🐦 CUCKOO v1.0.0 — Copy-the-Copiers (Meta-Strategy Follower)

Let the best strategies do the work. A cuckoo lays its eggs in other birds' nests. Cuckoo lets the platform's top-performing strategies find the trades, then rides whatever they agree on most — weighted by how well each one is actually doing.

Why this strategy exists

Following a single trader (Remora) or a leaderboard universe (Raptor/Jackal/Spider) is one layer. Cuckoo is a layer above that: it follows the top-performing strategies — which are themselves copy/algo/trader-following strategies — so it captures the consensus of the consensus. When the best strategies on the platform are independently piling into the same asset+direction, that agreement is a strong, self-cleaning signal: underperformers fall out of the top-N automatically, so the pool refreshes toward whatever is working.

CRITICAL RULES

RULE 1: Auto-discovered, not hand-picked

Cuckoo pulls the top topN strategies by realized performance every tick (discovery_get_top_strategies). There is no operator whale list (that's Remora) — the pool is discovered and refreshes as performance changes.

RULE 2: Performance-weighted consensus

Each top strategy casts one vote per (asset, direction) it holds above minNotionalUsd, weighted by performance_weight(roi) = clamp(1 + roi/50, 0.5, weightCap). A flat strategy weighs 1.0; a +100% strategy hits the cap; a losing strategy floors at 0.5. The cap stops one outlier from dominating.

RULE 3: Consensus is the gate

Votes are tallied into (asset, direction) candidates. Entry requires at least minStrategies (default 2) top strategies in agreement — one strategy's position is never enough.

RULE 4: Producer enters. DSL exits.

No close_position call site. The top strategies hold conviction positions, so the DSL is the let-winners-run preset — wide ladder (T0 +10% / lock 0), max_loss 18%, time-cuts OFF except a 96h hard_timeout staleness cap. ⚠️ Cuckoo re-evaluates the consensus each tick but does not yet mirror an EXIT — the hard_timeout prevents holding a stale consensus indefinitely. Exit-mirroring is a planned v1.1 enhancement.

How Cuckoo scores a trade

Gate: >= minStrategies top strategies hold the same (asset, direction).

Score components (max ~6):

Signal	Points
Held by at least one top strategy	+2
Consensus: 2 strategies agree	+1
Consensus: 3 strategies agree	+2
Consensus: 4+ strategies agree	+3
Aggregate consensus weight ≥ highWeight (default 6.0)	+1

Floor: minScore: 4 (so a 2-strategy consensus needs the high-weight bonus, or 3+ strategies, to clear).

DSL preset (let-winners-run — with a staleness cap)

Phase	Component	Setting
Phase 1	max_loss_pct	18%
Phase 1	retrace_threshold	10
Time cuts	hard_timeout	96h (staleness cap)
Time cuts	weak_peak_cut	DISABLED
Time cuts	dead_weight_cut	DISABLED
Phase 2	T0 → T4	+10/0 · +20/45 · +35/65 · +55/78 · +90/88

Scanner pattern

Introduces the Meta-strategy follower / copy-the-copiers archetype (#14) — see senpi-trading-runtime/references/producer-patterns.md. Primary MCP calls: discovery_get_top_strategies (the producer signature; defensive multi-key unwrap of the strategy list), leaderboard_get_trader_positions (each top strategy's positions; unwraps the nested data.positions.positions shape). The pure functions (performance_weight, mirror_direction, tally_consensus, consensus_score) are unit-tested in tests/test_signal.py.

Operator install

See README.md.

Changelog

v1.0.0 (2026-05-26) — initial release

First fleet agent to follow the auto-discovered top STRATEGIES (not individual traders) and trade their performance-weighted consensus — the copy-the-copiers meta-layer, a new archetype. Let-winners-run DSL class (wide ladder, time-cuts off except a 96h staleness cap), taker-true entry, no null numeric signal fields, defensive multi-key shape unwrapping on discovery_get_top_strategies + leaderboard_get_trader_positions, and unit-tested pure functions. Exit-mirroring is a planned v1.1 enhancement.

License

MIT — Copyright 2026 Senpi (https://senpi.ai).

Skill Attribution

When creating a strategy, include skill_name and skill_version. See references/skill-attribution.md.