CyGuru Gold Strategy & Audit Lab

The reference candidate closes beyond a confirmed symmetric pivot and the execution EMA. Pivots become available only after the selected number of right-hand bars. EMA uses an initial first-close seed with 50-bar warm-up; ATR is Wilder-smoothed after its initial mean. This initialization is specified for this model, not certified Pine parity.

Reference quality is 50% candle body fraction plus 50% directional closing location. Structure stop uses the opposite confirmed pivot with an ATR buffer; a missing or invalid pivot stop uses one ATR. TP distances are multiples of frozen initial risk. Adaptive mode offsets entry and scales the stop buffer and targets by at most ±10%, weighted only by prior completed nominal TP1 observations. Baseline and filtered use identical candidate levels and the same prior baseline decision cohort.

1H context uses complete four-bar UTC hours only, with 20/50 hourly EMA alignment and 50 complete hours of warm-up. Missing context excludes a filtered candidate. Cooldown is measured between accepted proposal bars. The unresolved guard blocks until a prior accepted record ends; this includes an ambiguous terminal classification and the chosen evaluation horizon.

Observation begins on the bar after proposal confirmation. Entry and exit ranges on one bar are ambiguous unless entry equals that bar’s open. A bar containing both stop and the next target is ambiguous. No tick path is invented. Gaps stop the audit. Expiry at the dataset end is unresolved, not a fabricated exit. Bar labels identify the containing interval, not exact execution times.

Every output is conditional on the supplied data, warm-up, parameters and cutoff. Comparing filtered and baseline counts is not proof of an improved strategy. Synthetic exercises are never evidence of historical market performance.