Automating Dead Endpoint Detection: Deleting 16,000 Lines of Legacy Node.js Code

How I automated dead endpoint detection and removed 16,000 lines from our Node.js codebase

Miguel Lopes developed an automated detection system to identify and prune unused endpoints in a legacy Express API. By cross-referencing access logs with registered routes, the team successfully deleted 16,000 lines of code, representing 35% of the total codebase.

Why This Matters

Static analysis tools like ESLint are insufficient for identifying dead endpoints because they only detect unreachable code, not the absence of live traffic. In legacy systems, this technical debt accumulates over years of product iterations, leading to bloated codebases that are difficult to refactor and slow to onboard new engineers.

Key Insights

• Production log ingestion: Miguel Lopes (2026) used access logs to extract routes receiving live traffic, providing a signal that static analysis cannot replicate.
• Automated route mapping: The system maps log data against every route registered in the Express application to identify discrepancies and identify candidates for removal.
• Metadata-driven verification: The detector ranks candidates by last-seen dates and call volume to help engineers distinguish between dead code and low-frequency internal jobs.
• The 12-month observation window: To avoid breaking legacy clients or quarterly tasks, a long-term verification checklist is used across production, staging, and canary environments.
• Impact of feature abandonment: Eight years of iterations resulted in a 45,000-line codebase where 50 endpoints were found to be completely unused.

Practical Applications

• Legacy API Pruning: Using the detector to rank endpoints by traffic volume and last-seen date before manual verification. Pitfall: Deleting endpoints that look dead but are actually triggered by infrequent internal scripts or quarterly jobs.
• Codebase Refactoring: Reducing the surface area of Express apps to improve maintenance speed. Pitfall: Relying purely on static analysis which results in ‘zombie’ code remaining in production.

References:

• https://dev.to/miguel_lopes_966962540a43/how-i-automated-dead-endpoint-detection-and-removed-16000-lines-from-our-nodejs-codebase-1fan