Release Claim-Evidence Ledger

This ledger is the release-prep scratchpad for matching user-visible claims to validation evidence. Use it before writing release notes: keep claims narrow, name the evidence that backs them, and call out plausible but unvalidated combinations directly.

Status vocabulary:

• Release-validated: covered by current documented local commands and CI jobs for the named toolchain or mode.
• Evidence-backed: covered by source, tests, examples, or docs, but not a broad compatibility claim.
• Plausible but unvalidated: may work, but is not part of the release-validated support boundary.
• Non-goal: intentionally outside the current release claim.

V1.0 Release-Note Claim Map

The v1.0 release note may claim only the support areas mapped to the ledger rows below:

• serial timing: disciplined serial wall-clock timing with structured summaries, reports, CSV, scoped guards, callbacks, mock-clock-friendly tests, and benchmark evidence;
• pure MPI: summaries and reports through the mpi_f08 communicator contract;
• OpenMP compatibility: existing ftimer/ftimer_core users through the master-thread-only carve-out;
• ftimer_openmp_t: opt-in true OpenMP worker timing through the explicit object API;
• strict/sparse hybrid output: strict and sparse union MPI+OpenMP rank/lane summary, report, and CSV output, with OpenMPI as the routine CI-covered path and MPICH as focused local-evidence-backed caveated support from #353;
• stable CSV/export claims: stable CSV/export schemas within the documented schema lines;
• installed CMake package behavior: installed CMake package consumption for supported feature modes and same-major 1.x package-version matching;
• public symbols: a curated source-level public symbol boundary; and
• benchmark evidence: harness evidence for release-readiness sweeps and performance-risk PRs.

The same release note should not claim broad compiler portability, package-manager availability, profiler-backend integration, hardware-counter support, FPM support, benchmark pass/fail thresholds, traces, dashboards, device synchronization, or automatic MPI barriers unless a later release issue adds matching evidence. The current Spack and EasyBuild readiness note records out-of-tree recipe guidance without making fTimer a package-recipe owner.

V1.0 Deferred And Non-Goal Classification

Topic	V1.0 classification	Evidence or disposition
Package-manager availability and recipe ownership	Non-goal for v1.0; post-v1.0 strategic decision tracked by #355.	docs/package-manager-readiness.md is readiness guidance only: no in-repository Spack/EasyBuild recipes, no direct spack/eb execution, and intentional prototype placeholders.
Fixed-team or no-explicit-region OpenMP worker timing	Post-1.0 design issue #294.	Current ftimer_openmp_t support remains the explicit timed-region, level-1 worker model documented in docs/openmp-timing-modes.md and docs/semantics.md.
MPICH MPI+OpenMP permanent CI coverage	Non-goal for v1.0 under the #353 maintainer decision.	OpenMPI MPI+OpenMP is routine CI-covered. MPICH MPI+OpenMP remains local-evidence-backed caveated support, not permanent PR CI coverage.
Ubuntu 24.04 MPICH pFUnit runner migration	Post-release runner maintenance tracked by #259.	Current pure-MPICH MPI CI/pFUnit coverage remains on the validated hosted Ubuntu 22.04 path with a raw launcher probe.
NVHPC serial smoke/install-consumer support	Plausible but unvalidated; post-release investigation #256.	Hosted-runner NVHPC 26.3 built but smoke/install-consumer executables aborted at runtime, so README and release notes must not claim NVHPC validation.
Profiler backends, hardware counters, traces, dashboards, accelerator/device synchronization, automatic MPI barriers, FPM, binary packages	Non-goals for v1.0 unless a later release issue accepts ownership and evidence.	Current docs describe wall-clock summaries and CSV exports only; docs/release.md owns artifact policy and release-note guardrails.

Claim Ledger

Release claim area	Current status	Evidence to cite	Caveats and compatibility boundary
Serial timing	Release-validated for GNU Fortran and LLVM Flang smoke/library builds; pFUnit behavioral coverage is GNU Fortran.	build-serial, build-serial-flang, and test-serial CI jobs; tests/test_*.pf; tests/test_phase0_smoke.F90; examples/basic_usage.F90; README.md "Supported toolchain matrix".	Other serial compilers may work, but are plausible and unvalidated until release evidence exists. Wall-clock timing does not synchronize accelerator/device queues.
Pure MPI	Release-validated for GNU Fortran wrapper builds with OpenMPI and MPICH.	build-mpi, build-mpi-mpich, test-mpi, and test-mpi-mpich CI jobs; tests/mpi/test_mpi_*.pf; examples/mpi_example.F90; configure-time mpi_f08 probe in CMakeLists.txt; docs/semantics.md MPI contract.	Requires MPI lifetime discipline and the mpi_f08 type(MPI_Comm) contract. Legacy integer communicators, mpif.h, divergent collectives, and automatic MPI barriers are outside the claim.
OpenMP compatibility	Release-validated for GNU Fortran OpenMP and LLVM Flang OpenMP smoke/example paths.	build-openmp, build-openmp-flang, and test-openmp CI jobs; tests/test_openmp_guards.pf; tests/check_openmp_option_off_global_flags.cmake; examples/openmp_example.F90; docs/openmp-timing-modes.md.	This is the master-thread-only carve-out for existing APIs. It is not general thread safety, thread-local timing, nested worker timing, or a hybrid timing model by itself.
ftimer_openmp_t	Release-validated as the opt-in serial-lane and level-1 worker timing API in OpenMP builds; serial-context lifecycle/catalog/timing is available in non-OpenMP packages.	tests/test_openmp_api_smoke.F90; tests/test_openmp_summary_smoke.F90; OpenMP diagnostic smoke tests; tests/check_installed_openmp_hybrid_api_surface.cmake; examples/openmp_worker_example.F90; docs/installed-api.md.	Worker timing requires FTIMER_USE_OPENMP=ON, an opened level-1 timed region, pre-registered ids for the hot path, and stopped-run summary/report calls. Global downstream OpenMP flags do not retrofit OpenMP support into a non-OpenMP fTimer package.
Strict/sparse hybrid output	Release-validated for OpenMPI wrapper MPI+OpenMP smoke/install-consumer coverage with GNU Fortran and OpenMP; evidence-backed local MPICH wrapper smoke/install-consumer coverage is caveated by #353 and is not permanent CI coverage.	build-mpi-openmp CI job; #306 local MPICH 5.0.1/GNU Fortran 15.2.0 ctest --test-dir build-codex-306-mpich-hybrid --output-on-failure run, 34/34 passed; tests/test_openmp_mpi_summary_smoke.F90; tests/test_openmp_mpi_summary_3rank_smoke.F90; tests/check_openmp_mpi_union_descriptor_contract.cmake; examples/mpi_openmp_example.F90; tests/install-consumer/mpi_openmp_main.F90; README.md hybrid contract.	MPICH hybrid is local-evidence-backed caveated support for v1.0, not routine PR CI coverage. Treat other MPI/compiler/OpenMP runtime combinations as plausible but unvalidated. Strict hybrid output rejects descriptor, eligible-lane, and mixed-epoch unknown missing-lane precision mismatches. Sparse union hybrid output is a separate participation-aware API and schema with explicit unknown missing-sample flags; it is not append-compatible with strict hybrid CSV.
Stable CSV/export claims	Evidence-backed for the documented schema lines and append-validation behavior.	tests/test_file_output.pf; tests/check_openmp_summary_smoke.cmake; tests/check_openmp_mpi_summary_smoke.cmake; tests/check_bench_csv_output.cmake; CSV sections in README.md, docs/semantics.md, docs/csv-schema.md, and docs/troubleshooting.md.	CSV is the first stable machine-readable export, not a trace/event/profiler format. Local/strict MPI, sparse MPI union, OpenMP, strict hybrid, and sparse hybrid schemas are dedicated and intentionally not append-compatible with one another.
Installed CMake package behavior	Release-validated for same-major 1.x package-version matching and the supported serial, MPI, OpenMP, and MPI+OpenMP installed-consumer paths.	ftimer_installed_package_consumer* CTest entries; tests/check_installed_package_consumer.cmake; tests/installed-consumer-contract/package-version-probes.cmake; tests/install-consumer/*.F90; installed docs check for docs/installed-api.md and LICENSE; docs/installed-api.md.	A 1.x install may satisfy same-major requests at or older than the installed package version. Future-major requests and same-major requests newer than the installed package are rejected. Installed .mod artifacts are compiler-, wrapper-, and feature-mode-specific. FPM, binary packages, cross-compiler module reuse, and broad package-manager support are non-goals unless separately validated.
Spack/EasyBuild readiness	Preliminary out-of-tree recipe guidance inferred from fTimer's CMake install contract, not direct package-manager execution or maintained package-manager ownership. Serial, MPI, OpenMP, and MPI+OpenMP variants appear package-manager friendly when the package manager follows fTimer's install contract and feature-mode boundaries.	docs/package-manager-readiness.md; fTimerConfig.cmake dependency exports; cmake/install_ftimer_modules.cmake.in; tests/check_installed_package_consumer.cmake; installed-consumer CTest entries.	No Spack or EasyBuild recipe is committed in this repository. spack and eb were not available in the #311 local validation environment, so direct package-manager execution remains future upstream/site-recipe work. #355 owns any later package-manager ownership decision.
Public symbols	Evidence-backed by an allowlisted source-level API boundary.	tests/public_symbol_allowlist.txt; tests/check_public_symbol_allowlist.cmake; docs/installed-api.md; src/ftimer.F90, src/ftimer_core.F90, src/ftimer_openmp.F90, and src/ftimer_types.F90.	New module-level public names must be intentionally classified as stable, unstable public-by-necessity, or test-only. Installed implementation module artifacts are not stable import targets.
Benchmark evidence	Evidence-backed for trend review and release-readiness sweeps, not pass/fail thresholds. Durable comparisons require the benchmark CSV plus the provenance sidecar policy in docs/release.md.	build-bench, build-openmp-bench, and build-mpi-openmp-bench CI jobs; bench/ftimer_bench.F90; bench_csv_smoke; serial CI artifact ftimer-bench-serial-<sha>; benchmark commands in README.md and docs/release.md.	GitHub-hosted runner timings are noisy. Serial CSV artifacts are uploaded by CI; OpenMP and MPI+OpenMP benchmark CI jobs currently smoke-check parseable CSV but do not upload durable feature-enabled artifacts. Feature-enabled trend evidence is local CSV plus sidecar unless a future issue adds sidecar-aware CI artifact upload.

Release Prep Update Rules

• Add a row only for a release-significant claim that maintainers may repeat in release notes, README support text, or downstream compatibility guidance.
• Keep evidence concrete: prefer CI job names, CTest names, examples, and source-of-truth docs over broad phrases like "covered by tests".
• When a new combination looks likely but lacks CI or local release evidence, mark it Plausible but unvalidated instead of widening the support matrix.
• When evidence changes because a CI job or smoke test is renamed, update this ledger in the same PR as the rename.
• Do not turn benchmark evidence into a threshold or gate here. This file records what evidence exists; docs/release.md owns release execution policy.