The thesis
momentum-code runs a simple, legible idea: on each scheduled run, buy the strongest positive movers in a 24-name large-cap universe (AAPL, MSFT, NVDA, JPM, XOM and peers), capped per position so no single name dominates. This is classic cross-sectional momentum, and its readability is a genuine strength — you can reason about why the book holds what it holds.
Backtest: strong headline, softer engine
Over 451 days the backtest returns +19.76% (10.6% CAGR), lifting a notional book to $11,976. But the quality metrics temper the headline. Sharpe is a modest 0.65 against a 20.48% max drawdown — roughly one unit of drawdown for each unit of annual return. Turnover near 93% on just 5 trades marks this as a concentrated, low-frequency book, so the outcome hinges on a handful of positions. Note the reported win rate of 0: with only buys and no closed round-trips in the sample, read it as "no realized winners recorded" rather than a strategy that loses every trade.
Validation: encouraging out-of-sample, but the gate said no
The walk-forward looks encouraging on the surface — all 4 folds positive (+15.38%, +3.27%, +12.18%, +13.71%), with fold Sharpes as high as 2.47 and a most-recent out-of-sample Sharpe of 1.93. Yet the validation gate failed. The reason lives in the deflated statistics: the Probabilistic Sharpe Ratio is a healthy 0.811, but the Deflated Sharpe Ratio — which penalises for the 6 trials searched — collapses to 0.338. Searching across trials inflates apparent skill, and once you discount for it, the edge is not statistically convincing. The gap between the per-fold Sharpes (often above 2) and the full-period 0.65 is a second warning: stitched end to end, the equity path endured a drawdown the individual windows never saw.
Live: fully invested and stalled
Live behaviour matches a fully deployed book. The executed trades were a late-May/early-June cluster — MSFT, HON, BAC, NVDA and XOM — and since then every scheduled run from 2026-07-06 through 07-13 has executed 0 orders and rejected 1–2, with cash pinned at $608.79. The paper account sits around $9,720 total, drifting modestly below its $10k start. In short, the strategy is currently holding, not trading, and can't add new positions without first freeing cash.
Verdict
momentum-code is a clean, interpretable momentum book with a respectable backtest and consistently positive out-of-sample folds — real points in its favour. But the failed validation gate, the trials-deflated Sharpe, the 20%+ drawdown, and a live account that has been flat-to-down while stuck fully invested are all reasons for caution. Promising raw material; not yet a validated edge.