Field notes
Notes on duplicate detection, AI-era codebases, and the decisions behind Deslop.
A detector produces evidence. A skill supplies procedure — a read-only first pass, a stop, a verdict, a test gate, and a reproducible command in the pull request. Here is why that layer matters and how to build one.
Read Article arrow_forwardDeslop's push toward 100% accuracy is public: the issue graph exposes known failures, dependencies, verification work, and the research behind every accuracy decision.
Read Article arrow_forwardDeslop now understands Go. It parses .go syntax trees, catches renamed clone shapes, ranks the worst duplication first, and lets coding agents search before they write another copy.
Read Article arrow_forwardMCP gives coding agents tools. Deslop gives them live repository memory: call find-similar before writing, then reuse the code that already exists.
Read Article arrow_forwardFlutter projects repeat widget trees, repositories, mappers, and test setup. Here is how Deslop compares their structure and reports the highest-impact duplicates first.
Read Article arrow_forwardAI-generated code can multiply duplicate logic before review catches it. This guide explains what to check, why duplicate code becomes technical debt, and where to read the full Deslop research background.
Read Article arrow_forwardCoding agents do not automatically know that similar code already exists elsewhere in a repository. This is why duplicate-code feedback belongs in the editing loop.
Read Article arrow_forwardDeslop ranks clusters by fragment size, additional copies, and spanned bytes so higher-impact findings appear first.
Read Article arrow_forwardDeslop uses parsed syntax trees instead of source-text patterns. Here is what that enables and where language-specific normalization fits.
Read Article arrow_forward
