State of the checked-out commit f42360e, dated 2026-09-24, updated for the FhSim Jacobian checker of the jacobian-checker-coverage branch (trajectory samples, shifted state offsets, per-scope overrides); see the CHANGELOG entry under [Unreleased]. Updated by the 2026-09-24 fix pass for the smoke tests of MARE-0028/0029 and the resolution of MARE-0035 and MARE-0038, by the resolution of MARE-0029 and MARE-0042 on 2026-09-25, and by the RMCable smoke test of MARE-0028 on 2026-09-25, and by the resolution of MARE-0044, 0046, 0056, 0059, 0065, 0067 and 0072 on 2026-09-25, and by the tests of MARE-0030, MARE-0032 (a, c) and MARE-0027 (a) on 2026-09-25, and by the review fixes of MARE-0077 and MARE-0078 on 2026-09-25, and by the resolution of MARE-0048 to 0055 and 0061 to 0063 on 2026-09-25, and by the MARE-0045 and MARE-0079 review follow-ups and the resolution of MARE-0047 on 2026-09-25, and by the resolution of MARE-0057, 0058, 0060, 0066, 0068, 0070, 0071 and 0074 on 2026-09-25 and its review, and by the regression baselines of MARE-0033 and MARE-0032 (b) on 2026-09-25, and by the resolution of MARE-0082, 0083 and 0084 on 2026-09-25, which regenerated three of them, and by the owner's review and approval of all five baselines on 2026-09-25, which resolved MARE-0032 and narrowed MARE-0033 to Cable/Branched, and by the resolution of MARE-0085 on 2026-09-25, which regenerated the NetStructure baseline.

1. How the library is tested

The gtest binary test_fhsim_marine_elements (tests/CMakeLists.txt) builds 24 source files: RunTests.cpp, TestUtils_Test.cpp, one file per Jacobian-tested SimObject (Disk_Test.cpp, Cable_Test.cpp, CableBranched_Jacobian_Test.cpp, CableConnector_Test.cpp, NetStructure_Jacobian_Test.cpp, Sphere_Test.cpp, TrawlBeam_Test.cpp, which also holds TrawlBeam's behaviour checks), two more NetStructure Jacobian files (NetStructure_PortJacobian_Test.cpp, NetStructure_JacobianOptOut_Test.cpp), NetStructure_DiskDrag_Test.cpp, NetStructure_PanelWeight_Test.cpp, NetStructure_MaxAcceleration_Test.cpp (NetStructure, NetStructureArray and NetStructureWakeEffect), three smoke-test files (RbCable_Test.cpp, RigidBarCable_Test.cpp, TrawlMooring_Test.cpp, which also holds the trawl/mooring behaviour checks) and three files of behaviour checks for net siblings (NetStructureWithConstraints_Test.cpp, NetStructureWakeEffect_Test.cpp, NetStructureArray_Test.cpp), NetStructure_Regression_Test.cpp (the NetStructure regression baseline), NetStructure_FreeDrift_Test.cpp (a free panel in a current, NetStructure and NetStructureWakeEffect), NetElement3NForces_Test.cpp (the net panel force core called directly) and CatenarySolve_Test.cpp (MooringCable's catenary solve). tests/CMakeLists.txt also builds the test-only SimObject library marine_elements_test_objects (tests/objects/) into SimObjectLibraries, with two objects for the port-Jacobian tests: Test/ForceSource, a force source with an analytical OutputPortJacobian, and Test/SignalSink, a consumer with an analytical InputPortJacobian. No install rule installs the library, and the bundle component excludes it. The binary is registered as the ctest target fhsim_marine_elements (tests/CMakeLists.txt, add_test(NAME "fhsim_marine_elements" ...)); run it with ctest -R fhsim_marine_elements from the build directory, or invoke the built test_fhsim_marine_elements[_d] binary directly (it accepts a working directory and a gtest filter as its own arguments, see tests/RunTests.cpp).

Inputs live under tests/in/<SimObjectName>/ (one folder per tested object, e.g. tests/in/Disk/Disk_in.xml, Disk_ref.csv, Disk.ref); a handful of auxiliary fixtures referenced by File= attributes are copied to the binary's working directory by tests/CMakeLists.txt (SimpleNet.xml, SimplePanelNet.xml, LowStrainPanelNet.xml, BranchableCable.xml, among others).

Six objects have a regression baseline, a <ID>.ref/<ID>_ref.csv pair compared by RunAndCompareRegressionXmlTiming: Disk, and Sphere_Regression, Cable_Regression, CableConnector_Regression, TrawlBeam_Regression and NetStructure_Regression. The five new ones were reviewed and approved by the owner on 2026-09-25, with plots and hand checks in Regression baselines (doc/user/baselines/); they compare every column at an absolute RMS of 1e-8. Shared helpers: tests/TestUtils.h (RunAndCompareRegressionXmlTiming, timing parsing) and tests/JacobianTestUtils.h (ExpectJacobiansMatch, a thin wrapper around fhsim::test::CheckJacobians that fails if nothing was checked). No DISABLED_ tests and no Ogre/visualization-gated tests exist in this suite.

Auto-mode Jacobian checking

fhsim::test::TestSpec.jacobianMode defaults to Auto (fhsim/include/fhsim/testtools/TestSpec.h). Any RunTest() call left at that default runs CheckJacobians on the whole scenario once the simulation finishes, and fails the test on a mismatch, whenever any SimObject in the scenario reports HasJacobians()==true (fhsim/src/testtools/TestRunner.cpp:328-332) — not only when a test explicitly asks for a Jacobian check. The default check config is relTol=1e-4, absTol=1e-10, central-FD step 1e-8, and it samples the initial condition plus 3 random states perturbed by 1% (fhsim/src/testtools/JacobianChecker.h:22-29).

None of this library's tests set jacobianMode=No. The seven smoke tests leave it at Auto, a no-op for their objects, which have no analytic Jacobian. Six of the other seven RunTest-based tests in this suite set jacobianMode=Yes explicitly with their own tolerance override, so for them "dedicated" and "implicit-Auto" coincide. Disk_Test.cpp's RunAndCompareRegressionXmlTiming (regression test) is the exception: it builds its TestSpec without touching jacobianMode, so it is also, implicitly, a Jacobian check — at the framework's default relTol=1e-4, which the explicit Disk_Jacobian test now uses as well — so Disk's Jacobian is checked twice, at the same tolerances. CableBranched_Jacobian_Test.cpp and NetStructure_Jacobian_Test.cpp instead call ExpectJacobiansMatch (tests/JacobianTestUtils.h), which calls fhsim::test::CheckJacobians directly, bypassing TestSpec/jacobianMode entirely. TrawlBeam_TrawlForce and TrawlBeam_Controller also leave jacobianMode at Auto, so they check TrawlBeam's Jacobian implicitly, at the default tolerances, in the force-saturated branch too.

What the checker covers since the jacobian-checker-coverage FhSim branch: besides the initial condition and its perturbations, it checks at 3 states the scenario's own integrator reaches (trajectorySamples), and repeats every per-SimObject check with a neutral stateful object placed first (shiftStateOffsets), so every object sits at a non-zero state offset. The second found that all seven Jacobians in this library indexed their local block and state slice with the global offset AddState returns; they now subtract their first state. The first found MARE-0036 in Cable/Branched, so CableBranched_Jacobian runs without trajectory samples. NetStructure_PortJacobian_Test.cpp checks NetStructure::InputPortJacobian in the system scope of CheckJacobians, and with PortJacobianBlockChecker, which checks only that a port-Jacobian callback is called and writes inside its buffer.

2. Coverage table

One row per ADD_SIMOBJECT_REF/ADD_SIMOBJECT_REF3 line in src/fhsim_marine_elements.cpp that is not commented out (22 rows; the 23rd registration, RbCable/Collision, is commented out at :22 — see 0001 — `RbCable/Collision` is registered only in a commented-out line). Rating: A, behaviour asserted and the Jacobian verified where one exists; B, partly; C, smoke/regression only or indirect; D, untested.

SimObject	Input name	Tests	What is asserted	Jacobian implementation	Jacobian verified	Rating	Issues
Disk	`Net/Disk`	`Disk_Test.cpp`: 2 gtest	Regression vs `.ref`/`_ref.csv` (also, implicitly, an Auto-mode Jacobian check); dedicated Jacobian analytic-vs-numeric	Dense, hand-derived: pos-vel identity, smooth-region buoyancy z-gradient, drag with symmetric subgradients at kinks (`src/net/Disk.cpp:124-219`); seabed friction and the surface step discontinuity omitted	Yes, twice: dedicated (default tolerances, 3 samples) + implicit Auto-mode via the regression test	A	—
Sphere	`Net/Sphere`	`Sphere_Test.cpp`: 3 gtest	Jacobian analytic-vs-numeric; a neutrally buoyant sphere in a current follows the closed-form drift (`Sphere_Current_AcceleratesTowardsTheCurrent`, MARE-0084, resolved); regression baseline `Sphere_Regression` (sinking to terminal velocity in a current; reviewed and approved by the owner, 2026-09-25)	Dense, hand-derived: pos-vel, buoyancy z-gradient, drag on the velocity relative to the water (`src/net/Sphere.cpp`, `OdeJacobian`)	Yes, dedicated, relTol=1e-4, 3 samples	B	—
Cable (LumpedMass)	`Cable/LumpedMass`	`Cable_Test.cpp`: 3 gtest	Jacobian; the end forces of a cable hanging in still water carry its weight in water (`Cable_StillWater_EndForcesCarryTheWeight`, MARE-0082, resolved); regression baseline `Cable_Regression` (hanging in a cross current, no Jacobian check as the Jacobian omits drag; reviewed and approved by the owner, 2026-09-25)	Dense, hand-derived via `InternalCable::ComputeStructuralJacobian`: spring + damping + numerical damping (`src/cable/subroutines/InternalCable.cpp:220-345`)	Yes, dedicated, default tolerances, 3 samples	B	0036 — The InternalCable structural-damping Jacobian ignores the compression-speed clamp, so it is wrong wherever that clamp is active
CableBranched	`Cable/Branched`	`CableBranched_Jacobian_Test.cpp`: 1 gtest (`ExpectJacobiansMatch`, bypasses `TestSpec`/Auto)	Jacobian only	Dense; reuses `InternalCable` blocks plus junction cross-coupling; `HasJacobians()` gated off unless `maxAcceleration>=1e9` (`src/cable/CableBranched.cpp:203-207`)	Yes, dedicated, relTol=1e-2/absTol=0.1 scoped to the object, 3 samples, no trajectory samples (MARE-0036); the IC checks see a slack cable (MARE-0037)	B	0033 — Cable/Branched has Jacobian tests only, no behavioural baseline, 0036 — The InternalCable structural-damping Jacobian ignores the compression-speed clamp, so it is wrong wherever that clamp is active, 0037 — Until an output port runs CalcOutput, a CableBranched element keeps its default relaxed length, so the fixture's IC and shifted-offset Jacobian checks see a slack cable
CableConnector	`RbCable/CableConnector`	`CableConnector_Test.cpp`: 2 gtest	Jacobian; regression baseline `CableConnector_Regression` (point mass under `Test/ForceSource`; reviewed and approved by the owner, 2026-09-25)	Dense, trivial exact: pos-vel identity block only (forces are frozen ports, gravity constant)	Yes, dedicated, relTol=1e-8, 5 samples	B	—
NetStructure	`Net/NetStructure`	`NetStructure_Jacobian_Test.cpp`: 2 gtest (`ExpectJacobiansMatch` x2); `NetStructure_PortJacobian_Test.cpp`: 4 gtest (`Test/ForceSource` into a force port, `Test/SignalSink` on `Pos`, `Vel` and `VelNorm`, each in tag and node-number form); `NetStructure_JacobianOptOut_Test.cpp`: 5 gtest (baseline and the four `HasJacobians()` opt-outs); `NetStructure_DiskDrag_Test.cpp`: disk drag of two short cables, with the disks in the middle and at `DiskPos` 0.25, a thick cable (`NetStructure_DiskDragThickCable`, which tells the MARE-0074 face-on area from the former one), and the end shares of a sphere; `NetStructure_PanelWeight_Test.cpp`: panel weight in water; `NetStructure_MaxAcceleration_Test.cpp`; `NetStructure_Regression_Test.cpp`; `NetStructure_FreeDrift_Test.cpp`; `NetElement3NForces_Test.cpp`: the force core called directly	Jacobian, both dense-state and panel sub-block, and the input and output port blocks; `HasJacobians()` false exactly for its four opt-outs; disk drag (face-on and edge-on) against its formula and its end shares; sphere mass and weight end shares; panel weight 9.81 × `CalcWeight`; `MaxAcceleration` limits the magnitude; a free panel drifts at the current speed (MARE-0083, resolved); the twine drag coefficient is taken at the Reynolds number of the flow through the meshes (MARE-0085, resolved); regression baseline `NetStructure_Regression` (a panel drifting in a current; reviewed and approved by the owner, 2026-09-25)	Dense; cable structural+drag analytic, panel block via forward-mode AD (`sfh::ad::Dual<18>`) over `NetElement3NForces`; `InputPortJacobian` and `OutputPortJacobian` (node ports, `VelNorm`) analytic, offered only while `HasJacobians()` is true (MARE-0077, resolved); `StrainEnergy` has none (MARE-0078); `HasJacobians()` (`src/net/NetStructure.cpp:821-836`) gated off for mass/velocity/acceleration adaptations and cables carrying an attached sphere/disk	State: yes, dedicated (relTol 1e-4 cable; panel relTol 1e-3, absTol 1e-3 at step 1e-6). Port Jacobian: yes, system scope	B	0018 — The OpenCL sources are never compiled, 0027 — Three follow-ups left by the scalar-templated panel force core, 0064 — The BottomContact output of the NetStructure family is always zero, 0078 — NetStructure reports analytical port Jacobians but has no OutputPortJacobian for its state-dependent output ports
TrawlBeam	`TrawlBeam`	`TrawlBeam_Test.cpp`: 4 gtest (`TrawlBeam_Jacobian`, `TrawlBeam_TrawlForce`, `TrawlBeam_Controller`, `TrawlBeam_Regression`)	Jacobian; a backward warp pull slows the beam at any heading; initial states with `StartTravel`; a braking force limited by `SaturationForce`; regression baseline of the controller tracking its speed against a trawl pull (reviewed and approved by the owner, 2026-09-25)	Dense; PID-controller ODE, hand-derived branch-by-branch (power-saturated / force-saturated / unsaturated), matching `OdeFcn` exactly (`src/trawl_mooring_interaction/TrawlBeam.cpp`)	Yes, dedicated, relTol=1e-5, 5 samples	B	—
RMCable	`RbCable/RMCable`	`RbCable_Test.cpp`: `RMCable_Smoke` (own fixture, the hard-coded planar chain started on its constraints, 1 s)	Run completes, outputs finite; the last link, to the moving end point, keeps unit length to 2e-6 (MARE-0047, resolved)	None (full FD fallback; no `HasJacobians()` override)	N/A	C	0028 — The whole RbCable family (RMCable, RBCable, Cable6Dof, CableRM, Ring) has zero automated test coverage
RBCable	`RbCable/RBCable`	`RbCable_Test.cpp`: `RBCable_Smoke` (`examples/input/RigidBodyCable.xml`, 1 s)	Run completes	None (FD fallback)	N/A	C	0010 — `RbCable/RBCable` has no force output for end A, and the RbCable group's force-port case is unsettled, 0028 — The whole RbCable family (RMCable, RBCable, Cable6Dof, CableRM, Ring) has zero automated test coverage
Cable6Dof	`RbCable/6DofCable`	`RbCable_Test.cpp`: `Cable6Dof_Smoke` (`examples/input/C6DofCable.xml`, 1 s)	Run completes	None (FD fallback)	N/A	C	0028 — The whole RbCable family (RMCable, RBCable, Cable6Dof, CableRM, Ring) has zero automated test coverage
CableRM	`RbCable/Cable`	None	—	None (FD fallback)	N/A	D	0010 — `RbCable/RBCable` has no force output for end A, and the RbCable group's force-port case is unsettled, 0025 — `RbCable/Cable` documents the `RBCABLE_CLASSIC_INTERFACE` branch, but no build defines that macro, 0028 — The whole RbCable family (RMCable, RBCable, Cable6Dof, CableRM, Ring) has zero automated test coverage
Ring	`RbCable/Ring`	`RbCable_Test.cpp`: `Ring_Smoke` (`examples/input/Ring.xml`, 1 s)	Run completes	None (FD fallback)	N/A	C	0017 — `RbCable/Ring` reads `NumInputsPerConnector` and `Cm_IsAddedMassHack` without using or documenting them, 0028 — The whole RbCable family (RMCable, RBCable, Cable6Dof, CableRM, Ring) has zero automated test coverage
RigidBarCable	`Cable/RigidBarCable`	`RigidBarCable_Test.cpp`: `RigidBarCable_Smoke` (`examples/input/RigidBarCable.xml`, 1 s)	Run completes	None (FD fallback)	N/A	C	0014 — `Cable/RigidBarCable` binds both `ForceB\<i\>a` and `ForceB\<i\>b` to the same slot, and reads neither
NetStructureArray	`Net/NetStructureArray`	`NetStructureArray_Test.cpp`: `NetStructureArray_Smoke` and `NetStructureArray_InitialPositions` (`ArrayNet.xml`, two panels and a cable, 0.01 s, nodes placed 2 % apart, velocities unset); `NetStructureArray_MaxAcceleration` in `NetStructure_MaxAcceleration_Test.cpp`	Run completes, outputs finite; `NumExternalPos` and the position ports match `NodeOutputs`; the given positions are kept (MARE-0070, resolved) and the twine tension pulls a node with no other load along the panel diagonal; `MaxAcceleration` limits the magnitude	None; direct `SimObject` subclass, inherits none of `NetStructure`'s Jacobian (FD fallback)	N/A	C	0064 — The BottomContact output of the NetStructure family is always zero, 0080 — NetStructureArray gives its cable elements no inertia, so a node on cables only has zero mass
NetStructureWakeEffect	`Net/NetStructureWakeEffect`	`NetStructureWakeEffect_Test.cpp`: `NetStructureWakeEffect_ForcePorts`, `NetStructureWakeEffect_NodeNumberPorts` (cable-only `SimpleNet.xml`, 0.01 s), `NetStructureWakeEffect_ExternalAndBottomForces` (`WakeEffectCableNet.xml`, 2 ms); `NetStructureWakeEffect_MaxAcceleration` in `NetStructure_MaxAcceleration_Test.cpp`; `NetStructureWakeEffect_FreeDrift_DoesNotOutrunTheCurrent` in `NetStructure_FreeDrift_Test.cpp`	`<tag>Force`/`ExternalForce<node>` outputs echo their inputs exactly, `NetTotalForceCheck` is zero, `UseTagForExternalNodes` picks the port names; an input force accelerates its node at F/m and the seafloor pushes a cable up (MARE-0068, resolved); `MaxAcceleration` limits the magnitude; a free panel drifts at the current speed (MARE-0083, resolved)	None; direct `SimObject` subclass (FD fallback)	N/A	C	0064 — The BottomContact output of the NetStructure family is always zero
NetStructureWithConstraints	`Net/NetStructureWithConstraints`	`NetStructureWithConstraints_Test.cpp`: 3 input-validation cases, `NetStructureWithConstraints_Smoke` (`SimplePanelNet.xml`, 0.01 s), `NetStructureWithConstraints_CablesThreadIndependent` (`CablePanelNet.xml`, 116 cables), `NetStructureWithConstraints_SolverFrequency`	Invalid input fails at setup with its message; run completes, outputs finite; one- and eight-thread runs bit-identical; `SolverFrequencyCoefficient` is read, its default 100 gives the former result	None; direct `SimObject` subclass (FD fallback)	N/A	C	—
MooringCable	`MooringCable`	`TrawlMooring_Test.cpp`: `MooringCable_Smoke` (own fixture with a `TrawlCable` and constant end points, 1 s), `MooringCable_BottomChainContact`, `MooringCable_ContactOrder`, `MooringCable_Buoy`, `MooringCable_BuoyOutsideMidRope`, `MooringCable_Reversed`, `MooringCable_EmptySegment`, `MooringCable_Straight`, `MooringCable_NearlySlack`; `CatenarySolve_Test.cpp`: `CatenarySolve.*` (the initial-layout solve, against fhlib and the small-sag approximation)	Run completes, outputs finite; contact radius of the bottom chain; the same contact whatever the `<Lib>` order; buoy element and its bounds; a line and its reverse give the same end forces; a chain or mid rope without an element fails at setup; a line as long as its span starts straight and fast, and lines 2e-9 and 1e-8 longer start fast with the small-sag shape	None (FD fallback)	N/A	C	0043 — MooringCable segfaults with NumElements="10"
TrawlCable	`TrawlCable`	`TrawlMooring_Test.cpp`: `TrawlCable_Smoke` (own fixture with constant winch and bridle-end inputs, 1 s), `TrawlCable_NoClumpWeight_Smoke` (the same without a clump weight), `TrawlCable_BottomBridleArea`	Run completes, outputs finite; the bottom bridle's default area follows its own diameter	None (FD fallback)	N/A	C	0003 — `TrawlCable`'s `Heun_i` branch is commented out while `RundDorg`'s is live
RundDorg	`RundDorg`	`TrawlMooring_Test.cpp`: `RundDorg_Smoke` (own fixture, clump weight and constant line ends, 1 s; also checks the force vectors), `RundDorg_UnknownWeightType`, `RundDorg_NumCpuCoreZero`, `RundDorg_StartCpuCoreNegative`, `RundDorg_Mirrored`, `RundDorg_StillWater`	Run completes, outputs finite; an unknown `WeightType`, a `NumCpuCore` below 1 or a negative `StartCpuCore` fails at setup with its message; mirror symmetry with Line1 and Line2 swapped; force vectors against magnitudes and directions; still water against a negligible current	None (FD fallback)	N/A	C	—
NetStructureTest	`Net/NetStructureTest`	None; Windows-only registration	—	None (FD fallback)	N/A	D	0031 — The Windows-only NetStructureTest family (NetStructureTest, NetStructureABAQUS, NetStructureConstrained) has no test coverage, 0081 — NetCableElementTest still puts sphere and disk mass and weight on the wrong end and splits disk drag as −pos and 1 − pos (Windows only)
NetStructureABAQUS	`Net/NetStructureABAQUS`	None; Windows-only registration	—	None (FD fallback)	N/A	D	0031 — The Windows-only NetStructureTest family (NetStructureTest, NetStructureABAQUS, NetStructureConstrained) has no test coverage, 0081 — NetCableElementTest still puts sphere and disk mass and weight on the wrong end and splits disk drag as −pos and 1 − pos (Windows only)
NetStructureConstrained	`Net/NetStructureConstrained`	None; Windows-only registration	—	None (FD fallback)	N/A	D	0031 — The Windows-only NetStructureTest family (NetStructureTest, NetStructureABAQUS, NetStructureConstrained) has no test coverage

Summary: 1 A, 6 B, 10 C, 5 D. 7 of 22 registered SimObjects have an analytic Jacobian; all 7 are verified against finite differences (state Jacobian). NetStructure's port Jacobians, the only ones in the library, are verified through the system Jacobian.

3. Jacobian implementation notes

Disk (src/net/Disk.cpp:124-219) — dense, hand-derived. Pos-vel identity block; surface-buoyancy z-gradient included under an explicit smoothRegion guard (position, the min(1,submersion) clip, and the seabed all checked); drag block computes symmetric subgradients at the MagN=0/MagT=0 kinks with an explanatory comment. Seabed friction and the belowSurface step discontinuity at z=DiskD/2 are omitted (documented at :218, a non-smooth term by construction).
Sphere (src/net/Sphere.cpp, OdeJacobian) — same pattern as Disk: pos-vel identity, smooth-region buoyancy z-gradient, drag block on the velocity relative to the water. The belowSurface step is likewise omitted.
Cable (LumpedMass) — the shared kernel, InternalCable::ComputeStructuralJacobian (src/cable/subroutines/InternalCable.cpp:220-345): pos-vel identity, a numerical-damping diagonal term, and per-segment structural spring + damper blocks (dFA_dposA, dFA_dposB, dFA_dvelA, dFA_dvelB), each contributed only while 0 < totalForce < maxT (the Bound() saturation region is zeroed, matching OdeFcn).
CableBranched — reuses the same InternalCable blocks for its segments plus AddEndpointJacobianContributions for junction cross-coupling. HasJacobians() (src/cable/CableBranched.cpp:203-207) is gated to true only when maxAcceleration>=1e9, because the acceleration Bound() clamp is otherwise non-smooth.
CableConnector (src/RMCable/CableConnector.cpp:53) — trivial and exact: only the pos-vel identity block, because its forces are frozen input ports and a constant gravity term, neither of which depends on the connector's own state.
NetStructure (src/net/NetStructure.cpp) — the most invested Jacobian in the library. Cable blocks are hand-derived (shared InternalCable kernel); panel blocks are computed by forward-mode automatic differentiation (sfh::ad::Dual<18>) over the templated force core NetElement3NForces, evaluating the exact derivative of the force code as written, with no step or truncation error (src/net/JACOBIAN_RECIPE.md records the derivation history, including the MARE-0004/MARE-0026 issue trail that replaced an inexact forward-difference panel block with this AD path). HasJacobians() (:821-836) falls back to full FD whenever a mass/velocity/acceleration adaptation is active or any cable element carries an attached sphere or disk, and HasPortJacobians() follows it (MARE-0077, resolved). InputPortJacobian() (rows local since MARE-0038, resolved) is tested through Test/ForceSource, OutputPortJacobian() (node ports and VelNorm, MARE-0078) through Test/SignalSink. StrainEnergy has no port Jacobian; FhSim's numerical pass fills the zero block that leaves.
TrawlBeam (src/trawl_mooring_interaction/TrawlBeam.cpp:126) — dense, hand-derived branch-by-branch to match the PID-controller ODE's own saturation branches (power-saturated / force-saturated / unsaturated), so the Jacobian tracks OdeFcn's regime exactly rather than approximating across the switch.

4. Prioritised improvement list

NetStructure test gaps — resolved (MARE-0032): the port-Jacobian test and a fixture per HasJacobians()==false branch exist, and so does a regression baseline, whose plots (Regression baselines) the owner reviewed and approved on 2026-09-25. The port-Jacobian tests found MARE-0077 (resolved) and MARE-0078. 0078 — NetStructure reports analytical port Jacobians but has no OutputPortJacobian for its state-dependent output ports
CableRM has zero test coverage (its fixture waits on MARE-0025); RMCable, RBCable, Cable6Dof and Ring have smoke tests only. 0028 — The whole RbCable family (RMCable, RBCable, Cable6Dof, CableRM, Ring) has zero automated test coverage
MooringCable, TrawlCable and RundDorg had zero test coverage — resolved: they and RigidBarCable now have smoke tests (MARE-0029, resolved).
NetStructureArray, NetStructureWakeEffect and NetStructureWithConstraints get none of NetStructure's Jacobian investment. Each now has a fixture that runs its own code path: behaviour checks of specific fixes for WakeEffect and WithConstraints (MARE-0056, 0059, 0060, 0066, 0067, 0068, 0072) and for NetStructureArray (MARE-0030, 0066, 0070, resolved). An analytic Jacobian for them was out of that issue's scope and has no issue of its own.
Sphere, Cable, CableBranched, CableConnector and TrawlBeam have Jacobian tests only — regression baselines now exist for all but CableBranched, and the owner reviewed and approved their plots (Regression baselines) on 2026-09-25. Their hand checks found defects in the Cable weight, the net's solidity-scaled current and the Sphere's unused current (MARE-0082, 0083, 0084, resolved); the three baselines are regenerated after the fixes. MARE-0033 stays open, narrowed to CableBranched, which waits for the deferred MARE-0036 and MARE-0037. 0033 — Cable/Branched has Jacobian tests only, no behavioural baseline
The Windows-only NetStructureTest family (NetStructureTest, NetStructureABAQUS, NetStructureConstrained) has no test coverage on any platform. The owner decided to keep and document them (MARE-0005, resolved); the tests need a Windows build. 0031 — The Windows-only NetStructureTest family (NetStructureTest, NetStructureABAQUS, NetStructureConstrained) has no test coverage
Stale Jacobian-test tolerance comments on Cable and Disk — resolved: both tests now use the default tolerances (MARE-0034, resolved).
Root JACOBIAN_ANALYSIS.md — refuted: the file is untracked and ignored (.gitignore), not part of the repository, and the Disk defect it describes is fixed in src/net/Disk.cpp:124-219 (MARE-0035, refuted).