Market-Data Quality, Splits and the Latest Candle
A chart is only as reliable as its inputs. Learn the checks that prevent one bad candle or split from corrupting every downstream level.
How to validate OHLCV candles, corporate-action adjustments, timestamps and incomplete bars before calculating technical evidence. This article explains the reasoning framework used by Trading.Red and the limitations a reader should keep in view.
OHLCV is structured evidence
Each candle contains an open, high, low, close and usually volume for a defined interval. Basic integrity requires the high to be at least as large as the open, close and low, and the low to be no greater than the other prices. Values must be finite, positive where appropriate and ordered by a valid timestamp.
These checks catch impossible candles but not every provider error. A plausible-looking extreme can still be a bad tick, a mismatched exchange or a corporate-action adjustment problem. Neighbor comparisons and cross-provider review are needed when a move is far outside recent volatility.
The latest candle is different
A historical candle is fixed after its interval closes. The live candle is still accumulating trades, so its close, high, low and volume can change. Calculating confirmation from that partial state creates signals that disappear later and makes backtests look cleaner than real-time behavior.
Trading.Red uses completed bars for confirmation-sensitive calculations. Live prices may still be shown separately, but they should not overwrite the latest completed candle or inherit its timestamp.
Splits, dividends and adjusted prices
A stock split changes the numerical price and share count without creating the same economic loss or gain implied by an unadjusted chart. Providers may offer adjusted close fields or adjust the full historical OHLC series. Indicators must use a consistent basis; mixing adjusted closes with raw highs and lows creates artificial gaps and false levels.
Cash dividends can also affect the opening price. The analytical question is whether the chart intends to describe traded prices, total return or an adjusted historical comparison. The choice should be consistent and documented rather than silently changing by feature.
Timestamp and timezone controls
Daily bars depend on the exchange calendar, not the server's local midnight. Daylight-saving changes, holidays and half sessions can shift expected boundaries. The provider timestamp, exchange timezone and selected session must travel with the data so the interface can describe what was actually observed.
Fail closed when evidence is unreliable
A robust engine quarantines an implausible observation, records why it failed and returns an unavailable state when the remaining history is insufficient. Substituting a convenient value preserves the layout but contaminates support, resistance, volatility, scores and every label derived from them.
Worked example: a suspicious final daily bar
- The last bar is still open.
- Its quoted low is several ATR below neighboring trades.
- A second source does not show the same print.
Reading: The bar should be marked partial and the extreme quarantined for confirmation-dependent logic. The system should not redraw support or a stop from an unverified observation.
Key takeaway
A technical label is a compressed description of market data, not knowledge of the future. Use the label to organize questions: which timeframe produced it, what confirmed it, which observation would invalidate it, and what data might be missing?