Switches combine (AND) when more than one is on. An asset with no rule for a given gate passes through unfiltered on that dimension rather than being dropped.
Combined bankroll over time
Since 2026-09-16, when the sim-tracked markets in this portfolio started being measured -- copper's own real trading history goes back further (see its own section below), but combining that with the others' shorter history would understate how new most of this portfolio is.
The 0.85 favorite strategy on these three markets added together. Starts 2026-09-23 19:45 UTC, when Platinum began trading, with each market entering at the bankroll it had at that moment. Copper runs as a live paper bot; Platinum and Natural Gas are paper-simulated from the book watcher.
Combined bankroll over time
—
The same fast maker as below, placing real post-only orders, 1 contract a side, from AWS Ohio. Run in short supervised sessions, not continuously. Figures come straight from the Kalshi account (balance, fills and settlements), so they include settlement results, not just spread.
Account value after each settled 15-minute period.
Recent periods
| Period ending (UTC) | Markets | Fills | Contracts | PnL |
|---|
Quotes both sides at the best price on 8 crypto markets (all but BTC), 10 contracts a side, and pulls the side in the path when the coin jumps on Coinbase. Paper only: fills are decided from real Kalshi trades with the same queue model as the simulation. Two identical copies run at different speeds: AWS Ohio (~15 ms, next to Kalshi) and New York (~30 ms, our droplet). The simulation's out-of-sample result was about +$9.50 per 15-minute period at 15 ms and +$6.75 at 30 ms. "Spread earned" values leftover contracts at the mid when quoting stops.
AWS Ohio (~15 ms)
Running total of spread earned net of the measured live penalty (0.42c per contract), one point per 15-minute period.
New York droplet (~30 ms)
Running total of spread earned net of the measured live penalty (0.42c per contract), one point per 15-minute period.
Stopped on 2026-09-29 after 255 trades: won 76.1% while paying 76.3¢ on average, so no edge (−$33.82 on $100). It briefly looked positive, then faded to zero, matching the 11-day historical replay. Results below are final.
Crypto only (BTC, ETH, XRP, SOL, DOGE, HYPE, ZEC, BNB, NEAR). If the favorite is priced 60–85¢ with 50, 40, 30 or 20 seconds left, buy it at the next quote. Hypothesis: crypto settles on a 60-second average, so late favorites are underpriced. Forward-tested since 2026-09-26 with no backfill. It only works if the win rate beats the average price paid, so edge below is win rate minus price. About 1,100 trades are needed to tell a 2.5¢ edge from luck.
Bankroll over time
Starting from $— paper capital, the same as the other simulated markets. 10 contracts per trade, net of Kalshi fees. Each point is one settled trade.
By asset
Each asset alone is far too small to read; this shows whether one asset is carrying the total.
| Asset | Trades | Win rate | Avg price | Edge | PnL |
|---|
Control panel: every checkpoint, crypto vs commodities
Logs every 60–85¢ favorite at 120s through 10s before close, not just the first one per window, so rows from the same window overlap. Commodities settle on a single closing price with no averaging, so they are the control. If the averaging idea is right, crypto edge should grow toward the close and commodities should show none.
| Time left | Obs | Win rate | Avg price | Edge | PnL |
|---|
| Time (UTC) | Window | Side | Entry | Size | Status | PnL |
|---|