[matt-pharr]

Release note

• Audience: users
• Numerical impact: none (harness @ 4d2a8cb)
• Migration: A deck run with gal_match_flag no longer writes SingularSurfaces/Match/Inner/; the composite inner-region profiles are a scripting diagnostic (Galerkin(; cut_solution=true) then solve(MatchProblem(...))). gal_cut_solution is no longer a deck key; keep the cut solution from a script with Galerkin(; cut_solution=true). solve(EulerLagrangeProblem; gal_match_flag=true) now errors; match the result with solve(MatchProblem(ffs; eta, rho, rotation), GGJModel()).

A deck now matches the inner layer through the same MatchProblem a script poses, run by main after the outer solve. That also lets a deck match a Riccati solve, where gal_match_flag used to be silently ignored, and makes a forward solve with gal_match_flag fail with a clear error instead.

Regression report

Stacked on #552, so the baseline is refactor/stage-timing-in-main.

SDCC fan-out at 4d2a8cb7f (vs refactor/stage-timing-in-main @ 4d75b8047) is running after the rebase onto develop 2842dfc2d;
the full report replaces this block when it lands.

gal_resistive_pe is N/A on both refs: a matched result publishes no ideal δW, so PerturbedEquilibrium skips it. That is the existing state, not a change here.

Notes for reviewers

Stacked on #552, which sits on #546 (stack B). Closes 7d-match-into-main, match-riccati and match-resistive.

• The gal_match_flag block moves out of run_force_free_states into run_deck_match(result, ctrl). main calls it right after the outer solve and before the HDF5 write, so SingularSurfaces/Match is written exactly as before. It builds the same MatchProblem and GGJModel a script would.
• The integrator == "galerkin" gate is gone. A Riccati deck with vac_flag = true now matches, and a Forward deck errors through MatchProblem ("no Δ′") instead of ignoring the flag.
• The deck's duplicate checks (gal_rpec_flag, missing gal_eta/gal_rho/gal_rotation) are dropped; MatchProblem already rejects both. The deck-only gal_inner_solver check stays.
• solve(EulerLagrangeProblem) rejects gal_match_flag=true, so matching has one route: a MatchProblem posed on the result.
• Cut solution is API-only. This follows InnerLayer.GGJ - REFACTOR! - Solve the :ray BVP with the Galerkin assembly and use one parity convention #536's decision that inner-region profiles are a diagnostic not exposed in the deck. gal_cut_solution is removed from ForceFreeStatesControl; Galerkin(; cut_solution) hands it to galerkin_solve as a solve option, which no longer infers it from gal_match_flag. Nothing in the examples, harness or other tests read Match/Inner from a deck run.
• A deck match with gal_inner_solver = "galerkin" logs MatchProblem's existing one-time warning that no cut solution was retained. Since InnerLayer.GGJ - REFACTOR! - Solve the :ray BVP with the Galerkin assembly and use one parity convention #536, :ray returns layer-centre values only, so :ray decks don't see it.

Tests (runtests_solve_api.jl):

• Test - TEST - Add deck/API parity tests with the known drifts marked broken #546's cut-solution test now passes, so it is no longer marked broken. The deck and the API without the cut agree on bpen, the matched ξ and the (empty) inner profiles, and the API with cut_solution=true still produces them.
• New: deck/API parity for a Riccati match (bpen, cout), a Forward deck with gal_match_flag erroring, and solve(...; gal_match_flag=true) erroring.
• runtests_solve_api, runtests_galerkin and runtests_matching_models pass locally.

The gal_* match keys are now misnamed, since Riccati matches too. A follow-up (roadmap PR 8b) moves them into a [Match] section named after the MatchProblem keywords, with a warn-and-map shim for one release.

Overlap: #536 edits the gal_inner_solver comment line in CoreTypes.jl, next to the gal_cut_solution and gal_match_flag comment lines changed here. Whichever lands second resolves a one-line conflict.

🤖 Generated with Claude Code

[logan-nc]

Read in full. Ready to approve from my read, with one migration nit for you to decide.

gal_cut_solution leaves ForceFreeStatesControl but is not added to _DEPRECATED_FFS_KEYS or the toml-no-deprecated-keys pre-commit regex. A deck that still sets it fails as an unknown keyword (MethodError) instead of the deprecation message, and once #569 lands, instead of its stop. One tuple entry plus one regex token would fix it. I did not push it because it touches src/ (stamp) and #569 edits the same lists. Nothing else: Match/Inner has no remaining readers, run_deck_match builds the same MatchProblem/GGJModel the old block did, and the Riccati-match and forward-error tests cover the two behaviour changes.

[logan-nc]

Approving at 769685c on the comment review above. Whether gal_cut_solution gets a deprecation entry is your call and does not block this.