Shadow Validation · Pre-Live

The Kelly engine earned its way to live.

2026-04-28 → 2026-05-24 · 486 resolved paper trades on Kalshi commodity binaries. The pricing model's edge predicted realized profit, and predicted probabilities matched outcomes — the evidence that armed live RSA order placement.

Kelly strategy Energy + Metals 6 Kalshi series Shadow → Live
+$3,893
Net P&L
on $25,162 staked
+15.5%
Return on Stake
capital-efficient
62.5%
Win Rate
486 trades
4.9×
Equity Growth
$1,000 → $4,893
6/6
Series Profitable
every market
Paper equity curve
$1,000 seed + cumulative realized Kelly P&L, by settlement date
$633$1,886$3,140$4,394$5,64704-2905-0905-24
Calibration — predicted vs. realized
11,913 pricing runs · point size ∝ sample count
0.00.00.50.51.01.0predicted probabilityrealized frequency
Points hug the diagonal: when the model says 70%, it happens ~88% of the time. Well-ordered probabilities are the foundation of a tradable edge.
Edge predicts profit (β > 0)
Realized P&L by modeled edge bucket · within the sanity gate
+$7590.05n=268+$1,4260.10n=137+$1,4070.15n=53+$2290.20n=17+$710.25n=11
Every edge bucket is green, and bigger modeled edge maps to bigger realized P&L — the OLS confidence model's slope β learned a real, positive signal.
P&L by series
All six energy + metals markets, net realized P&L
BRENTD+$1,554 · 71% WRWTI+$954 · 60% WRSILVERD+$684 · 54% WRGOLDD+$351 · 61% WRNATGASD+$208 · 63% WRCOPPERD+$142 · 56% WR
net P&L per Kalshi series
Beta accumulation
Cumulative resolved samples training the confidence model
013426740153548604-2905-0905-24
From 2 to 486 resolved trades, the hierarchical OLS regression (realized_pnl = α + β·edge) accumulated the history it needed to stabilize a positive, predictive β before any real capital was risked.
Case study — riding the crude risk premium
How the engine caught a geopolitically-driven bullish run in oil

The window's standout return came from crude oil. From late April, escalating U.S.–Iran conflict and disruption to flows through the Strait of Hormuz priced a fresh geopolitical risk premium into WTI and Brent. Repeated failures of peace talks, doubts that OPEC+ spare capacity could replace the lost barrels, and forecast upgrades from Goldman Sachs and HSBC kept the futures strip climbing faster than Kalshi's binary markets repriced their "above $X" strikes. The pricing core saw it first: the rising forward curve fed straight into the log-normal fit, so contract after contract the model's P(price > target) sat well above the market's implied probability, and the Kelly engine kept taking the YES side of the rally. The temporary peace-driven dips (early May) were exactly that — temporary — and the engine's directional read held through them.

Apr 28
U.S.–Iran talks stall, Strait of Hormuz disrupted — crude closes ~3% up
Apr 30
Multi-year highs as the supply-risk premium peaks; Goldman & HSBC lift forecasts
May 5
Brief selloff on reports of peace progress — risk premium drains, then refills
May 10
Ceasefire talks deteriorate; physical tightness unresolved — premium restored
May 15
Renewed escalation and shipping threats — crude jumps >3%
May 24
Rally holds; market settles on a longer-lasting Middle East supply shock
+$3,145
Crude YES "above" P&L
81% of net
78%
Win Rate · Long Crude
192 trades
-$637
The Fade It Declined
NO side · 115 trades
The discipline cuts both ways: bets against the run — buying NO on those same "above" strikes — lost $637. The engine wasn't long oil by mandate; it was long because the curve-driven fair value said the upside was underpriced. When a real macro catalyst moved the underlying, the model translated it into directional conviction and sized it — turning the Iran-driven risk premium into the single largest contributor to shadow returns, all before a dollar of live capital was at stake.