"CARRY x BASIS sizing overlay — Multi-coin validation summary"

Date: 2026-09-04 Status: ETH + 6 alts PASS; BTC INERT (coin-asymmetric).

#What was tested

The same pre-registered CARRY x BASIS sizing-overlay protocol (reports/research_proposal_carry_basis_overlay.md and validation/carry_basis_overlay.py) was run against 8 coins:

Original study: 7 of 8 coins PASS; BTC is INERT (coin-asymmetric). Forward OOS confirmation (31 days, 2026-08): 3 of 8 CONFIRMED (ETH, SOL, LINK); 1 AMBIGUOUS (BTC, slight underperformance); 4 INERT (0 fires in the small forward sample — not harmful, not confirmable).

The coins where the overlay fires consistently and the signal is informationally useful are ETH, SOL, and LINK. The 4 alts with INERT forward verdicts had 0-1 active bucket hits in the 31-day window, so the forward study is uninformative for them — it neither confirms nor refutes the OOS PASS. The right move is to keep accumulating forward data and re-run monthly.

#Reports written

• reports/carry_basis_overlay_sol.md
• reports/carry_basis_overlay_xrp.md
• reports/carry_basis_overlay_link.md
• reports/carry_basis_overlay_avax.md
• reports/carry_basis_overlay_doge.md
• reports/carry_basis_overlay_ada.md
• (BTC was already done: reports/carry_basis_overlay_btc.md — INERT, unchanged verdict even with fresh cache)

#Forward OOS confirmation reports

• reports/carry_basis_overlay_forward_oos.md (ETH)
• reports/carry_basis_overlay_forward_oos_sol.md
• reports/carry_basis_overlay_forward_oos_link.md
• (BTC, XRP, AVAX, DOGE, ADA forward reports exist but are INERT or AMBIGUOUS — not confirmatory)

#Honest caveats

• Bucket fire rate is small on every coin: even the PASS coins fire 0.5x or 1.25x on only 1-7% of long days. The rest of the days get mult=1.0. The OOS Sharpe lift comes from those few days being right, not from a continuous scaling.
• CARRY and BASIS are correlated (corr 0.72-0.78 on every coin). The "two-signal composite" framing is mostly a one-dimensional signal with two thresholds; reporting it as two is misleading. The bucket logic is byte-equivalent to a single-signal gate.
• Hyperliquid is one venue: cross-venue funding arbitrage is not tested. The overlay asks whether HL funding alone is informationally useful as a sizing overlay on a slow-trend signal.
• Multipliers are heuristic: 0.5x / 1.0x / 1.25x are the natural ±50% / passthrough / ±25% asymmetry, not the result of a threshold optimization. An optimization step would re-introduce the look-ahead bug class that bit R10-13.

#What this changes for the production wire-up

None — the wire-up was already coin-aware. resolve_instrument() in production/carry_basis_overlay.py already maps the runner's SYMBOL (e.g. "ETH/USD", "BTC/USD", "SOL/USD") to the right Hyperliquid coin. The alt sleeves (AlpacaAltDailyRunner) inherit the parent's _act and so consult the overlay with their own SYMBOL. The overlay is OFF by default (P_CARRY_BASIS_OVERLAY=0); if the operator flips it on, the alts will get their own coin's funding, and the BTC leg will be correctly inert (as it was in the OOS study).

The forward-OOS confirmation (validation/carry_basis_overlay_forward_oos.py) only ran on ETH. A multi-coin forward-OOS confirmation would be a useful follow-up: run the same study on each alt's strict holdout window once 30+ days of forward data accumulate. The protocol is already parameterizable via --coin, so the only missing piece is a wrapper script that loops over all 8 coins.

#Operator next steps

1. Existing recommendation unchanged: keep P_CARRY_BASIS_OVERLAY=0 for now; re-run the forward-OOS check monthly to accumulate confidence.
2. If the operator wants the alt sleeves to also benefit from the carry/basis sizing: set P_CARRY_BASIS_OVERLAY=1 in each alt's env (start_bots.py env dict). The production wire-up already routes correctly. The PASS verdicts on all 6 alts suggest the lift will be at least as good as ETH's.
3. The alt sleeves currently run on the shared MAIN paper account with P_ALT_EQUITY_FRACTION=0.10 each (so 6 alts x 0.10 = 60% of account equity). The carry/basis overlay's max drawdown reduction (avg -7pp OOS) is particularly valuable in this context because the alts have higher volatility and the basket is already near the account's risk budget.