtime-side reconciliation; the producer has no authority over DSL state.

v4.0.0 (2026-05-12) — plumbing-only migration

NO thesis change. v3.4 scoring tables, hard gates, score-scaled sizing tiers, and DSL preset all preserved verbatim. Six-layer plumbing flip:

1. MCP transport: mcporter subprocess (2.5-5s cold-start per call) → senpi_runtime_helpers.SenpiClient.mcp_call() in-process HTTPS (~280ms).
2. Signal emit: scanner called create_position directly; producer now emits via push_signal() to runtime /signals. Runtime LLM-gated condor_entry action opens via FEE_OPTIMIZED_LIMIT, passing marginUsd + leverage through from signal data for score-tier sizing.
3. Reentrancy: no v3 lockfile to replace (Condor v3.x ran on openclaw cron with no producer-side lock); producer_daemon owns the per-tick scanner_lock with stale-PID auto-recovery.
4. Scheduler: openclaw cron (3 min) → producer_daemon(interval_seconds=180). Long-lived process.
5. Risk gates: Python MAX_POSITIONS, dynamic daily cap (1 healthy / 0 at -25%), post-exit cooldown state files → declarative risk.guard_rails. state/trade-counter.json is vestigial in v4.0 (only the last_entry_ts field stays for legacy log compat; cooldown enforcement moves to runtime).
6. Exit fee: DSL exits switched from MARKET (taker, 0.045%) to FEE_OPTIMIZED_LIMIT (maker-first, 0.015%, 60s ALO timeout, taker fallback). Entries keep ensure_execution_as_taker: false per v3.x patience rule.

⛔ CRITICAL AGENT RULES

RULE 1: Install path is /data/workspace/skills/condor-strategy/

RULE 2: THE PRODUCER DOES NOT EXIT POSITIONS — DSL only.

RULE 3: MAX 1 POSITION (the one amazing trade)

RULE 4: Verify runtime + daemon on every session start

RULE 5: Never modify scoring or gate parameters

RULE 6: MAX 1 ENTRY per 24h (runtime max_entries_per_day=1)

RULE 7: 120-min post-exit cooldown (runtime per_asset_cooldown_minutes=120)

RULE 8: HARD_STOP circuit breaker at -25% drawdown (runtime drawdown_halt_pct=25)

Thesis

Pure trend continuation, never counter-trend.

From Kodiak's lifetime top 3 winners (all SOL, +$133 / +$87 / +$78):

"The absolute highest predictor of a massive directional swing is when the 4H, 1H, 15m, and 5m price momentum are perfectly unified in a single direction, AND the Smart Money leaderboard is heavily lopsided (>65% directional consensus) in that exact same direction."

From Wolverine's HYPE SHORT post-mortem (-$160 loss, 2026-04-16):

"We stepped in front of a runaway 32% freight train to catch a 1% micro-retrace. The historical winners traded with the massive macro shift, not against it."

Condor enforces both insights as hard gates.

Hard gates (all must pass — fail = skip asset)

1. Not XYZ, not stablecoin
2. OI > $1M USD (context-aware read)
3. trader_count >= 50 (signal validity)
4. 3TF ALIGNMENT — 4h_price + 1h_price + 15m SM velocity all aligned in entry direction, each clearing magnitude threshold (4h ≥ 1.0%, 1h ≥ 0.3%, 15m velocity ≥ 0.1)
5. MACRO TREND GATE — if |4h_move| > 10% in OPPOSITE direction of entry, BLOCK
6. SM consensus >= 70% in entry direction (v3.4 calibration)

Scoring (max ~18 pts, MIN_SCORE = 12 since v3.4)

Signal	Points
4h move magnitude: >2% (+2), >4% (+3), >6% (+4)	1-4
1h confirmation: >0.5% (+1), >1% (+2)	1-2
15m SM velocity: >1.0 (+1), >2.0 (+2)	1-2
3TF_ALIGNED bonus	+3
SM consensus: >=70% (+2), >=75% (+3), >=80% STRONGLY_TILTED (+4)	2-4
trader_count >= 100 (DEEP_CONSENSUS)	+1
Funding pays direction	+1
BTC macro confirms (aligned, >1.5%)	+1
Peak session (13-19 UTC or 00-05 UTC)	+1

Position sizing (score-scaled, 10x leverage cap)

Score	Leverage	Margin
11-12	10x	50% of equity
13-14	10x	70% of equity (HIGH)
15+ (APEX)	10x	80% of equity

Never above 10x leverage. Both Kodiak and Wolverine confirmed 10x is the empirical ceiling — above this, fees + wick vulnerability destroy the edge.

Leverage auto-clamped to per-asset Hyperliquid max via strategy_get_asset_trading_limits.

Producer computes marginUsd from score tier; signal data carries it; LLM gate passes through to runtime payload.

DSL preset (mid-beta default, preserved from v3.3+/v3.4)

Mechanism	Value	Rationale
hard_timeout	1440 min (24h)	One trade per day discipline
weak_peak_cut	DISABLED	v3.3: apex thesis needs 4-8h to develop; 2h @ 3% was pre-empting
dead_weight_cut	60 min	No movement = dead
Phase 1 max_loss	20%	Same level that killed Wolverine's HYPE SHORT — but that was counter-trend. Kodiak's SOL trades didn't hit Phase 1.
Phase 1 retrace	8%
Phase 2 tier 1	+8% / 30% HW lock	Kodiak's Tier 1
Phase 2 tier 2	+15% / 50%	Tier 2
Phase 2 tier 3	+25% / 70%	Kodiak's #3 exit tier
Phase 2 tier 4	+40% / 85%	Kodiak's #1/#2 exit tier
Phase 2 tier 5	+60% / 90%	Infinite trail on monster winners
Phase 2 tier 6	+100% / 94%	Monster-class winners

Risk gates (runtime.yaml risk.guard_rails)

Gate	Setting	Replaces
max_entries_per_day	1	v3.x dynamic daily cap
per_asset_cooldown_minutes	120	v3.x POST_EXIT_COOLDOWN_MINUTES
daily_loss_limit_pct	15	new (fleet-standard)
max_consecutive_losses	3	new (fleet-standard)
cooldown_minutes (post-loss)	120	new (fleet-standard)
drawdown_halt_pct	25	v3.x get_dynamic_daily_cap() hard-stop
drawdown_reset_on_day_rollover	false	fleet-standard (Roach lesson)

Hardcoded constants (not configurable)

• MAX_LEVERAGE: 10
• MIN_OI_USD: 1,000,000
• MIN_TRADER_COUNT: 50
• XYZ_BANNED: true
• STABLECOINS_BANNED: USDT, USDC, DAI, USDE, FDUSD, TUSD, BUSD
• UNIVERSE_SIZE: 50

Hard rule for user-conversation Claude sessions

User-conversation Claude sessions MUST NOT call: create_position, close_position, edit_position, ratchet_stop_add, ratchet_stop_edit, ratchet_stop_delete, cancel_order, strategy_close, strategy_close_positions.

These tools are reserved for the producer daemon (entry path) and the DSL ratchet engine (exit path). User-conversation sessions are read-only.

Operator install

See README.md for fresh-install + migration commands from v3.4.

Changelog

v4.0.0 (2026-05-12) — plumbing-only migration

Migrated to helpers-native (senpi_runtime_helpers). Thesis preserved verbatim from v3.4.

v3.4 (2026-05-05) — gate calibration fix

• v3.2 over-tightening produced ZERO trades in 13 days. MIN_SCORE 13→12, MIN_SM 75→70, MIN_15M_VELOCITY 0.2→0.1.

v3.3 — disable weak_peak_cut

• 2h @ 3% ROE was pre-empting apex thesis development.

v3.2 (2026-04-22) — gate tightening (reverted in v3.4)

v3.1 (2026-04-16) — BTC hard-gate removed

• BTC macro alignment moved from hard gate to scoring bonus only (HYPE/HIP-3 decoupling).

v3.0 (2026-04-16) — COMPLETE REWRITE

• Thesis flipped: multi-asset thesis picker → single-position trend-continuation apex sniper
• Universe expanded: 4 majors → top 50 HL assets by 24h volume
• Added 3TF_ALIGNMENT hard gate + MACRO_TREND_GATE
• Sizing scales with score; leverage hard-capped at 10x
• DSL profile calibrated from Kodiak's SOL winner exit tiers
• ONE TRADE PER DAY discipline

License

MIT — Built by Senpi (https://senpi.ai).

Skill Attribution

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