diff --git a/03-DESIGN/01-to-be/47-milestones/H-lead-answers.md b/03-DESIGN/01-to-be/47-milestones/H-lead-answers.md index df8cc7f0..db750d64 100644 --- a/03-DESIGN/01-to-be/47-milestones/H-lead-answers.md +++ b/03-DESIGN/01-to-be/47-milestones/H-lead-answers.md @@ -1735,3 +1735,13 @@ Where two preparers disagreed, the chosen one is named. H-A-294's neighbour, A-474); (2) a take-in writes a journal line, and the indexer's take-in state says it, for every wait longer than a minute on another part (the forge, a package publish), naming the part. P3, P-failure-loud, nothing-moved-in-five-minutes. +- **H-A-299 (records and stores that outgrow the bus; the deliveries read, 19:47)** Met live after the 36-mover pipeline: + the gate's ledger records for two main revisions exceed the bus's message size and are refused; the forge bridge's + store of performed acts is full ("maximum bytes exceeded"); the deliveries read tool cannot answer per repository + (its filter runs after the answer is sized, 4 MB). Owed, H2: (1) a ledger record is one key per module and + revision, never one record for a whole revision; (2) every store a module keeps has a cap it prunes toward (the + oldest performed acts first) and a tool that reads its fill and applies its limits (tools-are-how-a-module-is- + spoken-to); (3) a read tool filters before it sizes and answers paged or capped, never "too big". And a fact: hq is + not a source repository of the gate, so an hq pull request gets no delivery and no check from the mesh; it merges + on hq's own checks run before every commit (ADR 0527), as the day's earlier hq pull requests did; whether hq should + be a watched source with mesh/check is a question for the operator (it changes how they work). P3, P-failure-loud.