Backtesting & results

The Edge Rating

What the letter on a backtest means — a band of the edge we can be confident of after removing market exposure, worse costs, luck and the versions you tried — and how to move it.

Updated 2026-11-19·Rev. 2026.11

The letter beside a deep backtest answers one question: how much skill-driven edge can we be confident this strategy has? It is not a score of how good the backtest looks. A strategy can show a Sharpe of 2.5 and still rate C, because most of that came from holding a rising market.

What the letter means

The letter is read from the conservative edge: the edge (an annual Sharpe ratio) we are 80% confident the strategy has at least, after everything below is taken out.

LetterConservative edgeMeaning
A+1.5 or moreStrong edge, strongly evidenced
A1.0 or moreClear edge
B+0.6 or moreLikely edge, moderate size
B0.3 or moreLikely small edge
C+0 or moreProbably positive, not enough to rely on
Cbelow 0Coin-flip territory
D—Probably no edge
F—No edge, or the strategy failed a test

What gets taken out

  1. The market's own move. If your strategy was long while the market rose, part of its return is the market's, not the strategy's. We measure how much of the return moved with simply holding the asset, and rate only what is left. A long strategy that just sat in a bull market rates near zero.
  2. Worse costs. Every trade is charged at least the exchange's standard fee (the maker fee when it entered with a limit order) and at least 2 bps of slippage, or 1.5× your slippage if that is more, however you set your costs. A backtest set to zero cost is charged like any other. An edge that only exists at perfect fills is not one.
  3. Luck. A short or noisy record could have produced its Sharpe by chance. The fewer days and the wilder the returns, the more we take off.
  4. The versions you tried. The best of many tries looks better than it is. Every version you ran counts, including the ones tried on the strategy you copied this one from, and a rerun counts again when its results changed. Trying more never raises the letter.
  5. Decay. Edges shrink once they leave the window they were found in. We assume a quarter of it goes.

Then new data adds back confidence: the out-of-sample stretch (at half weight, because you saw it while choosing), a research study's sealed window, a Forward Check's days since the version was saved, and the strategy's results on other markets. Those are the only things that can raise the letter, and each counts only for days after the backtest's own.

The Edge Check can only lower it

Running the Edge Check does two things. It tests the strategy unchanged on two other markets, which counts as new evidence. And it runs tests that can fail a strategy but never pass one:

FindingEffect
The account was wiped out in an unseen periodF
Random entries with the same habits returned more than it 95% of the timeF
Nearby parameter values lose (a knife-edge)F
Entries returned less than an average random entryAt most C
It did not beat shuffled versions of its own marketAt most B+
The edge lived in a minority of the window's periodsAt most B+

Until the Edge Check runs, the letter is provisional (dashed outline) and stops at B+. A funding floor counts a provisional letter as at most B, so a floor of B+ or higher needs the Edge Check.

Other states

  • A range instead of a letter ("B–C+") on a quick run. A quick run can't separate skill from holding the market, so it shows where a deep run would likely land.
  • — when there are fewer than 30 days to judge (the badge says about how many more you need), when the returns never moved, or when the market's own returns could not be lined up with the run.
  • F for a deep backtest that never traded. A quick run that never traded says No trades.
  • Rating… for a moment while the versions you tried are counted.

Reading the card

Click the letter for the rating card:

  • Why this letter: up to four reasons, largest first, each measured by recomputing with that one thing changed.
  • From backtest Sharpe to conservative edge: each step above, as a number.
  • What would change it: next steps with the letter each is predicted to reach if the new evidence matches today's estimate. Re-tuning parameters is never on the list: every version is counted.
  • Details: market exposure, cost headroom, the trade size the market can take, and more.
  • Risk profile: drawdown, win rate and the rest. Shown beside the letter, never part of it.

Tip

The fastest way up is usually not a better backtest. It is new evidence: run the Edge Check, extend the window, or let a Forward Check run.

Backtest results are hypothetical. They are made with hindsight, and no rating says how a strategy will do in future. The Edge Report and the rating card say when the letter was rated, over which window and by which grader.

Related