I checked five COLA suites on Linux x86-64 using the headers from e13d0309557594c0becee3c1a5e32ce1f3b4771d, with the test-only oracle correction 9336a9ce4ad166c1aa7147fef2e31701f9194d91. No production header changed for this check. The same test correction is available on main as a66f7de471695628cc00048a1a4efd554bb80b3d. This is correctness and portability evidence, not a benchmark.

Environment and scope

Host: Linux x86-64, Intel Core i9-12900K.
Linux 6.8.0-100-generic, x86-64, glibc 2.35.
Ubuntu Clang 20.0.0, build b74e588e1f46, using the installed libstdc++ 12 headers.
C++20, -O2 -g -Wall -Wextra -Wpedantic -Werror -UNDEBUG, -fsanitize=address,undefined -fno-omit-frame-pointer -pthread.
Runtime: ASAN_OPTIONS=detect_leaks=1:halt_on_error=1 and UBSAN_OPTIONS=halt_on_error=1:print_stacktrace=1.
Default compiler ISA target; no -march=native, AVX2 or AVX-512 override.

I copied the headers and five standalone test sources directly by SCP to a fresh temporary directory. The snapshot contained 55 files and 686,068 bytes before the oracle correction. Compilation and execution used the host's cpu-heavy resource gate and a reserved build directory, with at most two concurrent compiler processes. The test programs ran sequentially.

Results

Suite	Result
`cola_index`	Passed ASan/UBSan
`mapped_cola`	Passed ASan/UBSan
`mapped_cola_builder`	Passed ASan/UBSan
`cola_terminal`	Passed ASan/UBSan
`cola_local_merge`	Passed ASan/UBSan

The initial strict cola_index compile failed on a deprecation diagnostic in libstdc++ 12's std::stable_sort implementation: its temporary-buffer helper was marked deprecated. The independent oracle had already appended each origin's entries in ordinal order. I made that tie break explicit and used std::sort on key, origin, then ordinal, preserving its occurrence order. The original compiler diagnostic is retained; no warning suppression was used.

The suites exercise owning and mapped dual-route indexes, exact targets, metadata-only terminal construction, malformed sections, query results and local merge stages. This check does not cover SQLite, the complete package or Doxygen build, every Everett component, or every x86 SIMD target.

Reproduction and evidence

The original and corrected source manifests, exact compiler commands, toolchain output, executable hashes and test output are retained in original results and corrected results. I verified every remote source hash against the pinned Git objects; only tests/cola_index.cc differs between the manifests. The original compiler diagnostic is also retained. Paths in the evidence use the current Everett spelling. The host check did not edit a checkout; its resource gate wrote the normal run logs.

Each compiler invocation uses this form, with the full argument arrays recorded in the JSON evidence:

/usr/bin/clang++ -std=c++20 -O2 -g -Wall -Wextra -Wpedantic -Werror -UNDEBUG \
  -fsanitize=address,undefined -fno-omit-frame-pointer -pthread \
  -I/tmp/everett-world-linux-YR37tcNZ/include \
  /tmp/everett-world-linux-YR37tcNZ/tests/cola_index.cc \
  -o /tmp/everett-world-linux-YR37tcNZ/build/cola_index
ASAN_OPTIONS=detect_leaks=1:halt_on_error=1 \
UBSAN_OPTIONS=halt_on_error=1:print_stacktrace=1 \
  /tmp/everett-world-linux-YR37tcNZ/build/cola_index

Compilation and execution ran serially under an exclusive host resource lease. Test runtimes in the raw records are verification metadata and are not comparisons with another implementation.