How do you migrate the last 3% of files that automation can't fully finish?
Treat the long tail differently from the bulk: stamp every file with a machine-readable status of what has and hasn't succeeded, build a tool to rerun just the failing step on just the affected files, then run a tight loop — sample a handful of similar failures, tune the approach, sweep it across the category, repeat for the next failure pattern. LLM output on the hardest files still served as a usable starting point for the small remainder of manual cleanup.
Answered in
How to Migrate Thousands of Legacy Test Files With LLMs, Not RewritesA step-based pipeline, retry loops instead of perfect prompts, and rich context injection turn a year-long manual migration into a six-week automated one.
Read the full analysisOther questions this article answers
More development best practices questions
- Why is a large-scale test framework migration usually so slow to do by hand?
- Why did retry loops with error feedback outperform carefully engineered prompts?
- When does a migration pipeline need rich prompt context instead of just retries?
- Does this approach only work for test framework migrations?
- What is Project Polaris and how does it relate to GitHub Copilot?
- What architecture does Project Polaris use?
- How does Project Polaris perform compared to GPT-4 Turbo on coding benchmarks?
- Can teams still use GPT-4 Turbo in Copilot after the Polaris rollout?
Every answer on Crashtech is written by the editor of the article it comes from — never auto-summarised. Browse all answers or the Development Best Practices beat.