Validation matrix

VALIDATION.md at the repository root is the single structured source of truth for cross-checked numerics. Every validated case carries:

• the reference being matched (citable publication, BModes Fortran solver output, closed-form analytical formula);

• the quantity being compared (mode frequency, mode-shape MAC, polynomial coefficient, derived quantity);

• the tolerance (relative or absolute, with units);

• the worst-observed margin at the time the matrix was last regenerated;

• the test file that enforces the tolerance in CI;

• the external-data flag — whether the test runs on a fresh clone (self-contained) or only with upstream data staged under external/.

Mechanical audit

scripts/audit_validation_claims.py parses every test-file link in VALIDATION.md, asserts the path exists, and asserts the file (or directory glob) contains at least one def test_… method. Runs as a required CI step alongside ruff and mypy, plus step 4.5 of Release checklist.

Claims cannot drift ahead of tests. If you remove a test, you have to remove (or replace) its row in the matrix in the same PR; if you add a row with no test, the audit blocks the PR.

Tracks

The matrix splits validation work into three tracks:

Track A — frequency accuracy

Direct mode-frequency comparisons against a reference. The golden cases:

• NREL 5MW on OC3 Monopile vs BModes JJ (CS_Monopile.bmi) — ≤ 0.005 % across 10 modes.

• NREL 5MW on OC3 Hywind floating spar vs BModes JJ (OC3Hywind.bmi) — ≤ 0.0003 % across 9 modes.

• BModes CertTest Test03 + Test04 (82.4 m tower with top mass + tension-wire support) — < 0.005 %.

• Rotating uniform blade vs Wright 1982 / Bir 2009 Table 3a closed form — synthetic, self-contained.

• Rotating blade with tip mass vs Bir 2010 Table 5 — same.

• Rotating pinned-free cable vs Bir 2009 Eq. 8 (analytical Legendre polynomial) — same.

Track B — coefficient consistency

Whether the polynomial blocks shipped in industry ElastoDyn .dat files are reproducible from the structural-property blocks in the same files.

They are not — see cases/ECOSYSTEM_FINDING.md. The worst observation: TwSSM2Sh on NREL 5MW land deck, file-RMS 5.90 vs pyBmodes-RMS 0.0023 (ratio 2,529×). pybmodes validate surfaces this as a per-block PASS / WARN / FAIL verdict; pybmodes patch rewrites the blocks from the structural inputs.

Six patched reference decks ship in the wheel under src/pybmodes/_examples/reference_decks/ — every coefficient block reaches PASS or WARN; no FAIL after patching. The before_patch.txt snapshots preserve the original drift for reference.

Track C — supporting pipeline

Component-level regressions for everything that supports the two main tracks:

• BMI / ElastoDyn / SubDyn / WAMIT / MoorDyn parser round-trips

• Mooring catenary closed forms (Jonkman 2007 Appendix B)

• Hydrostatic restoring vs cylinder closed form

• Mode-shape classifier (FA / SS / twist labelling)

• Degenerate-pair resolver (symmetric-tower 2-fold eigenspaces)

• Pre-solve sanity checks

• Serialisation round-trips

• CLI smoke tests (every subcommand, every flag)

What “self-contained” means

The default pytest run is self-contained: every test in it runs from numbers either constructed inline in the test or validated against published closed-form formulas. No third-party reference data is bundled in the repo for this default run.

Any commit that re-introduces a .bmi / .dat / .out file under tests/data/ for a default-run test should be questioned. See Data sources for how external data is staged under external/ and gated behind the integration pytest marker.

The full matrix

pyBmodes validation matrix

This document is the single structured source of truth for what pyBmodes is validated against, at what tolerance, with what worst observed error, and which test file enforces it. Prose-heavy reports elsewhere in the repo (cases/ECOSYSTEM_FINDING.md, src/pybmodes/_examples/reference_decks/VALIDATION_SUMMARY.md, the README’s Validation section) refer back to this matrix.

Enforcement. The Validation workflow (weekly cron + workflow_dispatch) clones the upstream OpenFAST / IEA-Task-37 repositories on the fly and runs pytest -m integration hard-fail — no exit-5 tolerance, unlike the per-PR ci.yml. The verifier report uploaded by every run is the machine-checkable record that the workflow’s coverage holds at the tagged commit.

The workflow validates the cases in the Needs external data column whose upstream is clonable from a public GitHub repository at a manifest-pinned SHA (every clone in external/MANIFEST.toml got a real commit SHA in 1.8.1). Cloning is manifest-driven — verify_external_data.py --clone fetches every entry not marked optional = true, so the workflow and the manifest can’t drift. The required upstreams actively cloned and checked by validation.yml are:

• OpenFAST/r-test — the NREL 5MW land + monopile + OC3 Hywind regression decks.

• IEAWindTask37/IEA-3.4-130-RWT — IEA-3.4 reference turbine.

• IEAWindTask37/IEA-10.0-198-RWT — IEA-10 reference turbine.

• IEAWindTask37/IEA-15-240-RWT — IEA-15 (v1.1 tower; Allen 2020 free-decay reference).

• IEAWindTask37/IEA-22-280-RWT — IEA-22 reference turbine.

• WISDEM/WISDEM — the WindIO ontology examples used by the geometry-corpus rows.

verify_external_data.py --strict then hard-fails if any of these required clones is missing or off its pin, so the verifier report can’t look green while a pinned, non-optional entry went unchecked. The only entries left out of CI are the ones marked optional = true — MoorPy and RAFT (cross-comparison references) and the BModes archive (not a public clone) — which stay maintainer-local until a corresponding test lands. Adding a turbine to, or removing one from, the required set is a deliberate maintainer action (flip optional) documented under Changed in the CHANGELOG.

The two BModes CertTest cases (Test03, Test04) depend on external/BModes, the BModes v3.00 distribution — Fortran source + CertTest decks, also on GitHub at old-NWTC/BModes, but government-funded reference data pyBmodes does not bundle or auto-fetch in CI. Those tests skip cleanly when the data is absent (module-level pytestmark = pytest.mark.integration plus per-file .is_file() guards) and stay a maintainer-local enforcement until the BModes archive is mirrored somewhere CI can clone. The matrix row’s worst-observed columns reflect the local run; the public CI workflow does not currently re-verify them.

The validation work is split into two tracks with different metrics:

• Track A — frequency accuracy. Eigenvalue agreement against an external reference (closed-form formula, published table, or the BModes Fortran solver). Tolerance is a relative frequency error.

• Track B — coefficient consistency. The mode-shape polynomial blocks shipped in OpenFAST ElastoDyn .dat files. Tolerance is the RMS residual of the polynomial against the FEM mode shape.

Track A — frequency-accuracy cases

Case	Source / reference	Quantity	Tolerance	Worst observed	Test file	Needs external data
Uniform Euler-Bernoulli cantilever, modes 1-3	\(\beta_n L = \{1.875,\,4.694,\,7.855,\ldots\}\) (textbook)	flap frequency	< 0.5 %	< 0.005 %	tests/fem/test_cantilever.py	no
Same, mode 4	textbook	flap frequency	< 1 %	(within tol)	tests/fem/test_cantilever.py	no
Same, mode 5	textbook	flap frequency	< 2 %	(within tol)	tests/fem/test_cantilever.py	no
Hermite cubic mesh-convergence	analytical \(h^4\) rate	error ratio at h × 2	> 5 (i.e. ≥ 5×)	confirmed	tests/fem/test_cantilever.py	no
Cantilever + tip mass, \(\mu \in [0, 5]\)	Blevins (1979) / Karnovsky & Lebed (2001) frequency equation	1st bending frequency	< 0.5 %	(within tol)	tests/fem/test_uniform_tower_analytical.py	no
Rotating uniform blade, flap modes 1-3, Ω ∈ [0, 12] rad/s	Wright et al. (1982) — Vibration Modes of Centrifugally Stiffened Beams, J. Appl. Mech. (transcribed from Bir 2009 Table 3a)	flap frequency	< 0.5 %	(within tol)	tests/fem/test_rotating_uniform_blade.py	no
Rotating uniform blade + tip mass, flap modes 1-2, Ω ∈ [0, 12] rad/s	Wright et al. (1982) (Bir 2010 Table 5)	flap frequency	< 0.1 %	(within tol)	tests/fem/test_rotating_blade_with_tip_mass.py	no
Inextensible spinning cable, flap modes 1-3, Ω ∈ [2, 30] rad/s	Bir 2009 §III.B / Eq. 8: \(\omega_k = \Omega\sqrt{k(2k-1)}\)	flap frequency	< 0.5 %	(within tol)	tests/fem/test_rotating_cable.py	no
Uniform steel tube cantilever via Tower.from_geometry (D, t, L → derived EI / mass)	Euler-Bernoulli closed form, \(\beta_1 L = 1.875104\)	1st bending frequency	< 0.1 %	machine-exact (forward-derived from geometry)	tests/test_geometry_windio.py	no
n_nodes mesh refinement is bias-free + self-convergent (issue #35)	uniform tube: identical to native grid + Euler-Bernoulli; taper: self-convergence \(n{=}200\) vs \(400\)	1st & 4th bending frequency; mode-shape sample count	uniform = EB < 0.1 %; taper 1st < 0.2 %, 4th < 2 % (\(200\) vs \(400\))	(within tol)	tests/test_geometry_windio.py	no
BModes v3.00 CertTest Test01 — non-uniform rotating blade, 60 rpm	BModes Fortran solver .out	per-mode frequency, modes 1-6	< 1 %	< 0.005 % across 20 modes	tests/test_certtest.py	yes
BModes v3.00 CertTest Test01, modes 7+	BModes Fortran solver .out	per-mode frequency	< 3 %	(within tol)	tests/test_certtest.py	yes
BModes v3.00 CertTest Test02 — rotating blade + tip mass + offsets	BModes Fortran solver .out	per-mode frequency, modes 1-6 / 7+	< 1 % / < 3 %	< 0.005 % across 20 modes	tests/test_certtest.py	yes
BModes v3.00 CertTest Test03 — 82.4 m tower with top mass + c.m. offsets	BModes Fortran solver .out	per-mode frequency, modes 1-6 / 7+	< 1 % / < 3 %	< 0.005 % across 20 modes	tests/test_certtest.py	yes
BModes v3.00 CertTest Test04 — Test03 + tension-wire support	BModes Fortran solver .out	per-mode frequency, modes 1-6 / 7+	< 1 % / < 3 %	< 0.005 % across 20 modes	tests/test_certtest.py	yes
CS_Monopile — NREL 5MW Reference Turbine on the OC3 Monopile (hub_conn = 3, soft monopile, mooring stiffness)	BModes JJ (v1.03.01) .out	per-mode frequency, first 10 modes	< 0.01 %	< 0.005 %	tests/test_certtest.py	yes
OC3Hywind — NREL 5MW on the OC3 Hywind floating spar (hub_conn = 2, full hydro + mooring + 6×6 platform inertia)	BModes JJ (v1.03.01) .out	per-mode frequency, first 9 modes	< 0.01 %	≤ 0.0003 % across surge / sway / yaw / roll / pitch / heave + 1st-2nd tower bending	tests/test_certtest.py	yes
Degenerate-pair resolver, symmetric tower	construction (EI_FA == EI_SS)	post-rotation FA / SS purity	\(p_{\text{FA}}, p_{\text{SS}} > 0.99\)	(within tol)	tests/test_classifier.py	no
IEA-3.4 modes 1-2 — degenerate-pair resolver fires no warning	IEA Wind Task 37 RWT deck	classifier verdict	no RuntimeWarning	clean	tests/test_classifier.py	yes

Citations (full author / year forms used in the table above). NREL reports link to the canonical docs.nrel.gov PDF; journal / conference papers link by DOI; textbooks have no DOI and are cited by title + publisher:

• Blevins (1979). Formulas for Natural Frequency and Mode Shape. Krieger Publishing. (Textbook; no DOI.)

• Karnovsky & Lebed (2001). Formulas for Structural Dynamics. McGraw-Hill. (Textbook; no DOI.)

• Wright, Smith, Thresher & Wang (1982). Vibration Modes of Centrifugally Stiffened Beams. Journal of Applied Mechanics 49(1), 197–202. DOI 10.1115/1.3161966.

• Bir (2005). User’s Guide to BModes (Software for Computing Rotating Beam-Coupled Modes). NREL/TP-500-39133. PDF.

• Bir (2009). Blades and Towers Modal Analysis Code (BModes): Verification of Blade Modal Analysis Capability. AIAA 2009-1035. DOI 10.2514/6.2009-1035.

• Bir (2010). Verification of BModes: Rotary Beam and Tower Modal Analysis Code. NREL/CP-500-47953. PDF.

• Jonkman (2007). Dynamics Modeling and Loads Analysis of an Offshore Floating Wind Turbine (Appendix B, catenary mooring). NREL/TP-500-41958. PDF.

• Jonkman, Butterfield, Musial & Scott (2009). Definition of a 5-MW Reference Wind Turbine for Offshore System Development. NREL/TP-500-38060. PDF.

• Jonkman & Musial (2010). Offshore Code Comparison Collaboration (OC3) for IEA Wind Task 23. NREL/TP-5000-48191. PDF.

• Jonkman (2010). Definition of the Floating System for Phase IV of OC3. NREL/TP-500-47535. PDF.

• Bortolotti, Tarrés, Dykes, Merz, Sethuraman, Verelst & Zahle (2019). IEA Wind TCP Task 37: Systems Engineering in Wind Energy — WP2.1 Reference Wind Turbines. NREL/TP-5000-73492. PDF.

• Gaertner et al. (2020). IEA Wind TCP Task 37: Definition of the IEA 15-Megawatt Offshore Reference Wind Turbine. NREL/TP-5000-75698. PDF.

• Allen et al. (2020). Definition of the UMaine VolturnUS-S Reference Platform Developed for the IEA Wind 15-Megawatt Offshore Reference Wind Turbine. NREL/TP-5000-76773. PDF.

Track B — coefficient-consistency cases

The metric here is the RMS residual of an ElastoDyn polynomial block evaluated against the FEM-computed mode shape produced by the deck’s own structural inputs. See pybmodes.elastodyn.validate for the implementation. Per-block verdicts: PASS < 0.01, WARN < 0.10, FAIL ≥ 0.10.

Case	Source / reference	Quantity	Tolerance	Worst observed (file_rms)	Test file	Needs external data
Validator on stock NREL 5MW r-test deck — TwFAM2Sh / TwSSM2Sh	OpenFAST r-test (commit dd5feaaa)	file polynomial RMS vs pyBmodes mode shape (detection target)	verdict = FAIL	5.08 (TwFAM2Sh) / 5.90 (TwSSM2Sh)	tests/test_validate.py	yes
Validator on same deck — 1st tower modes + blade modes	OpenFAST r-test	file polynomial RMS	verdict = PASS	0.0081 / 0.0075 / 0.0020-0.0090	tests/test_validate.py	yes
Patch round-trip on staged NREL 5MW copy	self (post-patch_dat)	file polynomial RMS after patching	verdict = PASS, ratio ≈ 1.0	ratio drift < 1 % (text-precision artefact)	tests/test_validate.py	yes
reference_decks/nrel5mw_land/ patched deck	committed deliverable	per-block verdict	PASS or WARN, no FAIL	all PASS	tests/test_reference_decks.py	no (artefact tracked)
reference_decks/nrel5mw_oc3monopile/ patched deck	committed deliverable	per-block verdict	PASS or WARN, no FAIL	all PASS	tests/test_reference_decks.py	no
reference_decks/iea34_land/ patched deck	committed deliverable	per-block verdict	PASS or WARN, no FAIL	all PASS	tests/test_reference_decks.py	no
reference_decks/nrel5mw_oc3spar/ patched deck (OC3 Hywind floating spar; cantilever basis)	committed deliverable	per-block verdict	PASS or WARN, no FAIL	all PASS	tests/test_reference_decks.py	no
reference_decks/nrel5mw_oc4semi/ patched deck (OC4 DeepCwind semi; cantilever basis)	committed deliverable	per-block verdict	PASS or WARN, no FAIL	all PASS	tests/test_reference_decks.py	no
reference_decks/iea15mw_umainesemi/ patched deck (UMaine VolturnUS-S; cantilever basis)	committed deliverable	per-block verdict	PASS or WARN, no FAIL	WARN on TwSSM2Sh (1.6 % RMS — ElastoDyn-basis representation limit; see footer in the deck’s validation_report.txt)	tests/test_reference_decks.py	no
Pre-patch sanity — at least one before_patch.txt shows FAIL or WARN	committed before-patch reports	per-deck overall verdict	≥ 1 FAIL/WARN	6/6 FAIL	tests/test_reference_decks.py	no

Track C — supporting-pipeline behavioural cases

These tests gate the workflow layers that sit between the FEM core and the user — pre-solve sanity checks, mode-by-mode comparison, result serialisation, bundled report generation, batch directory processing, sparse-solver dispatch, Campbell-diagram orchestration, the ElastoDyn-compatibility blade adapter, polynomial- fit conditioning, and parser / writer round-trips. They don’t have a separate external-reference frequency to compare against; the gate is behavioural / contract-style.

Case	Source / reference	Quantity	Tolerance	Worst observed	Test file	Needs external data
check_model — non-monotonic span detected	construction	ModelWarning(severity='WARN') raised	always	(within tol)	tests/test_checks.py	no
check_model — zero / negative mass density detected	construction	ModelWarning(severity='ERROR') raised	always	(within tol)	tests/test_checks.py	no
check_model — stiffness jump > 5× detected	construction	ModelWarning(severity='WARN') raised	always	(within tol)	tests/test_checks.py	no
check_model — EI_FA / EI_SS extreme ratio detected	construction	ModelWarning(severity='INFO') raised	always	(within tol)	tests/test_checks.py	no
check_model — RNA mass > tower mass detected	construction	ModelWarning(severity='INFO') raised	always	(within tol)	tests/test_checks.py	no
check_model — singular PlatformSupport matrix detected	construction	ModelWarning(severity='ERROR') raised	always	(within tol)	tests/test_checks.py	no
check_model — n_modes > 6 × n_nodes detected	construction	ModelWarning(severity='ERROR') raised	always	(within tol)	tests/test_checks.py	no
check_model — polynomial-fit cond > 1e4 detected	construction	ModelWarning(severity='WARN' or 'ERROR') raised	always	(within tol)	tests/test_checks.py	no
Tower.run(check_model=True/False) auto-run integration	construction	UserWarning emitted on True / suppressed on False	always	(within tol)	tests/test_checks.py	no
mac_matrix — identity test (shapes vs themselves)	construction	diagonal entries equal 1, off-diag < 1 for distinct shapes	exact / np.allclose	(within tol)	tests/test_mac.py	no
mac_matrix — orthogonal shapes (FA-only vs SS-only)	construction	every entry = 0	exact / np.allclose	(within tol)	tests/test_mac.py	no
compare_modes — frequency-shift sign matches input direction	construction	sign of frequency_shift matches sign of (f_B − f_A)	exact / np.allclose	(within tol)	tests/test_mac.py	no
compare_modes — Hungarian-optimal mode pairing	construction	output pair = identity when shapes match	exact	(within tol)	tests/test_mac.py	no
plot_mac — matplotlib smoke test	construction	Figure has ≥ 1 Axes; title carries labels	structural	(within tol)	tests/test_mac.py	no
ModalResult.save / load — NPZ round-trip	self	per-field np.allclose vs original	rtol = 1e-12	(within tol)	tests/test_serialize.py	no
ModalResult.to_json / from_json — JSON round-trip	self	per-field equality vs original	exact / np.allclose	(within tol)	tests/test_serialize.py	no
ModalResult — metadata capture (version + timestamp + git hash)	self	metadata dict populated with non-empty pybmodes_version	always	(within tol)	tests/test_serialize.py	no
CampbellResult.save / load — NPZ round-trip	self	per-field np.allclose vs original	rtol = 1e-12	(within tol)	tests/test_serialize.py	no
CampbellResult.to_csv — spec column order	self	header = [rpm, <labels>, <labels>_mac]	exact match	(within tol)	tests/test_serialize.py	no
NPZ load is pickle-free on modern archives; legacy dtype=object __meta__ is refused by default (object-array unpickling can execute code) and loads only via the opt-in allow_legacy_pickle=True (F2)	construction (forged legacy archive)	modern load emits no warning; default legacy load raises ValueError (“refusing to load a legacy …”); opt-in load raises UserWarning + round-trips	exact / structural	(within tol)	tests/test_serialize.py	no
_validate_lengths / CampbellResult._validate reject malformed schemas before any export (F5)	construction (bad participation / label shapes)	ValueError before save / to_json / to_csv	always raises	(within tol)	tests/test_serialize.py	no
check_model n_modes guard uses exact n_free_dof, not 6·n_nodes — no false ERROR in (6·n_nodes, n_free_dof] (F3)	construction (nselt=10 ⇒ n_free_dof=90)	n_modes=80 clean; n_modes=200 ERROR	always	(within tol)	tests/test_checks.py	no
pybmodes patch rejects conflicting --output / --output-dir (F4); silent-agree preserved for equal / single	construction	exit 2 + clear message only when paths differ	exact match	(within tol)	tests/test_cli_patch_args.py	no
kaimal_spectrum — IEC 61400-1 longitudinal closed form	textbook: S(0)=4σ²L/U, strictly decreasing in f	low-freq plateau + monotonicity	exact / pytest.approx	(within tol)	tests/test_environmental.py	no
jonswap_spectrum — peak at 1/Tp; m0 = Hs²/16 significant-wave-height identity	textbook JONSWAP; the Hs identity is exact by construction	f_peak, ∫S df	peak abs=2e-3; m0 rel=2 %	(within tol)	tests/test_environmental.py	no
plot_environmental_spectra — structure (bands + tower lines + spectra + legend) and input guards	construction	axvspans / vlines / legend entries; ValueError on bad freq_max / harmonics	structural / always raises	(within tol)	tests/test_environmental.py	no
read_out(strict=True) — fail-loud on short / non-numeric / non-finite / duplicate-mode / empty .out; default stays tolerant	construction (malformed .out fixtures)	BModeOutParseError with file/line/mode context vs tolerant skip	exact match / always raises	(within tol)	tests/test_out_parser.py	no
generate_report — markdown contains frequencies	construction	every 4-dp frequency appears in body	exact match	(within tol)	tests/test_report.py	no
generate_report — HTML is well-formed	construction	DOCTYPE + balanced <table> / <tr> / <td> tags	structural	(within tol)	tests/test_report.py	no
generate_report — CSV has coefficient columns	construction	second header row has C2..C6, rms_residual, cond_number	exact column match	(within tol)	tests/test_report.py	no
pybmodes batch — discovery filter excludes Tower / Blade / SubDyn .dat	self (committed reference_decks/)	6 main decks found; 0 aux files	exact	(within tol)	tests/test_batch.py	no
pybmodes batch — summary CSV column set	construction	header = [filename, overall_verdict, TwFAM2Sh_ratio, TwSSM2Sh_ratio, n_fail, n_warn]	exact column match	(within tol)	tests/test_batch.py	no
pybmodes batch --patch — every block reaches PASS / WARN	OpenFAST r-test 5MW deck	per-deck overall_verdict post-patch	no FAIL	(within tol)	tests/test_batch.py	yes
pybmodes examples --copy --kind samples — vendors cases/sample_inputs/	self (committed bundle)	four analytical-reference subdirs + reference_turbines/ present at dest	exact match	(within tol)	tests/test_examples_cli.py	no
pybmodes examples --copy --kind decks — vendors reference_decks/	self (committed bundle)	nrel5mw_land/ present at dest; sample_inputs/ not vendored	exact match	(within tol)	tests/test_examples_cli.py	no
pybmodes examples --copy (default --kind all) — vendors both bundles	self (committed bundle)	both sample_inputs/ and reference_decks/ present at dest	exact match	(within tol)	tests/test_examples_cli.py	no
pybmodes examples — destination conflict guard	construction (pre-existing target subdir)	exit code 2 + preexisting file preserved	exact match	(within tol)	tests/test_examples_cli.py	no
pybmodes examples --force — overwrites stale destination	construction (pre-existing target subdir + stale file)	stale file removed; real bundle present	exact match	(within tol)	tests/test_examples_cli.py	no
WamitReader.read — surge A_inf vs manual redimensionalisation	IEA-15-240-RWT-UMaineSemi WAMIT .1	A_inf[0,0] (dim, kg)	rel = 1 %	(within tol)	tests/test_wamit_reader.py	yes
WamitReader.read — heave A_inf vs manual redim	IEA-15-240-RWT-UMaineSemi WAMIT .1	A_inf[2,2] (dim, kg)	rel = 1 %	(within tol)	tests/test_wamit_reader.py	yes
WamitReader.read — pitch A_inf vs manual redim	IEA-15-240-RWT-UMaineSemi WAMIT .1	A_inf[4,4] (dim, kg·m²)	rel = 1 %	(within tol)	tests/test_wamit_reader.py	yes
WamitReader.read — heave C_hst vs manual redim	IEA-15-240-RWT-UMaineSemi WAMIT .hst	C_hst[2,2] (dim, N/m)	rel = 1 %	(within tol)	tests/test_wamit_reader.py	yes
WamitReader.read — pitch C_hst positive (stable semi)	IEA-15-240-RWT-UMaineSemi WAMIT .hst	C_hst[3,3] > 0 and ≈ 2.193e9 N·m/rad	rel = 1 %	(within tol)	tests/test_wamit_reader.py	yes
WamitReader.read — A_inf and C_hst symmetric (no transposed-index bug)	IEA-15-240-RWT-UMaineSemi WAMIT pair	max(|M − M.T|) < 1e-3 · max(|M|)	always	(within tol)	tests/test_wamit_reader.py	yes
WamitReader._resolve_pot_path — quoting / separator robustness	construction (4 spellings of one path)	all forms resolve to the same Path	exact match	(within tol)	tests/test_wamit_reader.py	yes
WamitReader.read — clear error on missing PotFile	construction (non-existent root)	FileNotFoundError naming <root>.1	exact match	(within tol)	tests/test_wamit_reader.py	yes
HydroDynReader — ULEN / PotMod / PotFile / PtfmRefzt / ρ / g	IEA-15-240-RWT-UMaineSemi _HydroDyn.dat	6 scalar fields	exact / default	(within tol)	tests/test_wamit_reader.py	yes
WamitReader — upper-triangle-only .hst / .1 mirrored	synthetic 2-entry input	C[2,4] == C[4,2] and A_inf[0,2] == A_inf[2,0] after parse	exact	(within tol)	tests/test_wamit_reader_synth.py	no
WamitReader — Fortran D / d exponents accepted	synthetic 4.430486D+02 / 1.233416d+04	dimensional values match the C-form equivalents	rel = 1e-12	(within tol)	tests/test_wamit_reader_synth.py	no
_scan_platform_fields — raises on malformed critical scalars	synthetic deck with not_a_number   PtfmMass	ValueError naming the offending field	exact match	(within tol)	tests/test_mooring.py	no
Line.solve_static — inextensible limit matches Irvine 1981 §2.3	construction (EA → ∞)	residuals on Irvine eq. (2.27)–(2.29)	rel = 1e-4	(within tol)	tests/test_mooring.py	no
Line.solve_static — horizontal-line symmetry	construction (ΔZ = 0)	V_F = W·L/2 at fairlead	rel = 1e-6	(within tol)	tests/test_mooring.py	no
Line.solve_static — Newton residuals satisfied to ≤ solver tol	construction (typical slack-line geometry)	both catenary residuals after solve	< 1e-6 m	(within tol)	tests/test_mooring.py	no
MooringSystem.restoring_force — 3-fold-symmetric zero offset	construction (synthetic 3-line layout)	F_x, F_y, M_x, M_y, M_z all < 1e-3·|F\_z|	rel = 1e-3	(within tol)	tests/test_mooring.py	no
MooringSystem.stiffness_matrix — diagonal positive	construction (synthetic 3-line)	every diag entry > 0	always	(within tol)	tests/test_mooring.py	no
MooringSystem.stiffness_matrix — symmetric after symmetrisation	construction (synthetic 3-line)	max(|K − K.T|) < 1e-10 · max(|K|)	always	(within tol)	tests/test_mooring.py	no
MooringSystem.from_moordyn — OC3 Hywind layout round-trip	OpenFAST r-test 5MW_OC3Spar MoorDyn .dat	3 lines / 6 points / fairlead radius 5.2 / anchor radius 853.87 / L 902.2	rel = 1e-3	(within tol)	tests/test_mooring.py	yes
MooringSystem.stiffness_matrix — OC3 Hywind surge stiffness	OC3 r-test MoorDyn .dat	K[0,0] ≈ 41,180 N/m (Jonkman 2010 Table 5-1)	rel = 5%	within 0.01 %	tests/test_mooring.py	yes
MooringSystem.stiffness_matrix — OC3 yaw stiffness (catenary-only)	OC3 r-test MoorDyn .dat	K[5,5] ≈ 1.156e7 N·m/rad (catenary lines only; OC3’s delta-line crowfoot adds the bulk of the published 9.83e7)	rel = 5%	(within tol)	tests/test_mooring.py	yes
MooringSystem.stiffness_matrix — 3-fold-symmetry signatures	OC3 r-test MoorDyn .dat	K[0,0] = K[1,1], K[3,3] = K[4,4], K[0,4]·K[1,3] < 0	rel = 1e-4	(within tol)	tests/test_mooring.py	yes
Tower.from_elastodyn_with_mooring — end-to-end OC3 coupled solve	OC3 ElastoDyn + r-test MoorDyn .dat	hub_conn=2, PlatformSupport populated, 1st tower-bending in 0.4–0.6 Hz	rel = 5%	within 1.2 % of Jonkman 2010’s 0.482 Hz	tests/test_mooring.py	yes
MudlineFoundation.from_soil_properties — Yu & Amdahl (2023) Table 9 DTU 10 MW flexible pile, E_SO = 30 MPa	Yu and Amdahl (2023) Marine Structures 92, 103482, Table 9 row 2	K_hh, K_hr, K_rr via Shadlou and Bhattacharya (2016) formulas	rel = 10%	within 3.0 % on all three springs	tests/test_foundation.py	no
MudlineFoundation.from_soil_properties — Yu & Amdahl (2023) Table 9 DTU 10 MW rigid pile, E_SO = 30 MPa	Yu and Amdahl (2023) Table 9 row 1	K_hh, K_hr, K_rr via Shadlou and Bhattacharya (2016) rigid-pile formulas	rel = 5%	within 2.9 % on all three springs	tests/test_foundation.py	no
MudlineFoundation.as_mooring_K — cross-coupling sign matches Jonkman 2010 OC3 convention	construction	K[0,4] < 0, K[1,3] > 0, K[0,4]·K[1,3] < 0	always	(within tol)	tests/test_foundation.py	no
Randolph (1981) pile classifier — auto dispatch to flexible / rigid / intermediate	construction (synthetic geometries)	high L/D → flexible; low L/D → rigid; intermediate emits UserWarning and falls back to flexible	always	(within tol)	tests/test_foundation.py	no
Sparse eigsh shift-invert — matches dense eigh	construction (SPD problem at threshold + 100 DOFs)	lowest-k eigenvalues	rtol = 1e-8	(within tol)	tests/fem/test_sparse_solver.py	no
Sparse path triggered above threshold	construction	log message announces “sparse shift-invert”	always	(within tol)	tests/fem/test_sparse_solver.py	no
Sparse path skipped below threshold	construction	log message does not mention “sparse”	always	(within tol)	tests/fem/test_sparse_solver.py	no
Asymmetric problems fall back to dense general eig	construction (asymmetric K)	log message announces “dense general”; sparse not invoked	always	(within tol)	tests/fem/test_sparse_solver.py	no
pybmodes patch --dry-run writes nothing (mtime check)	construction	mtime of every staged file unchanged	exact match	(within tol)	tests/test_validate.py	yes
pybmodes patch --diff PR format contains × improvement ratio	construction	stdout contains ×, RMS improvement:, better literals	exact match	(within tol)	tests/test_validate.py	yes
pybmodes patch --output leaves source byte-identical	construction	source-file sha256 unchanged	exact match	(within tol)	tests/test_validate.py	yes
Default in-place pybmodes patch emits first-time-run hint	construction	stdout contains --dry-run --diff	exact match	(within tol)	tests/test_validate.py	yes
Tower torsion-contamination filter — rejects T_tor ≥ 10 %	construction (synthetic torsion-contaminated mode)	rejected_modes carries the contaminated mode	always	(within tol)	tests/test_classifier.py	no
Tower torsion-contamination filter — accepts pure bending	construction (synthetic clean modes)	rejected_modes is empty	always	(within tol)	tests/test_classifier.py	no
IEA-3.4 deck — torsion participations populated, summing to 1	OpenFAST IEA-3.4-130-RWT deck	per-mode (T_FA, T_SS, T_tor) triple	Σ = 1 to abs_tol = 1e-9	(within tol)	tests/test_classifier.py	yes
Campbell sweep — Hungarian MAC tracking on bundled NREL 5MW reference deck	committed reference_decks/nrel5mw_land/	mac_to_previous ≥ 0.90 between consecutive steps on a smooth sweep	always	(within tol)	tests/test_campbell.py	no
Campbell sweep — input validation (NaN / inf / negative / unsorted RPM)	construction	ValueError raised	always raises	(within tol)	tests/test_campbell.py	no
Campbell sweep — restores bbmi.rot_rpm after sweep (clean + on-exception)	construction	post-sweep / post-exception BMI state	unchanged	(within tol)	tests/test_campbell.py	no
Campbell sweep — tower modes constant across all rotor speeds	construction	tower frequency vs rotor speed	exactly constant (no rpm dependence in tower modal eigenproblem)	(within tol)	tests/test_campbell.py	no
plot_campbell — 6-DOF platform overlay: degenerate-pair merge + Hz/period labels + log axis (issue #39); default-off byte-identical; non-finite/≤0 skipped	construction (synthetic CampbellResult + platform_modes)	horizontal refs + right-margin labels + yscale; regression invariant w/o platform_modes	structural / exact	(within tol)	tests/test_campbell.py	no
ElastoDyn-compat blade adapter — strips str_tw, tw_iner, offsets	Jonkman 2015 NREL forum guidance	resulting BMI fields	zeroed for compat-on, preserved for compat-off	(within tol)	tests/test_elastodyn_compatible.py	no
ElastoDyn-compat — frequency drift on flap modes is small	construction	rel freq diff vs compat-off	≤ ~ few % on flap modes	(within tol)	tests/test_elastodyn_compatible.py	no
ElastoDyn .dat parse → write → parse round-trip	self	per-field np.allclose vs original	rtol = 1e-12	(within tol)	tests/test_elastodyn_reader.py	yes
Polynomial-fit design-matrix cond-number reporting	construction	RuntimeWarning above thresholds	WARN > 1e4, FAIL > 1e6	(within tol)	tests/test_fitting.py	no
BMI / section-properties parser primitives	construction (synthetic fixtures)	round-trip equality	exact / np.allclose	(within tol)	tests/test_io.py	no
FEM building blocks — boundary conditions, eigensolver, normalisation	construction	per-DOF / per-mode invariants	exact / np.allclose	(within tol)	tests/fem/test_*.py	no
tubular_section_props — closed-form tube identities	\(A=\pi(R_o^2-R_i^2)\), \(I=\tfrac{\pi}{4}(R_o^4-R_i^4)\), \(J=2I\), \(G=E/2(1+\nu)\) (textbook)	mass_den / flp_stff / tor_stff / axial_stff / flp_iner	exact	pytest.approx	tests/test_geometry_windio.py	no
outfitting_factor — scales mass + rotary inertia only, never stiffness	construction (factor 1.07)	mass × 1.07; EI / GJ / EA / structural ρI unchanged	exact	pytest.approx	tests/test_geometry_windio.py	no
Geometry guard — rejects D ≤ 0 / t ≤ 0 / 2t ≥ D	construction (degenerate sections)	ValueError raised	always raises	(within tol)	tests/test_geometry_windio.py	no
WindIO dialect equivalence — modern outer_shape/structure vs older outer_shape_bem/internal_structure_2d_fem	construction (numerically identical fixtures)	same WindIOTubular from both key dialects	exact / np.allclose	(within tol)	tests/test_geometry_windio.py	no
WindIO duplicate-anchor tolerance — IEA-10’s redefined-anchor habit	construction (&g redefined; stock SafeLoader raises ComposerError)	parses (last-wins) where strict PyYAML fails	exact	(within tol)	tests/test_geometry_windio.py	no
WindIO orthotropic wall material — clear out-of-scope error	construction (list-valued composite E)	ValueError naming “orthotropic”, not bare float(list) TypeError	exact match	(within tol)	tests/test_geometry_windio.py	no
WindIO thickness interp — linear vs piecewise-constant differs on a taper	construction (smoothly tapered wall)	mid-station wall thickness diverges by > 0.01 m	exact	0.035 vs 0.05 m	tests/test_geometry_windio.py	no
WindIO PyYAML-absent guard — names the [windio] extra	construction (monkeypatched import)	ModuleNotFoundError matching pybmodes\[windio\]	exact match	(within tol)	tests/test_geometry_windio.py	no
IEA-15 base WindIO yaml → derived mass / EI vs IEA-15 Monopile ElastoDyn tower table (RNA-independent like-for-like; that deck was generated from this geometry)	IEA Wind Task 37 IEA-15-240-RWT ontology + OpenFAST Monopile deck	distributed mass_den & FA EI, station-by-station	< 0.5 %	7.5 × 10⁻¹² (machine-exact)	tests/test_geometry_windio.py	yes
Upstream WindIO corpus — IEA-3.4 / 10 / 15 / 22 + WISDEM examples (both dialects + IEA-10 dup-anchor) parse to a physically sane tube	IEA Wind Task 37 RWT ontologies + WISDEM example yamls	grid ∈ [0,1] monotone, D / t > 0, 2t < D, steel E / ρ / ν range, span > 0	structural	12 yamls clean	tests/test_geometry_windio.py	yes
Tower.from_windio — older-dialect modal smoke (IEA-3.4 / 10 / 22)	IEA Wind Task 37 RWT ontologies	full FEM solve → bare-member spectrum	finite, positive, ascending	(within tol)	tests/test_geometry_windio.py	yes
Older-dialect yaml-derived mass / EI vs same-turbine ElastoDyn tower table (ballpark — those decks were not 1:1 geometry round-trips, unlike IEA-15)	IEA-3.4 / 10 / 22 ontology + own OpenFAST tower deck	distributed mass_den & FA EI envelope	mass < 25 %, EI < 30 %	(within tol)	tests/test_geometry_windio.py	yes
CLT laminate primitives — reduced stiffness / ply transform / ABD / membrane condensation	Jones, Mechanics of Composite Materials (2nd ed.) §2.5–4.3 closed forms	\(Q\), \(\bar Q(\theta)\), \(A/B/D\), \(\tilde A\)	exact	pytest.approx (12 cases)	tests/test_windio_blade.py	no
Airfoil nd_arc profile — arc spine / LE-TE landmarks / chord↔arc / blend	construction (analytic circle & ellipse)	s_le, tc, landmark (x,y), blend weights	exact (≤ 2e-3)	(within tol)	tests/test_windio_blade.py	no
WindIO web/layer nd_arc resolver — older-explicit ≡ modern-anchor-registry dialect equivalence	construction (numerically identical fixtures)	identical resolved bands from both dialects	exact / np.allclose	(within tol)	tests/test_windio_blade.py	no
Thin-wall reduction — isotropic tube & box vs exact closed form	textbook thin-ring / thin-wall-box EA / EI / GJ / mass	section properties	exact	< 2 % (discretisation)	tests/test_windio_blade.py	no
Multi-cell Bredt–Batho — symmetric interior web carries zero shear flow	analytic (a symmetric diametral web ⇒ GJ unchanged vs webless)	GJ, n_cells	exact	< 2 %	tests/test_windio_blade.py	no
windio_blade dual-dialect end-to-end	construction (tube blade, both WindIO dialects)	identical SectionProperties; physically sane	exact / np.allclose	(within tol)	tests/test_windio_blade.py	no
6×6 cross-sectional decoupling (issue #50) — rigid-offset congruence → tension centre → principal-axis eigen-decomposition → shear centre / GJ; analytic round-trips	construction (build a 6×6 from known decoupled props at a known offset / rotation)	recovered EA/EI_flap/EI_edge/GJ/mass/centres + defining decoupling property	exact (rtol≤1e-6) round-trip	(within tol)	tests/test_decouple.py	no
WindIO published distributed elastic properties → decoupled beam (issue #48/#50) — elastic="auto" prefers them over PreComp; both dialects parsed onto the span grid as the full 6×6 then decoupled (elastic/shear centre + principal elastic axes — not raw diagonal); "file"/"precomp"/bad-value guards	construction (tube blade with embedded published properties, both dialects)	SectionProperties from the decoupled 6×6; mode-string dispatch	exact / np.allclose; ValueError on bad/missing	(within tol)	tests/test_windio_blade.py	no
WindIO published elastic properties vs the turbine’s own BeamDyn 6×6, IEA-15, span 0.15–0.90 (issue #48/#50) — decoupled-vs-decoupled (the BeamDyn 6×6 is also referenced at the blade axis, so the apples-to-apples oracle is the decoupled BeamDyn, not its raw diagonal); also pins that raw≠decoupled so the case genuinely exercises #50, and that the published path stays far tighter than PreComp	IEA Wind Task 37 IEA-15 ontology + companion *_BeamDyn_blade.dat	decoupled mass / EA / EI_flap / EI_edge / GJ	mass & EA med < 3 %; EI med < 5 %; GJ med < 8 %; raw-vs-decoupled EI > 5 %; strictly tighter than elastic="precomp"	(within tol)	tests/test_windio_blade.py	yes
WindIO composite blade → distributed beam props vs the turbine’s own BeamDyn 6×6 (WISDEM-PreComp-generated; elastic="precomp" — the PreComp-fallback path’s true scope), IEA-3.4 / 10 / 15 / 22, span 0.15–0.90	IEA Wind Task 37 RWT ontology + companion *_BeamDyn_blade.dat	mass / EA (PreComp-class) ; GJ & EI (diagonal-reduction approximate — documented limitation)	mass med < 6 % / max < 15 %; EA med < 10 % / max < 18 %; GJ med < 22 % / max < 47 %; EI med < 35 % / max < 55 %	mass ≈ 1.5–4 % med ; EA ≈ 1–8 % med ; GJ ≈ 3–18 % med (IEA-10 worst, composite multi-cell torsion) ; EI ≈ 2–27 % med (weak axis worst — omits spar-cap-offset / bend-twist coupling)	tests/test_windio_blade.py	yes
WindIO elastic_properties_mb ecosystem-drift caveat — a turbine’s published _mb 6×6 is not always mutually consistent with its own companion BeamDyn deck (IEA-10 _mb ≈ 50 % from its BeamDyn 6×6); auto treats the WindIO ontology as canonical per #48 intent	IEA-10 ontology _mb vs *_BeamDyn_blade.dat	documented divergence (not a parser bug — same pattern as cases/ECOSYSTEM_FINDING.md)	n/a (documented, not gated)	≈ 50 % (intrinsic to upstream data)	cases/ECOSYSTEM_FINDING.md	yes
RotatingBlade.from_windio — modal smoke, all 4 RWTs (both dialects + parametric-layer blades)	IEA Wind Task 37 RWT ontologies	full FEM → parked-blade spectrum	finite, positive, ascending	(within tol)	tests/test_windio_blade.py	yes
WindIO floating reader — joints (cartesian + cylindrical [r,θ°,z]), member geometry, axial-joint resolution, transition	construction	parsed WindIOFloating	exact / np.allclose	(within tol)	tests/test_windio_floating.py	no
Hydrostatic restoring — single surface-piercing cylinder & off-axis column vs closed form	textbook waterplane + buoyancy (WAMIT/.hst convention)	C_hst 6×6	exact	< 1 % (discretisation)	tests/test_windio_floating.py	no
Morison added mass + RAFT end-cap & rigid-body inertia vs closed form	thin-ring transverse + RAFT Ca_End end disc; thin-wall steel	A_inf / mass 6×6	exact	pytest.approx	tests/test_windio_floating.py	no
from_windio_mooring — topology / props / 120°-symmetric stiffness / regression-warn / bad-ref	construction	MooringSystem + 6×6 K	exact / symmetry	(within tol)	tests/test_windio_floating.py	no
from_windio_floating — yaml-only coupled modal smoke (stable single-column FOWT)	construction	full coupled FEM spectrum	finite, ascending, soft RB ≪ tower	(within tol)	tests/test_windio_floating.py	no
WindIO floating C_hst vs the turbine’s own potential-flow WAMIT .hst (geometry-exact anchor)	IEA-15 UMaine VolturnUS-S ontology + companion HydroDyn/WAMIT	heave / roll / pitch restoring	< 3 % / < 4 %	heave 0.8 %, roll/pitch 1.6 %	tests/test_windio_floating.py	yes
WindIO screening-preview Morison A_inf & structural mass vs WAMIT / ElastoDyn (the yaml-only tier — NOT industry-grade, emits a UserWarning)	IEA-15 UMaine VolturnUS-S + companion HydroDyn / ElastoDyn	added-mass diagonal / PtfmMass	surge/sway/yaw < 45 %; heave/roll/pitch factor ~2; mass a lower bound	surge 22 %, heave 53 % (Morison ≠ BEM, as RAFT/WISDEM also find — supply a HydroDyn deck for industry grade); struct+fixed mass ≈ 0.36·PtfmMass (trim ballast excluded by design)	tests/test_windio_floating.py	yes
from_windio_mooring vs from_moordyn (the yaml-only screening mooring tier)	IEA-15 UMaine VolturnUS-S ontology + companion MoorDyn	deck-fallback line props; 6×6 K	props exact (rel<1e-9); roll/pitch<15 %, heave<15 %, yaw<20 %, surge/sway<40 %	props exact; roll/pitch ~3 %, heave ~9 %, surge/sway ~32 % (WindIO centreline vs MoorDyn surface-attachment radius — the screening path; the coupled industry path below uses the full MoorDyn system)	tests/test_windio_floating.py	yes
Tower.from_windio_floating (companion decks present) vs the BModes-JJ-validated from_elastodyn_with_mooring — the industry-grade tier (full deck model: MoorDyn system + WAMIT A_inf/C_hst + ElastoDyn PtfmMass/RIner+RNA; only the tower is the Phase-1 machine-exact WindIO one)	IEA Wind Task 37 ontology + companion HydroDyn/MoorDyn/ElastoDyn	coupled rigid-body + tower-bending frequencies	every platform RB mode + 1st tower bending < 1 %; 2nd+ tower harmonics < 8 %	all 6 platform rigid-body modes + 1st tower bending 0.0–0.3 % (reference grade); 2nd+ tower harmonics ≤ 6 % (the Phase-1 WindIO-vs-ElastoDyn tower-discretisation residual, orthogonal to the platform)	tests/test_windio_floating.py	yes
Tower.from_windio_floating(..., platform_support=...) — injected-platform tier (WindIO tower + caller-supplied PlatformSupport; floater designed separately, issue #35)	construction (synthetic FOWT yaml + stable diagonal PlatformSupport)	byte-equivalence to the manual WindIO-tower-BMI + attach-PlatformSupport recipe; no screening warning; decks mutually-exclusive ValueError; soft RB ≪ tower	rtol=1e-9 / exact / structural	(within tol) — adds no new numerics; the free-free PlatformSupport FEM is the OC3-Hywind-certified one (test_certtest)	tests/test_windio_floating.py	no
pybmodes windio one-click — companion-deck auto-discovery scoping (bare yaml ⇒ no decks/screening; RWT layout ⇒ scoped, config-matched, aux-excluded) + fixed-tower / floating-screening / bad-input	construction	discovery dict + report file + exit code + SCREENING warning	exact / structural	(within tol)	tests/test_windio_cli.py	no
pybmodes windio <IEA-15 VolturnUS-S> end-to-end — discovery picks the UMaineSemi floating decks ⇒ industry-grade deck-backed coupled model	IEA Wind Task 37 IEA-15 RWT tree (ontology + companion HydroDyn/MoorDyn/ElastoDyn)	discovered config + exit code + report + no screening warning	exact / structural	(within tol)	tests/test_windio_cli.py	yes

What “needs external data” means — and how integration coverage is gated

yes rows skip in the default pytest run (they carry the integration marker). Run them locally with pytest -m integration once you have the upstream sources:

• external/OpenFAST_files/r-test/ — clone of the OpenFAST regression-test corpus (any recent commit; pyBmodes was last validated against dd5feaaa).

• external/OpenFAST_files/IEA-3.4-130-RWT/, IEA-10.0-198-RWT/, IEA-15-240-RWT/, IEA-22-280-RWT/ — clones of the IEA Wind Task 37 reference-turbine repositories (each ships a WindIO ontology .yaml plus matching OpenFAST/ElastoDyn decks; the WindIO validation rows consume both).

• external/OpenFAST_files/WISDEM/examples/ — clone of the WISDEM examples tree (modern-dialect WindIO ontology yamls for the corpus-parse coverage).

• external/BModes/CertTest/ — BModes v3.00 CertTest reference outputs.

• external/BModes/docs/examples/ — the bundled CS_Monopile.bmi and OC3Hywind.bmi example decks plus their BModes JJ .out files.

These directories are gitignored under the Independence stance. The data is not bundled in the repo because the licence terms of the upstream NREL / IEA Wind Task 37 packages include attribution / indemnification obligations that pyBmodes can’t inherit by republication. The contributor clones them locally.

Integration-track coverage is split by whether the upstream data is publicly clonable:

• Public required set — CI-gated. The Validation workflow clones the required public upstreams (OpenFAST r-test + IEA-3.4 / 10 / 15 / 22 + WISDEM) at their manifest-pinned commit SHAs and runs pytest -m integration hard-fail — no exit-5 tolerance — on every workflow_dispatch and the weekly cron. It first runs verify_external_data.py --strict (pinned commit + line-ending- normalized content hashes; a missing required clone is a hard FAIL) and uploads the verifier report. Publishing to PyPI is gated on a green run of this workflow for the tagged commit. So the public integration cases carry re-runnable CI evidence, not just a maintainer claim.

• BModes CertTest + optional cross-comparisons — maintainer-local. The BModes Fortran solver and its CertTest decks are on GitHub (old-NWTC/BModes), but they are government-funded reference data pyBmodes does not bundle or auto-fetch in CI; the CS_Monopile / OC3Hywind rows additionally compare against .out files from the patched BModes_JJ binary (distributed only via the author’s personal drive, not an official release). MoorPy / RAFT are optional = true cross-comparison clones; those rows stay maintainer-local until/unless mirrored somewhere CI can fetch. Their tests skip cleanly when the data is absent.

• Per-PR ci.yml runs the self-contained suite and tolerates exit code 5 (“no tests collected”) on its integration step, because the default PR runner has no upstream decks — the hard-fail enforcement lives in validation.yml above, not in the per-PR job.

• scripts/audit_validation_claims.py (run in CI and the release checklist) scans this matrix and asserts that every test-file path named in a row exists and contains at least one collected test method — the gate that catches “claim ahead of test” drift.

For users: the public integration cases (the NREL 5MW family and the IEA Wind Task 37 turbines) carry re-runnable CI evidence via the Validation workflow, pinned to immutable upstream commits and hashed for content. Only the BModes CertTest guarantee remains “the maintainer ran these locally before tagging,” because that reference data isn’t publicly clonable.

Reproducing every row

Track A (no external data):

pytest tests/fem/test_cantilever.py
pytest tests/fem/test_uniform_tower_analytical.py
pytest tests/fem/test_rotating_uniform_blade.py
pytest tests/fem/test_rotating_blade_with_tip_mass.py
pytest tests/fem/test_rotating_cable.py
pytest tests/test_classifier.py        # 3 of 4 pass without external data
pytest tests/test_geometry_windio.py   # closed-form tube + dialect/anchor robustness
pytest tests/test_windio_blade.py      # CLT + thin-wall + multi-cell closed forms
pytest tests/test_windio_floating.py   # hydrostatic / Morison / mooring closed forms

Track A (external data needed):

pytest tests/test_certtest.py -m integration
pytest tests/test_geometry_windio.py -m integration  # WindIO corpus + IEA-15 anchor
pytest tests/test_windio_blade.py -m integration     # blade vs BeamDyn 6×6 (4 RWTs)
pytest tests/test_windio_floating.py -m integration  # IEA-15 VolturnUS-S vs WAMIT/MoorDyn/ED

Track B:

pytest tests/test_validate.py -m integration   # validator round-trip
pytest tests/test_reference_decks.py            # always self-contained

Or run the full matrix at once:

pytest -m ""        # default + integration combined