Backtesting.

TradingView's own migration guide confirms it: integer division and a flipped margin default change a script's numbers with zero compile errors.

Original walk-forward data on BTC and SOL showing that trading edge is monotonic in timeframe: daily produces real modest edges, 4H is thin to zero, and 1H is uniformly negative once retail fees are charged.

Your equity curve looks perfect because a broker emulator guessed the inside of every candle. The documented assumptions that inflate a backtest.

Thirty trades cannot tell a winning system from a losing one. The arithmetic that sets your real sample size, and why every variant you tried raises it.

AI assistants are good at producing Pine Script that compiles. They are bad at producing Pine Script that does what you meant. Here is the workflow we use to keep that gap from costing real money.