Min Thresholds
The results-based cut: minimum MAR, CAGR, win rate and confluence, and a maximum drawdown — every threshold slider explained.
Min Thresholds are the second half of narrowing: where Filters select by what a strategy is, thresholds select by what it did over the selected period. The panel is collapsed by default — click Min Thresholds under the filters to expand it.
All thresholds start neutral (nothing filtered) and only take effect when you move them.
The sliders
| Threshold | Range | What it floors (or caps) |
|---|---|---|
| Min MAR | 0–50, steps of 0.1 | MAR = CAGR ÷ Max Drawdown over the period — return per unit of drawdown pain. The workhorse threshold: a floor of 1.5–2.0 already removes most of the catalog. |
| Min CAGR (%) | 0–80 | Compound annual growth rate. Note that CAGR alone says nothing about the path — pair it with the drawdown cap. |
| Max Drawdown (%) | 0–10, steps of 0.5 | The cap: deepest peak-to-trough decline allowed. At its default (10%) it filters nothing. |
| Min Confluence (0–5) | 0–5, steps of 0.5 | Neighbourhood robustness — see below. |
| Min Win % | 0–100 | Per-trade win rate floor. High win rates are typical for credit spreads; the stop-loss level shapes what a "loss" costs. |
Confluence, properly explained
Confluence answers one question: do this strategy's parameter neighbours confirm its result, or is it an isolated spike?
For the 10 nearest parameter neighbours (entry time ±3 steps, premium ±2 steps, same width and stop loss), each neighbour contributes up to 1 point depending on how its MAR compares to this strategy's MAR; missing neighbours count half. The sum is scaled to a 0–5 score:
- ≥ 4 — the neighbourhood confirms the result: nearby parameter sets perform comparably, so the result is less likely to be a "magic minute".
- 2–4 — mixed neighbourhood.
- < 2 — an isolated result its neighbours don't confirm. With 200,000+ variants, some parameter combination will always look spectacular by chance; low confluence is the tell.
Setting Min Confluence to 3+ is the cheapest robustness filter in the Browser: it costs you the lottery tickets and keeps the setups whose whole neighbourhood worked.
A workflow that holds up
- Filter parameters first (width, stop, premium, entry window).
- Floor MAR moderately (1.5+), cap drawdown where you can honestly sit through it.
- Add Min Confluence 3+ to drop isolated spikes.
- Only then sort and shortlist — the ordering reflects the statistic you sort by, not a recommendation.