A timeframe change is a new test

A timeframe is the length of one bar. Changing that length changes the clock that every lookback, every stop distance in bars, and every entry count is measured on. The strategy file can stay the same. The test is not the same.

The interval is a listed length

What a candlestick shows is the four prices inside one bar. This page is the length of that bar. Binance's spot WebSocket stream document lists kline intervals that include 1h and 4h. Those labels are durations. They are not two views of one series.

Tradee stores three candle tiers: 1m, 1h, and 1d. A 4h series is built from hourly candles. The 4h close is not a 1h close drawn larger. A forming candle is not a closed candle still applies: the current 4h bar can change until that four-hour interval ends.

Twenty bars, two clocks

These figures are a worked example, not a recorded run. The lookback length is 20 bars on both sides.

window hours = bar count x hours per bar

20 x 1 = 20 on the 1 hour chart

20 x 4 = 80 on the 4 hour chart

The panel shows 20 hours beside 80 hours. Trade count usually falls on the longer bar, because fewer bars close in the same calendar range. Holding time in bars can look similar while holding time in hours grows. Neither change is a defect. It is the experiment you just ran.

A bot instance binds its own timeframe. Two instances of one strategy version, one on 1h and one on 4h, are two tests that share a rule file. Strategy, bot, and execution is the split between those objects.

What Tradee will not mix

Every source and indicator node has a timeframe setting. The default is match, which follows the running instance. You can point a node at a coarser interval, so a 4h average can feed a 1h instance. You cannot point it at a finer interval. A 1h node on a 4h instance raises an error before any candle is processed, because the 1h close inside a still-open 4h bar would be a partial reading.

When a coarser series is allowed, Tradee only exposes a higher-timeframe value on the running bar whose close is at or after that higher bar's true close. The 4h average on a 1h instance therefore steps once per four hours, at the 4h close, rather than leaking the forming 4h bar into earlier 1h decisions during a backtest.

The learn path's second backtest is the mechanical next step: same asset, different timeframe, then compare trade count, holding time, and equity shape. Treat a better number on the second run as a new sample, not as proof that the first run was confirmed.