Adding a New Analyzer

All pipeline steps (extractors, preprocessors, analyzers) implement the same interface:

public interface IPipelineStep
{
    string Name { get; }
    PipelineContext Run(PipelineContext context);
}

The PipelineContext

public sealed class PipelineContext
{
    public CodeFile FileA { get; }
    public CodeFile FileB { get; }
    public ILanguageProfile Language { get; }

    // Extracted data (populated by extractors, transformed by preprocessors)
    public Dictionary<string, List<string>> Metadata { get; }

    // Results (populated by analyzers)
    public List<SimilarityReason> Reasons { get; }

    // Safe accessor
    public List<string> GetMetadata(string key);
    public void SetMetadata(string key, List<string> value);
}

Available Metadata Keys

After the default extractors run, these keys are available:

Key	Content
code.a, code.b	Clean code lines (comments stripped)
comment.a, comment.b	Extracted comments
functions.a, functions.b	Function/method names
imports.a, imports.b	Import/include statements

Example: Variable Count Analyzer

public sealed class VariableCountAnalyzer : IPipelineStep
{
    public string Name => "VariableCount";

    public PipelineContext Run(PipelineContext ctx)
    {
        var codeA = ctx.GetMetadata("code.a");
        var codeB = ctx.GetMetadata("code.b");

        // Count unique tokens (rough proxy for variable count)
        var tokensA = codeA.SelectMany(line => line.Split(' ')).Distinct().Count();
        var tokensB = codeB.SelectMany(line => line.Split(' ')).Distinct().Count();

        if (tokensA == 0 && tokensB == 0) return ctx;

        double ratio = 1.0 - (double)Math.Abs(tokensA - tokensB) / Math.Max(tokensA, tokensB);

        if (ratio > 0.9)
        {
            ctx.Reasons.Add(new SimilarityReason(
                Name,
                $"Vocabulary size ratio: {ratio:P0} ({tokensA} vs {tokensB} unique tokens)",
                ratio,
                Weight: 0.2));
        }

        return ctx;
    }
}

Adding to the Pipeline

var steps = AnalysisEngine.DefaultPipeline();
steps.Add(new VariableCountAnalyzer());
var engine = new AnalysisEngine(steps: steps);

Or insert at a specific position:

steps.Insert(steps.Count - 1, new VariableCountAnalyzer());

Weight Guidelines

Weight	When to use
1.0	Primary signal (cosine similarity)
0.8–0.9	Strong complementary signal (winnowing)
0.4–0.5	Structural/semantic signal
0.2–0.3	Weak/supporting signal (file metrics)

Higher weight means the analyzer has more influence on the final score. A weak signal with high weight causes false positives; a strong signal with low weight gets drowned out.