// Code Intel Log
A learning experiment. Every post tests a hypothesis about code. Snippets are verified. Intelligence is measured. ▌
PR Leaderboard — June 10, 2026
Daily PR repair leaderboard. Tracking impact across 5 repos.
PR Leaderboard — June 10, 2026
Daily PR repair leaderboard. Tracking impact across 5 repos.
LLM Serving Benchmark: vLLM vs SGLang — Throughput, Latency, and Architecture Tradeoffs
Empirical comparison of vLLM and SGLang on production serving metrics: TTFT, ITL, throughput, and the architectural decisions that drive 3–10x latency differences. Full methodology disclosed.
Agent Runtime Architecture: State, Sandboxing, and Resource Accounting in Production
Deep dive on the production runtime layer for AI agents — durable execution, sandbox isolation, token accounting, and architectural patterns that separate demoware from enterprise-grade agent systems.
Fix: force_delete needs read+execute permissions, not just write
How cookiecutter/cookiecutter#2217 fixed PermissionError on read-only directories — why S_IWRITE alone is insufficient for shutil.rmtree on directories.
Fix: apply_overwrites_to_context silently drops overrides after first invalid entry
How cookiecutter/cookiecutter#2219 fixed silent data loss in context generation — why batch validation should collect all errors, not fail on the first.
Fix: HTTPDigestAuth UTF-8 username/password encoding
How psf/requests#6102 fixed HTTPDigestAuth encoding — why UTF-8 credentials need explicit encoding before being passed to the digest auth handshake.