A single-instrument, factor-based strategy on Gold futures (1OZ, COMEX). Three alpha factors chosen by information-coefficient screening, validated walk-forward, and wrapped in a risk rulebook built for a funded-account evaluation.
Real-time paper-trading results, published unedited — the research and architecture behind them follow below. Holdings are disclosed as portfolio weights.
One instrument. Three factors. Zero improvisation.
Depth beats breadth. ProjectA began life as a multi-instrument signal system — BTC order flow, forex carry, cross-asset spreads. That architecture was shelved for something harder to do well: a single market, understood deeply, traded only when independently validated factors agree.
The three alphas earned their place through IC screening: COT commercial positioning (the dominant signal — what hedgers are doing in the futures market), gold carry via the rate of change in real yields, and DXY momentum. Instrument choice was itself a research decision — Micro Gold was rejected because contract-sizing math against the account rules produced zero viable trades.
Validation caught what sloppy pipelines miss: a COT parser silently corrupting data by matching the bare word “GOLD” across unrelated contracts, and a release-lag timing bug that failed under realistic conditions. Both fixed before a single paper trade — that is the point of the process.
Final validated configuration: a 0.67/0.33 COT/DXY factor blend, entry-alpha threshold 0.3, with the calendar filter, HMM regime-stability gate, partial-profit taking, and regime-conditional exits all enabled. This is the fitted-regime-model read — the most conservative, most honest number the validation produced. Every gate for the Phidias evaluation cleared with margin (mean OOS Sharpe > 1.5, worst window > −0.5, ≥6/8 positive windows, WF efficiency > 0.70, zero account failures). Hypothetical, walk-forward simulated performance; does not represent live trading with real capital.
Factor pipeline
Weekly COT commercial positioning, real-yield carry, and DXY momentum are computed on a validated data pipeline with release-lag handling — signals only ever see data that existed at decision time.
IC-screened signal stack
Factors survive on measured information coefficient, not narrative. COT positioning carries the most weight; carry and dollar momentum complete the ensemble.
Semi-automated execution
Deployed on AWS EC2 (Ubuntu). An automated job fires every weekday at 4:15 PM CT, sizing and generating each entry/exit signal — Phidias Premium doesn't permit fully automated order routing, so the job alerts and the trade is placed by hand. The equity feed on this page updates once that trade is confirmed.
Constraint-first risk engine
Every signal is checked against the rulebook before it exists. The strategy fires roughly 2–3 times a month; patience is enforced by code.
EOD trailing drawdown — $2,500
Trails the peak end-of-day balance only, never the intraday mark, per Phidias Premium's EOD-basis rule. An internal alert fires at $2,000 and new entries block at $2,375 — inside the firm's hard $2,500 limit, since breaching it fails the evaluation outright.
Profit target — $4,000 (8%)
The pass threshold on the 50K account. Progress is tracked continuously and feeds directly into the sizing logic.
Contract cap — 100 micro / 10 mini
Phidias Premium's own position-size ceiling on gold futures; the strategy's own sizing rarely approaches it.
CVaR ceiling — $200 per trade
Tail risk is bounded directly: expected loss in the worst-case scenarios is capped at $200 per trade, about 8% of the full drawdown buffer.
Overnight & weekend holds allowed
The reason for the move from Apex 25K to Phidias 50K Premium — Apex flattened every position by the close and forbade holding through the weekend; Phidias allows both, which the gold factors need to express fully.
No daily loss limit, no consistency rule
Two constraints Apex enforced that Phidias Premium doesn't — the tradeoff for giving up fully automated order routing in exchange for a materially looser rulebook.