diff --git a/03-DESIGN/01-to-be/47-milestones/H2-lead-answers.md b/03-DESIGN/01-to-be/47-milestones/H2-lead-answers.md index c0649e9c..fa964303 100644 --- a/03-DESIGN/01-to-be/47-milestones/H2-lead-answers.md +++ b/03-DESIGN/01-to-be/47-milestones/H2-lead-answers.md @@ -961,3 +961,9 @@ The lead's one-line answers to the questions H2's develop and review raise (impl the engine composed the setting as 5 (the manifest's default), so a resolve with source "default" does not reach the node's layer today. The config seat composing a key again when its declaration changes is the next milestone's item, with a journal line in the engine if a stale resolve ever wins. P-evidence. +- **H-A-478 (grants-for answers from what the pass already composed, 17:10)** The grants part's `grants-for` reads + the inputs again and composes every module twice per ask (3 to 4 s against the runner's 5 s bound; cut + three times today). It answers from what it keeps: the newest revision's grants composed once per revision at the + pass and kept, the candidate's composed once per candidate revision and kept, the last pass's inputs reused; each + stage's time in the journal. The runner's `widens-read-bound` default becomes 30 s, so a cold ask is not cut. + P3, P-failure-loud.