Number of Winners

The count of profitable trades a strategy produced. For low-win-rate strategies (most trend-followers), this is often the binding statistical constraint — more binding than total trade count.

In plain English

A trend-following strategy might have 1000 trades but only 100 winners. PnL comes from those 100 winners (the 900 losers each cost a small fixed amount). So when you ask "is the edge real?", what you're really asking is: "Do those 100 winners reflect a true pattern, or are they luck?"

You have 100 samples of "winning trade." That's the real sample size for purposes of validating the winner distribution. The 900 losers are characterized in detail by their sheer number, but the winning side — the side that determines whether the strategy is profitable — is where statistics get thin fast.

Why it matters for this fleet

Most variants in the 210-fleet are trend-followers with low win rates — the 21/50 swing pair (the only pair with any edge) sits around 30%. A strategy with win rate 15% over 50 trades has only ~7 winners. Even a strategy with win rate 30% over 80 trades has only ~24 winners — below the "directional" threshold.

This means total trade count overstates how reliable a strategy's metrics are. The fleet looks more statistically sound than it actually is when you only consider N. Because the win rates are low (~30%), winners accumulate slowly: you need a lot of trades before the winning side is well-characterized.

Recommended thresholds

These map roughly to the same tiers as sample size, but applied specifically to the winning subset.

Examples from the live fleet

Both rows below cleared the edge test — but on very different winner counts, which is exactly the constraint this concept is about.

• id522 (EMA 21/50 · ETH · 4h · 50× · long): N=117, win rate (the share of trades that close in profit) 22.2%, W=26. The edge-significant verdict rests on just 26 winners. Thin: the winning side is barely characterized, and the headline "beat buy-and-hold" came from 50× leverage (amplified market exposure), not a rich winning distribution.
• id523 (EMA 21/50 · SOL · 1h · 2× · long): N=436, win rate 31.9%, W=139. With 139 winners, the winning distribution is far better described. This is the most credible edge-significant row in the fleet (still in-sample, and it trailed simply holding SOL — but statistically the winning side is real).

Note both win rates hover around 30%, so winners accumulate slowly: id523 needed 436 trades to reach 139 winners.

The rare-event problem (for low-WR strategies)

A 13% WR strategy with 100 trades has ~13 winners. Each additional winner moves the WR estimate by 1 percentage point. The estimate is brittle.

A 50% WR strategy with 100 trades has ~50 winners. Each additional winner moves WR by 1 pp, but the distribution of winners (avg size, variance) is much better characterized.

This is why high-WR strategies need fewer total trades to validate than low-WR trend-followers.

How to mitigate small W

• Pool family variants — but only across genuinely different conditions. Siblings at different leverage share the identical trades (id523 at 2× ≡ id659 at 1× — the same 139 winners), so pooling across leverage adds zero new winners.
• Longer backtest window — more years = more winners.
• Lower interval — smaller candles = more crossings = more winners, but watch for whipsaw degradation.
• Avoid the trap of "selective" strategies — they look picky, but their winner count is often the lowest in the fleet.

Related

• sample size
• statistical significance
• confidence interval
• edge
• trend following
• win rate
• profit factor

Sources

• wiki/qa-sessions/2026-05-17-session.md#q3 (first asked here)
• /api/analytics winCount field