# Known limitations, with measured magnitudes Every figure below was measured from data inside this table, not asserted. ## 1. Anisotropy is collapsed to isotropy — 19,299 rows Wood and fibre-reinforced laminates are orthotropic. This table stores one E (the longitudinal / fibre-direction modulus) and one ν, so an isotropic solver will get the **deformation shape** wrong even when the magnitude is right. The table contains 287 rows whose shear modulus is source-reported, which lets the isotropic identity G = E/(2(1+ν)) be checked directly: | material group | n | closure vs measured G | |---|---|---| | isotropic classes (metals, ceramics, glass, rock) | 273 | **0.24 % median error** — identity holds | | wood (Populus tremula veneer, oak consensus) | 2 | over-predicts **3.3–5.2×** | | fibre laminates (carbon / aramid / basalt / glass) | 7 | over-predicts **3.8–8.9×** | | paper, machine direction | 13 | over-predicts 1.4–1.6× | The isotropic control group at 0.24 % shows the method is sound; the error is entirely the anisotropy collapse. Transverse stiffness of wood and laminates is therefore too high by roughly 3–9×. For rigid-body contact this is irrelevant; for FEM/MPM deformation it is the dominant error. Particulate composites (24 glass-sphere/epoxy rows) are genuinely isotropic and carry no such caveat. ## 2. E on wood is a density regression — 19,033 rows Only 1.4 % of wood rows have a species-measured E. The rest are computed from density by the USDA Wood Handbook specific-gravity regression. - **13,265 rows (ρ ≥ 500) carry a calibrated regression value** (×1.226). The factor comes from `fill_log.csv`, which retains both the round-1 regression value and the round-5 USDA measured MOE for the same cell. Leave-one-out validation, n = 10: median bias **−18.4 % → +0.0 %**, share of \|error\| > 20 % **30 % → 0 %**. - **5,768 rows (ρ < 500) carry the uncalibrated regression** — the required factor scatters 0.50–1.29 there, so no single factor is supportable. Residual bias ≈ −18 %. - **672 rows are flagged for a low-density over-prediction**: below ρ ≈ 400 the power law over-predicts; measured over-estimate on 5 reference species is **+54 % median** (worst: *Ceiba pentandra*, measured 2.8 GPa vs regression 5.56 GPa, +99 %). The residual disagreements are structural rather than calibration error: the exponent 0.72 compresses the dense end, and the 17 calibration species are all tropical imports from USDA Table 5-5a, so temperate softwoods remain slightly low. ## 3. Moisture basis — all wood is on a 12 % MC basis GWDD v.2 is an ecology database: it stores **basic density** (oven-dry mass / green volume) and the matched USDA values are **green** MOE. CIRAD stores 12 %-MC density and 12 %-MC MOE. Those two conventions are not comparable, and a species present in both databases would otherwise appear twice on different scales. The 18,046 GWDD rows are therefore expressed at 12 % MC: ρ × **1.213**, E × **1.179**. Both factors were measured inside the table — 793 species present in both databases give the density ratio, 11 species with USDA measured MOE in both give the modulus ratio (e.g. *Ceiba pentandra* 2.8 / 3.7, *Terminalia superba* 5.3 / 7.0). Prior-knowledge pass rate on 68 reference species rose from 78 % to 94 %. The original basic density is recoverable by dividing by 1.213, and is in the source file `gwdd_v2.2.csv`. Three species are absent from this table because their converted density would exceed the density of wood cell-wall substance (~1500 kg m⁻³) — two of them are palms, which are not true wood. ## 4. ν is a group median on 19,138 rows Wood ν comes from USDA Wood Handbook Table 5-2 softwood/hardwood group medians (0.341 / 0.3665). The 156 species-level values in this table span **0.229–0.495** (σ = 0.051), so a single species can be off by ±38 %. This cannot currently be improved: ν shows no exploitable density trend (r = +0.202 against log ρ), so the group median is the best available estimate. A further **1,144 rows** carry a ν whose upstream reference literally reads *"this dataset (self-referential prior)"* — agreement with other rows of the same class is therefore **not** independent validation. ## 5. Values whose citation does not report them — 636 rows On these rows the cited paper does **not** report the quantity; an alloy-family ASM value was substituted upstream after the original was judged corrupt (e.g. AZ40 raw 325.8 GPa, Hastelloy X raw 5.07 GPa). They are labelled `family_standard`, never `source`. Within-family spread of measured values is ±4–6 % for aluminium, stainless, carbon steel and titanium, and ±32 % for magnesium. ## 6. Volumetric locking — 48 rows 38 elastomer rows at ν = 0.49 and 10 wood rows at ν = 0.485–0.495 would lock linear tetrahedra and cause pressure oscillation in MLS-MPM. `poissons_ratio_fem_safe` clamps them to 0.45. The cost is real: K/G falls from 49.7 at ν = 0.49 to 9.7 at ν = 0.45. With mixed / F-bar elements or a near-incompressible constitutive model, use the original column. ## 7. Unit-of-measurement caveats on individual rows - `Twaron aramid yarn [fibre-direction, …]` (2 rows): ρ = 1440 is **fibre** density and E is the fibre-direction modulus — not fabric-scale values. Renamed to make this explicit. - `Epoxy|Polyester / …wt% phosphate glass fibre` (4 rows): the wt % in the grade name is inconsistent with the reported density (which implies v_f ≈ 0.29–0.48, not 0.07–0.13). ρ and E are mutually consistent and inside Voigt–Reuss bounds; do not use the wt % as a fibre fraction. - `quartzite`: 23 GPa is a jointed **rock-mass** deformation modulus; intact quartzite is 70–90 GPa. - `POM` (openpoly): 7.36 GPa is a literature mean that mixes glass-filled and neat grades; a neat-resin row (2.075 GPa, takayama2024) is also present. ## 8. Coverage gaps - 848 source-reported cells (3.8 %) across ~50 small sources have not been round-tripped to an original file — mostly single PMC papers contributing 2–96 cells each. - The `textile_woven` moduli derived from the Sperl et al. SIGGRAPH 2022 dataset were computed from the published force–elongation curves rather than taken as published scalars. - Wood ν for 19,138 rows and the anisotropy of all wood rows would both be resolved by USDA Wood Handbook Table 5-1 (E_R/E_L, E_T/E_L, G_LR); that table is not yet ingested.