# How the Tactfolio Score works

> How Discover is ordered, from 1 to 100: expected Sharpe ratio, Calmar ratio, and growth once hindsight is taken out, compared on a Pareto frontier.

The Tactfolio Score orders the published strategies in [Discover](https://tactfolio.com/strategies), from 1 to 100. It asks one question: once we take out what a backtest could owe to hindsight, which strategies are expected to do better from here on, on risk-adjusted return, on return for the drawdowns you sit through, and on growth, all at once?

A score of **100**, the best, puts a strategy on the frontier: no other public strategy is expected to beat it on all three. A score of **80** means some strategy is expected to be at least 25% better on every one of them. A score of **10 or less** usually means we do not expect it to beat Treasury bills: those strategies score from 1, the worst, to 10 by their expected Sharpe ratio alone. Many still have an edge that their volatility eats, so a smaller position in one, the rest in T-bills, could do better.

## Why not just sort by backtest results?

A backtest is written with the past in view. The author picked the rules, the assets, and the numbers while looking at the history the backtest replays, so the best-looking backtests are often the best-fitted ones.

We checked. We took 56 classic strategies from published research and scored them on the history their authors had, then looked at how they did after publication. We did the same with 61 strategies built by AI models on 2016–2026, looking at 2011–2015, which they never saw. In both groups, **the strategies with the best backtest Sharpe ratios did worse afterward**, not better. What carried over was risk: how volatile a strategy was, and how deep its drawdowns went.

So the score leans on what lasts and discounts what does not.

## What goes into it

Every figure is measured against Treasury bills, so parking in cash does not look like skill.

**Complexity.** We count the choices an author could have tuned: each distinct rule and each distinct number, such as a 200-day window or an RSI below 30. The same rule on five assets counts once, and adding assets does not count at all. Every choice lets a backtest fit history a little better, so the score assumes each one hid a little of the drawdown to come.

**Fragility.** We run the strategy again 24 times with every number nudged by up to 25%: 200 days becomes 170 or 240, 30 becomes 26 or 35. A strategy whose results depend on its exact numbers is fitted to history. We use the middle result of the whole family, not the strategy alone.

**Market exposure.** A strategy that holds stocks earns some of the stock market's return without any skill; one that holds T-bills earns none. Before any other evidence, we expect a strategy to earn what its correlation with the market has historically paid.

**The backtest and the live record.** Both then move that starting point, each by how much it can be trusted. Sharpe ratios are noisy, and the backtest is fitted, so both move it only a little: ten years of backtest count for about 8%, ten live years about the same. The live record counts at once for drawdowns: a loss that already happened live is one the strategy can have.

## How the three measures are compared

From that evidence we work out three expected figures over T-bills:

- **Sharpe ratio:** return per unit of volatility. Our [Sharpe ratio guide](https://tactfolio.com/learn/sharpe-ratio.md) explains it.
- **Growth:** yearly return above T-bills, after the drag that volatility puts on compounding.
- **Calmar ratio:** yearly return above T-bills for each unit of worst drawdown.

Growth and drawdown come from replaying the ten years before publication for the strategy and its nudged family, with each copy's returns lowered to the expected Sharpe ratio. The drawdown is then deepened a little for each choice the author made.

No one of the three is weighted over the others. A strategy scores 100 if no other strategy beats it on all three; several can, since without weights none of them beats another, and Discover lists them by expected Sharpe ratio. A copy of a strategy never counts against the original. The score changes when other strategies are published or retired, and after every market close.

Your workspace scores your private strategies the same way, as if each were published today: against every public strategy, without a live record, and without joining Discover, so a private strategy never changes anyone else's score.

## An example

A trend strategy holds stocks above their 10-month average and T-bills otherwise: two choices, one rule and one window. Its backtest earned a Sharpe ratio of 0.75 over T-bills, and its nudged copies about the same. Its correlation with the stock market is 0.56, so the score starts from 0.56 × 0.35 ≈ 0.20, and the backtest lifts that to about 0.25. Replayed at 0.25, its last ten years show growth of about 2% a year above T-bills with a worst drawdown near 17%.

A strategy tuned with fifteen choices to a backtest Sharpe ratio of 1.4 gets little extra for it: its expected Sharpe ratio lands near its market exposure too, and its expected drawdown is six points deeper than its backtest's.

## What the score does not measure

- **Your situation.** It is one score for everyone. It does not know your taxes, costs, or how much of a drawdown you can live with.
- **The future.** A high score says a strategy's risk and evidence compare well by these rules so far. It is not a forecast you can rely on, and it is not investment advice.

The rules may change as Discover and its live records grow. When they do, this page will say what changed.

## Learn the ideas behind it

- [Max drawdown](https://tactfolio.com/learn/max-drawdown.md): Max drawdown is the biggest fall from a peak to a low.
- [Calmar ratio](https://tactfolio.com/learn/calmar-ratio.md): The Calmar ratio divides yearly growth by the worst drop.
