Sample
Dorn/cuda/66
Human rating uses the original dataset scale.
Scores
| Method | Score | Scale | Dataset rank (0–100) |
|---|---|---|---|
| Posnett | 0.008 | 0–1; higher is more readable | 67.1Δ 53.5 |
| Dorn | 0.24 | 0–1; higher is more readable | 19.2Δ 5.6 |
| Mi | 0.674 | 0–1; higher is more readable | 77.2Δ 63.5 |
| LLOC | 6 | Logical lines; fewer is more readable | 72.8Δ 59.2 |
| DeepSeek V4 Pro | 3.67 | 0–20; mean of three runs | 6.1Δ 7.5 |
| GPT-6.1 Sol | 10 | 0–20; mean of three runs | 10.2Δ 3.5 |
| BTTF | 0.27 | 0–1; higher is more readable | 12.5Δ 1.1 |
| Human | 2.58 | 1–5; original rating | 13.6 |
Views
Select the downloaded ZIP or its extracted folder. Files stay in your browser.
Loading feature diagnosis…
Source regions
Feature contributions
Feature contributions
| Feature | Value | Contribution (log-odds) |
|---|---|---|
| Halstead volumeHow much symbolic information the code contains. It combines the total number of operators and values with the number of distinct ones. | 316.362 | −10.44 |
| Byte entropyHow varied the bytes in the source text are. Repeated bytes give a lower value; a more even mixture gives a higher value. | 4.839 | −7.259 |
| Source linesThe number of non-empty source lines. | 10 | +4 |
Intercept 8.87 + feature contributions = -4.829 (log-odds).
Score = sigmoid(-4.829) = 0.008.
Feature contributions
| Feature | Value | Contribution (log-odds) |
|---|---|---|
| Operator visual areaThe share of the code’s visual area occupied by operators such as +, =, and *. | 0.301 | −1.98 |
| Identifier layout (horizontal)How rapidly the visual pattern of identifier characters changes from left to right. It measures spatial frequency, not the meaning of the names. | 53 | +1.42 |
| Comment layout (horizontal)How rapidly the visual pattern of comment characters changes from left to right. | 0 | −0.804 |
| Comment layout frequencyHow frequently the layout switches between lines containing only comments or whitespace and lines containing code. | 11 | +0.647 |
| Indentation frequencyHow rapidly indentation changes from line to line. | 9 | −0.426 |
| Identifier layout (vertical)How rapidly the visual pattern of identifier characters changes from top to bottom. | 9.852 | −0.272 |
| Number layout frequencyHow rapidly the number of numeric values per line changes through the code. | 10 | +0.191 |
Intercept 0.07 + feature contributions = -1.153 (log-odds).
Score = sigmoid(-1.153) = 0.24.
| Score | 0.674 |
|---|
| Score | 6 |
|---|
Run 1
The code snippet is very difficult to evaluate in isolation, lacking context such as function signatures, variable declarations, or surrounding logic. It begins with orphaned closing braces, suggesting it's a fragment mid-function, which hurts clarity of intent. The structure relies on implicit assumptions about thread indexing and data layout (e.g., sumH array, WMATRIX macro). Nesting is shallow, which is positive, but the code length is extremely short and leaves many operations unexplained. Action granularity is mixed—some lines perform multiple operations (e.g., the macro use with division). Reading flow is disrupted by missing context and abrupt conditionals, making it hard to follow the intended parallel computation pattern. Overall, the fragment is unreadable without substantial surrounding context.
Run 2
The code snippet is a partial fragment, making it difficult to evaluate structure and intent fully. It begins with closing braces and a thread synchronization call, indicating a break from a larger loop or conditional. The remaining logic is a single conditional with matrix operations, but it contains multiple actions per line and unclear variable names (WMATRIX, sum1, sum2, sumH) without context. Nesting is minimal here, but the fragment's abrupt start disrupts reading flow. Action granularity is poor, as multiple operations occur in single lines. Without surrounding code, clarity of intent is low, and overall readability suffers due to missing context, compact expressions, and ambiguous naming.
Run 3
This code snippet is difficult to read because it is out of context, lacking surrounding function definitions, variable declarations, and control-flow logic. It begins with orphaned closing braces and ends abruptly, making the structure unclear. The intent of the normalization operations is partially discernible—adjusting matrix values by a ratio of sums—but variable names like WMATRIX, sum1, sum2, and sumH[x] are cryptic without definitions. Nesting is shallow, but action granularity is mixed: multiple operations are packed into single lines, while others are spread out. Reading flow is poor due to missing context and an abrupt final block. Overall, the fragment is not self-contained and would require significant surrounding code to assess properly.
Run 1
Code Structure: The bounds check groups related matrix updates, but unmatched closing braces and missing context obscure the overall structure. Nesting: The visible conditional nesting is shallow and manageable. Clarity of intent: Synchronization and matrix scaling are recognizable, but WMATRIX, sum1, sum2, sumH, and the stride of 16 are unexplained. Code length: The fragment is concise without obvious redundancy. Action granularity: Most statements perform one action, although the second conditional and update share a line. Reading flow: The first update, index increment, and guarded second update form a logical sequence, but the abrupt opening and omitted surrounding code interrupt comprehension.
Run 2
Code Structure: The update block is logically grouped, but leading closing braces and missing context make the overall structure difficult to assess. Nesting: The visible conditional nesting is shallow and manageable. Clarity of intent: Bounds checks and matrix scaling are apparent, but WMATRIX, sum1, sum2, sumH, and the stride of 16 are unexplained. Code length: The snippet is concise without obvious redundancy. Action granularity: Most actions occupy separate lines, although the second conditional and update share one line. Reading flow: The main block reads sequentially, but unmatched opening context and the mutation of x interrupt comprehension.
Run 3
Code Structure: The visible update block is logically organized, but unmatched closing braces and missing context make the overall structure difficult to assess. Nesting: The main block has only one nested conditional, so nesting is manageable. Clarity of intent: Bounds checks and matrix scaling are recognizable, but WMATRIX, sum1, sum2, sumH, and the offset 16 lack explanation. Code length: The snippet is concise without obvious redundancy. Action granularity: Most lines contain one action, although the inline conditional combines a check and an update. Reading flow: Synchronization followed by guarded updates provides a reasonable sequence, but the abrupt opening, macro-based access, and mutation of x interrupt readability.