Design 47 H2: H-A-411, M1's store rows and the validator's home

This commit is contained in:
2026-11-22 00:57:16 +01:00
parent f987fbe3f2
commit 2ba6b946f0
@@ -328,3 +328,16 @@ The lead's one-line answers to the questions H2's develop and review raise (impl
age, at most 768 KB with capped), and the taken asks come from the holder's own state `taken`; no read of the broker's
show for it and no new seat verb. Built in mesh-sdk (S16) and mesh-tools (T4) now; the assigner and the builder build
their tools against those heads. P9, P-NATS-native, tools-are-how-a-module-is-spoken-to.
- **H-A-411 (M1's store rows: the findings stream's numbers, the caps of existing stores, the prune's reach, the
validator's home, 00:58)** §783, §812, §814, §820; corrects H-A-378 and H-A-394 in part. (1) The finding-events row is
`max-mib` 8 and `max-age-days` 7 (the events template's when-full is drop-oldest already); no setting findings-max-age
is declared, since nothing reads one (a setting nothing reads says what is not true, P-failure-loud). (2) Every state
of a delivery module that names no cap gets `max-mib` 8 (as §820's states); each log and objects store names its present
engine default explicitly. (3) M1's prune reach: each delivery module's prunable state is bound through the prune helper
and its two tools served, with prune-mark and prune-target, by the developer whose share the module is (the gate and
artifact-store-tools on feat/47-H2-c; the three build modules and distribution with BM/RG on feat/47-H2); the forge
module is TypeScript and mesh-sdk has no TypeScript prune helper, so its states refuse when full, said in its journal,
until one exists (H2b: the TypeScript prune helper). (4) The manifest validator is mesh-judge in mesh-delivery's
indexer, not in mesh-catalog: the §814 rules (a kept store with no cap, the finding-events numbers, `class` present on
a seat-claiming manifest) are the indexer developer's I6, and H-A-378's and H-A-394's "mesh-catalog developer builds
the rule" read so. P3, P-failure-loud, ADR 0595 §3.