Design 47 H2: H-A-415 to H-A-418

This commit is contained in:
2026-11-22 01:11:00 +01:00
parent d78897ee1d
commit 702f226920
@@ -386,3 +386,33 @@ The lead's one-line answers to the questions H2's develop and review raise (impl
header of failing tests from the results. (b) `output` answers no-longer-kept only when this node's steps key shows the
ask ended here more than the objects' max-age ago; an ask never run here is refused naming it (failure must be loud).
P-failure-loud, P3.
- **H-A-415 (the validator's three findings at the H2 heads, 01:10)** (1) The gate and artifact-store-tools are refused
"finding-events sets max-mib to 8, not 16 (G §529)": H-A-411 (1) sets 8 MiB, so the indexer's rule (I6, mesh-judge)
takes 8 and the gate's and the sweep's rows stand. (2) the node-engine (`mesh-host`) is refused for invoking `versions` on the two
registry seats, which H2's definitions no longer declare (D4a folded a version read into `list` with `module`, §813
(7), DMC-10): the node-engine reads a module's versions with `list {module}` and its manifest invokes `list`; one E
row more, the node-engine's developer. (3) gitea is refused in both cases because its package seats' claim serves
publish, publish-part, remove and versions that H2's package-registry definitions do not hold: the forge module's
claim names only the verbs the seat defines (`list` for versions, as in (2)); the mesh-catalog developer of
feat/47-H2-c. §812's prune of the gate's `source` ends: `source` is truth (a branch's counter), written by
compare-and-set; only `heads` is pruned (the developer's reading, confirmed). P3, a rule states how it is checked.
- **H-A-416 (a want's run; the composer writes the Version the node's engine reads, 01:13)** A4. (1) A want's `run` is
the source revision (decimal) of the pipeline whose passed end wrote it; a want from a put or a placement with no
pipeline carries the newest source the assigner holds; "a run whose pipeline ended writes nothing more" means only
that the engine refuses a want with a lower run than the one it applied (H-A-355, H-A-384 (2)). (2) The composer is
the node-engine's own build, run by the assigner's composition run, and mesh-sdk's composition inside it writes the
declaration; so the node-engine's manifest states both `declaration-reads: 2` and `declaration-writes: 2`, the assigner's
neither, and H-A-388's plan edge falls away, since the writer and the reader are one module: instead the composer
writes node X's declaration at the lower of its own Version and the Version X's running engine states it reads (from
X's have, 1 when it states none), leaving out the fields that Version lacks (drain-bound takes the engine's default),
so an older engine never meets a Version it refuses and the engine updates itself through its own composer. P3,
design-it-as-it-should-be; H-A-388 (2) read so.
- **H-A-417 (keep-now is a verb of mesh-wants, 01:13)** A5. The acting tool assigner_keep_now reaches the process by the
seat verb `keep-now` on mesh-wants (input why; answer the revision kept, or the fault), added to the seat's definition
in mesh-delivery-contracts by the assigner's developer, regenerated. H-A-398's rule. P-NATS-native.
- **H-A-418 (a withdrawn reservation re-pins through pipeline-ended; a reserved version pins through the series key,
01:13)** §805, I4. (1) The indexer hears a reservation's end as the runner's pipeline-ended not passed whose revision
is a version-reserved's revision, and re-pins that module's importers to the series' newest released version; no new
seat event. (2) version-reserved writes the series key's numbered with the contents read at the commit of the
module's entry in the event's revision; acceptedSeries pins the numbered when newer than the newest; the release's
package-published then moves no fingerprint and keepRepinned keeps no revision. P3.