What was regenerated

Every regression test of the library (RunAndCompareRegressionXmlTiming, tests/TestUtils.h) was run on the integration head 9aca09d (after F8) against its committed reference. Four pass unchanged and were not touched; two differ and were regenerated:

Test	Fixture	Before (integration head)	Regenerated	Differs because of
`SimObject.Sphere_Regression`	`tests/in/Sphere`	passes	no	–
`SimObject.Cable_Regression`	`tests/in/Cable`	passes	no	–
`SimObject.CableConnector_Regression`	`tests/in/CableConnector`	passes	no	–
`SimObject.TrawlBeam_Regression`	`tests/in/TrawlBeam`	passes	no	–
`SimObject.NetStructure_Regression`	`tests/in/NetStructure`	`DISABLED_` since A7 (R23); fails: RMS up to 0.87 m on the positions, 0.19 m/s on the velocities (limit 1e-8)	yes	A7 (default law LocalThroughFlow, MeshOpening solidity, waked flow), G1 (twine friction in the panel plane)
`SimObject.Disk`	`tests/in/Disk`	`DISABLED_` since A11; fails (limit 1e-10)	yes	A11 (Disk drag projections, owner answer Q2, lead ruling R24)

The .ref limits are unchanged (output_rms_max 1e-8 for NetStructure, 1e-10 for Disk; wall_max 30 s). Both tests are re-enabled (DISABLED_ removed). Regeneration followed the baselines README: delete <ID>.ref and <ID>_ref.csv, run the test (ReferenceMode::CreateIfMissing), restore the limit. Determinism: a second process run of each test passed against the new reference, and the new _ref.csv files are byte-identical to the outputs of the same tests in the F8 worktree (a separate build). Red: the old references fail the current code (table above), so each test detects a change of this size.

After WP-J1 (MaxAcceleration removed, MARE-0133, lead ruling R40). The branch was rebased onto the integration head 2a7a436, which removes the acceleration limit from the nets and cables. No regression fixture set a nonzero limit (LowStrainPanelNet.xml, the NetStructure_Regression net, had MaxAcceleration="0", now deleted; the Sphere, Cable, CableConnector, TrawlBeam and Disk fixtures did not set it), and both fixtures use the explicit RK45_i, so the Jacobian gate that J1 also removed does not act. All six regression tests pass against the references above without regeneration (ctest 3/3, main 261 tests, 14 disabled; VALIDATION 30, 9 disabled). Nothing was regenerated for J1.

Which WPs changed the numbers

NetStructure_Regression. The fixture has no cables, no wake (CastWake off) and no current profile, so D1/F4 (wakes), R32 (a) (cable Rho), H1 (NetStructureArray damping) and F5 (bodies) do not reach it. Recorded measurements on this fixture: A7 moved the steady drift relative to the current from 0.1399 m/s to 0.1377 m/s (CHANGELOG, "Size of the change"), G1 to 0.1376 m/s (STATUS, E6 rerun); G2's clamp r ≤ 1 does not act (Sₙ = 0.160, r = 0.925), G3–G5 concern ScreenMF2022 only. A7 is therefore the change; G1 adds the twine friction in the panel plane, visible in the transient (below).
Disk. Only A11 (MARE-0098) changed the disk's force: v_n = (n·v)n, v_t = v − v_n, each drag term paired with its own speed. R32 (b) (DiskMesh) and F5 (CastWake, off here, byte-identical when off) do not change the states.

Net/NetStructure: <tt>NetStructure_Regression</tt>

A 1 m × 1 m panel of two triangles (LowStrainPanelNet.xml, t = 4 mm, L₀ = 50 mm, 20 meshes a side) across a 0.5 m/s current, held open by 0.5 N floats and sinkers and 0.2 N edge pulls, pulled 0.5 N upstream at each corner; 15 s, outputs every 0.1 s. The regenerated page with every node trace is NetStructure_Regression (doc/user/baselines/).

Drag. The fixture has no force output; the drag is fixed by the steady drift, where it balances the 2 N pull and the panel's 0.37 N weight in water (2.034 N). The new steady relative flow is 0.1376 m/s against 0.1399 m/s, so the default law gives the same drag at a 1.6 % lower relative speed: its drag coefficient is 3.4 % higher than the 3.x twine model's here. Hand check of the new baseline (LocalThroughFlow, Sₙ 0.160, r 0.925, Re_t 530, cd 0.2202, θ = 10.7°, A = 0.983 m²): ½ρA cd((1 − a3)cos θ + a3 cos 3θ)U² = 2.045 N against 2.034 N (+0.5 %).

Tensions. The fixture has no cables and NetStructure writes no twine tension; the external node forces are constant (0.5 N, 0.2 N, 0.5 N per node), so the tension state shows only through the shape: the edge width y(N2) − y(N1) and the height z(N3) − z(N1).

Key quantities, old (approved 2026-09-25) against new (means over the four nodes; NED, z down):

t = 1 s

Quantity	old (approved 2026-09-25)	new	change
mean x position [m]	0.3457	0.1866	-46.0 %
mean y position [m]	0.5131	0.5058	-1.4 %
mean z position [m]	5.5361	5.5835	+0.9 %
mean x velocity [m/s]	0.3004	0.1622	-46.0 %
drift relative to the current, 0.5 − v_x [m/s]	0.1996	0.3378	+69.3 %
mean z velocity [m/s]	0.0301	0.0926	+207.9 %
top edge width, y(N2) − y(N1) [m]	0.3709	-0.1269	-134.2 %
height, z(N3) − z(N1) [m]	0.6086	0.1784	-70.7 %
tilt, x(N3) − x(N1) [m]	0.1718	0.1674	-2.6 %

t = 2 s

Quantity	old (approved 2026-09-25)	new	change
mean x position [m]	0.6572	0.3088	-53.0 %
mean y position [m]	0.5242	0.4917	-6.2 %
mean z position [m]	5.5724	5.6667	+1.7 %
mean x velocity [m/s]	0.3499	0.0358	-89.8 %
drift relative to the current, 0.5 − v_x [m/s]	0.1501	0.4642	+209.2 %
mean z velocity [m/s]	0.0470	0.0784	+66.9 %
top edge width, y(N2) − y(N1) [m]	0.6325	0.0249	-96.1 %
height, z(N3) − z(N1) [m]	0.9509	0.7107	-25.3 %
tilt, x(N3) − x(N1) [m]	0.2662	0.3530	+32.6 %

t = 5 s

Quantity	old (approved 2026-09-25)	new	change
mean x position [m]	1.7347	0.7611	-56.1 %
mean y position [m]	0.5802	0.4542	-21.7 %
mean z position [m]	5.7296	5.8903	+2.8 %
mean x velocity [m/s]	0.3631	0.3461	-4.7 %
drift relative to the current, 0.5 − v_x [m/s]	0.1369	0.1539	+12.4 %
mean z velocity [m/s]	0.0538	0.0489	-9.1 %
top edge width, y(N2) − y(N1) [m]	0.9498	0.9119	-4.0 %
height, z(N3) − z(N1) [m]	0.9757	0.9451	-3.1 %
tilt, x(N3) − x(N1) [m]	0.1886	0.3017	+60.0 %

t = 10 s

Quantity	old (approved 2026-09-25)	new	change
mean x position [m]	3.5400	2.5818	-27.1 %
mean y position [m]	0.6686	0.4547	-32.0 %
mean z position [m]	5.9883	6.1329	+2.4 %
mean x velocity [m/s]	0.3603	0.3628	+0.7 %
drift relative to the current, 0.5 − v_x [m/s]	0.1397	0.1372	-1.9 %
mean z velocity [m/s]	0.0498	0.0489	-1.9 %
top edge width, y(N2) − y(N1) [m]	0.9782	0.9825	+0.4 %
height, z(N3) − z(N1) [m]	0.9836	0.9802	-0.3 %
tilt, x(N3) − x(N1) [m]	0.1708	0.1906	+11.6 %

t = 15 s

Quantity	old (approved 2026-09-25)	new	change
mean x position [m]	5.3407	4.3946	-17.7 %
mean y position [m]	0.7427	0.4700	-36.7 %
mean z position [m]	6.2316	6.3773	+2.3 %
mean x velocity [m/s]	0.3601	0.3624	+0.6 %
drift relative to the current, 0.5 − v_x [m/s]	0.1399	0.1376	-1.6 %
mean z velocity [m/s]	0.0478	0.0489	+2.4 %
top edge width, y(N2) − y(N1) [m]	0.9848	0.9844	-0.0 %
height, z(N3) − z(N1) [m]	0.9832	0.9817	-0.2 %
tilt, x(N3) − x(N1) [m]	0.1747	0.1848	+5.8 %

Node positions at 15 s [m]

Node	old (x, y, z)	new (x, y, z)	Δ (x, y, z)
N1	5.2999, 0.2339, 5.6844	4.3351, -0.0377, 5.8469	-0.9647, -0.2716, +0.1625
N2	5.4255, 1.2187, 5.8047	4.4886, 0.9467, 5.9333	-0.9369, -0.2720, +0.1286
N3	5.4746, 0.2868, 6.6677	4.5200, 0.0076, 6.8287	-0.9546, -0.2791, +0.1610
N4	5.1628, 1.2315, 6.7695	4.2345, 0.9633, 6.9003	-0.9283, -0.2681, +0.1309

Steady-drift drag coefficient ratio, new/old at the same drag (the 2 N pull): (u_old/u_new)² = (0.1399/0.1376)² = 1.0337 (+3.4 %).

Reading the numbers. The steady state (10–15 s) agrees within 2 % in drift and shape; the positions differ by about 0.95 m in x and 0.27 m in y at 15 s because of the first 5 s. There the default law folds the panel much further: at 1 s the top edge has crossed over (width −0.127 m against +0.371 m) and the height is 0.18 m against 0.61 m, the panel drifts at 0.16 m/s against 0.30 m/s, and it reopens by 5 s. This is the screen laws' small in-plane load (twine friction only, MARE-0111, owner answer QG3 "friction only until the MARE-0131 refit"): while the panel is folded, most of it is along the flow and carries little load. The lower early drift explains the lag in x; the y offset stays from the asymmetric fold.

Diagnostic, not a baseline: the same fixture with HydroModel="TwineCrossFlow" KnotFactor="1.1" (the setting that reproduces the 3.x force at normal incidence, CHANGELOG) run with the same binary stays close to the old baseline also in the transient (top edge 0.315 m against 0.371 m at 1 s, mean x −1.1 % at 15 s). At 15 s, old against TwineCrossFlow K 1.1:

Quantity	old (approved 2026-09-25)	new	change
mean x position [m]	5.3407	5.2837	-1.1 %
mean y position [m]	0.7427	0.8183	+10.2 %
mean z position [m]	6.2316	6.3736	+2.3 %
mean x velocity [m/s]	0.3601	0.3574	-0.8 %
drift relative to the current, 0.5 − v_x [m/s]	0.1399	0.1426	+1.9 %
mean z velocity [m/s]	0.0478	0.0582	+21.8 %
top edge width, y(N2) − y(N1) [m]	0.9848	0.9878	+0.3 %
height, z(N3) − z(N1) [m]	0.9832	0.9865	+0.3 %
tilt, x(N3) − x(N1) [m]	0.1747	0.1521	-12.9 %

Steady-drift drag coefficient ratio, new/old at the same drag (the 2 N pull): (u_old/u_new)² = (0.1399/0.1426)² = 0.9624 (-3.8 %).

So the transient difference is the default law, not the panel core, the solidity or the flow lookup.

Mean x velocity of the four nodes (dashed: old baseline, solid: new)

Mean x position of the four nodes (dashed: old baseline, solid: new)

Mean z velocity (NED, + down) (dashed: old baseline, solid: new)

Top edge width y(N2) − y(N1) (dashed: old baseline, solid: new)

Tilt x(N3) − x(N1) (bottom behind top) (dashed: old baseline, solid: new)

Net/Disk: <tt>SimObject.Disk</tt>

A 1 m × 0.1 m disk of 10 kg (cable diameter 0.01 m, ρ 1025) in a 1 m/s current along x, released at the surface level z = 0 with no external force, so its axis is z (edge-on to the current). Outputs at 0, 1 and 4 s only (the fixture's TOutput).

t [s]	Output	old	new	change
0	Position_0	0	0	–
0	Position_1	0	0	–
0	Position_2	0	0	–
0	Velocity_0	0	0	–
0	Velocity_1	0	0	–
0	Velocity_2	0	0	–
1	Position_0	-0.00479125	0.0125013	+360.9 %
1	Position_1	0	0	–
1	Position_2	-0.303812	-0.385203	-26.8 %
1	Velocity_0	-0.00773711	0.0247397	+419.8 %
1	Velocity_1	0	0	–
1	Velocity_2	-0.120622	0.007487	+106.2 %
4	Position_0	-0.0324413	0.190211	+686.3 %
4	Position_1	0	0	–
4	Position_2	-0.375331	-0.375977	-0.2 %
4	Velocity_0	-0.00956742	0.0921254	+1062.9 %
4	Velocity_1	0	0	–
4	Velocity_2	-0.000306971	0.0125524	+4189.1 %

Hand checks of the new baseline.

Along the current (x). Only the tangential drag acts: m dv/dt = k(1 − v)² with k = ½ρ C_t (D − d)T = ½·1025·0.005·0.099 = 0.2537 kg/m, so v(t) = 1 − 1/(1 + k t/m): 0.0247 m/s at 1 s and 0.0921 m/s at 4 s; the baseline has 0.02474 and 0.09213. The old baseline moved the disk upstream (−0.0077 m/s at 1 s): the old code paired the tangential drag with |v_n| and formed n_i v_i element-wise (MARE-0098), so the drag in x depended on the vertical speed and had the wrong sign here.
Vertical (z). The disk floats: the net vertical force 9.81(m − L + L s) with the displaced mass L = ρπD²T/4 = 80.5 kg and the submerged share s = (D/2 − z)/D vanishes at z = D/2 − (1 − m/L)D = −0.3758 m; the baseline has −0.3760 m at 4 s (old −0.3753 m). The new normal drag ½ρ C_n A_n |v_n| v_n (R24) damps the heave less than the old term, which used the current-driven |v_t| ≈ 1 m/s, so the disk overshoots to −0.385 m at 1 s (old −0.304 m, still rising at −0.12 m/s) and still heaves at 0.0126 m/s at 4 s.

Disk x velocity (along the 1 m/s current) (dashed: old baseline, solid: new; outputs at 0, 1 and 4 s only)

Disk z position (NED, + down) (dashed: old baseline, solid: new; outputs at 0, 1 and 4 s only)

What the owner is asked to approve

NetStructure_Regression: the new baseline, including the deeper fold in the first 5 s under the default law (a known consequence of MARE-0111 / QG3, not a new defect).
SimObject.Disk: the new baseline (the disk now drifts downstream and heaves about its floating level).

Approved by the owner on 2026-09-28 (both items). The branch was merged into feature/net-hydrodynamics after WP-J1 and MARE-0099 is resolved.

Table of Contents

What was regenerated

Which WPs changed the numbers

Net/NetStructure: <tt>NetStructure_Regression</tt>

Net/Disk: <tt>SimObject.Disk</tt>

What the owner is asked to approve