TEST - IMPROVEMENT - Re-pin Delta' diagonal for the two-pass grid

The pinned delta_prime_matrix diagonal in runtests_parallel_integration.jl was
measured on the old auto-mpsi grid. Adopting develop's two-pass measured-curvature
grid moves dpm[1,1] from 6.1887 to 7.7036, which failed the pin and aborted the
whole suite at that include, so the 20 test files after it never ran.

Re-pinned to the new grid's values. Tolerances are unchanged (rtol=1e-1) — only
the reference point moved, because the grid generator did. The full suite now
runs to completion: 54 testsets, no failures.

Also records the underlying problem in docs/src/developer_notes.md: these pins
are single points on a curve that varies ~50% across a psi_accuracy scan, so they
must be re-pinned whenever the grid changes. The note sets out the plateau-based
criterion (scan psi_accuracy x truncation, pin where the result is stationary)
that would make the reference a property of the physics rather than of the
discretization, along with what implementing it requires.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>

Author: d-burg

docs/src/developer_notes.md

@@ -20,6 +20,38 @@ Where CODE is the module name (EQUIL, ForceFreeStates, VAC, PERTURBED EQUILIBRIU
 
 The regression harness **must be run on every pull request before merging into `develop`**. It is the project's primary safeguard for tracking how numerical results evolve across changes, so it is only useful if every PR exercises it. When you open a PR, paste the regression report into the PR thread so reviewers can see what moved (and what did not). If your change touches a quantity that is not yet tracked, add a new regression case — or extend an existing one — in the same PR.
 
+### Open problem: pinning grid-sensitive Δ′ robustly
+
+The ideal-MHD Δ′ values pinned in `test/runtests_parallel_integration.jl`
+(`delta_prime_matrix` diagonal) and tracked by the harness are **single-point
+snapshots**: one value at one `psi_accuracy` on one grid. They exist to catch
+unintended changes, and are explicitly *not* converged Δ′. The extraction is
+intrinsically grid-sensitive — a `psi_accuracy` scan from 2e-3 to 2.5e-4 swings
+the DIII-D-like `dpm[1,1]` by roughly 50% (about 6.2 to 9.9), and switching the
+grid generator moves it again. Any such pin therefore encodes an arbitrary point
+on a varying curve, and re-pinning is required whenever the grid changes, which
+weakens it as a regression signal.
+
+A more defensible criterion is to pin the **plateau**: scan Δ′ across
+`psi_accuracy` and the integration-truncation controls, and take the value where
+the result is stationary — the mode of the resulting distribution rather than any
+single sample. Where a plateau exists it is a property of the physics rather than
+of the discretization, so it would survive grid changes and would be a genuine
+convergence statement.
+
+This is not implemented. Doing it properly needs:
+
+  - a scan driver over `psi_accuracy` × truncation (`psihigh`, `dmlim`, `qhigh`)
+    that records the Δ′ diagonal per configuration;
+  - a plateau/mode detector with an explicit stationarity tolerance, plus a
+    defined failure mode for surfaces where no plateau exists (the
+    near-separatrix surfaces are expected to fall in this class);
+  - a harness case that pins plateau values and their scan width, replacing the
+    single-point pins.
+
+Until then, treat the pinned diagonal as an order-of-magnitude and sign
+diagnostic only, and expect to re-pin it whenever the equilibrium grid changes.
+
 ### Regression Harness: Quick usage guide
 
 Set up an alias for convenience (optional):

test/runtests_parallel_integration.jl

@@ -544,20 +544,22 @@ using TOML
         end
 
         # Pinned diagonal `delta_prime_matrix` REAL parts, PEST3-convention self-response Δ' from
-        # the STRIDE BVP with vacuum coupling, on the two-pass auto grid (rational surfaces pinned
-        # as mandatory knots). These pin the value at the default psi_accuracy on a fixed grid so
-        # the test catches unintended changes; they are NOT converged Δ'. The ideal-MHD Δ'
-        # extraction is intrinsically grid-sensitive: a psi_accuracy scan (2e-3→2.5e-4) swings
-        # dpm[1,1] by ~50% (6.2–9.9), and a finer ldp mpsi=512 grid gives ≈8.5, so treat the
-        # diagonal as an order-of-magnitude/sign diagnostic pending the resistive-layer Δ' work.
+        # the STRIDE BVP with vacuum coupling, on the two-pass measured-curvature grid (rational
+        # surfaces pinned as mandatory knots). These pin one point at the default psi_accuracy so
+        # the test catches unintended changes; they are NOT converged Δ'. The extraction is
+        # intrinsically grid-sensitive — a psi_accuracy scan (2e-3→2.5e-4) swings dpm[1,1] by ~50%
+        # (6.2–9.9) and a finer ldp mpsi=512 grid gives ≈8.5 — so treat the diagonal as an
+        # order-of-magnitude/sign diagnostic, and expect to re-pin whenever the grid generator
+        # changes (as here). See the "pinning grid-sensitive Δ′ robustly" open problem in
+        # docs/src/developer_notes.md for the plateau criterion meant to replace single-point pins.
         # (et[1], NTV torque, and ‖resonant flux‖ stay grid-robust to <1% — the sensitivity is
         # local to the singular-layer matching, not the global response.) Only real parts are
         # pinned; the imaginary parts are dominated by the PEST3 four-term cancellation and are
         # FP/platform-sensitive. Near-separatrix surfaces q=5,6 keep only the finiteness/non-zero
         # checks above. Values use this testset's mode range (mpert=27, vs full-pipeline mpert=35).
-        @test isapprox(real(dpm[1, 1]), +6.188700e+00; rtol=1e-1)   # q=2
-        @test isapprox(real(dpm[2, 2]), -5.554900e+00; rtol=1e-1)   # q=3
-        @test isapprox(real(dpm[3, 3]), -1.578700e+01; rtol=1e-1)   # q=4
+        @test isapprox(real(dpm[1, 1]), +7.703609e+00; rtol=1e-1)   # q=2
+        @test isapprox(real(dpm[2, 2]), -5.344199e+00; rtol=1e-1)   # q=3
+        @test isapprox(real(dpm[3, 3]), -1.590034e+01; rtol=1e-1)   # q=4
     end
 
 end