Edge Log · Wave 03
Do add-ons make the gold rule better?
No validated edgeNo improvement established
The idea.
We took the gold rule picked after Wave 2 and tested individual add-ons—a breakeven stop, a midpoint reclaim and a stricter range filter—plus a version combining all three. Each version was compared with the rule it came from, after costs.
Base rule: gold, H1 fade with the trend, signals from 08:00 to before 12:00 New York, stop and target at 1R and 2R from the signal close, and a timed exit at the next open once a close reaches 12:00, or after 16 bars. Two parents: COST (assumed cost small next to the planned risk) and BOTH (also a previous H1 range at least 2.0 × ATR wide). Breakeven: once a 15-minute close is a full planned risk in profit, the stop moves to the entry price from the next bar. Midpoint: the signal close must also be back past the middle of the previous H1 range. Stricter range: BOTH's minimum rises from 2.0 to 2.2 × ATR.
The test.
Gold CFD · Dukascopy gold bid prices, 15-minute candles, January 2018 to September 2026; paired test January 2023 to September 2026 · Assumed round-trip cost: gold 0.40 points; stress test at 2×.
The result.
Each row is an add-on version minus its parent: the mean daily difference in summed R over 967 observed New York dates, Jan 2023 to Sep 2026, days without trades included. Dot: base cost. Ring: double costs. Bar: descriptive 95% block CI at base cost. Every row is below zero at base cost. The BOTH + stricter range interval sits wholly below zero; the others include zero.
What each add-on changed in the BOTH rule.
Mean R per trade: +0.179 → +0.167
Jan 2023 to Sep 2026, base cost. Same 175 trades: the stop moved to entry on 43 of them and changed 4 exits. The total moved by −2.004 R.
Mean R per trade: +0.179 → +0.489
Jan 2023 to Sep 2026, base cost. It removed 155 of 175 trades, which had summed +22.977 R (77 winners, 78 losers), and added 1. Mean R per trade rose from +0.179 to +0.489; the total fell from +31.269 to +10.259 R.
Mean R per trade: +0.179 → +0.100
Jan 2023 to Sep 2026, base cost. It removed 37 trades that had summed +18.083 R (23 winners, 14 losers) and added 2. The total fell from +31.269 to +14.044 R.
What it taught us.
No tested add-on established an improvement over its parent. In Jan 2023–Sep 2026 at base cost, BOTH's midpoint version raised mean R per trade from +0.179 to +0.489. It removed 155 parent trades totaling +22.977 R and added one trade totaling +1.967 R; summed R fell by 21.010.
Evidence and limits
- Retrospective: every date here had been studied before, and this gold rule was picked after seeing earlier results, so this is not an untouched out-of-sample test.
- Costs are assumptions, not measured broker quotes. No swap or financing.
- R is the planned signal-to-stop risk, not an account return.
- The paired 95% block CIs use 2,000 moving-block replicates of 20 observed dates. They describe the mean daily difference, not R per trade, and do not adjust for the six comparisons here or for earlier searches.
- A breakeven stop is not risk-free: a gap through the moved stop fills at the open, and costs still apply.
- The stricter range threshold was set after we had seen two price feeds disagree near the old threshold.
- BOTH + all three applies the three add-ons together; it does not separate their effects.
- A trade an add-on removes counts with its parent's result: a removed winner is a missed gain, not an avoided loss.
Provenance
- Research journal record
RVD-W3-RESULT-IMPORT wave3/results/study.json· sha2562fc05b17cb3e…wave3/results/claim-ledger.json· sha256d01e2c128ca4…- Claim review record
RVD-W3-WEB-EDGE-LOG-NOTE-001