write an enum-keyed map as z.partialRecord, which accepts the subsets serde writes

[eburgos]

z.record over an enumerated key schema demands every member be present, so a
partially populated HashMap<EnumKey, V> — and an empty one — was rejected by
its own generated schema. Measured against real serde payloads: a partial map
failed with one missing-key issue per absent member per field, an empty map
with all of them, while the full map alone parsed. TypeScript had said
Partial<Record<...>> all along, and the JSON document carries no required,
so Zod was the one surface rejecting what the type system and the document both
admit — and what a Rust HashMap plainly holds.

The record call is chosen off a new wire answer for an enumerated key, recorded
where a plain enum registers and forwarded by the existing brand and alias
forwarding — the constructor's question being whether the key's published Zod
binding enumerates, which a brand over an enum does even though its JSON stays
an open object. A registry miss answers as enumerated, which is what a key
naming an enum declared later reads: measured safe, since partialRecord is
runtime-identical to record over every non-enumerating key schema this crate
emits, and a forward-declared key that is not an enum is already refused
elsewhere.

TypeScript and the JSON document do not move. A stray key is still refused.
Verified on the verbatim emission with real payloads: partial, full and empty
all parse byte-identically, ajv agreeing before and after — the before-state
verdicts are recorded in the pinning test, showing Zod alone was wrong.

just fmt, just check, just quick, just lint-all across all 68 toggles, and the
test powerset across all 68 in four partitions — all green.