Sample

Dorn/cuda/77

Human rating uses the original dataset scale.

DatasetDorn
Human3.1
Lines10

Scores

MethodScoreScaleDataset rank
(0–100)
Posnett0.20–1; higher is more readable
76.0Δ 40.7
Dorn0.6580–1; higher is more readable
60.2Δ 24.8
Mi0.5410–1; higher is more readable
57.7Δ 22.3
LLOC5Logical lines; fewer is more readable
78.0Δ 42.6
DeepSeek V4 Pro3.330–20; mean of three runs
4.6Δ 30.8
GPT-6.1 Sol9.670–20; mean of three runs
8.8Δ 26.6
BTTF0.5820–1; higher is more readable
69.6Δ 34.3
Human3.11–5; original rating
35.4

Views

Download official ZIP ↗

Select the downloaded ZIP or its extracted folder. Files stay in your browser.

Loading feature diagnosis…

Source regions

Feature contributions

Feature contributions

FeatureValueContribution
(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.216.644−7.149
Byte entropyHow varied the bytes in the source text are. Repeated bytes give a lower value; a more even mixture gives a higher value.4.737−7.106
Source linesThe number of non-empty source lines.10+4

Intercept 8.87 + feature contributions = -1.385 (log-odds).
Score = sigmoid(-1.385) = 0.2.

Feature contributions

FeatureValueContribution
(log-odds)
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.53.909+1.382
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.9+0.718
Indentation frequencyHow rapidly indentation changes from line to line.1−0.703
Identifier layout (vertical)How rapidly the visual pattern of identifier characters changes from top to bottom.9.855−0.272
Number layout frequencyHow rapidly the number of numeric values per line changes through the code.10+0.191
Operator visual areaThe share of the code’s visual area occupied by operators such as +, =, and *.0.09+0.072

Intercept 0.07 + feature contributions = 0.654 (log-odds).
Score = sigmoid(0.654) = 0.658.

Score0.541
Score5
Run 1

This code snippet is extremely difficult to read due to multiple issues. The structure is fragmented, beginning mid-expression with 'resultImageGradient.y' and 'resultImageGradient.z', suggesting disconnected code. The nesting is unclear because the opening brace for the conditional is absent from the visible context. The intent is opaque: the self-comparison checks (e.g., 'x==x') appear to be NaN detection but are not explained, and the declared variables are never assigned or used within the visible scope. Code length is wasted on verbose but cryptic conditions. Action granularity is reasonable per line but meaningless without surrounding context. Reading flow is completely broken because the snippet starts mid-statement, omits the opening brace, and lacks any comments explaining the purpose of the entropy derivatives. Overall, this fragment would be nearly impossible to evaluate or maintain without seeing the full function and surrounding logic.

Run 2

The snippet is severely unreadable due to incomplete context, excessive indentation, and a nonsensical self-comparison condition (NaN check written as x==x, etc.). The code lacks logical structure, as it opens with orphaned function arguments and unexplained variables. The NaN check, while functionally valid, is opaque without comments or a helper like std::isnan. The nesting is not visible but the deep indentation suggests it resides within multiple blocks. Variable names are long and repetitive, reducing clarity. Action granularity is reasonable with one declaration per line, but overall reading flow is broken by the missing surrounding code and ambiguous intent.

Run 3

The code is poorly readable due to several factors. The nesting is unclear without surrounding context, but the structure appears disorganized with misaligned indentation and inconsistent formatting. The purpose is ambiguous—variables like `resultImageGradient` and multiple entropy derivatives are not explained, making intent unclear. The self-comparison `x==x` is a NaN check but is not immediately obvious, reducing clarity. There are multiple actions and declarations on single lines, violating the one action per line guideline. The snippet appears to be a fragment of a larger function, further reducing readability. Overall, the lack of context, poor formatting, and unclear logic make it difficult to read and maintain.

Run 1

Code Structure: The derivative initializations are logically grouped, but the snippet begins mid-expression and ends inside an open block, obscuring its organization. Nesting: Only one conditional is visible, though heavy indentation suggests surrounding nesting. Clarity of intent: Descriptive variable names help, but comparing each gradient component with itself is a non-obvious NaN check that would benefit from an explicit predicate or comment. Code length: The visible code is concise, although the repeated comparisons add verbosity. Action granularity: Each initialization occupies its own line, making individual operations easy to distinguish. Reading flow: The component checks and initializations follow a consistent sequence, but inconsistent spacing, excess indentation, and incomplete context interrupt readability.

Run 2

Code Structure: The derivative initializations are logically grouped, but the snippet starts mid-expression and ends inside an open block, limiting structural clarity. Nesting: Only one visible conditional level keeps nesting shallow, although excessive indentation adds visual noise. Clarity of intent: Variable names describe their roles, but self-equality comparisons are an obscure NaN check without explanation. Code length: The fragment is concise, though the repeated component checks are verbose. Action granularity: Each variable initialization occupies its own line, and each comparison is separated clearly. Reading flow: The checks followed by initializations form a reasonable sequence, but incomplete context, inconsistent spacing, and extra blank lines interrupt readability.

Run 3

Code Structure: The fragment has a simple conditional block, but starts mid-expression and ends before the block closes, obscuring its organization. Nesting: Only one visible conditional level keeps nesting shallow. Clarity of intent: Descriptive variable names help, but self-comparisons apparently checking for NaN are non-obvious without a comment or named helper. Code length: The snippet is concise, although repeating the component checks adds some verbosity. Action granularity: Each derivative initialization occupies its own line, and the condition is split by component, making individual operations easy to distinguish. Reading flow: Consistent initialization patterns help, but excessive indentation, uneven spacing, blank lines, and incomplete context interrupt the flow.