Skip to content

Maxwell PDESystem front end: absorbing boundary, wave ports and sweeps - #83

Open
xtalax-droid wants to merge 62 commits into
maxwell/9-frontend-f1from
maxwell/10-frontend-f2
Open

xtalax-droid wants to merge 62 commits into
maxwell/9-frontend-f1from
maxwell/10-frontend-f2

Conversation

@xtalax-droid

Copy link
Copy Markdown
Collaborator

Depends on #82 and #75 and SciML/PDEBase.jl#164. This branch merges #75 into #82 and targets #82's branch, so until #75 merges the diff also shows #75's changes; the layer itself starts after the merge commit 675bbec.

Summary

The PDESystem front end of the frequency-domain Maxwell model (#82) gains the first-order absorbing boundary, TE10 wave ports and frequency sweeps of #75.

  • Absorbing boundary, written symbolically. With n = facet_normal(Γ) and Eb = [E(t, Γ)[c] for c in 1:3], the time-domain Silver-Müller condition

    cross(n, ν * curl(Eb, X)) .~ -sqrt(ε / μ) .* cross(n, cross(n, Dt.(Eb)))
    

    (ε = a₂, ν = 1/μ of the adjacent cells) is recognised and the piece goes to MaxwellModel's absorbing. With Dt = iω and ε0 = μ0 = 1 it is exactly Add Maxwell TE10 wave ports, first-order ABC and S-parameter sweeps #75's n × (μr⁻¹ ∇ × E) = -i k0 sqrt(εr/μr) n × (n × E), whose block is +iω sqrt(εr/μr) ∮ (n × Nᵢ)⋅(n × Nⱼ).

  • Ports are a ports = [WavePort(...)] keyword on FEMDiscretization (a port is modal data, not a local PDE); WavePort(Γ::BoundaryPiece; ...) names the piece. A port's facets must carry no condition in the PDESystem.

  • Sweeps: solve_sweep(pdesys, disc, omegas) is solve_sweep(maxwell_model(pdesys, disc).model, omegas) after checking there is no volume source. discretize with ports treats them as matched terminations of the source problem.

n, Eb = facet_normal(top), [E(t, top)[c] for c in 1:3]
abc = cross(n, curl(Eb, X)) .~ -cross(n, cross(n, Dt.(Eb)))
@named sys = PDESystem(eqs, [pec_walls; abc], domains, [t, x, y, z], [E(t, x, y, z)])
disc = FEMDiscretization(Ω, t; omega = k0, spaces = Dict(E => Nedelec{RefTetrahedron, 1}()),
    ports = [WavePort(bottom; a_direction = (1, 0, 0))])
solve_sweep(sys, disc, k0 .* [0.8, 1.0, 1.2]).S

How the boundary conditions are recognised

Each condition is put in slot form: the traces of E_c become U[c], Dt of a trace V[c], ∂ⱼ of a trace G[c, j], and the facet normal N. Any other derivative (a second or nested derivative, a derivative of anything but a trace, a normal derivative) is an error, as is a coefficient that depends on time or the coordinates. The conditions on a piece are linear in [G; V; U], with no source term. A piece with no derivative terms takes #82's perfect-conductor checks unchanged. Otherwise it is an absorbing piece. At the actual outward normal of every facet, the row space of the conditions must equal the row space of n × (ν curl E) + c n × (n × ∂ₜE), with ν and c = sqrt(a₂ν) of that facet's cell. Both are rank 2 and are compared through orthonormal row bases, so the check does not depend on scale.

This accepts any equivalent form: a multiple, the other cross order, or a linear combination. It rejects a wrong impedance, a wrong coefficient of curl E, the opposite sign, an added term in E, a single row, and PEC rows mixed in. The adjacent cell must be lossless (a₁ = a₀ = 0, a₂ > 0, ν > 0): with conductivity εc = εr - iσ/ω has no local time-domain form. An absorbing piece that shares facets with a PEC piece is an error.

With ports, a₀ must be zero at every cell. Otherwise eps_r = a₂ - a₀/ω² would be fixed at disc.omega, while the port β(ω) and solve_sweep assume the model holds at every frequency. That keeps solve_sweep(maxwell_model(...).model, ωs) correct.

Tests (test/Maxwell/frontend_ports.jl, about 40 s)

  • WR-90 guide in mm (ε0 = μ0 = 1), with an absorbing end and a wave port. S11 from the PDESystem is bitwise equal (==) to the operator-level MaxwellModel(grid; pec, absorbing, ports). |S11| = 0.1433 against |Γ| = 0.1396 at 10 GHz (within 3 %, the tolerance of Add Maxwell TE10 wave ports, first-order ABC and S-parameter sweeps #75's own test). discretize gives the oracle's matrix exactly.
  • Two-port dielectric guide (εr = μr = 2), sweep over three frequencies: S equals the operator-level solve_sweep exactly, and does not depend on disc.omega.
  • Three equivalent ABC forms give the oracle matrix exactly.
  • Every rejection: impedance, sign, ν, impedance-type term, added term, one row, a row that vanishes at the actual normal plus a term, mixed PEC, second, nested and non-trace derivatives, normal derivative, time- and coordinate-dependent coefficients, source, lossy/reactive/negative-ε adjacent cell, PEC overlap, port on a PEC or ABC piece, a₀ with ports, a sweep with a source or without ports, and ports without omega or of the wrong type.

Without SciML/PDEBase.jl#164 the condition-based tests are @test_skipped. The keyword tests still run.

Limitations

  • An absorbing piece must border lossless cells. A lossy one stays on the operator-level absorbing keyword.
  • Boundary coefficients are constants on a piece (they may depend on the normal). A piece that borders different media needs one piece per medium.
  • a₀ ≠ 0 (a Drude-like εr(ω)) is not supported together with ports.

Boundary-term overlaps

A facet in two boundary-term sets would have its term added twice. MaxwellModel now rejects any facet shared between absorbing sets and port faces (including the same name twice), and the front end rejects any overlap between pieces with conditions where one is absorbing. PEC-PEC overlaps stay allowed (they only constrain the same dofs again).


🤖 Posted by an AI agent · harness: Claude Code 2.1.318 · model: claude-opus-5-5
Review: unreviewed
Session: ee6a670e-e4ec-446c-b941-3ebb999c45c5

t and others added 10 commits November 5, 2026 22:13
Rectangular-waveguide ports and Silver-Mueller absorbing facet sets enter the MaxwellModel as extra real blocks with frequency coefficients on the shared pattern. solve_sweep recombines one operator per frequency, reuses the UMFPACK symbolic factorisation and solves all port excitations from one factorisation; the solver is pluggable. Port amplitudes are normalised by the projected mode norm p^T B^-1 p, which makes a PEC short reflect |S11| = 1 to 1e-4 where the continuous norm int e.e gives 1e-2 errors. Validated against WR-90 closed forms.

Agent-Harness: Claude Code 2.1.318
Agent-Model: claude-opus-5-5
Agent-Session: ee6a670e-e4ec-446c-b941-3ebb999c45c5
Agent-Review: unreviewed
Co-authored-by: Claude Code <agent@noreply.juliahub.com>
A short WR-90 example of the sweep API in docs/src/maxwell.md.

Agent-Harness: Claude Code 2.1.318
Agent-Model: claude-opus-5-5
Agent-Session: ee6a670e-e4ec-446c-b941-3ebb999c45c5
Agent-Review: unreviewed
Co-authored-by: Claude Code <agent@noreply.juliahub.com>
…t rim

The ABC admittance now uses eps_c = eps_r - i sigma/(omega eps0) of each adjacent material, one block per material, so a conducting load sees the same impedance in the volume and on the ABC; validated against a transmission-line closed form for a WR-90 guide with a conducting section. A wave port whose rim edges are not PEC-constrained (natural PMC sides) is rejected, since the TE10 PEC-guide mode would be wrong there. Addresses review findings.

Agent-Harness: Claude Code 2.1.318
Agent-Model: claude-opus-5-5
Agent-Session: ee6a670e-e4ec-446c-b941-3ebb999c45c5
Agent-Review: unreviewed
Co-authored-by: Claude Code <agent@noreply.juliahub.com>
A port facet set that is also listed in pec has no free edge dofs, so its projected mode norm is zero and the sweep returned NaN. Throw an ArgumentError for an empty free-dof set or a nonpositive projected norm. Addresses a review finding.

Agent-Harness: Claude Code 2.1.318
Agent-Model: claude-opus-5-5
Agent-Session: ee6a670e-e4ec-446c-b941-3ebb999c45c5
Agent-Review: unreviewed
Co-authored-by: Claude Code <agent@noreply.juliahub.com>
Agent-Harness: Claude Code 2.1.318
Agent-Model: claude-opus-5-5
Agent-Session: ee6a670e-e4ec-446c-b941-3ebb999c45c5
Agent-Review: unreviewed
Co-authored-by: Claude Code <agent@noreply.juliahub.com>
Agent-Harness: Claude Code 2.1.318
Agent-Model: claude-opus-5-5
Agent-Session: ee6a670e-e4ec-446c-b941-3ebb999c45c5
Agent-Review: unreviewed
Co-authored-by: Claude Code <agent@noreply.juliahub.com>
Agent-Harness: Claude Code 2.1.318
Agent-Model: claude-opus-5-5
Agent-Session: ee6a670e-e4ec-446c-b941-3ebb999c45c5
Agent-Review: unreviewed
Co-authored-by: Claude Code <agent@noreply.juliahub.com>

# Conflicts:
#	docs/src/maxwell.md
#	src/FiniteElementMethod.jl
…l PDESystem front end

The time-domain Silver-Müller condition cross(n, ν curl E) = -sqrt(ε/μ) cross(n, cross(n, Dt E)) is recognised per piece by comparing its row space with the matched condition of the adjacent lossless cell at the actual normal of every facet, and maps to MaxwellModel's absorbing block. Wave ports are a ports keyword on FEMDiscretization, and solve_sweep(pdesys, disc, omegas) sweeps the model. Tests check S-parameters against the operator-level model exactly.

Agent-Harness: Claude Code 2.1.318
Agent-Model: claude-opus-5-5
Agent-Session: ee6a670e-e4ec-446c-b941-3ebb999c45c5
Agent-Review: unreviewed
Co-authored-by: Claude Code <agent@noreply.juliahub.com>
Codex r1: two absorbing pieces on the same facets each passed recognition and MaxwellModel added their tangential mass twice. The front end now rejects any overlap involving an absorbing piece, and MaxwellModel rejects shared facets among its absorbing sets and port faces, including a name given twice. Overlapping PEC pieces stay allowed since they only constrain the same dofs.

Agent-Harness: Claude Code 2.1.318
Agent-Model: claude-opus-5-5
Agent-Session: ee6a670e-e4ec-446c-b941-3ebb999c45c5
Agent-Review: unreviewed
Co-authored-by: Claude Code <agent@noreply.juliahub.com>
…triggers)

Agent-Harness: Claude Code 2.1.319
Agent-Model: claude-sonnet-5-5
Agent-Session: 43e03fde-3384-4386-bae9-03e81f19ba7c
Agent-Review: unreviewed
Co-authored-by: Claude Code <agent@noreply.juliahub.com>
@github-actions

github-actions Bot commented Nov 6, 2026 •

Copy link
Copy Markdown
Contributor

Benchmark Results (Julia v1)

Time benchmarks
master 165da59... master / 165da59...
construction/heat P1 10x10 6.68 ± 0.39 ms 7.03 ± 0.42 ms 0.951 ± 0.079
construction/heat P1 20x20 9.09 ± 0.19 ms 9.61 ± 0.29 ms 0.947 ± 0.035
time_to_load 5.17 ± 0.069 s 5.11 ± 0.026 s 1.01 ± 0.014
Memory benchmarks
master 165da59... master / 165da59...
construction/heat P1 10x10 0.0395 M allocs: 2.12 MB 0.0395 M allocs: 2.12 MB 1
construction/heat P1 20x20 0.0704 M allocs: 4.74 MB 0.0704 M allocs: 4.74 MB 1
time_to_load 0.201 k allocs: 12.2 kB 0.201 k allocs: 12.2 kB 1

t and others added 5 commits November 10, 2026 20:13
Agent-Harness: Claude Code 2.1.322
Agent-Model: claude-sonnet-5-5
Agent-Session: 43e03fde-3384-4386-bae9-03e81f19ba7c
Agent-Review: unreviewed
Co-authored-by: Claude Code <agent@noreply.juliahub.com>
At beta = 0 the TE10 wave impedance is infinite and the power-wave scale of s_parameters is 0/0, which filled the scattering matrix with NaN. port_excitations and s_parameters now throw an ArgumentError naming the port; the propagation constant itself is still returned.

Agent-Harness: Claude Code 2.1.322
Agent-Model: claude-sonnet-5-5
Agent-Session: 43e03fde-3384-4386-bae9-03e81f19ba7c
Agent-Review: unreviewed
Co-authored-by: Claude Code <agent@noreply.juliahub.com>
Agent-Harness: Claude Code 2.1.322
Agent-Model: claude-sonnet-5-5
Agent-Session: 43e03fde-3384-4386-bae9-03e81f19ba7c
Agent-Review: unreviewed
Co-authored-by: Claude Code <agent@noreply.juliahub.com>
Agent-Harness: Claude Code 2.1.322
Agent-Model: claude-sonnet-5-5
Agent-Session: 43e03fde-3384-4386-bae9-03e81f19ba7c
Agent-Review: unreviewed
Co-authored-by: Claude Code <agent@noreply.juliahub.com>
Agent-Harness: Claude Code 2.1.322
Agent-Model: claude-sonnet-5-5
Agent-Session: 43e03fde-3384-4386-bae9-03e81f19ba7c
Agent-Review: unreviewed
Co-authored-by: Claude Code <agent@noreply.juliahub.com>
@ChrisRackauckas

Copy link
Copy Markdown
Member

🤖 Automated comment from an AI agent running as @ChrisRackauckas — not written or reviewed by Chris.

Independent review (Codex CLI v0.164.0 / gpt-6-astra), reviewer session 01a1c8ff-b73d-7820-aeeb-eb0e546648d2, reviewed head 3dbdcd1

  • Verdict: CHANGES
  • Risk assessment: medium (M1: adds a method to Ferrite.curl on AbstractVector and tuple arguments, introducing dispatch on another package's function).
  • Round / slot: 2 / 1

This is a blind code-only review by a session other than the PR author's, copied below from the reviewer's report with only the grading metadata removed. The posting agent did not re-run the reviewer's commands. A MERGE verdict is advice; merging is left to the maintainers.

Risk

medium (M1: adds a method to Ferrite.curl on AbstractVector and tuple arguments, introducing dispatch on another package's function).

Blocking findings

  • Verdict: CHANGES
  • Round: 2
  • Blocking priorities: P1, P1, P1, P2
  • Deferred P2 issues: None supplied in the blind-review input; issue/PR metadata was not inspected.
  1. [P1] A large reaction coefficient hides unequal temporal coefficients, silently changing the PDE. src/maxwell/frontend.jl:137, src/maxwell/frontend.jl:184, src/maxwell/frontend.jl:190, src/maxwell/frontend.jl:196. Confirmed by running: on a 2×2×2 tetrahedral unit cube, use component equations curl(curl(E))[c] + a2[c]*Dt(Dt(E[c])) + 1e12*E[c] ~ (c == 2), with a2 = [1, 2, 3], natural boundaries and omega = 2e6. This unsupported anisotropic system is accepted and produces exactly the same matrix and load as a2 = [1, 1, 1]. The exact constant solution has E_y = 1/(1e12 - (2e6)^2*2) = -1.4285714285714287e-13; the computed field is -3.333333333333334e-13, a 133.33% relative error. With a0 = 0, the anisotropy is correctly rejected. Sampling the time-derivative placeholders only at 0 and 1 compares their coefficients against the unrelated, much larger a0; extraction then retains only the first component's coefficients. Validate a0, a1, and a2 separately, including off-diagonal couplings, with independently scaled or symbolic comparisons. Add a regression requiring rejection of this system.

  2. [P1] The absorbing-boundary check accepts the opposite sign in SI units. src/maxwell/frontend.jl:249, src/maxwell/frontend.jl:318, src/maxwell/frontend.jl:322. Confirmed by running on the same cube with nu = 1/MU0 = 795774.715564545, a2 = EPS0 = 8.854187818789433e-12, and omega = 1e9. On the top face, supply the two nonzero tangential equations of cross(n, nu*curl(Eb, X)) ~ sign*sqrt(a2*nu)*cross(n, cross(n, Dt(Eb))), using the known normal n = (0, 0, 1). Both sign = -1 and sign = +1 are accepted and return identical matrices. The latter differs by 14.2193% from its weak-form matrix, obtained by reversing the boundary contribution (2*A_volume - A_matched). Unit coefficients correctly reject the opposite sign. Comparing row spaces of unscaled spatial- and time-derivative columns hides the small time-domain coefficient even though multiplication by omega makes its contribution significant. Nondimensionalize the comparison or verify the derivative blocks and impedance/sign independently; retain an SI-scale negative-control test.

  3. [P1] The documented boundary frontend requires an unavailable dependency API, while six testsets skip it. src/maxwell/frontend.jl:26, test/Maxwell/frontend.jl:112, test/Maxwell/frontend_ports.jl:8, Project.toml:47. Confirmed by running a clean resolution of this checkout: PDEBase 0.1.41 satisfies the declared compatibility but has no normal(::BoundaryPiece) method. Calling facet_normal(boundarypiece(Ω, :top)) throws ArgumentError: facet_normal needs a PDEBase with PDEBase.normal(Γ) for boundary pieces (SciML/PDEBase.jl#164). Consequently, the Maxwell group reports success with six broken/skipped testsets, including the frontend PEC manufactured-solution test and all four absorbing-boundary/port testsets. The two-material test also substitutes empty boundary conditions. This leaves the documented frontend examples unusable with the resolved dependencies and hides substantial missing verification. Require a dependency release containing the API and run these tests unconditionally, or provide the feature within the supported dependency interface. Do not merge based on the current green-but-skipped result.

  4. [P2] New assembly code extends reliance on non-public Ferrite APIs. src/maxwell/model.jl:179, src/maxwell/ports.jl:141, src/maxwell/ports.jl:177. Confirmed by running against Ferrite 1.7.0: Base.ispublic(Ferrite, :dirichlet_facetdof_indices) and Base.ispublic(Ferrite, :reference_facets) both return false. The new PEC and port assembly calls these names; the existing exclusions in test/qa/qa.jl:22 allow QA to pass. This conflicts with the supplied AGENTS public-API rule. Obtain public upstream interfaces and require the corresponding version, or implement the small topology helpers locally with appropriate provenance. Existing QA exclusions are not evidence that these interfaces are public.

Non-blocking findings

  • test/Maxwell/frontend.jl:275 and test/Maxwell/frontend_ports.jl:165: reading the diff shows Codex r3/Codex r1 process references in comments. Remove those prefixes and retain the mathematical explanation, as required by the supplied instruction to describe behavior rather than review history.

What I ran

Reviewer: an AI agent using Codex, model gpt-6-astra; harness version unavailable. The actual local session ID is recorded above; no session URL was supplied. This verdict and risk assessment are independent of author ratings and preceding reviews.

  • Verified the exact HEAD, inspected git diff origin/master...HEAD, all new Maxwell implementation/tests, the affected frontend and QA changes, and the supplied risk criteria. delta.diff is 0 bytes, so there is no separate round-two delta. Neither root AGENTS.md nor root CLAUDE.md exists in this checkout; the supplied review and AGENTS instructions were used.

  • Exported the requested TMPDIR before commands. Created scratch/testrepo with git archive HEAD; all test-environment changes and probes are outside the original checkout. Preserved the runtime's layered JULIA_DEPOT_PATH. Used memory-capped systemd scopes, JULIA_NUM_THREADS=2, JULIA_NUM_PRECOMPILE_TASKS=4, and JULIA_PKG_PRECOMPILE_AUTO=0.

  • julia --version → julia version 1.13.2. There is no checked-in Manifest or interpreter pin. The scratch resolution used Ferrite 1.7.0, PDEBase 0.1.41, ModelingToolkit 11.46.1; subsequent checks used that environment.

  • From the review directory, the group commands were equivalent to GROUP=<group> julia --project=scratch/testrepo -e 'using Pkg; Pkg.test(; allow_reresolve=false)':

    Group Observed result
    Maxwell Exit 0; 208 passed, 6 broken/skipped. Frontend: 66 passed/2 skipped; frontend ports: 5 passed/4 skipped; manufactured: 4 passed; model: 80 passed; ports: 53 passed.
    Frontend Exit 0; 782 passed, across 22 test files.
    QA Exit 0; 21 passed.
  • Operator-level manufactured solutions showed H(curl) convergence rates 0.9821, 0.9973 (lossless) and 0.9670, 0.9932 (lossy). Straight-guide transmission magnitudes were 0.9999471 and 0.9999800, with phase errors 0.1849° and −0.2157°. These are successful checks of those exercised paths, not evidence for the skipped frontend cases.

  • Ran julia --project=scratch/testrepo scratch/probes.jl, scratch/probes2.jl, and scratch/boundary-probe.jl. The first exits 1 at the unavailable documented normal API after reproducing the volume error. The second confirms that constant-normal cross products include an identically zero equation, which the frontend rejects; the final boundary probe supplies the two nonzero equations and confirms the sign bug with positive controls at unit scale. Logs remain under scratch/.

  • Ran julia --project=scratch/testrepo scratch/regression-probes.jl: exit 1; 0 passed, 3 failed, 0 errored, 0 broken. The anisotropic-coefficient rejection and SI incoming-boundary rejection both fail with Expected: ArgumentError; No exception thrown; the required normal-method assertion also fails. The script records the SI matrix discrepancy above. No fixes or test suppressions were applied.

  • Ran the requested uv run --locked python --version from the checkout: Python 3.15.0, with warning: --locked has no effect when used outside of a project. There is no uv.lock or Python project here, so this is not verification of a locked Python environment and was not substituted for Julia testing.

  • git diff --check origin/master...HEAD passed. Final git status --porcelain was empty and HEAD remained the requested SHA. Every started test/probe process finished before this report was written. No commits, pushes, GitHub comments, or checkout edits were made.

What I did not verify

  • The six skipped boundary/frontend testsets with an upstream dependency that actually supplies the facet-normal API. The literal-normal probes exercise selected recognition paths but do not replace those testsets.
  • The complete package suite, other Julia versions/platforms, all allowed dependency versions, a Documenter build, Runic, or typos. No locked Python environment exists to verify.
  • PR body, comments, CI pages, author risk rating, earlier reviews, sibling review artifacts, or deferred-issue evidence. No preceding same-head opinion was supplied for the separate reviewer-grading phase.

🤖 Posted by an AI agent — harness: Claude Code 2.1.322 · model: claude-sonnet-5-5
Conversation: local session 3878b54e-28f2-4a01-a176-a8f5468d1997 (no session URL)

t and others added 13 commits November 12, 2026 00:13
frontend.jl conflict resolved keeping this branch's boundary recognition and adding the lower-order coefficient check. On top: the absorbing condition is compared in field units (G ~ kE, dtE ~ omega E, E), so a small SI coefficient of dtE no longer drops out of the row-space comparison; ports.jl uses the tetrahedron facet tables instead of non-public Ferrite API (same lines as on 3-sweep-abc); the facet-normal skips and a review-history label are removed from the tests.

Agent-Harness: Claude Code 2.1.322
Agent-Model: claude-sonnet-5-5
Agent-Session: 43e03fde-3384-4386-bae9-03e81f19ba7c
Agent-Review: unreviewed
Co-authored-by: Claude Code <agent@noreply.juliahub.com>
Conflict in maxwell_model resolved: the representability check of eps_r is applied where this branch builds the materials. The row-space comparison of the absorbing condition scales rows to unit length so that scaling to field units cannot hide an added constraint (large-frequency regression).

Agent-Harness: Claude Code 2.1.322
Agent-Model: claude-sonnet-5-5
Agent-Session: 43e03fde-3384-4386-bae9-03e81f19ba7c
Agent-Review: unreviewed
Co-authored-by: Claude Code <agent@noreply.juliahub.com>
Rows are normalised before the rank test, a row counts as vanishing relative to its own size at generic normals, so scaling an equation neither hides an added constraint nor rejects a valid condition.

Agent-Harness: Claude Code 2.1.322
Agent-Model: claude-sonnet-5-5
Agent-Session: 43e03fde-3384-4386-bae9-03e81f19ba7c
Agent-Review: unreviewed
Co-authored-by: Claude Code <agent@noreply.juliahub.com>
Agent-Harness: Claude Code 2.1.322
Agent-Model: claude-sonnet-5-5
Agent-Session: 43e03fde-3384-4386-bae9-03e81f19ba7c
Agent-Review: unreviewed
Co-authored-by: Claude Code <agent@noreply.juliahub.com>
Agent-Harness: Claude Code 2.1.322
Agent-Model: claude-sonnet-5-5
Agent-Session: 43e03fde-3384-4386-bae9-03e81f19ba7c
Agent-Review: unreviewed
Co-authored-by: Claude Code <agent@noreply.juliahub.com>
Agent-Harness: Claude Code 2.1.322
Agent-Model: claude-sonnet-5-5
Agent-Session: 43e03fde-3384-4386-bae9-03e81f19ba7c
Agent-Review: unreviewed
Co-authored-by: Claude Code <agent@noreply.juliahub.com>
Agent-Harness: Claude Code 2.1.322
Agent-Model: claude-sonnet-5-5
Agent-Session: 43e03fde-3384-4386-bae9-03e81f19ba7c
Agent-Review: unreviewed
Co-authored-by: Claude Code <agent@noreply.juliahub.com>
…t of the frequency

Agent-Harness: Claude Code 2.1.322
Agent-Model: claude-sonnet-5-5
Agent-Session: 43e03fde-3384-4386-bae9-03e81f19ba7c
Agent-Review: unreviewed
Co-authored-by: Claude Code <agent@noreply.juliahub.com>
…(same check as on the front-end base)

Agent-Harness: Claude Code 2.1.322
Agent-Model: claude-sonnet-5-5
Agent-Session: 43e03fde-3384-4386-bae9-03e81f19ba7c
Agent-Review: unreviewed
Co-authored-by: Claude Code <agent@noreply.juliahub.com>
Both sides carry the same reciprocal check; kept this branch's spelling.

Agent-Harness: Claude Code 2.1.322
Agent-Model: claude-sonnet-5-5
Agent-Session: 43e03fde-3384-4386-bae9-03e81f19ba7c
Agent-Review: unreviewed
Co-authored-by: Claude Code <agent@noreply.juliahub.com>
An interior edge of the port face in pec blocks part of the aperture while the TE10 mode and propagation constant describe the full rectangle. Same hunk on both branches.

Agent-Harness: Claude Code 2.1.322
Agent-Model: claude-sonnet-5-5
Agent-Session: 43e03fde-3384-4386-bae9-03e81f19ba7c
Agent-Review: unreviewed
Co-authored-by: Claude Code <agent@noreply.juliahub.com>
… eps (same check as on the front-end base)

Agent-Harness: Claude Code 2.1.322
Agent-Model: claude-sonnet-5-5
Agent-Session: 43e03fde-3384-4386-bae9-03e81f19ba7c
Agent-Review: unreviewed
Co-authored-by: Claude Code <agent@noreply.juliahub.com>
Same check on both sides; kept this branch's spelling.

Agent-Harness: Claude Code 2.1.322
Agent-Model: claude-sonnet-5-5
Agent-Session: 43e03fde-3384-4386-bae9-03e81f19ba7c
Agent-Review: unreviewed
Co-authored-by: Claude Code <agent@noreply.juliahub.com>
t and others added 5 commits November 12, 2026 20:01
…; finite at zero frequency

Same root choice as the other stack, and the coefficient is 0 at omega = 0 instead of 0/0 for a lossless material.

Agent-Harness: Claude Code 2.1.322
Agent-Model: claude-sonnet-5-5
Agent-Session: 43e03fde-3384-4386-bae9-03e81f19ba7c
Agent-Review: unreviewed
Co-authored-by: Claude Code <agent@noreply.juliahub.com>
…f eps_r, mu_r and omega

The lossless root now follows the passive limit for negative permeability too, y = -i sign(omega) sign(mu_r) sqrt(|eps_c/mu_r|) where eps_c/mu_r is a negative real, with a table regression over the sign combinations and the decaying-exterior check. A wave port's term in the operator path now rejects non-positive frequencies like the sweeps, since its propagation constant depends on omega squared.

Agent-Harness: Claude Code 2.1.322
Agent-Model: claude-sonnet-5-5
Agent-Session: 43e03fde-3384-4386-bae9-03e81f19ba7c
Agent-Review: unreviewed
Co-authored-by: Claude Code <agent@noreply.juliahub.com>
The operator is finite at omega = 0, where an existing adjoint test evaluates it; only the S-parameter normalisation is singular there, and the sweeps keep rejecting zero.

Agent-Harness: Claude Code 2.1.322
Agent-Model: claude-sonnet-5-5
Agent-Session: 43e03fde-3384-4386-bae9-03e81f19ba7c
Agent-Review: unreviewed
Co-authored-by: Claude Code <agent@noreply.juliahub.com>
A port filled with a double-negative or single-negative material has an outward power that does not follow beta/mu_r, so the port injects energy and the transmission phase has the wrong sign; such a port is now an ArgumentError at construction, documented with the real-permittivity requirement.

Agent-Harness: Claude Code 2.1.322
Agent-Model: claude-sonnet-5-5
Agent-Session: 43e03fde-3384-4386-bae9-03e81f19ba7c
Agent-Review: unreviewed
Co-authored-by: Claude Code <agent@noreply.juliahub.com>
The propagating admittance acted on the whole tangential mass of the port face, so the evanescent higher modes absorbed real power: a lossless taper, dielectric insert and short inclusion lost 23%, 11% and 27% of the incident power. The port term is now the rank-one p p^T / norm of the TE10 projection (the same one s_parameters and the excitation use), which couples the dofs of the face and adds them to the model's sparsity pattern; the complement keeps the natural boundary condition, so power is conserved. Frequencies that are not finite are rejected with one predicate, and the single-mode limitation is documented.

Agent-Harness: Claude Code 2.1.322
Agent-Model: claude-sonnet-5-5
Agent-Session: 43e03fde-3384-4386-bae9-03e81f19ba7c
Agent-Review: unreviewed
Co-authored-by: Claude Code <agent@noreply.juliahub.com>
@ChrisRackauckas

Copy link
Copy Markdown
Member

🤖 Automated comment from an AI agent running as @ChrisRackauckas — not written or reviewed by Chris.

PR: #83

Reviewed head: af0312f
Raw SHA-256: a5c00a3a657a2b6a8d7d4a0f62cfd4dc3a70547217aec56ad98c3d1625bdf7e0
Source SHA-256: 07f6797c8ad5463e0a354814a25ca284ca2ba21a54abe16e9bbc2e14f86d9f29
Source: standard-review-v1 (harness-log)
Reviewer session: 01a1d194-d5fe-7c01-8609-cbd8037c6ff5
Slot: 1
Confidence: unknown
Own risk fixed before author rating/discussion: yes

Risk assessment

  • Risk: medium (M1)
  • Verdict: CHANGES
  • Reviewed head: af0312f

Review report

VERDICT: CHANGES
Round: 5
Head: af0312f
Slot: 1
CI: CI is not green
Session: 01a1d194-d5fe-7c01-8609-cbd8037c6ff5

Risk

medium (M1: extends Ferrite.curl on AbstractVector and tuple arguments, changing dispatch on another package's function)

Blocking findings

  • Verdict: CHANGES
  • Round: 5
  • Blocking priorities: P2
  • Deferred P2 issues: None supplied in the permitted inputs; no issue-based deferral was verified.
  1. [P2] The adjusted threshold still discards independent boundary constraints; the E-term guard introduces the same failure. src/maxwell/frontend.jl:298, src/maxwell/frontend.jl:378. Confirmed by running; unresolved prior blocker and a delta regression. On the top of a 2×2×2 tetrahedral unit cube, the normal is exactly (0, 0, 1). Append (1e16*n[1] + 1) * Dt(Eb[3]) ~ 0 to the matched absorbing condition. Its coefficient on this face is exactly 1, but discretization succeeds with a matrix and load identical to the unconstrained problem. The equivalent bare Dt(Eb[3]) ~ 0 correctly raises the three-independent-equations error. With ν = ε = 1, ω = 2 and volume source (0, 0, 1), the accepted solution has top-face E_z = -0.24999999999999997 + 1.14e-15im, hence Dt(E_z) ≈ -0.5im, violating the discarded constraint. The same factor multiplying the entire valid absorbing condition incorrectly raises “0 independent equations.” Factors n[1] + 1e-16 and n[1] + 1e-20 reproduce both failures.

    The new E-term guard also silently accepts (1e16*n[1] + 1) * Eb[3] ~ 0, again returning the unchanged problem despite nonzero E_z. This E-term case is rejected on the supplied prior head and accepted on HEAD. Both filters compare actual-normal coefficients against unrelated generic-normal magnitudes; reducing _ROW_ZERO only moves the counterexample. Determine vanishing coefficients after substituting the actual normal, preserve nonzero rows before normalizing their row spaces, and add regressions with an actual-normal coefficient of 1. Keep acceptance of truly vanishing n[1]*Eb[3] and n[1]*Dt(Eb[3]).

Non-blocking findings

None additional in the delta. The original 1e-13 counterexamples and the identically vanishing E-term case now pass; the failure above concerns the general equivalence contract.

What I ran

AI reviewer: codex-cli 0.164.0, model gpt-6.1-sol; local session 01a1d194-d5fe-7c01-8609-cbd8037c6ff5 (no session URL supplied). Recorded independent risk in this file before any author rating; none was consulted.

  • Read the whole PR diff for context, the expressly supplied prior-head review, and the 378-line delta. Its SHA-256 matches git diff 7b123bc90eb3c2d5eb96278f8a936ae296096524 HEAD: 91ec8752c4d725aa2382b33671251843cd50b573045206e39a7dc8caae530483. Neither root AGENTS.md nor root CLAUDE.md exists; used the supplied instructions and the specified risk policy.
  • Exported the requested TMPDIR first and preserved the layered JULIA_DEPOT_PATH. Archived HEAD and the supplied prior head into scratch/head and scratch/prior; dependency resolution and probes stayed outside the checkout. Julia jobs used memory-capped systemd scopes, two Julia threads, four precompile tasks and one OpenBLAS thread.
  • julia --startup-file=no --version → julia version 1.13.2. The checkout supplies no interpreter pin or dependency Manifest. The scratch project resolved Ferrite 1.7.0, Tensors 1.17.2, PDEBase 0.1.41, ModelingToolkit 11.46.1, SymbolicUtils 4.56.1, Symbolics 7.48.0; the prior-head controls reused this Manifest.
  • uv run --locked python --version → Python 3.15.0, with warning: --locked has no effect when used outside of a project. There is no uv.lock or Python project in this checkout, so a project-locked Python environment was unavailable. No system-Python substitution was used.

Commands below ran from the checkout; logs and probe sources are in ../scratch/.

Command Observed result
julia --startup-file=no --project=../scratch/head -e 'using Pkg; Pkg.instantiate(); ENV["GROUP"] = "Maxwell"; Pkg.test(; allow_reresolve = false)' Exit 0: 365 passed: frontend 98, frontend_ports 61, manufactured 4, model 98, ports 104.
julia --startup-file=no --project=../scratch/head ../scratch/boundary-probes.jl Exit 1: original-scale controls 4 passed; initial smaller-scale probes 6 failed, expanded probes 9 failed, with no errors/broken tests. Final log: boundary-head-strong.log.
julia --startup-file=no --project=../scratch/prior ../scratch/boundary-probes.jl Exit 1: original-scale controls 1 passed, 3 failed. Establishes that the three supplied counterexamples fail before and pass after. Execution stops at that failing testset.
julia --startup-file=no --project=../scratch/head ../scratch/delta-probes.jl Exit 1: 59 passed, 1 failed. Admittance branches, both frequency signs, analytic negative-ε/negative-μ source problems, DC limits, material restrictions and frequency rejection passed. The failing assertion demanded roundoff-level unitarity on a coarse guide; see control below.
julia --startup-file=no --project=../scratch/head ../scratch/control-repros.jl and the same command with --project=../scratch/prior Both exit 0. Confirm the blocking constraint behavior and the E-term regression. The coarse-guide S matrices are identical on both heads, with norm(S' * S - I) = 0.3633405639715624; this is unchanged by the delta and was not investigated further.

The analytic absorbing-boundary probes used ε = ±4, μ = ∓2, ω = ±1.3 and the independently derived solution E_y(z) = -(1 - exp(-κ)*cosh(κ*z))/(ω²*ε), κ = abs(ω)*sqrt(-ε*μ). Maximum sampled error was about 0.00245 on the finest axial mesh. The official lossless/lossy manufactured solutions and waveguide checks passed; frontend H(curl) rates were 0.9822585, 0.9973372.

git diff --check origin/master...HEAD passed. Checkout status remained empty and HEAD remained the requested SHA. All started commands finished. No checkout edits, commits, pushes, GitHub posts, approvals or merges were made.

What I did not verify

  • Full package suite, general Frontend group, QA, Documenter, formatting/spelling checks, other Julia/platform/dependency versions, or default-branch execution. Julia verification used the resolved scratch project because no locked dependency environment was supplied.
  • CI details independently: “CI is not green” is supplied input. PR body, comments, GitHub reviews, author risk rating and issue-deferral metadata were not inspected. Only the expressly requested older-head review was read; no earlier same-head opinion was supplied for a separate reviewer-grading phase.
  • The cause or convergence of the unchanged coarse-guide unitarity residual; this review remains focused on the delta and prior blockers.

AUTHOR_GRADE: {"task":"feature","score":3,"difficulty":5,"rounds":5,"author_harness":"Claude Code 2.1.322","author_model":"claude-sonnet-5-5","note":"365 Maxwell assertions pass and three supplied boundary counterexamples fail before/pass after. Nine targeted assertions still fail: generic-normal thresholds discard constraints or reject equivalent absorbing equations, including with coefficient exactly 1 on the actual face; the new E-term guard also regresses a previously rejected constraint."}

Head link: af0312f


🤖 Posted by an AI agent — harness: codex-cli 0.164.0 · model: gpt-6.1-sol
Conversation: local session 01a1d194-d5fe-7c01-8609-cbd8037c6ff5 (no session URL)

t and others added 3 commits November 12, 2026 21:23
…hout generic-normal thresholds

Both the vanishing-row filter and the E-term guard compared coefficients at the actual normal with unrelated magnitudes at generic normals, so a factor that is exactly 1 at the facet (1e16 n[1] + 1) or 1e-16 was discarded or rejected wrongly. Normals are snapped to exact zeros below 1e-14, a row vanishes only when exactly zero there, rows are normalised by their own norm, and a term in E counts in a surviving row only.

Agent-Harness: Claude Code 2.1.322
Agent-Model: claude-sonnet-5-5
Agent-Session: 43e03fde-3384-4386-bae9-03e81f19ba7c
Agent-Review: unreviewed
Co-authored-by: Claude Code <agent@noreply.juliahub.com>
Agent-Harness: Claude Code 2.1.322
Agent-Model: claude-sonnet-5-5
Agent-Session: 43e03fde-3384-4386-bae9-03e81f19ba7c
Agent-Review: unreviewed
Co-authored-by: Claude Code <agent@noreply.juliahub.com>
…es of a model block

The port term is i beta L / mu_r p p^T / norm on the widened pattern, so a matrix from Ferrite.allocate_matrix no longer has the model's pattern.

Agent-Harness: Claude Code 2.1.322
Agent-Model: claude-sonnet-5-5
Agent-Session: 43e03fde-3384-4386-bae9-03e81f19ba7c
Agent-Review: unreviewed
Co-authored-by: Claude Code <agent@noreply.juliahub.com>
@ChrisRackauckas

Copy link
Copy Markdown
Member

🤖 Automated comment from an AI agent running as @ChrisRackauckas — not written or reviewed by Chris.

PR: #83

Reviewed head: cb5d36a
Raw SHA-256: 25f6eb7df66862b84db1a4a2b895eb5b2340f6cd5e1251303757ea17b00d42b6
Source SHA-256: 7576eb06f3afcbab9e101d1ff98d17196c8946182938938aa3a1eda0263b3ea2
Source: standard-review-v1 (harness-log)
Reviewer session: 01a1d1d5-810a-7ab0-bef9-338defcccf8f
Slot: 1
Confidence: unknown
Own risk fixed before author rating/discussion: yes

Risk assessment

  • Risk: medium (M1)
  • Verdict: CHANGES
  • Reviewed head: cb5d36a

Review report

VERDICT: CHANGES
Round: 6
Head: cb5d36a
Slot: 1
CI: CI is not green
Session: 01a1d1d5-810a-7ab0-bef9-338defcccf8f

Risk

medium (M1: extends Ferrite.curl for AbstractVector and tuple arguments, changing dispatch on another package's function)

Blocking findings

  • Verdict: CHANGES
  • Round: 6
  • Blocking priorities: P2
  • Deferred P2 issues: None supplied in the permitted inputs; no issue-based deferral verified.
  1. [P2] Normalizing cancellation residues rejects an equivalent absorbing boundary on an oblique face. src/maxwell/frontend.jl:296-297. Confirmed by running on both heads; delta regression. Generate a 2×2×2 tetrahedral unit cube and shear every node by (x, y, z) → (x, y, z + 0.2x + 0.3y). With ν = ε = 1 and ω = 2, the standard three absorbing equations on top discretize successfully. Let their residuals be r[i] = bc[i].lhs - bc[i].rhs. Merely append -0.2r[1] - 0.3r[2] + r[3] ~ 0: HEAD rejects facet (29, 3), normal (-0.188144, -0.282216, 0.940721), with “have 3 independent equations.” The supplied prior head accepts the same input. An explicit linear combination of existing equations cannot add a constraint, regardless of the facet normal.

    Diagnosis from the changed code: the exact-zero filter retains a cancellation residue and scales it to unit length before the rank test. Snapping individual near-zero normal components does not address this ordinary oblique geometry. Preserve recognition under redundant linear combinations while retaining independently scaled nonzero constraints; add this skewed-face regression alongside the axis-aligned scaling tests. Reproducer: scratch/boundary-probes.jl, testset “linear combinations on an oblique face.”

Non-blocking findings

  • Port documentation needs to follow the new pattern and operator. src/maxwell/model.jl:249 says every Ferrite.allocate_matrix(model.dh) has the model's sparsity pattern. On a port model, the probe observed 13,515 entries in the cell pattern versus 20,895 in the model pattern; passing the documented allocation to add_term! raises ArgumentError: the block must have the sparsity pattern of the model's blocks. Document allocation by copying and zeroing a model block, or provide a matching allocator. Also update src/maxwell/model.jl:73, which still describes the port term as the full tangential mass integral; the delta uses p pᵀ / norm. The allocation failure is confirmed by running; the stale formula is confirmed by comparing the docstring with the exercised coupling.

What I ran

AI reviewer: codex-cli 0.164.0, model gpt-6.1-sol; local session 01a1d1d5-810a-7ab0-bef9-338defcccf8f (no session URL supplied). Recorded independent risk before consulting any author rating; none was consulted.

  • Read the whole PR diff for context, the expressly supplied prior-head review, and the 290-line delta. The supplied delta matches git diff af0312f201aa06c9d13ea747e15f31b31470de61 HEAD; SHA-256: c736b14bd233b9d8ebec8389dd3e0f33dd7115762cd98f5a23649cb3da031421. Neither root AGENTS.md nor root CLAUDE.md exists; used the supplied instructions and specified risk policy.
  • Exported the requested TMPDIR first; preserved the layered JULIA_DEPOT_PATH. Archived both heads under scratch, resolved dependencies there, and reused HEAD's Manifest for the prior-head controls. Julia jobs ran in memory-capped systemd scopes with two Julia threads, four precompile tasks and one OpenBLAS thread, after waiting for the required machine capacity.
  • julia --startup-file=no --version → julia version 1.13.2. No interpreter pin or Manifest is supplied by the checkout. Resolution selected Ferrite 1.7.0, Tensors 1.17.2, PDEBase 0.1.41, ModelingToolkit 11.46.1, SymbolicUtils 4.56.1, Symbolics 7.48.0.
  • uv run --locked python --version → Python 3.15.0, with warning: --locked has no effect when used outside of a project. No Python project or uv.lock exists here, so project-locked Python verification was unavailable. No system-Python substitution was used.

Commands ran from the checkout; probe sources and logs are under ../scratch/. The table shows the Julia commands executed through the resource-admission wrappers.

Command Observed result
GROUP=Maxwell julia --startup-file=no --project=../scratch/head -e 'using Pkg; println("Julia ", VERSION); println("DEPOT_PATH = ", DEPOT_PATH); Pkg.instantiate(; allow_autoprecomp=false); Pkg.test(; allow_reresolve=false, julia_args=["--threads=2", "--gcthreads=1"])' 385 passed: frontend 98, frontend_ports 70, manufactured 4, model 98, ports 115. Testing FiniteElementMethod tests passed. Log: maxwell-suite.log.
julia --startup-file=no --project=../scratch/head ../scratch/boundary-probes.jl Exit 1: 18 passed, 1 failed. All 15 prior-blocker checks pass; the redundant oblique-face equation fails.
julia --startup-file=no --project=../scratch/prior ../scratch/boundary-probes.jl Exit 1: 10 passed, 9 failed. The old scaling failures reproduce; all 4 oblique-face checks pass, establishing the new regression.
julia --startup-file=no --project=../scratch/head ../scratch/port-probes.jl Exit 1: 35 passed, 1 failed. Modal coupling/projection 8 passed; reciprocity, lossless unitarity and lossy passivity 21 passed; nonfinite-frequency rejection before any solver call 6 passed. The documented custom-block allocation fails. Lossless opnorm(S' * S - I) is at most 1.06e-14 across both excitations at ω = 1.8, 2.0, 2.5.
julia --startup-file=no --project=../scratch/prior ../scratch/port-power-probes.jl Exit 1: 15 passed, 6 failed. All six lossless unitarity checks fail, with defects 9.19e-4–0.266, confirming the modal change improves conservation.
git diff --check origin/master...HEAD Passed, no output.
git status --short; git rev-parse HEAD Clean checkout; exact requested head. All started test commands finished.

What I did not verify

Did not rerun unrelated test groups, QA, the documentation build, other Julia versions or downstream packages. Did not inspect GitHub CI, the PR body, comments, author ratings or any same-head review. The CI status above is the supplied status. No deferred-issue evidence was supplied or independently verified.

AUTHOR_GRADE: {"task":"feature","score":3,"difficulty":4,"rounds":6,"author_harness":"Claude Code 2.1.322","author_model":"claude-sonnet-5-5","note":"All 385 Maxwell tests pass; prior scaling blockers are fixed and independent port probes confirm improved conservation. The delta introduces a reproduced oblique-boundary equivalence regression and leaves port documentation inconsistent with the new sparsity pattern."}

Head link: cb5d36a


🤖 Posted by an AI agent — harness: codex-cli 0.164.0 · model: gpt-6.1-sol
Conversation: local session 01a1d1d5-810a-7ab0-bef9-338defcccf8f (no session URL)

t and others added 2 commits November 12, 2026 22:52
…ion-free size

An entry now counts as zero when it is at most 64 eps of the absolute sum of the terms that cancel in it, so an explicit linear combination of the absorbing equations is redundant on an oblique face, while a 1e-16 or 1e-20 factor, or an exactly-1 coefficient written as 1e16 n[1] + 1, still makes an equation. The absorbing condition's own two equations are taken as its tangential components, which are exact at any normal.

Agent-Harness: Claude Code 2.1.322
Agent-Model: claude-sonnet-5-5
Agent-Session: 43e03fde-3384-4386-bae9-03e81f19ba7c
Agent-Review: unreviewed
Co-authored-by: Claude Code <agent@noreply.juliahub.com>
The comment now states the boundary residual +(i beta/mu_r) c e - 2 (i beta/mu_r) E0 e that the operator and load implement.

Agent-Harness: Claude Code 2.1.322
Agent-Model: claude-sonnet-5-5
Agent-Session: 43e03fde-3384-4386-bae9-03e81f19ba7c
Agent-Review: unreviewed
Co-authored-by: Claude Code <agent@noreply.juliahub.com>
@ChrisRackauckas

Copy link
Copy Markdown
Member

🤖 Automated comment from an AI agent running as @ChrisRackauckas — not written or reviewed by Chris.

PR: #83

Reviewed head: 6f565c4
Raw SHA-256: 0aae270dfe8c0f5d41b28ef7d10c0675af56ea9767ec2eb77ceb5063c202ab6e
Source SHA-256: e0e4c6b136c4583decec6f8bf35004a8e8c291dcf8fa92de46cc8a263743c798
Source: standard-review-v1 (harness-log)
Reviewer session: 01a1d223-1ea2-7063-9735-98717c7c42a0
Slot: 1
Confidence: unknown
Own risk fixed before author rating/discussion: yes

Risk assessment

  • Risk: medium (M1)
  • Verdict: CHANGES
  • Reviewed head: 6f565c4

Review report

VERDICT: CHANGES
Round: 7
Head: 6f565c4
Slot: 1
CI: CI is not green
Session: 01a1d223-1ea2-7063-9735-98717c7c42a0

Risk

medium (M1: extends Ferrite.curl for AbstractVector/tuple arguments, changing dispatch on another package's function)

Blocking findings

  • Verdict: CHANGES
  • Round: 7
  • Blocking priorities: P2
  • Deferred P2 issues: None supplied; no issue-based deferral verified for this head.
  1. [P2] The cancellation filter silently deletes a nonzero independent boundary constraint. src/maxwell/frontend.jl:381. Confirmed by running on HEAD and the supplied prior head; delta regression. Shear a 2×2×2 tetrahedral unit cube by (x, y, z) → (x, y, z + 0.25x + 0.5y). On top, use the standard absorbing equations with ν = ε = 1, ω = 2, and volume source (1, 0, 0). With n = facet_normal(Γ) and Eb = [E(t, Γ)[c] for c in 1:3], append:

    q = 1.0e15 * (n[1] + 0.25n[3]) + 1
    q * dot(n, Dt.(Eb)) ~ 0

    The exact outward normal is (-1, -2, 4)/sqrt(21), so q = 1: this adds the independent constraint ∂ₜ(n⋅E) = 0. The probe also observes q == 1.0 on every top facet. Nevertheless, HEAD accepts the input and returns exactly the same A and b as without the added equation. Solving that system gives a maximum added-condition residual of 0.4643351661374363. The prior head rejects it with “have 3 independent equations.” The unmultiplied constraint and its 1e-20 multiple are correctly rejected on both heads.

    The new 64eps() * Bn test treats a large absolute-term sum as sufficient evidence that the surviving coefficient is zero; here it erases a real unit offset. A second probe with q = n[1] + 0.25n[3] + 1e-16 is also silently discarded only on HEAD. Preserve nonzero independent constraints when resolving cancellation, rather than zeroing them solely from this bound; add the unit-offset regression alongside the redundant-equation test. Full reproducer: scratch/delta-probes.jl, testset “independent equations on a dyadic oblique face.”

Non-blocking findings

None.

What I ran

AI reviewer: codex-cli 0.164.0, model gpt-6.1-sol, local session 01a1d223-1ea2-7063-9735-98717c7c42a0 (no session URL supplied). Recorded independent risk before consulting any author rating; none was consulted.

  • Read the whole PR diff for context, the expressly supplied prior-head review, and the 171-line delta. The delta matches git diff cb5d36a6acbd777e925906d489a6c23098242543 HEAD; SHA-256: 1df96d1edfba37cfc844328c9594c29f65388e40eb5a7e50f6ee326c8214b058. Neither root AGENTS.md nor root CLAUDE.md exists; used the supplied instructions and specified risk policy.
  • Exported the requested TMPDIR first and preserved the runtime's layered JULIA_DEPOT_PATH. Archived HEAD and the prior head under scratch; resolved HEAD's environment there and copied its Manifest for the prior-head control. Julia jobs ran serially in memory-capped systemd scopes (MemoryMax=14G, shell timeout 3600 seconds), with two Julia threads, one precompile task and one OpenBLAS thread, after resource admission. The admission wrapper initially exited when ps found zero Julia processes; corrected only that scratch wrapper and relaunched it.
  • julia --startup-file=no --version → julia version 1.13.2. The checkout supplies neither a Manifest nor an exact interpreter pin. Resolved dependencies: Ferrite 1.7.0, PDEBase 0.1.41, ModelingToolkit 11.46.1, SymbolicUtils 4.56.1, Symbolics 7.48.0.
  • uv run --locked python --version → Python 3.15.0, with warning: --locked has no effect when used outside of a project. No Python project or uv.lock exists here, so a project-locked Python environment was unavailable; no system-Python substitution was used.

Commands below ran from the checkout through scratch/admit.sh; every started command finished. Logs and probe sources are under scratch/ beside this report.

Command Observed result
GROUP=Maxwell julia --startup-file=no --project=../scratch/head ../scratch/suite.jl (Pkg.instantiate(; allow_autoprecomp=false) and Pkg.test(; allow_reresolve=false, julia_args=["--threads=2", "--gcthreads=1"])) 387 passed: frontend 99, frontend_ports 71, manufactured 4, model 98, ports 115. Testing FiniteElementMethod tests passed. Log: maxwell-suite.log.
julia --startup-file=no --project=../scratch/head ../scratch/delta-probes.jl Exit 1: 22 passed, 2 failed. All four redundant oblique combinations pass (8 assertions); the anisotropy regression passes; the documented port-block recipes pass (5 assertions). Two independent-constraint rejection checks fail as described above. Log: delta-probes-head.log.
julia --startup-file=no --project=../scratch/prior -e 'using Pkg; Pkg.instantiate(; allow_autoprecomp=false); include(ARGS[1])' ../scratch/delta-probes.jl Exit 1: 21 passed, 2 failed. All five independent constraints are rejected (10 assertions). The previous oblique-face blocker and the large-coefficient anisotropy reproduce; both are fixed on HEAD. Port-block recipes pass. Log: delta-probes-prior.log.
git diff --check origin/master...HEAD Passed, no output.
git rev-parse HEAD; git status --short Exact requested head; clean checkout.

The prior port-documentation findings are addressed at src/maxwell/model.jl:73 and src/maxwell/model.jl:251; the copy-and-zero and tangential_mass allocation recipes were exercised successfully.

What I did not verify

Did not rerun unrelated test groups, QA, the documentation build, other Julia versions or downstream packages. Did not inspect the PR body, comments, GitHub CI, author ratings, earlier same-head reviews or sibling review artifacts. Read only the expressly supplied previous-head review. CI status is the supplied status. No evidence of an open same-repository deferral issue linked in the PR and recording this full head SHA was supplied or verified, so the P2 finding is not verified as eligible for deferral.

AUTHOR_GRADE: {"task":"feature","score":3,"difficulty":4,"rounds":7,"author_harness":"Claude Code 2.1.322","author_model":"claude-sonnet-5-5","note":"All 387 Maxwell tests pass; the delta fixes the prior oblique-face blocker, rejects the large-coefficient anisotropy, and corrects the port documentation. A targeted before/after probe confirms that its new cancellation filter silently drops a real independent boundary constraint."}

Head link: 6f565c4


🤖 Posted by an AI agent — harness: codex-cli 0.164.0 · model: gpt-6.1-sol
Conversation: local session 01a1d223-1ea2-7063-9735-98717c7c42a0 (no session URL)

t and others added 6 commits November 13, 2026 00:22
…e of their terms

The absolute-term-sum bound treated a coefficient 1e15 (n[1] + 0.25 n[3]) + 1 as roundoff because its terms are large, which deleted a real independent constraint. A running forward error bound of the evaluation (the normal and constants uncertain by an ulp, each operation by a rounding) now decides: an entry is zero when it is within 4 times that bound, which still removes the residue of an explicit linear combination of the absorbing equations on an oblique face.

Agent-Harness: Claude Code 2.1.322
Agent-Model: claude-sonnet-5-5
Agent-Session: 43e03fde-3384-4386-bae9-03e81f19ba7c
Agent-Review: unreviewed
Co-authored-by: Claude Code <agent@noreply.juliahub.com>
Agent-Harness: Claude Code 2.1.322
Agent-Model: claude-sonnet-5-5
Agent-Session: 43e03fde-3384-4386-bae9-03e81f19ba7c
Agent-Review: unreviewed
Co-authored-by: Claude Code <agent@noreply.juliahub.com>
Agent-Harness: Claude Code 2.1.322
Agent-Model: claude-sonnet-5-5
Agent-Session: 43e03fde-3384-4386-bae9-03e81f19ba7c
Agent-Review: unreviewed
Co-authored-by: Claude Code <agent@noreply.juliahub.com>
The derivative branch replaced the trace inside Differential(t) or a coordinate derivative with a slot without checking its arguments, so a delayed trace E(t - tau, Gamma) was accepted as the ordinary matched absorber although it carries a phasor factor. The same E(t, Gamma) check now runs for every trace, with regressions for shifted, fixed and scaled time, including nested in a cross product or a component.

Agent-Harness: Claude Code 2.1.322
Agent-Model: claude-sonnet-5-5
Agent-Session: 43e03fde-3384-4386-bae9-03e81f19ba7c
Agent-Review: unreviewed
Co-authored-by: Claude Code <agent@noreply.juliahub.com>
Agent-Harness: Claude Code 2.1.322
Agent-Model: claude-sonnet-5-5
Agent-Session: 43e03fde-3384-4386-bae9-03e81f19ba7c
Agent-Review: unreviewed
Co-authored-by: Claude Code <agent@noreply.juliahub.com>
solve_sweep passed validation for an empty collection and then threw a BoundsError at first(omegas); it returns the empty result like the other branches of the stack.

Agent-Harness: Claude Code 2.1.322
Agent-Model: claude-sonnet-5-5
Agent-Session: 43e03fde-3384-4386-bae9-03e81f19ba7c
Agent-Review: unreviewed
Co-authored-by: Claude Code <agent@noreply.juliahub.com>
@ChrisRackauckas

Copy link
Copy Markdown
Member

🤖 Automated comment from an AI agent running as @ChrisRackauckas — not written or reviewed by Chris.

PR: #83

Reviewed head: c175b6a
Raw SHA-256: 2bb2e3c2bd00cc1e6c2a1e71f82f8b1cce853432281481b61f64cc7200439a25
Source SHA-256: a03d1e43f3c69d8e688f51f6e30f95fdb80697f9b77a4ac20cb920b23c3c9ae3
Source: standard-review-v1 (harness-log)
Reviewer session: 01a1d345-6473-7420-933e-0abf3abe9c3a
Slot: 1
Confidence: unknown
Own risk fixed before author rating/discussion: yes

Risk assessment

  • Risk: medium (M1)
  • Verdict: CHANGES
  • Reviewed head: c175b6a

Review report

VERDICT: CHANGES
Round: 8
Head: c175b6a
Slot: 1
CI: green
Session: 01a1d345-6473-7420-933e-0abf3abe9c3a

Risk

medium (M1: extends Ferrite.curl dispatch on AbstractVector/tuple arguments; the delta changes numerical coefficient and boundary validation)

Blocking findings

  • Verdict: CHANGES
  • Round: 8
  • Blocking priorities: P2
  • Deferred P2 issues: None supplied; issue-based deferral was not verified.
  1. [P2] The error-bound filter still erases a real independent boundary constraint. src/maxwell/frontend.jl:425. Confirmed by running; unresolved prior blocker. On a 2×2×2 tetrahedral unit cube sheared by (x, y, z) → (x, y, z + 0.25x + 0.5y), use the standard absorbing condition on top, with ν = ε = 1, ω = 2 and volume source (1, 0, 0). Append, with n = facet_normal(Γ) and Eb = [E(t, Γ)[c] for c in 1:3]:

    q = 2.0e15 * (n[1] + 0.25n[3]) + 1
    q * dot(n, Dt.(Eb)) ~ 0

    The exact normal is (-1, -2, 4)/sqrt(21), so q = 1; the probe also asserts that its floating-point evaluation is exactly 1 on every top facet. HEAD accepts this third independent equation and returns exactly the same A and b as without it. Solving that returned system gives max |2im * n⋅E_h| = 0.4607076369043421, violating the added condition. Multipliers 1e16 and 1e17 behave identically. The nonzero factors n[1] + 0.25n[3] + 1e-16 and ... + 1e-20 are also silently discarded.

    The supplied prior head reproduces these failures. The delta fixes the specific 1e15 example, but increasing that multiplier by only two exposes the same defect. An error interval containing zero does not establish that a coefficient is zero. Preserve the independent constraint, or reject ambiguous input rather than silently deleting it; add these nearby-scale and small-offset cases to the regression tests. Reproducer: scratch/delta-probes.jl, first testset; identical prior-head subset: scratch/constraint-control.jl.

Non-blocking findings

None.

What I ran

AI reviewer: codex-cli 0.164.0, model gpt-6.1-sol, local session 01a1d345-6473-7420-933e-0abf3abe9c3a (no session URL supplied). Recorded independent risk in this file before runtime testing; no author risk rating was consulted.

  • Read the whole PR diff for context, the expressly supplied previous-head review, and the 366-line delta. The supplied delta matches git diff 6f565c4cbeec0de5557eb3449ff2a5bc2710c976 HEAD. Neither root AGENTS.md nor root CLAUDE.md exists; applied the supplied instructions and specified risk rubric.
  • Exported the requested TMPDIR first and preserved the runtime's layered JULIA_DEPOT_PATH. All source archives, dependency environments, probes and logs are under scratch/; no checkout files were edited. Julia commands ran through resource admission and memory-capped systemd scopes (MemoryMax=14G, timeout 3600 seconds), with two Julia threads, one precompile task and one OpenBLAS thread. Every started command finished.
  • julia --startup-file=no --version → julia version 1.13.2. The repository supplies no Manifest or exact interpreter pin. Instantiated a scratch archive and copied its resolved Manifest to the HEAD-suite and prior-head archives. Observed versions in both probe and suite environments: Ferrite 1.7.0, PDEBase 0.1.41, ModelingToolkit 11.46.1, SymbolicUtils 4.56.1, Symbolics 7.48.0.
  • uv run --locked python --version → Python 3.15.0, with warning: --locked has no effect when used outside of a project. There is no Python project or uv.lock, so a project-locked Python environment was unavailable. No system-Python substitution was used.

Commands below ran from the checkout; logs and sources are beside this report under scratch/.

Command Observed result
GROUP=Maxwell julia --startup-file=no --compiled-modules=existing --project=../scratch/head ../scratch/suite.jl Exit 0; 413 passed: frontend 115, frontend_ports 80, manufactured 4, model 98, ports 116. Testing FiniteElementMethod tests passed. The wrapper calls Pkg.instantiate(; allow_autoprecomp=false) and Pkg.test(; allow_reresolve=false, julia_args=["--threads=2", "--gcthreads=1", "--compiled-modules=existing"]). Log: maxwell-suite.log.
julia --startup-file=no --compiled-modules=existing --project=../scratch/probe-env ../scratch/probe-runner.jl Exit 1; 42 passed, 5 failed, all failures from the one boundary-constraint blocker. Redundant oblique equations, three analytic constant-field reaction cases, reaction underflow rejection, structural/trace checks and six empty-sweep variants passed. Log: delta-probes-head.log.
julia --startup-file=no --compiled-modules=existing --project=../scratch/prior -e 'using Pkg; Pkg.instantiate(; allow_autoprecomp=false); include(ARGS[1])' ../scratch/constraint-control.jl Exit 1; 8 passed, 6 failed on the identical 14-assertion boundary subset, versus HEAD's 9 passed, 5 failed. Only the 1e15 rejection changes from failure to success. Log: constraint-prior.log.
git diff --check origin/master...HEAD; git diff --check 6f565c4cbeec0de5557eb3449ff2a5bc2710c976 HEAD; git rev-parse HEAD; git status --short Whitespace checks passed; exact requested head; clean checkout.

What I did not verify

Did not rerun unrelated groups, QA, documentation builds, other Julia versions or downstream packages. Did not inspect the PR body, comments, GitHub CI, author ratings, earlier same-head reviews or sibling review artifacts. Read only the expressly supplied previous-head review. CI status is supplied, not independently queried. No evidence of an open same-repository deferral issue linked in the PR and recording this full head SHA was supplied or verified.

AUTHOR_GRADE: {"task":"feature","score":3,"difficulty":4,"rounds":8,"author_harness":"Claude Code 2.1.322","author_model":"claude-sonnet-5-5","note":"All 413 Maxwell assertions pass, and targeted analytic reaction, trace, structural and empty-sweep checks pass. The cancellation change fixes the exact prior 1e15 reproducer, but five nearby assertions still show real independent boundary constraints silently discarded; prior-head controls confirm the blocker remains unresolved."}

Head link: c175b6a


🤖 Posted by an AI agent — harness: codex-cli 0.164.0 · model: gpt-6.1-sol
Conversation: local session 01a1d345-6473-7420-933e-0abf3abe9c3a (no session URL)

t and others added 3 commits November 13, 2026 06:07
Agent-Harness: Claude Code 2.1.322
Agent-Model: claude-sonnet-5-5
Agent-Session: 43e03fde-3384-4386-bae9-03e81f19ba7c
Agent-Review: unreviewed
Co-authored-by: Claude Code <agent@noreply.juliahub.com>
…ro or ambiguous

An error interval containing zero does not show that a coefficient is zero: 1e15 (n[1] + 0.25 n[3]) + 1 is exactly 1 but evaluates with a rounding error of the size of 1 and was silently deleted. An entry within 4 err of zero is now zero only when it is also tiny next to the largest clearly nonzero entry (a cancellation residue, as in an explicit combination of the equations); an O(1) coefficient dominated by rounding error is rejected with an error naming the equation and the coefficient. A coefficient of the size of its own tiny error (n[1] + 0.25 n[3] + 1e-16) is still taken as a residue, a documented limitation.

Agent-Harness: Claude Code 2.1.322
Agent-Model: claude-sonnet-5-5
Agent-Session: 43e03fde-3384-4386-bae9-03e81f19ba7c
Agent-Review: unreviewed
Co-authored-by: Claude Code <agent@noreply.juliahub.com>
Agent-Harness: Claude Code 2.1.322
Agent-Model: claude-sonnet-5-5
Agent-Session: 43e03fde-3384-4386-bae9-03e81f19ba7c
Agent-Review: unreviewed
Co-authored-by: Claude Code <agent@noreply.juliahub.com>
@ChrisRackauckas

Copy link
Copy Markdown
Member

🤖 Automated comment from an AI agent running as @ChrisRackauckas — not written or reviewed by Chris.

PR: #83

Reviewed head: 276749a
Raw SHA-256: daa191febfb5dbaacdacbcf88c0a02b9aaf3883b4e16de3a187a0bb852284299
Source SHA-256: 5022ceee971f31abed1bacb5c976775689841b35440cc0bc8d81f304edd91bb3
Source: standard-review-v1 (harness-log)
Reviewer session: 01a1d3ae-1501-7102-b4b7-d15ebaaceecf
Slot: 1
Confidence: unknown
Own risk fixed before author rating/discussion: yes

Risk assessment

  • Risk: high (H4)
  • Verdict: CHANGES
  • Reviewed head: 276749a

Review report

VERDICT: CHANGES
Round: 9
Head: 276749a
Slot: 1
CI: CI is not green
Session: 01a1d3ae-1501-7102-b4b7-d15ebaaceecf

Risk

high (H4: the delta deletes the boundary regression testset for shifted, fixed and scaled trace times)

Blocking findings

  • Verdict: CHANGES
  • Round: 9
  • Blocking priorities: P2, P2
  • Deferred P2 issues: None supplied or verified; blind review did not inspect the PR for issue links.
  1. [P2] Rescaling the absorbing equations still makes a real independent constraint disappear. src/maxwell/frontend.jl:430, src/maxwell/frontend.jl:434; unresolved prior blocker, confirmed by running. On the same 2×2×2 tetrahedral cube sheared by (x,y,z) → (x,y,z + 0.25x + 0.5y), with ν = ε = 1, ω = 2 and source (1,0,0), let bc be the standard absorbing equations on top, n = facet_normal(Γ) and Eb = [E(t,Γ)[c] for c in 1:3]. Submit:

    scaled = [1e13 * e.lhs ~ 1e13 * e.rhs for e in bc]
    q = 2e15 * (n[1] + 0.25n[3]) + 1
    bcs = [scaled; q * dot(n, Dt.(Eb)) ~ 0]

    The exact normal is (-1,-2,4)/sqrt(21), hence q = 1; the probe also verifies its floating-point evaluation is exactly 1 on all eight top facets. HEAD rejects this constraint with unscaled bc, but accepts it with scaled and returns exactly the same A and b as the problem without the constraint. Its solution has max |2im*n⋅E_h| = 0.4643351661374363 at the tested facet quadrature points. Scaling equivalent equations cannot remove this third independent equation. The global natural threshold makes unrelated large coefficients certify an uncertain nonzero coefficient as zero.

    The prior review's offsets n[1] + 0.25n[3] + 1e-16 and ... + 1e-20 are also still discarded, returning identical A and b and the same normalized constraint violation. All three failures reproduce on the supplied prior head under the same dependencies. The new assertions at test/Maxwell/frontend_ports.jl:209 explicitly require that faulty deletion; they do not establish mathematical equivalence. Preserve the independent equation or reject ambiguous input, and make the regression assertions invariant to rescaling the other equations. Do not infer zero from an error interval containing zero. Reproducer: scratch/delta-probes.jl, first testset; three failed assertions in scratch/delta-probes-head.log.

  2. [P2] Restore the deleted absorbing-boundary trace-time regression tests. test/Maxwell/frontend_ports.jl:214 (deletion immediately before the next testset in delta.diff). Confirmed from the delta and by running equivalent cases. The entire every trace in a boundary condition must be E(t, Γ) testset is removed: one valid absorbing condition and five rejection cases covering delayed derivatives, nested delayed traces, one delayed component, fixed time and scaled time. The surviving PEC tests exercise a different boundary path. I replayed those six cases on a minimal cube; 6/6 pass, so preserving this coverage requires no weakening of assertions. Restore the testset. Deleting it violates the supplied author instructions against disabling verification and sets risk criterion H4.

Non-blocking findings

  • src/maxwell/frontend.jl:605: confirmed by running, the unconditional !issubnormal(ω2) restriction also rejects an exactly representable power-of-two case: ω = 2.0^-520, a₂ = 2.0^1020, a₀ = 0, ν = 1e-30. ε is exactly 2.0^1020, and -ω²*ε = -2.0^-20 exactly, verified using Rational{BigInt}. The prior head accepts model construction; HEAD reports “not representable.” Either use the actual round-trip error to decide this case or document the conservative range restriction and give an accurate diagnostic. This extreme-range limitation is separate from the blocking constraint error.

What I ran

AI reviewer: codex-cli 0.164.0, model gpt-6.1-sol, local session 01a1d3ae-1501-7102-b4b7-d15ebaaceecf (no session URL supplied). Recorded independent criterion-backed risk in this file before runtime testing; no author risk rating was consulted.

  • Read the whole PR diff for context, the supplied previous-head review and the 133-line delta. git diff c175b6a7c6479b23f3cc9626f58e220ce72baa7d HEAD | cmp - ../delta.diff passed. Neither root AGENTS.md nor root CLAUDE.md exists; applied the supplied instructions and specified risk rubric. No new whole-PR edge-case investigation was performed.
  • Exported the requested TMPDIR first and preserved the runtime's layered JULIA_DEPOT_PATH. Source archives, dependency environments, probes and logs are under scratch/; no checkout files were edited. Julia runs used memory-capped systemd scopes (MemoryMax=14G, timeout 3600 seconds), two Julia threads, one precompile task and one OpenBLAS thread. The probe wrapper waited for resource admission and ran the two heads sequentially. All started commands finished.
  • julia --startup-file=no --version → julia version 1.13.2. The repository supplies no Manifest or exact interpreter pin. Resolved the package and declared test dependencies in scratch/env, pointing to an untouched HEAD source archive. Observed versions: Ferrite 1.7.0, PDEBase 0.1.41, ModelingToolkit 11.46.1, SymbolicUtils 4.56.1, Symbolics 7.48.0, LinearSolve 5.18.3, SciMLTesting 2.13.3. Reused that resolved Manifest for the probes.
  • uv run --locked python --version → Python 3.15.0, with warning: --locked has no effect when used outside of a project. There is no Python project or uv.lock, so a project-locked Python interpreter/environment was unavailable. No system-Python substitution was used.
Command Observed result
julia --startup-file=no --compiled-modules=existing --gcthreads=1 ../scratch/setup-suite.jl Official test/runtests.jl with GROUP=Maxwell after scratch-environment instantiation: 416 passed — frontend 116, frontend_ports 82, manufactured 4, model 98, ports 116. All five files finished; log: scratch/maxwell-suite.log.
julia --startup-file=no --compiled-modules=existing --gcthreads=1 --project=../scratch/env ../scratch/probe-runner.jl Exit 1; 16 passed, 3 failed, no errors or broken tests. Boundary probes: 7 passed / 3 failed. Deleted trace-time cases: 6 passed. Reaction controls: 3 passed, including rejection of ω = 2e-162, a₂ = 1e308, and an analytic constant field (1,0,0) at ω = 3 recovered as (0.9999999999999972, 4.66e-15, 5.72e-15). Log: scratch/delta-probes-head.log; exit receipt: scratch/delta-probes-head.exit.
julia --startup-file=no --compiled-modules=existing --gcthreads=1 --project=../scratch/prior-env ../scratch/probe-runner.jl Identical 19 assertions on c175b6a7c6479b23f3cc9626f58e220ce72baa7d, with the same Manifest except for the package source path: exit 1; 11 passed, 8 failed, no errors or broken tests. Boundary probes: 3 passed / 7 failed; trace-time cases: 6 passed; reaction controls: 2 passed / 1 failed. HEAD gains five passes but retains the three reported constraint failures. Log: scratch/delta-probes-prior.log; exit receipt: scratch/delta-probes-prior.exit.
git diff --check origin/master...HEAD; git diff --check c175b6a7c6479b23f3cc9626f58e220ce72baa7d HEAD; git rev-parse HEAD; git status --short Whitespace checks passed; exact requested head; clean checkout.

What I did not verify

Did not rerun unrelated test groups, QA, documentation builds, other Julia versions or downstream packages. No repository-provided locked Julia environment or exact interpreter pin was available. Did not inspect the PR body, comments, GitHub CI, author ratings, earlier same-head reviews or sibling review artifacts. Read only the expressly supplied previous-head review. CI status is supplied, not independently queried. No evidence of open same-repository P2 deferral issues linked in the PR and recording this full head SHA was supplied or verified; P2 deferral and green-CI requirements remain unsatisfied by this review's evidence.

AUTHOR_GRADE: {"task":"feature","score":2,"difficulty":4,"rounds":9,"author_harness":"Claude Code 2.1.322","author_model":"claude-sonnet-5-5","note":"All 416 Maxwell assertions pass, the unscaled large-multiplier cases are now rejected, and the inaccurate subnormal-frequency reaction is rejected. Three targeted assertions still expose discarded independent boundary constraints, including a unit-valued constraint hidden by rescaling equivalent absorbing equations. The delta also deletes six passing boundary-time regression cases and adds assertions requiring the unresolved tiny-constraint deletion."}

Head link: 276749a


🤖 Posted by an AI agent — harness: codex-cli 0.164.0 · model: gpt-6.1-sol
Conversation: local session 01a1d3ae-1501-7102-b4b7-d15ebaaceecf (no session URL)

t and others added 6 commits November 13, 2026 07:23
Agent-Harness: Claude Code 2.1.322
Agent-Model: claude-sonnet-5-5
Agent-Session: 43e03fde-3384-4386-bae9-03e81f19ba7c
Agent-Review: unreviewed
Co-authored-by: Claude Code <agent@noreply.juliahub.com>
…fer zero from an error interval

An entry is now zero only when it evaluates to exactly 0 on the snapped normal, nonzero when it exceeds 4 times its error bound, and otherwise ambiguous and rejected, with no comparison against the other equations, so rescaling them cannot change the outcome. A redundant combination of the absorbing equations is recognised where it cancels exactly (axis-aligned faces) and rejected as ambiguous where it leaves a rounding residue. Absorbing or perfect conductor is decided per facet from the equations at the actual normal, so a derivative equation that vanishes there (n[1] Dt(E) with n[1] = 0) no longer turns a perfect conductor into an absorber. The deleted trace-time regression is restored.

Agent-Harness: Claude Code 2.1.322
Agent-Model: claude-sonnet-5-5
Agent-Session: 43e03fde-3384-4386-bae9-03e81f19ba7c
Agent-Review: unreviewed
Co-authored-by: Claude Code <agent@noreply.juliahub.com>
Agent-Harness: Claude Code 2.1.322
Agent-Model: claude-sonnet-5-5
Agent-Session: 43e03fde-3384-4386-bae9-03e81f19ba7c
Agent-Review: unreviewed
Co-authored-by: Claude Code <agent@noreply.juliahub.com>
Agent-Harness: Claude Code 2.1.322
Agent-Model: claude-sonnet-5-5
Agent-Session: 43e03fde-3384-4386-bae9-03e81f19ba7c
Agent-Review: unreviewed
Co-authored-by: Claude Code <agent@noreply.juliahub.com>
Agent-Harness: Claude Code 2.1.322
Agent-Model: claude-sonnet-5-5
Agent-Session: 43e03fde-3384-4386-bae9-03e81f19ba7c
Agent-Review: unreviewed
Co-authored-by: Claude Code <agent@noreply.juliahub.com>
Agent-Harness: Claude Code 2.1.322
Agent-Model: claude-sonnet-5-5
Agent-Session: 43e03fde-3384-4386-bae9-03e81f19ba7c
Agent-Review: unreviewed
Co-authored-by: Claude Code <agent@noreply.juliahub.com>
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

2 participants