From 0252fffefda152647f9347b1c42ec85b738900a3 Mon Sep 17 00:00:00 2001 From: Hiroshi Shinaoka Date: Fri, 17 Jul 2026 22:29:20 +0900 Subject: [PATCH] rules: separate focused local checks from hosted validation --- rules/common/docs-and-tests.md | 21 +++++++++++++++++++++ 1 file changed, 21 insertions(+) diff --git a/rules/common/docs-and-tests.md b/rules/common/docs-and-tests.md index 5198645..d1bbbde 100644 --- a/rules/common/docs-and-tests.md +++ b/rules/common/docs-and-tests.md @@ -40,6 +40,27 @@ - Do not relax tolerances, skip failing coverage, or remove checks without a clear reason. +## Local And Hosted Validation + +- Ordinary local development and focused edit-test loops should use + non-release builds with incremental compilation enabled. Keep local checks + proportional to the changed surface instead of requiring every contributor + to rebuild and run the complete workspace before each pull request. +- Local pull-request preparation should run formatting and focused tests for + changed code, relevant documentation checks for documentation-only changes, + and focused CI-helper checks for CI-only changes. Unknown paths should fall + back to the conservative code-change policy. +- Hosted CI owns comprehensive workspace tests, coverage enforcement, feature + and backend matrices, documentation builds, hardware-dependent tests, and + clean builds with incremental compilation disabled. +- Use release mode locally when optimization semantics matter: benchmarks, + performance claims, release-only failures, unsafe or optimization-sensitive + behavior, or an explicit maintainer request. Release mode is not the default + for ordinary correctness-oriented edit-test loops. +- Repository-local policy may require stricter local validation for a specific + project or change class. Such overrides should explain why hosted CI alone is + insufficient for that risk. + ## Benchmarks - Use release-mode benchmarks for performance claims.