Skip to main content
The incubation pipelineBacktest & paper · zero real money

Strategies audition in public — and die in public.

The Graveyard is where hypotheses go once they fail. This is the room next door — everything currently on trial for the blend. Each candidate ships with the exact gate it must clear, written before we looked at a result, and its honest status today: soaking, validated but held dark, blocked on data, or still recording. Everything here is a backtest or a paper measurement. There is no real-money record.

How the audition works

From hypothesis to the model — or to the Graveyard

A strategy earns a place in the published blend by clearing one fixed, pre-registered gate — never by looking promising. Until it does, it lives here at one of four honest stages, and nothing speculative touches the model.

Live · soaking

Passed its gate and running on a paper account, accruing a forward record before any real capital.

Validated · dark

Cleared its gate in backtest but shipped flag-off — behavior-identical to the live model until proven twice.

Awaiting data

Built and self-tested, but the decisive run needs data we do not yet have. Result: not published.

Recording

A shadow book logging observations. No verdict may be read until the pre-set sample size is reached.

On trial right now

Four strategies auditioning for the blend. For each: the thesis, its pre-registered gate, the measured result so far, and where it actually stands. Where a number was never measured, it says so.

Live · soakingPAPERthe product · #54

Trend blend — index core + trend sleeve

The one strategy that cleared its own gate and became the product. A broad-index core plus a sleeve that holds ETFs only while they trend.

Pre-registered gate: Graduated. It IS the baseline every other candidate is measured against — Sharpe 0.89, maxDD −33.8%, ~9.4% CAGR (backtest 2005–2026, unlevered).
Where it stands: Running on a paper account since June 2026. The first rebalance at 1.0× leverage executed on 1 Aug 2026 — 12 of 12 orders filled — so from that date the paper record matches the published 60/40 model exactly. Next rebalance 1 Sep 2026; 14 of the 15 sleeve ETFs are currently trending up. Forward-soaking now; the earlier leveraged weeks are quarantined and never counted.
LIVE · paper — backtest 2005–2026
Validated · darkBACKTEST#51 · TREND_VIX_SCALER

VIX term-structure stress scaler

When the VIX term structure stays backwardated, cut the index core from 60% to 30% until it normalizes; the freed weight earns T-bills. The best-replicated signal in the alt-data review.

Pre-registered gate: Passed every clause: Sharpe 0.893 → 0.971, maxDD −33.8% → −25.3%, inside the trade budget, and no sub-period materially worse.
Where it stands: Shipped dark. It only fires in a genuine backwardation / stress episode, and the tape has been in contango — so it has never once fired. Awaiting a real stress window to soak against before it can go live.
BACKTEST — validated, held dark
Validated · darkBACKTEST#48 · TREND_BONDVOL_SCALER

Bond-vol × VIX combined scaler

Add a bond-volatility arm to the VIX scaler to catch bond-led stress — 2022 being the obvious case — that an equity-vol-only signal misses.

Pre-registered gate: Beat VIX-only: Sharpe 1.034, maxDD −19.0% (+0.062 Sharpe), improving exactly where the thesis predicted — 2022.
Where it stands: Held dark, blocked on data. It was tested on a free realized-vol proxy for 10y yields, not true ICE BofA MOVE (proprietary); and the circular-shift timing null came back at p ≈ 0.06, so we cannot yet claim the timing is what works. Needs true MOVE data before adoption.
BACKTEST — validated on a proxy; blocked on MOVE data
Awaiting dataBACKTESTTranche 2

Futures-tier trend sleeve

Run the trend logic on back-adjusted continuous futures — long AND short, the real managed-futures edge the long/flat ETF sleeve cannot reach.

Pre-registered gate: Net Sharpe must beat the ETF blend (0.89) by ≥ 0.10 AND stay stable across sub-periods.
Where it stands: not published. The harness is built and self-tests end-to-end, but a decisive run needs paid back-adjusted daily bars (Databento / Norgate). The free-data run is inconclusive by design — a preliminary read only.
not published — built, not yet run
The house gate

One gate, written before the result

Every candidate above is judged against the same four clauses, and each was written into the test script before a single line of output was read. A strategy clears the bar it set for itself, or it stays here.

Sharpe ≥ baseline + 0.05

A candidate must lift the full-period Sharpe from the blend's 0.89 to at least 0.94 — a real step, not noise. An improvement inside the ±0.05 band is treated as zero and does not promote.

Max drawdown ≥ 3 pts better

Peak-to-trough loss must improve by at least three points versus the −33.8% baseline, and be worse on neither Sharpe nor drawdown. A scaler that buys return by deepening the drawdown fails.

Turnover ≤ +6 round-trips / yr

A promotion may add at most six extra round-trip trades a year. Edge that only shows up by trading far more is edge eaten by cost, and it does not count.

Stable across three sub-periods

The improvement has to hold in each of 2005–2012, 2013–2019 and 2020–2026 — not be carried by one regime. A result that flips sign between halves of the sample is killed, not shipped.

The pipeline at a glance

Every candidate, its basis, its stage, and the honest one-line label. The same table the desk reads.

StrategyBasisStageHonest label
Trend blend — index core + trend sleeve
the product · #54
PAPER Live · soakingLIVE · paper — backtest 2005–2026
VIX term-structure stress scaler
#51 · TREND_VIX_SCALER
BACKTEST Validated · darkBACKTEST — validated, held dark
Bond-vol × VIX combined scaler
#48 · TREND_BONDVOL_SCALER
BACKTEST Validated · darkBACKTEST — validated on a proxy; blocked on MOVE data
Futures-tier trend sleeve
Tranche 2
BACKTEST Awaiting datanot published — built, not yet run

The moat is the discipline, not the secret

Anyone can claim an edge. What is hard to copy is a room where every hypothesis carries the gate it must clear, the winners graduate to the model in public, and the losers get a headstone. Watch both ends of it.

Read this honestly:every figure on this page is a backtest or a paper measurement, and where a strategy has not been run the result reads “not published.” Backtested and paper results are hypothetical, benefit from hindsight, and are not a promise of future performance. There is no real-money track record. Uptogain publishes one public model for everyone — it is not personalized advice, and we neither manage your money nor take custody of it.