> For the complete documentation index, see [llms.txt](https://www.degenbot.wiki/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://www.degenbot.wiki/performance-and-methodology/accuracy-null-and-edge.md).

# Accuracy, null, and edge

Understand the arithmetic behind the performance labels and why accuracy is not profitability.

### Accuracy

For a report defined over eligible scored alerts:

```
Accuracy = successful eligible alerts ÷ all eligible scored alerts × 100
```

The word **eligible** matters. The method must say how it handles open observations, expiries, ambiguous candles, missing data, and exclusions. Different denominators can produce different percentages from the same underlying messages.

### Null or baseline

The null baseline is a comparison result used to judge the method. It is not automatically 50%. Its construction may depend on the task, direction mix, market period, target/stop geometry, time window, and comparison policy.

For Degen's report, the exact baseline construction still requires release verification. A label alone does not explain whether the benchmark is matched fairly.

### Edge

When the report defines edge as a rate difference:

```
Edge in percentage points = accuracy percentage − baseline percentage
```

A fictional example:

| Measure                | Example value         |
| ---------------------- | --------------------- |
| Eligible scored alerts | 150                   |
| Successful alerts      | 108                   |
| Accuracy               | 72%                   |
| Matched baseline       | 54%                   |
| Difference             | +18 percentage points |

Here, `108 ÷ 150 × 100 = 72`, and `72 − 54 = 18` percentage points. The relative increase over 54% would be about 33.3%; it is a different calculation. Neither figure is an investment return.

If another 30 alerts remain open, they are not part of this example's 150 scored alerts. They should still be disclosed. You cannot infer the eventual full-period result until their treatment is known.

### Why a high success rate can still lose money

Consider an invented strategy that succeeds 70% of the time. Assume each success gains 1 unit and each failure loses 3 units:

```
Average result = 0.70 × 1 − 0.30 × 3 = −0.20 units
```

It loses on average under these assumptions before any costs, despite winning more often than it loses. This is arithmetic, not a statement about Degen's results.

Unequal target and stop distances are not automatically wrong. They change the economics and may change the chance of reaching one level first. Accuracy and a matched baseline must be interpreted with those rules in mind.

### What an edge figure cannot establish alone

A positive observed difference does not by itself prove statistical reliability, a durable forecasting advantage, or profitable execution. Sample size, overlapping observations, changing market conditions, and repeated strategy selection can affect the interpretation.

Treat the number as a starting point for review. Ask for the dates, method, complete outcome counts, and relevant uncertainty before making a stronger claim.