Chameleon Strategy

🦎 CHAMELEON v1.0.0 — Relative-Value / Pairs (Ratio Mean-Reversion)

Trade the spread between two coins, not the market. When ETH gets expensive relative to BTC (or SOL vs ETH), the ratio stretches — and ratios revert far more reliably than outright prices. Chameleon measures how stretched a pair's ratio is (z-score) and bets on the snap-back.

Why this strategy exists

Correlated majors trade as a pack, but their ratios oscillate around a mean. A 3-sigma ETH/BTC extension is a much cleaner edge than guessing ETH's absolute direction — the pair's correlation does the work. This is the first fleet agent to trade relative value rather than a single asset's trend, funding, or order flow.

Single-position note: a textbook pairs trade is two legs (long one, short the other) for true market-neutrality. The Senpi runtime is single-position, so Chameleon instead takes a directional bet on the high-beta leg in the reversion direction. It captures most of the reversion edge without needing spread-level position management.

CRITICAL RULES

RULE 1: The z-score is the gate

For each pair (numerator/denominator), Chameleon computes the latest ratio's z-score vs its mean/std over lookbackBars 1h candles. Entry requires |z| >= zEntryMin (default 2.0). No extension, no trade.

RULE 2: Which leg, which direction

It trades the configured high-beta leg (the more volatile asset — ETH in ETH/BTC, SOL in SOL/ETH). Direction profits from reversion:

• z high (numerator rich): leg == numerator → SHORT; leg == denominator → LONG
• z low (numerator cheap): leg == numerator → LONG; leg == denominator → SHORT

RULE 3: Reversion must be starting

A bonus, not a gate, but important: the position only scores its "turning" point when |z| this bar is smaller than last bar — i.e., the ratio is already snapping back, not still extending. Don't catch a ratio mid-blowout.

RULE 4: Producer enters. DSL exits.

No close_position call site. A ratio reversion is a bounded move, so the DSL is the mean-reversion preset — tight ladder (lock 30% at +5%), 48h hard_timeout + weak_peak_cut. Bank the snap-back; don't hold for a trend.

How Chameleon scores a trade

Gate: |z| >= zEntryMin on at least one pair.

Score components (max ~7):

| Signal | Points | |---|---| | |z| >= zEntryMin (gate-confirmed) | +2 | | |z| >= zStrong (default 3.0) | +2 | | Ratio turning (reversion starting) | +1 | | Smart Money on the leg confirms the direction (≥ smTiltMinPct) | +1 | | Leg volume rising (> 10%) | +1 |

Floor: minScore: 4.

DSL preset (mean-reversion — bank the snapback)

| Phase | Component | Setting | |---|---|---| | Phase 1 | maxlosspct | 15% | | Phase 1 | retracethreshold | 6 | | Time cuts | hardtimeout | 48h | | Time cuts | weakpeakcut | 120 min / 2% | | Time cuts | deadweightcut | DISABLED | | Phase 2 | T0 → T4 | +5/30 · +10/50 · +15/65 · +25/80 · +40/90 |

Scanner pattern

This introduces the Relative-value / pairs archetype — see senpi-trading-runtime/references/producer-patterns.md. Primary MCP calls: market_get_asset_data (1h candles for both legs of each pair), leaderboard_get_markets (SM on the leg). The pure functions (ratio_zscore, reversion_direction) are unit-tested in tests/test_signal.py.

Operator install

See [README.md](README.md).

Changelog

v1.0.0 (2026-05-22) — initial release

First fleet agent to trade relative value (ratio mean-reversion) rather than a single asset's direction. Introduces the relative-value/pairs archetype. Built with the mean-reversion DSL class (tight ladder + time-cuts on), taker-fallback entry (enter at the extreme; maker-only risks CREATEORDERRESTING and a missed reversion window), no null numeric signal fields, and unit-tested pure functions.

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.

Source & license

This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code.

• Author: Senpi-ai
• Source: Senpi-ai/senpi-skills
• License: MIT

Install and usage instructions live in the source repository linked above.