https://blog.chuanxilu.net/en/categories/classic-theory-meets-agent-practice/Recent content in Classic Theory Meets Agent Practice on Chuanxilu for Skilled Homo sapiensHugoen-USMon, 25 May 2026 10:00:00 +0800https://blog.chuanxilu.net/en/posts/2026/05/review-dimension-experiments/Mon, 25 May 2026 10:00:00 +0800https://blog.chuanxilu.net/en/posts/2026/05/review-dimension-experiments/Code review dimensions went from 8 to 11, design review from 11 to 14 — known-issue detection rose from 1/6 to 6/6 in both. The inspiration came from Axiomatic Design — a 36-year-old theory. But more dimensions are not always better — adding a math formula made results worse. The key is controlled experiments with known issues as reference.https://blog.chuanxilu.net/en/posts/2026/05/signal-purity-less-is-more/Sun, 24 May 2026 08:00:00 +0800https://blog.chuanxilu.net/en/posts/2026/05/signal-purity-less-is-more/A review prompt went from 317 lines to 135 lines, and review quality improved by 29%. What I removed was not useful procedure, but redundant content the model could infer on its own. What stayed were strategy genes: irreplaceable constraints, negative examples, and tone locks.https://blog.chuanxilu.net/en/posts/2026/05/dual-pass-review-recall-precision-tradeoff/Fri, 22 May 2026 10:00:00 +0800https://blog.chuanxilu.net/en/posts/2026/05/dual-pass-review-recall-precision-tradeoff/A design review agent needs to find every issue AND avoid false positives. One agent can't do both well. Borrowing cascade retrieval from information retrieval — a 15-year-old method — I split the agent into two: one for recall, one for precision. Real defects get caught earlier, and the risk of rework during development drops.