Prototype

CodeBLAST

Genomics has a tool for the question “what else looks like this?” — BLAST. This is the same idea aimed at capabilities: paste an implementation, get the genes in the registry that do the same job, ranked by how much of the fingerprint aligned.

Read this first

The matching engine here is a demonstration. It scores lexical overlap and distributes it across the six fingerprint facets deterministically, so the interface can be designed and argued about before the real comparison exists. The percentages are honest about their own method and should not be read as measured similarity. What a real implementation requires.

Or start from an example

How this differs from grep

You are not searching for text. You are searching for a job the code does.

Two implementations can share no identifiers, no language and no dependencies, and still be the same capability. CodeBLAST compares fingerprints rather than characters, then tells you which facet of the fingerprint carried the match — so you can disagree with it.

AST structure: Shape of the control and data flow, names stripped.
Public contracts: What goes in, what comes out, what can throw.
Semantic embedding: Vector proximity of documentation and identifiers.
Import graph: The neighbourhood the code lives in.
Test shape: What the suite asserts about behaviour.
Dependency closure: Transitive packages the capability needs.