RESEARCH AND PRACTICE
Closed candles versus live candles: avoiding hindsight in crypto analysis
Understand candle confirmation, publication time and full-bar evaluation so a live crypto observation is not mistaken for a closed-candle forecast.
A candle is a changing summary until its interval closes. A chart saved after the close can therefore tell a different story from the information available when a signal appeared. Reliable analysis preserves the original observation and distinguishes the market interval, the arrival of its data and the publication of a decision. Without those distinctions, a backtest can quietly credit a forecast for a move that happened before anyone could see it.
Separate three clocks
The interval clock defines which trades belong to the candle. The feed clock describes when the venue produced or transmitted an update. The application clock records when that update arrived and when the resulting analysis was published. Those moments are related but not interchangeable. A calculation started before a candle closes may finish afterwards; another may use a closed candle that arrived late. Save the timestamps needed to reconstruct actual availability. Do not assign an earlier issue time merely because a batch, cache or database uses a rounded interval label as its identifier.
Treat confirmation as part of the data contract
Bybit's documented WebSocket kline message includes a confirmation field that distinguishes a closed candle from an open, updating one. Other feeds require their own documented rule; an identical-looking timestamp field is not sufficient evidence. If a strategy intentionally uses live candles, preserve their snapshots and evaluate that live rule. If it requires closes, exclude unconfirmed bars and apply a clearly stated delivery allowance. Replacing the original live values with the final high, low and close changes the experiment. It should be treated as a different dataset, not a harmless chart refresh.
A hypothetical full-bar evaluation timeline
Suppose an observation becomes available at 10:00:20. The high of the minute starting at 10:00 may already contain a move from before publication. A deliberately defined next-full-minute proxy can instead start at 10:01, using that minute's open as a separate evaluation entry. A 60-minute horizon then needs the 60 complete bars from 10:01 through 11:00 and ends at 11:01. This is a bar-based analytical convention, not proof of an executable fill. If publication slips beyond the planned start, the protocol must select the next eligible start rather than backdating the signal.
Make gaps and unfinished horizons visible
Before settling a result, verify that the required bars are closed, unique and consecutive. Sixty rows are not necessarily sixty consecutive minutes: a duplicate can hide a missing interval. Do not fill a gap with an assumed flat price when the missing interval could have touched a target or invalidation boundary. Distinguish pending, incomplete and evaluated outcomes. A pending horizon has not failed, and an incomplete horizon should not be quietly excluded from every reliability report. Record coverage alongside performance so periods with poor data cannot make an experiment appear selectively accurate.
Report the resolution of a threshold hit honestly
Minute OHLC data can show that a price level was inside a bar's range. It usually cannot reveal the exact second of the first touch or the order of two levels touched within the same bar. Record a bar interval or a deliberately chosen upper bound for elapsed time, and disclose the resolution. If both the target and invalidation level appear in one candle, apply a predeclared ambiguity rule rather than choosing the favourable sequence. Tick data may answer a finer question, but only when its completeness and relationship to the relevant market are also established.
Reproduce the decision before reviewing the result
For a sample of observations, reconstruct the visible inputs from their saved versions and verify that every input was available by publication. Then replay settlement with synthetic cases: a pre-publication high, a missing middle candle, an unclosed final candle and two barriers touched in one bar. The expected outcomes should follow the written protocol. Keep signal price, evaluation-entry proxy and any actual fill in separate fields. This makes a chart useful evidence while preventing a precise-looking label from implying execution quality or foresight that the data cannot support.
Sources and example scope
Sources support the definitions and mechanisms. Numerical scenarios are hypothetical teaching examples, not live prices, forecasts or reported HOSTuvo returns. Images are editorial illustrations.