Design 47 H2: H-A-465, the take-in's manifest read sets an unknown word aside

This commit is contained in:
2026-11-22 04:05:59 +01:00
parent 5fd71bf9a2
commit aa4bd831b7
@@ -781,3 +781,12 @@ The lead's one-line answers to the questions H2's develop and review raise (impl
refused; H-A-464's stated reason does not hold, its result stands (the module's account is the simpler, correct shape,
and the one-process reading of ADR 0436 §6 stays). When scoped grants return (H2b), the builder's bucket delete and its
run-time consumers need their grants, as the review notes.
- **H-A-465 (the take-in's manifest read sets an unknown top-level word aside, 06:10)** A gap in the build and delivery
system found before H2's merge: the running indexer reads every manifest strictly when it takes a revision in (its
fingerprint inputs), so a manifest carrying a word the running indexer's SDK does not know yet (`class`) gets no
fingerprint and nothing of the revision delivers, the indexer that knows the word included. The take-in's read
(ParseManifestForInputs) sets an unknown top-level word aside, as the SDK's decoder does, and the fingerprint is taken
from the words it knows; the judgement's strict read stays (the validator built from the revision's own indexer
refuses what it does not know). Delivered on its own before H2, as the smallest change of main; the baseline-words
mechanism the engine has (ADR 0614) is the thorough shape for the indexer, noted for H2b. design-it-as-it-should-be,
P-failure-loud (the validator still refuses; only the read that must survive a new word is lenient).