Track records

Every cumulative return shown on pnl.xyz is paper-traded forward, not backtested.

A backtest is a claim about the past, made after the fact, by whoever is selling you the strategy. A track record is a claim the platform made before the outcome was known and cannot revise afterwards.

Fills walk the real order book

Paper trading here means real depth, synthetic fills: the order book comes from the venue’s live feed, the match is simulated, and no order is ever sent.

When visible depth cannot clear an order, the fill is rejected rather than filled at the mid.

That rejection is the whole point. A paper account that always fills at the mid flatters every strategy in exact proportion to how much size it uses — so the strategies it flatters most are precisely the ones that would have hurt most in production. Walking the book instead makes a published curve capacity-aware from its first day, which is what stops a launchpad becoming a leaderboard of strategies that only work at $10k.

Testnet runs are never shown as track records

Testnet and mainnet have separate order books, and a testnet book is thin and largely market-made by the venue itself. Walking one measures a fiction — the capacity property that is the entire reason for book-walking fills is simply absent.

So testnet runs exercise the pipeline and are not evidence about a strategy. Every record carries which network produced it, and only mainnet records can be published. The check lives on the data rather than in anyone’s memory, because once a curve is drawn the two are indistinguishable.

The order is gate, then paper, then listed

Paper performance does not substitute for the gate. A few weeks of live paper is a smaller sample than the holdout, and an algo that looks good over a few weeks by luck is exactly what the gate’s discipline dimension — the deflated Sharpe — exists to price.