Backtest draft — no wallet, no engine
Delta-neutral funding harvest routed by regime: carrywhen trailing-7d funding > 10% APR, reversewhen < −10%, lend otherwise — money never sleeps. Build order: advisory monitor first (same signal, alerts only), carry engine when the first hot regime fires, reverse + lend automation after. Study: reports/2026-08-06/funding-arb-study.md. Venue reality: Drift is dead (hacked April 2026), so live candidates are Hyperliquid (deep, off-Solana) and Pacifica (Solana-native, young).
HL SOL full rotation
+11.2%/yr
2023–26, the venue we'd trade
Binance SOL full rotation
+13.7%/yr
2020–26 incl. the 2022 panic
Lend floor (benchmark)
+5.0%/yr
USDC money market, always-on
Current regime
OFF
cold — SOL funding negative YTD 2026
How it works
One pool of USDC, always in exactly one of three modes. The dial is a single number: the trailing 7-day average funding rate on the perp market — what leveraged traders are currently paying to hold their positions. The pool itself never takes a view on price: both trading modes hold a long and a short of the SAME asset at the same time, so a price move gains on one leg exactly what it loses on the other. The income is the fee stream, not the direction.
Carry — bull seasons
When funding > +10% APR: buy SOL/BTC spot, short the same amount on the perp DEX. Leveraged longs pay shorts an hourly fee — we are the short collecting it. 2021 paid ~30%/yr, 2024 ~24%/yr (gross, on notional). Spot leg can sit in JitoSOL for ~7% extra on that half.
Reverse — panic seasons
When funding < −10% APR (crypto panics: 2022 SOL hit −38%/yr — shorts were paying longs): flip the machine. Long the perp, short spot via a money-market borrow. Same delta-zero shape, now collecting the negative funding. Insurance that PAYS during crashes.
Lend — cold seasons (now)
Funding near zero (like right now): nothing worth harvesting, so the USDC parks in a Solana money market at ~5%/yr until the dial swings. Money never sleeps, it just earns the floor.
Historically the pool was DEPLOYED (carry or reverse) 40–70% of all calendar time, in stretches of weeks to months — the median hot stretch on Hyperliquid BTC lasted 128 days. This is a seasons business, not a sniper trade. All returns here are USD-on-capital with zero net crypto price exposure — not comparable to SOL-denominated yields like staking or INF, which ride the full SOL rollercoaster underneath.
The first four rows below are NOT four strategies— they are the SAME strategy measured on four data series (venue × coin): Hyperliquid is the venue we would actually trade (deepest funding, hourly, 2023–26), Binance is the longest history we can measure (8h funding, 2020–26, including the 2021 mania and the 2022 panic that reverse mode harvests). The fifth and sixth lanes ARE different strategies: our live SPYx machine's regime clock, and the rejected DOV put-selling candidate — kept on the board for contrast (find 2022-11 in the returns view: the rotation's best month is the put seller's worst). The overlap matrix below the timeline quantifies it.
Who earned what, when — monthly backtested returns
Each cell is one month's realized return from the backtests — hover for the number. Deep green/red saturate at ±5%/mo (outliers pin to the ends: that's the point).
−5%/mo+5%/moLanes are data-driven (lib/data/funding-rotation-sim.json) — when we draft the next regime strategy, its would-have-been deployment windows appear here as a new lane, and the overlap (or the gap) against these is the diversification verdict.
So what do we rotate on? (71 shared months, 2020–26)
Months won: crypto rotation 11 · DOV puts 49 · SPYx+GLD machine 11 — and monthly win-rate is the WRONG metric: the strategy that wins the most months is the one with the −19.6% craters. Judge the rows by CAGR and drawdown together.
| policy | CAGR | maxDD | worst month |
|---|---|---|---|
| SPYx Dip + GLD Reservelive — self-rotating internally (dip → ladder → GLD rungs); never switched off | +11.5%/yr | 17.3% | -7.9% |
| Crypto rotation (carry/reverse/lend)draft — rotates ITSELF on the funding dial; floor makes red months ~impossible | +14%/yr | 0.1% | -0.1% |
| DOV put-sellingno-build — SOL beta in a premium costume: fat CAGR is the 48x window, 2022 realized −11%/yr | +89.3%/yr | 26.6% | -19.6% |
| 50/50 machine + rotation, in parallelthe real answer — separate wallets, both always on | +13%/yr | 8% | -3.8% |
| Hop to last month's winnerwhat naive rotation does — mostly held DOV, inherits its beta AND its craters | +69.2%/yr | 19.8% | -17.4% |
The rotation doctrine this board argues for: rotate each sleeve on the DIRECT price of its own trade, never on macro proxies.The crypto sleeve rotates itself on the trailing 7d funding APR (the actual cash flow — TVL, Fear&Greed, BTC moving averages are lagging correlates of it); the machine rotates itself on price vs its 20d high; idle treasury follows the monitor's state. Between strategies, don't hop — hold the complementary pair and let each sleeve's own dial do the switching.
Raced as ONE movable pool (2020–26, sim_one_pool.ts): funding-dial fusion +15.7%/yrvs BTC-200dMA +6.9% vs Fear&Greed +5.3% — the direct signal beats every proxy 2–3x. But fusion's edge over 50/50 (+2%/yr) costs 3x the drawdown (23% vs 8.4%) and a −11.5% worst month: the single-pool constraint itself is what's expensive. If the pool must move, move it on funding; better, split it and don't.
Regime overlap — when the machine is stressed, is the rotation earning?
Each cell: share of shared calendar days the crypto rotation spent in that mode, GIVEN the SPYx machine's state. The diversification signal to look for: reverse share rising in the crash rungsrow — equity crashes coinciding with crypto funding panics means the rotation's insurance leg pays exactly when the machine digs in.
vs Hyperliquid SOL (806d shared)
| machine state | carry | lend | reverse |
|---|---|---|---|
| harvest | 63% | 28% | 9% |
| dipLadder | 31% | 46% | 23% |
| crashRungs | 11% | 53% | 37% |
vs Binance SOL (1476d shared)
| machine state | carry | lend | reverse |
|---|---|---|---|
| harvest | 49% | 42% | 8% |
| dipLadder | 29% | 56% | 15% |
| crashRungs | 5% | 71% | 23% |
Reading (2020–26): in normal times both earn together (machine harvest → crypto carry 49–76%). When the machine goes to crash rungs, carry mostly switches OFF (5–11% on SOL) and the sleeve is in lend or reverse — on SOL, reverse jumps from ~9% of days to 23–37%. Different clocks, complementary stress modes: the crypto sleeve's panic leg fires in the same weeks the GLD vault funds the rungs.
APR on capital, all costs charged
| series | Lend floor only | Carry-when-hot + lend | Full rotation |
|---|---|---|---|
| Hyperliquid BTC 2023–26 | +5.0% | +10.4% | +10.1% |
| Hyperliquid SOL 2023–26 | +5.0% | +11.5% | +11.2% |
| Binance BTC 2020–26 | +5.0% | +9.0% | +8.8% |
| Binance SOL 2020–26 | +5.0% | +8.8% | +13.7% |
Assumptions: 67% capture (short at 2x), 5% lend floor, 8% borrow drag in reverse, 9.5bp per flip, gate 10%/5% hysteresis on trailing-7d funding. Reverse mode is what turns Binance SOL — the worst always-on series in the study (+0.1%/yr, 4.3y frozen) — into the best rotation performer: 2022's panic paid it.
Same window, the SPYx machines (2020-01 → 2026-07, GLD-study backtest at live economics)
| SPYx dip machine (base) | +9.3%/yr | maxDD 32% |
| SPYx dip + GLD reserve | +13.5%/yr | maxDD 20% |
| SPYx dip + gold hold | +12.8%/yr | maxDD 19% |
| SPY buy & hold | +13.4%/yr | maxDD 33% |
| GLD hold | +15.4%/yr | maxDD 28% |
Read together with the rotation rows above: all USD-denominated, same clock. The machines and the rotation are complements, not rivals — the overlap matrix shows their stress seasons interlock, so the portfolio question is allocation between them, not either/or.
Gross funding by year (APR on notional, short receives)
| year | Hyperliquid BTC | Hyperliquid SOL | Binance BTC | Binance SOL |
|---|---|---|---|---|
| 2020 | — | — | +17.2% | -12.6% |
| 2021 | — | — | +30.7% | +28.6% |
| 2022 | — | — | +4.2% | -38.1% |
| 2023 | +13.6% | +11.7% | +7.9% | +1.3% |
| 2024 | +24.2% | +28.2% | +11.9% | +13.6% |
| 2025 | +10.6% | +5.3% | +5.1% | +0.4% |
| 2026 | +4.1% | -2.1% | +1.9% | -2.6% |
2021 and 2024 are the “15–40%” seasons the pitch remembers; 2022 is the panic reverse mode harvests; 2026 is the cold season we're in right now.