albatross-strategy

Installation
SKILL.md

🐦‍⬛ ALBATROSS v1.0.0 — Multi-week Arena Conviction Mirror

Mirror the consistent winners, not the lucky-week winners. Albatross selects Arena leaders by a composite score that rewards multi-week persistence and penalizes one-shot lucky weeks.

Why this strategy exists

The Senpi Arena resets every Thursday. A trader can post a +98% weekly ROE from one good week and three flat weeks — and they'll top the leaderboard, but mirroring them is a survivor-bias trap.

Albatross corrects for this by pulling 4 weekly snapshots + 1 monthly snapshot and ranking by a composite that respects both magnitude and consistency:

conviction = 0.3 × monthly_roe
            + 0.7 × mean(weekly_roe across last 4 weeks)
            − 0.5 × stdev(weekly_roe)

A trader posting +30%/+25%/+20%/+15% across 4 weeks (mean +22.5%, stdev ±5.6%) scores conviction ≈ 14.0. A trader posting +98%/0%/-5%/-3% (mean +22.5%, stdev ±42.5%) scores conviction ≈ -5.8 despite the same mean. Albatross rejects the second; mirrors the first.

Installs
5
GitHub Stars
99
First Seen
May 29, 2026
albatross-strategy — senpi-ai/senpi-skills