plagiarism-sdk

SDK for plagiarism detection in Platform Network challenges.

Features

• AST Similarity: Structural code comparison with 97% threshold
• Agent Sandbox: Isolated workspace for agentic investigation
• Agentic Investigator: LLM-powered investigation with function calling

Usage

Add to your Cargo.toml:

[dependencies]
plagiarism-sdk = { git = "https://github.com/PlatformNetwork/plagiarism-sdk" }

AST-based Detection

use plagiarism_sdk::{check_plagiarism, PlagiarismReport};

let code_a = r#"
def fibonacci(n):
    if n <= 1:
        return n
    return fibonacci(n-1) + fibonacci(n-2)
"#;

let code_b = r#"
def fib(x):
    if x <= 1:
        return x
    return fib(x-1) + fib(x-2)
"#;

let report = check_plagiarism(&[code_a, code_b])?;

match report {
    PlagiarismReport::Results { comparisons, plagiarized, .. } => {
        println!("Highest similarity: {:?}", comparisons.iter().max_by_key(|c| c.score));
        if !plagiarized.is_empty() {
            println!("Plagiarism detected in submissions: {:?}", plagiarized);
        }
    }
    _ => println!("No comparison possible"),
}

Agentic Investigation

use plagiarism_sdk::{AgenticInvestigator, InvestigatorConfig};

let config = InvestigatorConfig {
    max_iterations: 20,
    timeout_seconds: 60,
    llm_endpoint: "http://localhost:11434/api/chat".to_string(),
    llm_model: "llama3".to_string(),
};

let investigator = AgenticInvestigator::new(config)?;
let result = investigator.investigate(code_a, code_b, 0.95).await?;

println!("Verdict: {}", result.verdict.is_plagiarism);
println!("Confidence: {:.2}%", result.verdict.confidence * 100.0);
println!("Reasoning: {}", result.verdict.reasoning);

Sandboxed Workspace

use plagiarism_sdk::AgentWorkspace;

let workspace = AgentWorkspace::new()?;
workspace.write_file("submission.py", code)?;
let content = workspace.read_file("submission.py")?;
let matches = workspace.grep(r"def\s+\w+")?;

Feature Flags

• std (default): Enables full standard library support including sandbox and agentic features
• alloc: Enables no_std mode with allocator support (limited to core AST types)
• python-parser: Enables Python AST parsing (pulls in rustpython-parser)

Plagiarism Thresholds

Score	Status	Action
< 30%	Clean	No action needed
30-96%	NeedsLlmVerification	Flag for agentic review
≥ 97%	Plagiarized	Mark as plagiarized

License

Apache-2.0