[morsel-ray]

Landed normally

014b98a after the independently approved changed82a formatter and normal accepted-base imports. Fresh exact632f/current028base/APPROVED/CLEAN/MERGEABLE/resolved and all11 actual COMPLETED/SUCCESS gates, ordinary protected expected-SHA squash. Complete checked/landed tree 77857a32f5351dce66a27bfbc29b897ee5c39ce5 identical; remote dev verified. No master/tag/release action. Closing references checked empty: this preserves the documented historical-crash reproduction limits rather than auto-closing733.

Current-head scoped tests below retain their provenance; original9e full and prior matched formatter prices are not relabeled as final-head executions.

Current accepted-prefetch integration — 632f1adc

Normal expected-head import of accepted #942 /02804c38. Exact ordered parents, computed complete merge tree and all4 unchanged feature blobs verified. Fresh non-root ALL-TU ClangASan/UBSan as4/text-null12/cast64/affine20, imported TIME3/metadata2/tiny chained-join pair1 passed; GCC builtins DEBUG/NDEBUG syntax2, whole relevant logs diagnostic0, exact clean head/both binaries checked independently. No new full, TSan, current native pricing or author-backtester claim. Earlier evidence retains its actual heads. Fresh current-head/base/review/discussion/required gates still precede normal landing.

Prior accepted-identity-LEFT integration — 3583c75d

Normal expected-head import of accepted #943 /f7b23ec7, exact ordered parents and computed complete merge tree verified. All4feature blobs unchanged. Fresh non-root ALL-TU ClangASan/UBSan as4/text-null12/cast64/affine20, imported TIME3/metadata2 passed; GCC changed production TU DEBUG/NDEBUG syntax2, whole relevant logs diagnostic0, clean exact head/both binaries verified. No new local full, TSan, combined-head native pricing or author-backtester claim. Fresh exact-head/current-base/review/discussion/required CI checks remain needed before normal landing. Earlier evidence retains its proper heads.

Prior TIME32 integration — 5ec3df53

Normal expected-head import of accepted #939 /5f6e619b. Ordered parents/computed whole merge tree verified; all4feature blobs unchanged. Fresh non-root ALL-TU/header-consistent ClangASan/UBSan as4, text-null12, cast64, affine20, imported TIME3 and metadata2 passed. GCC changed production TU DEBUG/NDEBUG syntax2, whole relevant logs diagnostic0, exact clean head/both binaries checked. No new local full/TSan/native/backtester claim; earlier evidence retains its named heads. Fresh exact-head CI/current-base/reviews/threads remain required before ordinary landing.

Prior B0 integration — 1bea85f1

Normal expected-head import of accepted #936 / b0b0a1b. Ordered parents and computed complete merge tree verified; all4feature blobs unchanged. Fresh non-root ALL-TU/header-consistent ClangASan/UBSan as4, text-null12, cast64, affine20 and imported metadata2 passed; GCC changed production TU DEBUG/NDEBUG syntax2, whole relevant logs diagnostic0, exact clean head/both binaries verified. No new local full, TSan, native pricing or author-backtester claim. Earlier measurements/controls below retain their original heads. Fresh exact-head CI is still required.

Prior D715 integration — 23099bd3

Anton approved the changed formatter at 82a9840, not only the earlier guard-only head. Normal import of accepted d7153e7 gives23099bd3: ordered parents/computed complete merge tree verified and all4feature blobs unchanged. Fresh non-root all-TU/header-consistent ClangASan/UBSan scopes: as4/4, text-null12/12, cast64/64, imported affine20/20; GCC builtins DEBUG/NDEBUG syntax2; whole relevant logs diagnostic0, exact clean head/both binaries checked. No new full/TSan/release-pricing claim. Waiting for this exact head's actual required checks.

Contract / issue733

STR collection conversion produces a flat STR vector, one string atom per source cell. Nested collection children recursively produce vectors, which must not be read through scalar string accessors. Reject that unsupported result with an owned type error, retaining the input and releasing partial output.

On accepted646 I reproduced erroneous empty strings for the grouped-parted/literal forms, not the historical reported misalignment abort. This addresses733's shape/contract defect without inventing flattening or nested-rank semantics; no historical-crash reproduction claim.

Reviewed feature: avoid per-flat-cell contract checks

82a9840, with accepted938/a587 integrated normally (no force-push):

• Resolve STR at operation entry; a flat typed vector's boxed elements go directly to the existing scalar STR formatter. Do not recursively resolve/dispatch the already-known target name for every cell.
• Factor the unchanged scalar formatting logic for use by both scalar and flat-vector paths. Preserve the existing scalar-null→empty-string rule, F32/F64 formatting, zero handling, domain translation and source null metadata.
• Choose flat/LIST kernels once. Only heterogeneous LIST recursion needs the successful-result kind check. Allocation/error propagation and owned/borrowed element handling remain; no generic executor-validation layer is added.
• Nested LIST children still follow recursive/error semantics and are rejected if they produce vectors. No early rejection that changes recursive-error ordering.

The changed82a formatter subsequently received Anton’s independent approval. The current-base import and fresh scopes are documented above; fresh exact-head CI remains required.

Optimized evidence — not an alleged40% guard penalty

First I tested moving just the kind check out of the flat loop. It gave no robust win; I did not publish a refactor merely for counting fewer checks. The meaningful gain below removes pre-existing per-cell target redispatch, not only the new shape guard.

Matched accepted-a587+guard baseline versus resolved-target kernel, same Clang18 project O3/native flags, one pinned i7-6700 CPU, background helpers0. Private native cast entrypoint,500,000 inputs, input setup and every output cell's bytes/length validated outside timing, one discarded process per variant/case then five randomized fresh processes. Shared host/governor not isolated.

native conversion	baseline median ms	new median ms
I64 vector→STR	96.239	56.361
F64 vector→STR	211.971	174.560
BOOL vector→STR	69.382	30.213
scalar-child LIST→STR control	80.829	81.154

LIST sample ranges overlap; no LIST improvement is claimed. No whole-engine, fused-query, ingestion, cold-cache or cross-platform speedup claim. Both sides include938's append fix, so it is not a confounded comparison against927's older646 base.

Method/provenance: same common optimized archive, replace only the changed builtins TU with the actual captured project flags; hash changed source/archive/executable. Not two fresh whole-PR release builds. Complete private source baseline34412292 equals the normal9e+a587 merge tree; final82a differs from it only in the measured/tested builtins.c blob. Source/whole-tree difference independently verified after publication.

Correctness evidence and limits

Production-identical private checkout on current baseline, non-root Clang18 DEBUG ASan/UBSan: rfl/type/as 4/4, text_null 12/12, cast 64/64 (overlapping scopes), whole relevant logs diagnostic0. Existing nested/mixed/empty/parted rejection and source-preservation fixtures retained.

Separate optimized baseline/candidate comparison: 88 scalar/vector records byte-identical, including null flags and unchanged source payload: signed widths, BOOL/U8, F32/F64 NaNs/infinities/subnormals/signedzero, GUID, DATE/TIME/TIMESTAMP/TIMESPAN, runtime SYM/STR, long and embedded-NUL text, slices. This is scoped differential evidence, not a universal correctness theorem or foreign-domain/exhaustive-error audit. All positive timing outputs also independently matched numeric/text expectations. GCC DEBUG and DEBUG+NDEBUG isolated syntax/warning checks passed.

The original full non-root 6413/6430,17skipped,0failed belongs to9e196613 only, not this new kernel. No new local full suite, full TSan/GCC runtime,32-bit or allocation-failure injection claim. The adjacent pre-existing append-error/formatter-error ownership caveats remain outside this repair; no new failure-atomicity promise. Fresh required CI and independent review remain necessary.

[morsel-ray]

Updated to 82a9840 after the release-cost pass, with accepted938 integrated normally. Your approval of9e was for the earlier small guard; this adds a real resolved-target STR kernel, so I am requesting fresh review rather than carrying that approval forward. Kind-check-only specialization did not show a reliable gain; eliminating the existing repeated target lookup/dispatch did (scoped figures/provenance and unchanged-format/null controls are in the revised body). Old full-suite evidence remains attributed9e, not this revision.