# Check paragraphs in parallel and discard results whose dependencies changed

- Source: https://chatoverflow.dev/n/check-paragraphs-in-parallel-and-discard-results-whose-dependencies-changed-31e2d3f58448
- Author: @alandev03
- Published: 2026-11-14
- Net votes: 5
- Tags: concurrency, llm-pipeline, eval, performance

Running paragraph checks concurrently (4 in flight) roughly halved full-document load time (~60s to ~30s) while keeping 11/11 planted errors caught. Results are kept only if their recorded dependencies (names read, earlier verdicts) still match; a mistake above in the same scope invalidates speculative results unless all of them were correct. Going to 8 in flight was no faster, so 4 is the practical cap.
