fix(ts): contextually type nested callbacks in Zod processors

Parent: #548

After #588, the untouched pinned Zod 4.5.2 production corpus retains nine false TS7006 diagnostics in src/v4/core/json-schema-processors.ts. They are all nested callbacks whose containing expressions have known library or array types:

• two Array.prototype.every predicates at lines 168-169
• one Array.prototype.map callback at line 245
• four Array.prototype.map callbacks at lines 373 and 416
• two JSON.stringify replacer parameters at line 558

Pinned tsc 6.0.3 and tsgo 7.0.2 remain silent on the unchanged strict NodeNext corpus.

Acceptance criteria:

• identify the real contextual-typing gap for these nested callbacks
• add minimal reference-validated regression coverage
• remove only semantically false diagnostics, with zero added diagnostics on the untouched 106-file corpus
• measure deterministic before/after diagnostic sets
• do not publish a timing or README row until the corpus is semantically admissible

First small chunk landed on origin/main: 90a9f5a (fix(ts): model complete array callback parameters (#589)).

The built-in Array model now matches the full lib callback parameter lists: (value, index, array) for map/every/filter/etc. and (accumulator, value, index, array) for reduce/reduceRight. Focused strict regressions verify both contextual types and the absence of the prior false TS2345; the full ts_checker suite, zig fmt --check, scoped Pickier, and git diff --check pass.

Untouched pinned Zod corpus: 1,166 -> 1,164 diagnostics, exactly 2 false diagnostics removed and 0 added, byte-identical across two runs (SHA-256 05b677ab53676ffdf922f486701060d54ab9ff8dc5da04230bafe8027d75b3eb). One removal is the directly explained full-arity .map TS2345 in classic/from-json-schema.ts; the other is a false TS2430. The nine target TS7006 diagnostics remain, so this issue stays open.

Two broader approaches were explicitly rejected before landing this chunk: richer JSON overload allocation produced one added false TS2430, and globally publishing qualified interface graphs produced two added false diagnostics. Neither was committed. This keeps the zero-addition benchmark gate intact.

Second #589 chunk landed on origin/main: aa659e5 (fix(ts): contextually type JSON transform callbacks (#589)).

Home now supplies the standard (key: string, value: any) context for built-in JSON.parse revivers and JSON.stringify replacers without changing the permissive lightweight overload surface. The implementation verifies the receiver is the cached built-in JSON object; a focused shadowed-JSON control continues to emit both expected TS7006 diagnostics.

Validation:

• focused built-in and shadowing regressions pass
• full ts_checker suite passes
• zig fmt --check, scoped Pickier, and git diff --check pass
• untouched pinned Zod corpus: 1,164 -> 1,162 diagnostics
• exact delta: the two intended json-schema-processors.ts:558 TS7006 diagnostics removed, 0 added
• two optimized runs byte-identical, SHA-256 fcfb4df6c0f270edc09a9a6287be10eb0b4c1690df1ce0d198fb63089b24bd4a

Seven nested array-callback TS7006 diagnostics remain, so #589 stays open.

Further root-cause work isolated the remaining imported-return case to Readonly<Record<string, EnumValue>>. An exact mapped-schema experiment passed ts_program but crashed the untouched Zod run with exit 139 and introduced a false reduction TS2345, so it was rejected and fully reverted; the empty crash output is not counted. The unsafe mapped-utility transfer path is now tracked in #591. Worktree and origin/main remain clean at aa659e5.

Safety update: 67adab666 on origin/main fully reverts the unsafe transferred utility/qualified-reference experiment tracked by #591. The admitted untouched Zod baseline is restored and deterministic at 1,162 diagnostics across two byte-identical ReleaseFast runs (SHA-256 0d1d85991dcbb73deb00dba10da1bbef60c10c48fd50b3819887096553110ab9). The seven target array-callback TS7006 diagnostics remain, so this issue stays open and no timing/README claim is being published from the rejected experiment.

Correction: the reported 1,162 -> 1,821 comparison was invalid because 82 generated .js files had contaminated the pinned src/v4 tree and changed NodeNext resolution. Those files are quarantined, the source tree matches the official Zod 4.5.2 tarball byte-for-byte, and an explicit --noEmit run restores the admitted 1,162-diagnostic baseline (SHA-256 0d1d85991dcbb73deb00dba10da1bbef60c10c48fd50b3819887096553110ab9). The utility experiment remains removed pending a clean retest; the seven target diagnostics still define this issue.

#609 now has a safe foundation on origin/main: 0bd698363 retains fully transferable scalar namespace aliases inside exported interfaces, with exact cross-file array callback context in the focused reduction and a zero-change Zod corpus gate (1,162 diagnostics, 0 added/0 removed, two byte-identical --noEmit runs). Rich structured and mapped projections remain deliberately unsupported, so the seven target Zod callbacks are unchanged and #589 stays open.

#591 landed an independently valid safety prerequisite as 6c4c9be01: object-literal mismatch checking now snapshots interner members before nested checks can relocate the pool. Full checker tests pass and the clean Zod gate remains exactly 1,162 diagnostics (0 added/0 removed, baseline-identical hash twice). A compact Readonly<Record<...>> candidate did remove the two target .every TS7006s, but also added 156 processor TS7006s and two false TS2430s, so it was fully rejected. Seven target diagnostics therefore remain.

Progress landed on origin/main via d8fc967a9 and 23a559690; child issue #591 is complete.

The readonly Record transfer removes the two target .every TS7006 diagnostics at json-schema-processors.ts:168-169. Together with the previously admitted JSON callback work, the #589 target set is now down from nine false TS7006 diagnostics to five. The same change also removes two false callbacks in schemas.ts:3516,3520, two false impossible-comparison diagnostics, two false generic-interface diagnostics, and one downstream false assignment.

Untouched Zod gate: 1,162 → 1,153 diagnostics, 0 added / 9 removed, two byte-identical runs at SHA-256 265739efe928fd81e15df0d5a7012cc8bb1aabe6d355903760b227559ede451e, explicit --noEmit, and zero generated JavaScript files. Full checker/program suites pass, and both focused reductions are accepted by version-verified TypeScript 6.0.3, native TypeScript 7.0.2, and Home.

#589 remains open for the five remaining processor callback diagnostics; no timing or README benchmark row is admitted from this correctness-only result.

Progress landed on origin/main in 2fdb7d0c9 (fix(ts): project qualified assertion array members (#609)).

On the admitted Zod 4.5.2 corpus this removes 4 more nested callback TS7006 diagnostics with 0 added diagnostics (1,153 -> 1,149): both x and i callbacks at json-schema-processors.ts lines 373 and 416. Two runs were byte-identical (9a171804efa705d3c9cc15053c5b6b02e87704b415af3c638b1e6e9ef3c27145) and emitted no JS.

One target callback remains: m in mime.map at line 245, tracked under #609.

Completed by a5ae69070 on origin/main (fix(ts): project indexed assertion array bindings (#609)).

The last target callback, m in mime.map at json-schema-processors.ts:245, now receives contextual typing. The admitted Zod corpus moves from 1,149 to 1,148 diagnostics with 0 added / 1 removed; two runs are byte-identical (a9747a7f3a4f28de0fc3737b17fd3af2cad83a72ff6fed03ed2aff4c11734b4a) and emit no JS.

Together with the earlier #591 and #609 commits, every nested-callback target tracked by this issue is resolved.