Horizon learning guide · English
Why Forecast Timing Matters: Early vs. Late Candle Predictions
Understand why early and late market forecasts need separate scoring, and how to keep a fair record when readings update during a candle.
The same candle can contain different forecasting tasks
A candle summarises price activity over a fixed interval. Near its opening, most of that interval is unknown. Near its end, much of the outcome has already been observed. Both readings can be useful for learning, but they have different information available. Their hit rates should be interpreted in that context.
A simple one-hour example
Imagine a candle that opens at 100. At minute 5, price is 100.2. At minute 55, it is 103. The late reading already knows that the market has spent most of the hour moving above its open. It may still reverse before closing, but the early reading had much less information. This is an invented example to explain timing, not a measured Horizon forecast.
Measure progress rather than clock minutes alone
Five minutes into a 15-minute candle is one third of the interval. Five minutes into a daily candle is a much earlier stage. When comparing timeframes, record elapsed time as a share of the candle duration. Horizon’s public record groups direction calls by progress through the candle and also provides results by timeframe.
Preserve forecasts when they change
Live readings can update as data changes. Horizon’s Pulse feature rechecks during the candle. Keep the time and contents of each reading you want to evaluate. If you overwrite the first prediction with the final one, you lose the ability to ask whether the early forecast was useful. A later improvement cannot retrospectively make an earlier miss correct.
Separate direction from range at the open
Horizon’s public methodology says that the forecast made at the candle open makes no direction call and is scored for its range only. Later direction calls are checked against the candle’s eventual close relative to its open. A candle closing exactly at its open counts neither way. Do not treat a missing directional call at the open as a bullish or bearish prediction.
Use a learning journal
For each observation, record the symbol, timeframe, candle open time, reading time, direction or no-call status, expected close, expected low and expected high. After the candle closes, add the actual close and apply the same scoring rule every time. Keep different timeframes and early-versus-late readings in separate groups.
Choose a comparison that matches your question
To study early forecasting, inspect early readings. To understand how analysis responds to new information, compare a sequence of readings and their explanations. For either question, include the number of observations and an appropriate baseline. The Horizon track record shows the always-up comparison alongside direction results and keeps no-call readings visible.
Turn the lesson into a better review
Timing is part of the forecast itself. Ask when a reading was issued before asking whether it was right. Read the companion accuracy guide for the other pieces of a fair comparison, and use the public record for current observed outcomes. Historical outcomes cannot guarantee the next candle.