Start with the market’s actual question
An up or down label is not a complete specification. Polymarket says markets resolve according to their predefined rules. Read those rules on the particular market before comparing any signal. Source: Polymarket Help Center.
Match evidence to the outcome
Write down the asset, reference price, start and end timestamps, timezone, resolution source and handling of equality. Keep these next to the model output. A Binance spot candle and another venue’s settlement reference can disagree. If the rule is unclear, the signal cannot answer the market’s question reliably.
A useful research record separates three times: the candle observation, the model’s generation time and the market’s expiry. A 15-minute input candle does not establish that a model predicts the next 15-minute contract. Likewise, a bullish category does not supply a calibrated probability for a specific settlement event.
Test a strategy before calling it an edge
Define the signal rule before examining its outcomes. Keep every eligible observation, including neutral views and missing data, with an explicit exclusion reason where appropriate. Compare results over the same period with a simple baseline; include execution costs and report sample size. A short winning streak is not enough to establish a repeatable advantage.
Use Coinpred as research context
The BTC 15 minute prediction page exposes available directional context and its time. The JEV evaluation disclosure explains why no validated contract win probability is currently claimed. Keep the research log separate from TradingView signal outcomes, which evaluate a different system.