feat(hex8): integrate paired HHT dynamic state and bounded verification

docs/hex8-structural-analysis/PRD.md

@@ -434,7 +434,7 @@ HHT 식·기준 consistent M·입력·고정 schedule·100회·Si/exact-zero·
 - [x] Solid Domain/parser/3N DOF/ordered assembly/amplitude 검증
 - [x] 선형·비선형 정적 accepted state/반력/잔차 검증
 - [x] Generalized consistent-mass frequency·최저 N·original-pencil 검증
-- [ ] 선형·비선형 HHT u/v/physical a·시간수렴·수지 검증
+- [x] 선형·비선형 HHT u/v/physical a·시간수렴·수지 내부 검증
 - [ ] 공식 HDF5/CLI/self-check/실패 final 보존 검증
 - [ ] 정적2·동적2 원본 필수 family 비교 PASS
 - [ ] 사용자 원본 최저10 modal Value 제공 및 필수 비교 PASS
@@ -446,9 +446,14 @@ Local kernel과 입력·3N mapping은 독립 검토 및 연결 후 관련180개
 native CTest exit0이며 독립 수치 검토를 통과했다. 두 정적 원본과442절점 고유치 원본은
 직접 API로 실행했지만 원본 결과값 비교를 수행한 것은 아니다. 연결 source `28be39f`의
 전체 MSVC Debug build는 exit0이며 이 build 자체를 전체 CTest 통과로 해석하지 않는다.
-HHT 내부 사례는 진행 중이다. 별도 구현 `f661864`의 관련76개 검사와 독립 수치 검토는
+과거 단일 Float64 구현 `f661864`는 관련76개 검사와 독립 수치 검토를 통과했지만,
-통과했지만, 두 동적 원본은 세 번째 증분100회 제한에서 엄격 잔차 기준을 만족하지 못했다.
+두 동적 원본의 세 번째 증분에서 strict gate를 만족하지 못했다. 이 실패는
-최신 잔차 비율은 선형 `1.77739e-8`, 비선형 `8.73572e-9`이며 기준은 `1e-10`이다.
+[lessons](lessons.md)에 보존하며 현재 방식의 결과와 구분한다.
-동적 검증은 미완료이며 원인과 추가 검증은 [lessons](lessons.md)에 기록한다.
+승인된 전체 high+low 구현 `c49cae4`는 소규모32/32 및 별도 성공한 LIN/TL·SDOF/MDOF
+시간수렴 사례와 독립 numerical/physics 검토를 확보했다. Root 연결 후 관련49/49와
+추가 직렬/TBB low 일치1/1 검사도 native build/CTest0/0이다. 두 원본의 초기 상태와
+증분1–3은 unchanged schedule로 관측했고, 원래 모든 free 성분의 전체 pair 물리 gate를
+독립 재계산했다. 이 관측은 네 번째 증분 전에 종료했으며 전체101 endpoint 완료가 아니다.
+상세 범위와 한계는 [수치 검토](numerical-review.md), [물리 검토](physics-review.md)에 있다.
 공식 결과·외부 비교·최신 전체 회귀도 남아 있다.
 원본 frequency 결과 부재는 구현 진행을 막지 않지만 기능 검증 완료를 막는다.

docs/hex8-structural-analysis/lessons.md

@@ -259,3 +259,16 @@ trial에서 계산한다. 독립 reviewer가 이 캐시와 native44/44의 원시
 독립 numerical reviewer는 실제 shared arithmetic·kernel·geometry의 한정된 범위를 PASS로
 판정했다. 재현 로그는 `.harness/hex8-evidence/precision/`에 보존한다. 이 결과는 동적 전체
 상태·시간수렴·원본 endpoint·공식 출력의 완료를 뜻하지 않는다. 현재 checklist는 미완료다.
+
+## 전체 상태의 제한적 원본 관측과 연결
+
+전체 high+low T6 `c49cae4`는 원래 시간 간격·INC를 바꾸지 않고 두 원본의 초기 상태와
+증분1–3을 accept했다. Test-only typed stop으로 다음 증분 전에 종료해 prefix 상태를
+관측했으며 native0·2/2를 전체 원본 완료로 보고하지 않는다. 메인이 전체 pair·full M/Mff,
+모든1275 free r/Si·zero branch·HHT update·weighted old-r·실제 solve 보정을 exact Fraction으로
+검산했다. 물리 비율 최대값은 선형3.12680e-20, 비선형1.3043731e-16이다.
+
+Root 연결의 related49/49와 두 brick의 LIN/TL 전체 low 직렬/TBB 일치1/1은 native0이다.
+최초 broad filter가 긴 전체 원본 smoke를 함께 선택한 실행은 의도적으로 중단했고
+CTest exit−1로 보존했다. 완료하지 않은 실행을 수렴 실패로 해석하거나 현재 PASS에 합산하지
+않는다. 공식 출력·101 endpoint·원본 비교·전체 회귀는 후속 검증이다.

docs/hex8-structural-analysis/numerical-review.md

@@ -20,6 +20,37 @@ T6 consumer·backend·worker 경계와 rollback은 해당 구현이 고정된
 동일 graph·선형 force/energy/recovery·비선형 trial Hessian·기존 F64 Create 지원을 유지하며
 native44/44 원시 로그를 확인했다. 이 캐시 변경도 bounded PASS다.
 
+## 2026-10-10 — 전체 동적 상태와 원본 세 증분의 제한적 검토
+
+동일 독립 reviewer가 고정된 T6 `c49cae4`의 predictor/rates, 기준 M의 전체 pair 작용,
+ordered force/tangent 누적, full residual·scale, 실제 old residual을 포함한 weighted 식,
+fresh solve와 실제 상태 변경의 네 보정 채널, backend/worker 이후 실행 전제 검사,
+accepted low의 복사·검증·rollback을 검토했다. 원시73개 inventory와 owned/shared source,
+binary·patch의 hash를 확인했고 해당 consumer 범위 bounded PASS, actionable finding0이다.
+Current32/32 검사와 과거 temporal4/5 실행의 성공 사례를 구분한다.
+
+이후 변경하지 않은 두 원본의 초기 상태와 증분1–3을 관측했다. 원래 h=0.1,T=10,INC=100,
+1326 full/1275 free/51 constrained를 유지하고, 네 번째 증분 local 계산 전 test-only 예외로
+종료했다. 관측 native0·2/2는 이 종료 조건의 성공이며 전체 해석 완료가 아니다.
+메인은51개 raw artifact의 hash/크기를 확인한 뒤 두 binary64 성분을 exact dyadic Fraction으로
+읽어 모든 상태의 normalization, 실제 Mff의 full M restriction, 원래 모든 free 성분의
+물리 r/Si와 exact-zero, HHT 세 update 식과 weighted old-r 항등식을 독립 재계산했다.
+
+Exact CSR M 작용으로 재계산한 물리 비율의 최댓값은 선형3.12680e-20,
+비선형1.3043731e-16으로 승인1e-10 이내다. Update·weighted identity는1e-12 이내이며
+초기1275개 free row는 모두 exact0이다. 실제 PARDISO 보정의 선형6회·비선형9회 기록으로
+Newton 시작값인 이전 accepted u부터 최종 u를 재구성하고, fresh/observed 네 보정 채널도
+각각 확인했다. 비선형 최대 보정 비율1.01046e-13은 승인1e-10 이내다.
+
+Storage graph 대조의 별도 절대 오차 budget과 물리 gate를 혼동하지 않는다. Exact M 작용과
+저장된 pair Ma의 상대 차이는 최대2.88e-32였다. Strict physical gate에 floor를 추가하거나
+high만으로 full 상태를 대신하지 않았다. 재현 자료는 로컬
+`.harness/hex8-evidence/precision/{prefix3-frozen,prefix-independent.json,check_prefix.py}`다.
+Root 연결 후 관련49/49 검사의 native build/CTest는0/0이다. 이어 두 brick의 LIN/TL 직렬·TBB
+전체 low 필드 일치와 실제 nonzero low를 확인한 focused1/1 검사도0/0이다.
+최초 넓은 filter가 전체 원본 smoke를 선택해 중단한 실행은 CTest exit−1로 보존하며
+수렴 실패 또는 통과로 해석하지 않는다. 전체101 endpoint·공식 HDF5·원본 비교는 아직 없다.
+
 ## HEX8-NR-001 — 영 회전 경계와 필수 원본 충돌
 
 검토 당시 PRD008/I/O는 BC/CLOAD DOF4–6을 값0이어도 거부했다. 그러나 모든 필수 INP에

docs/hex8-structural-analysis/physics-review.md

@@ -35,7 +35,30 @@ temporal 실행은4/5 PASS·CTest exit8이며 실패는 이전 principal trace a
 owned/shared source·원본 두 INP·binary·patch hash를 대조했다. Reviewer는 관련22개 raw와
 7개 owned source·patch를 별도로 대조했다.
 
-전체 high+low endpoint 배열의 독립 gate 재계산과 원본 prefix/101 endpoint·필수 reference
+첫32개 검토 이후 원본 prefix의 전체 pair를 아래와 같이 별도 재계산했다. 전체101 endpoint·
-비교는 아직 수행하지 않았다. Rounded observable 불변량 검사와 전체 pair의 production
+필수 reference 비교는 아직 없다. Rounded observable 불변량 검사와 전체 pair의 production
 gate 통과를 구분한다. Alpha−0.05의 장시간 에너지, warped dynamic mesh,
 near-incompressibility 및 큰 strain StVK 적용은 이번 evidence로 보증하지 않는다.
+
+## 원본 초기 상태와 세 증분의 독립 물리 계산
+
+변경하지 않은 LIN/TL 두 원본의 초기 상태와 증분1–3에 대해 reviewer가 모든 high+low를
+exact Fraction으로 읽었다. 기준 CSR M·reference X·실제 full force를 사용해 T/W/balance,
+p/pdot/L/Ldot, reference/current/free-residual moments와 global force/moment 수지를
+총240개 대조했다. 실제 pair Add의 discarded-tail bound를 전파한 별도 절대 graph budget
+내였으며 재생 graph의 high/low도240/240 일치했다. 구속 u/v/a/A와 support work는 exact0다.
+Source442절점·300연결·51구속·하중·원래 h/T/INC와 test-only 관측 종료 조건도 대조했다.
+선형 U와 독립0.5uᵀf의 최대 절대차는1.30e-29였다.
+
+Captured F64 M의 ordered coefficient에는7752개의 비대칭 entry가 있고
+최대|Mij−Mji|=2.84217e-14다. 이에 따른 exact Σv×Mv의 작은 비영 값은512eps same-unit
+coefficient-roundoff budget 이내이며 이 비대칭으로 정확히 설명된다. Pair 산술 오류와
+계수 오차를 구분하고 M을 대칭화하거나 모멘트를0으로 보정하지 않는다. Ldot=x×Ma는
+같은 endpoint의 계약 observable로 확인했다. LIN의 current internal moment는 O(u²)로
+비영이며 TL의 reference moment도 별도0 기대량이 아니다.
+
+Alpha−0.05의 frame별 energy balance는 관측량이지 exact conservation gate가 아니다.
+Prefix에서 나타난 balance와 진동에 따른 trapezoid work 감소를 그대로 보존한다.
+이 범위에서 물리 blocker는 없다. 재현 자료는 로컬
+`.harness/hex8-evidence/precision/physics-prefix-review/`와 `prefix3-frozen/`다.
+TL의 local polynomial U 검산은 별도 frozen reference cache에 대해 추가 확인한다.

docs/tracking/findings.md

@@ -3,7 +3,7 @@
 ## Hex8 동해석의 엄격 잔차와 내부 상태 정밀도
 
 - 조건·현상: Hex8 동적 구현 `f661864`에서 `reference/hexa/hexa_dynlinear.inp`와
-  `hexa_dynnonlinear.inp`를 실행하면 세 번째 증분100회 제한에서 실패한다. 최신 원본
+  `hexa_dynnonlinear.inp`를 실행하면 세 번째 증분100회 제한에서 실패했다. 이 과거 원본
   physical residual 비율은 `1.77739e-8`·`8.73572e-9`, 기준은 `1e-10`이다.
 - 영향: 원본101개 endpoint의 필수 검증과 동해석 완료 조건을 충족하지 못했다.
   현재 전체 요구와 checklist는 [기능 PRD](../hex8-structural-analysis/PRD.md)에 있다.
@@ -12,9 +12,10 @@
   Float64 해의 부재를 증명하지 않으며, 수렴 기준·원본을 변경할 근거로 쓰지 않는다.
 - 후속 방향: [실패 근거](../hex8-structural-analysis/lessons.md)를 보존하고, 전체 물리
   상태의 high+low 계산과 Float64 출력의 반올림 의미를 정하는
-  [승인된 변경안](../hex8-structural-analysis/precision-proposal.md)을 구현한다.
+  [승인된 변경안](../hex8-structural-analysis/precision-proposal.md)을 구현했다.
-  사용자가 2026-10-09 선택 A로 승인했다. 산술·요소의 첫 native 검사는 통과했으나
+  사용자가 2026-10-09 선택 A로 승인했고 소규모 상태·시간수렴·수지와 두 원본의 세 증분
-  전체 동적 state·시간수렴·원본 endpoint 검증은 남아 있다. Acceptance 기준은 유지한다.
+  관측을 검증했다. 독립 full-pair 재계산에서 원래 모든 free의 strict gate를 통과했다.
+  전체101 endpoint·공식 출력·필수 원본 비교는 남아 있으며 acceptance 기준은 유지한다.
 
 ## HHT shell의 시간수렴
 

include/fesa/analysis/solid_dynamic_analysis_state.h

@@ -0,0 +1,203 @@
+#ifndef FESA_ANALYSIS_SOLID_DYNAMIC_ANALYSIS_STATE_H_
+#define FESA_ANALYSIS_SOLID_DYNAMIC_ANALYSIS_STATE_H_
+
+#include <array>
+#include <cstddef>
+#include <cstdint>
+#include <optional>
+#include <string>
+#include <vector>
+
+#include "fesa/fem/dof_manager.h"
+#include "fesa/results/solid_result_records.h"
+
+namespace fesa {
+
+/// @brief Dimensionally explicit translation-only Newton evidence.
+struct SolidDynamicIterationEvidence {
+  std::uint32_t iteration_index{0};
+  double residual_numerator{0};  // Force infinity norm.
+  double residual_scale{0};      // Uncancelled force infinity scale.
+  double residual_metric{0};     // Maximum componentwise force ratio.
+  std::size_t worst_full_dof{0};
+  std::array<double, 5> worst_force_terms{};  // r,Ma,fint,F,uncancelled scale.
+  double weighted_residual_numerator{0};  // Original HHT force infinity norm.
+  double recombined_residual_numerator{
+      0};                                 // Actual solved weighted force norm.
+  double old_residual_numerator{0};       // Previous physical free force norm.
+  double weighted_equivalence_metric{0};  // Uncancelled force identity ratio.
+  std::array<double, 4>
+      worst_weighted_terms{};  // Original,recombined,old r,scale.
+  bool zero_scale_exact{true};
+  std::array<double, 4> correction_metrics{};  // du/L,h*dv/L,h²*da/L,h²*dA/L.
+  double update_metric{0};  // Maximum uncancelled update equality ratio.
+  bool residual_pass{false};
+  bool correction_pass{false};
+  bool update_pass{false};
+};
+
+/// @brief Failure-only shadow dot products; these never change force or
+/// acceptance.
+struct SolidDynamicLinearDotEvidence {
+  EntityIndex element_index{0};
+  std::size_t local_row{0};
+  double absolute_product_sum{0};
+  double kernel_force{0};
+  double compensated_force{
+      0};  // Shadow using rounded relative differences only.
+};
+
+/// @brief Failure-only coordinate arithmetic evidence, with length units.
+struct SolidDynamicCoordinateEvidence {
+  EntityIndex element_index{0};
+  std::size_t local_row{0};
+  double displacement{0};
+  double previous_displacement{0};
+  double fresh_correction{0};
+  double reference_displacement{0};
+  double previous_reference_displacement{0};
+  double relative_high{0};
+  double previous_relative_high{0};
+};
+
+/// @brief Failure-only arithmetic shadow; never substitutes physical forces.
+struct SolidDynamicForceSumEvidence {
+  double assembled_internal{0};
+  double compensated_internal{0};
+  double absolute_local_internal_sum{0};
+  double assembled_inertia{0};
+  double compensated_inertia{0};
+  double absolute_inertia_product_sum{0};
+};
+
+/// @brief Internal exact-candidate recurrence evidence, separate from gates.
+struct SolidDynamicNewtonRepeatEvidence {
+  std::size_t history_entry_count{0};
+  std::uint32_t matched_entry_iteration{0};
+  std::size_t changed_translations{0};
+  std::size_t changed_holding_mask{0};
+  std::size_t changed_correction_metrics{0};
+  bool repeat{false};
+};
+
+/// @brief Keeps failed attempt evidence outside the accepted physical history.
+struct SolidDynamicIterationRecord {
+  std::uint32_t increment_index{0};
+  double time{0};
+  std::string stage;
+  std::string outcome;
+  std::string failure_code;
+  SolidDynamicIterationEvidence evidence;
+  std::vector<SolidDynamicLinearDotEvidence> failure_linear_dot;
+  std::vector<SolidDynamicCoordinateEvidence> failure_coordinates;
+  std::optional<SolidDynamicForceSumEvidence> failure_force_sum;
+  std::optional<SolidDynamicNewtonRepeatEvidence> newton_repeat;
+};
+
+/// @brief Retains normalized physical low parts; never serialized as extra
+/// result fields. Corresponding endpoint fields contain the nearest Float64
+/// high parts. Both parts are copied and validated together at commit.
+struct SolidDynamicFrameLowParts {
+  Vector displacement{0};
+  Vector velocity{0};
+  Vector acceleration{0};
+  Vector algorithmic_acceleration{0};
+  Vector inertia_force{0};
+  Vector internal_force{0};
+  Vector external_force{0};
+  Vector residual{0};
+  Vector reaction{0};
+  Vector residual_scale{0};
+  double strain_energy{0};
+  double kinetic_energy{0};
+  double external_work{0};
+  double energy_balance_error{0};
+  std::array<double, 3> linear_momentum{};
+  std::array<double, 3> angular_momentum{};
+  std::array<double, 3> linear_momentum_rate{};
+  std::array<double, 3> angular_momentum_rate{};
+  std::array<double, 6> global_equilibrium{};
+  std::array<double, 3> linearization_current_internal_moment{};
+  std::array<double, 3> reference_internal_moment{};
+  std::array<double, 3> current_free_residual_moment{};
+};
+
+/// @brief Owns one complete Solid endpoint; physical a differs from auxiliary
+/// A.
+struct SolidDynamicAcceptedFrameState {
+  std::uint32_t frame_index{0};
+  std::uint32_t increment_index{0};
+  std::uint32_t iteration_count{0};
+  double time{0};
+  bool is_initial{false};
+  std::vector<EntityIndex> nodal_internal_order;
+  std::vector<SourceEntityId> nodal_source_ids;
+  Vector displacement{0};
+  Vector velocity{0};
+  Vector acceleration{0};
+  Vector algorithmic_acceleration{0};
+  Vector inertia_force{0};
+  Vector internal_force{0};
+  Vector external_force{0};
+  Vector residual{0};
+  Vector reaction{0};
+  Vector residual_scale{0};
+  std::vector<SolidIntegrationPointRow> solid_results;
+  double strain_energy{0};
+  double kinetic_energy{0};
+  double external_work{0};
+  double energy_balance_error{0};
+  std::array<double, 3> linear_momentum{};
+  std::array<double, 3> angular_momentum{};
+  std::array<double, 3> linear_momentum_rate{};
+  std::array<double, 3> angular_momentum_rate{};
+  std::array<double, 6> global_equilibrium{};
+  // Internal evidence, not additional official output fields.
+  std::array<double, 3> linearization_current_internal_moment{};
+  std::array<double, 3> reference_internal_moment{};
+  std::array<double, 3> current_free_residual_moment{};
+  std::vector<SolidDynamicIterationEvidence> iterations;
+  SolidDynamicFrameLowParts low;
+};
+
+/// @brief Owns only validated accepted Solid history; rejects partial
+/// candidates.
+class SolidDynamicAnalysisState {
+ public:
+  /// @brief Creates an empty state for exactly one zero-BC fixed dynamic
+  /// schedule.
+  static Result<SolidDynamicAnalysisState> Create(const Domain& domain,
+                                                  const DofManager& dofs);
+  /// @brief Validates shape, identity, schedule, finite values and Newton
+  /// evidence.
+  /// @note A failed candidate never changes accepted frames or current state.
+  Status CommitAcceptedFrame(const SolidDynamicAcceptedFrameState& frame);
+  /// @brief Returns an independent accepted endpoint copy for trial
+  /// preparation.
+  Result<SolidDynamicAcceptedFrameState> BeginTrialFromCommitted() const;
+  /// @brief Returns accepted endpoints in source increment order.
+  const std::vector<SolidDynamicAcceptedFrameState>& AcceptedFrames()
+      const noexcept;
+  /// @brief Returns the last endpoint; throws when no frame is accepted.
+  const SolidDynamicAcceptedFrameState& CommittedState() const;
+  /// @brief Reports whether schedule and inventory have been initialized.
+  bool IsInitialized() const noexcept;
+  /// @brief Appends attempt evidence without committing physical state.
+  void RecordIteration(SolidDynamicIterationRecord record);
+  /// @brief Returns attempt evidence, including a terminal failure if present.
+  const std::vector<SolidDynamicIterationRecord>& IterationRecords()
+      const noexcept;
+
+ private:
+  std::size_t full_dof_count_{0};
+  std::uint32_t increment_count_{0};
+  double time_increment_{0};
+  std::vector<std::size_t> constrained_dofs_;
+  std::vector<SourceEntityId> source_ids_;
+  std::vector<SolidIntegrationPointRow> point_inventory_;
+  std::vector<SolidDynamicAcceptedFrameState> accepted_frames_;
+  std::vector<SolidDynamicIterationRecord> iteration_records_;
+  bool initialized_{false};
+};
+}  // namespace fesa
+#endif  // FESA_ANALYSIS_SOLID_DYNAMIC_ANALYSIS_STATE_H_

include/fesa/analysis/solid_hht_dynamic_procedure.h

@@ -0,0 +1,44 @@
+#ifndef FESA_ANALYSIS_SOLID_HHT_DYNAMIC_PROCEDURE_H_
+#define FESA_ANALYSIS_SOLID_HHT_DYNAMIC_PROCEDURE_H_
+
+#include "fesa/analysis/solid_dynamic_analysis_state.h"
+
+namespace fesa {
+class LinearSolver;
+class ParallelFor;
+class TangentSolver;
+
+/// @brief Owns the validated 3N dynamic result independently of serialization.
+struct SolidDynamicExecutionResult {
+  DofManager dofs;
+  SolidDynamicAnalysisState state;
+};
+
+/// @brief Integrates linear or TL Solid mechanics with constant consistent
+/// mass.
+class SolidHhtDynamicProcedureExecutor {
+ public:
+  /// @brief Retains caller-owned adapters which must outlive execution.
+  SolidHhtDynamicProcedureExecutor(const ParallelFor& parallel_for,
+                                   LinearSolver& mass_solver,
+                                   TangentSolver& tangent_solver);
+  /// @brief Returns complete accepted history without opening output files.
+  Result<SolidDynamicExecutionResult> Execute(const Domain& domain) const;
+  /// @brief Executes into an empty initialized state, preserving accepted
+  /// history on every failure. Domain must match the state inventory and
+  /// schedule.
+  Status ExecuteInto(const Domain& domain,
+                     SolidDynamicAnalysisState& state) const;
+  /// @brief Internal verification only; alpha0 verifies conservative dynamics.
+  /// @note This does not extend the approved INP parameter subset.
+  Status ExecuteIntoForVerification(const Domain& domain,
+                                    SolidDynamicAnalysisState& state,
+                                    double alpha) const;
+
+ private:
+  const ParallelFor& parallel_for_;
+  LinearSolver& mass_solver_;
+  TangentSolver& tangent_solver_;
+};
+}  // namespace fesa
+#endif  // FESA_ANALYSIS_SOLID_HHT_DYNAMIC_PROCEDURE_H_

src/fesa/CMakeLists.txt

@@ -17,7 +17,9 @@ add_library(
     analysis/nonlinear_static_control.cpp
     analysis/nonlinear_static_procedure.cpp
     analysis/prepared_analysis.cpp
+    analysis/solid_dynamic_analysis_state.cpp
     analysis/solid_frequency_procedure.cpp
+    analysis/solid_hht_dynamic_procedure.cpp
     analysis/solid_static_procedure.cpp
     analysis/solid_static_state.cpp
     analysis/thermal_analysis_state.cpp

src/fesa/analysis/solid_dynamic_analysis_state.cpp

@@ -0,0 +1,280 @@
+#include "fesa/analysis/solid_dynamic_analysis_state.h"
+
+#include <algorithm>
+#include <cmath>
+#include <stdexcept>
+#include <string>
+
+#include "fesa/elements/element_factory.h"
+#include "fesa/elements/hex8_element.h"
+#include "math/high_low_arithmetic.h"
+
+namespace fesa {
+namespace {
+/// @brief Rejects a candidate without changing observable history.
+Status Failure(const std::string& message) {
+  return Status::Failure(FailureCategory::kSolver,
+                         {{Severity::kError,
+                           "solid-dynamic-invalid-state",
+                           {},
+                           "DYNAMIC",
+                           {},
+                           message}});
+}
+/// @brief Checks a physical record without replacing nonfinite quantities.
+template <std::size_t N>
+bool Finite(const std::array<double, N>& values) {
+  return std::all_of(values.begin(), values.end(),
+                     [](double x) { return std::isfinite(x); });
+}
+bool ValidPair(double high, double low) {
+  const math_internal::HighLow value(high, low);
+  return math_internal::IsFinite(value) && math_internal::IsNormalized(value);
+}
+bool ValidPair(const Vector& high, const Vector& low, std::size_t size) {
+  if (high.Size() != size || low.Size() != size) return false;
+  for (std::size_t i = 0; i < size; ++i)
+    if (!ValidPair(high[i], low[i])) return false;
+  return true;
+}
+template <std::size_t N>
+bool ValidPair(const std::array<double, N>& high,
+               const std::array<double, N>& low) {
+  for (std::size_t i = 0; i < N; ++i)
+    if (!ValidPair(high[i], low[i])) return false;
+  return true;
+}
+/// @brief Preserves canonical and raw source identity independently of indices.
+bool SameId(const SourceEntityId& a, const SourceEntityId& b) {
+  return a.instance_name == b.instance_name &&
+         a.source_label == b.source_label &&
+         a.source_label_text == b.source_label_text;
+}
+}  // namespace
+
+Result<SolidDynamicAnalysisState> SolidDynamicAnalysisState::Create(
+    const Domain& domain, const DofManager& dofs) {
+  const auto* step = domain.Dynamic();
+  if (!domain.IsSolid() || step == nullptr ||
+      dofs.Profile() != DofProfile::kSolidStructural ||
+      dofs.FullDofCount() != 3 * domain.Nodes().size() ||
+      !dofs.ValidateInvariants().IsOk() || step->increment_count == 0 ||
+      !std::isfinite(step->initial_increment) || step->initial_increment <= 0 ||
+      !std::isfinite(step->time_period) ||
+      static_cast<double>(step->increment_count) * step->initial_increment !=
+          step->time_period ||
+      step->increment_count > step->maximum_increments) {
+    return Result<SolidDynamicAnalysisState>::Failure(
+        Failure("Solid HHT requires a valid 3N exact fixed schedule."));
+  }
+  for (std::size_t i = 0; i < dofs.PrescribedValues().Size(); ++i)
+    if (dofs.PrescribedValues()[i] != 0)
+      return Result<SolidDynamicAnalysisState>::Failure(
+          Failure("Dynamic prescribed translations must be exactly zero."));
+  SolidDynamicAnalysisState state;
+  state.full_dof_count_ = dofs.FullDofCount();
+  state.increment_count_ = step->increment_count;
+  state.time_increment_ = step->initial_increment;
+  state.constrained_dofs_ = dofs.ConstrainedDofs();
+  for (const auto& node : domain.Nodes())
+    state.source_ids_.push_back(node.source_id);
+  ElementFactory factory;
+  for (std::size_t e = 0; e < domain.Elements().Size(); ++e) {
+    auto runtime = factory.Create(domain.Elements()[e], domain);
+    if (!runtime.HasValue())
+      return Result<SolidDynamicAnalysisState>::Failure(runtime.GetStatus());
+    const auto* solid = dynamic_cast<const Hex8Element*>(runtime.Value().get());
+    if (solid == nullptr)
+      return Result<SolidDynamicAnalysisState>::Failure(Failure(
+          "Dynamic Solid inventory contains a different element family."));
+    auto zero = solid->EvaluateLinear(Vector(24));
+    if (!zero.HasValue())
+      return Result<SolidDynamicAnalysisState>::Failure(zero.GetStatus());
+    state.point_inventory_.insert(state.point_inventory_.end(),
+                                  zero.Value().points.begin(),
+                                  zero.Value().points.end());
+  }
+  state.initialized_ = true;
+  return Result<SolidDynamicAnalysisState>::Success(std::move(state));
+}
+
+Status SolidDynamicAnalysisState::CommitAcceptedFrame(
+    const SolidDynamicAcceptedFrameState& f) {
+  if (!math_internal::ArithmeticEnvironmentSupported())
+    return Failure(
+        "Paired state commit requires nearest rounding and gradual "
+        "underflow.");
+  const std::size_t index = accepted_frames_.size();
+  if (!initialized_ || index > increment_count_ || f.frame_index != index ||
+      f.increment_index != index || f.is_initial != (index == 0) ||
+      f.time != static_cast<double>(index) * time_increment_ ||
+      f.iteration_count > 100 || (index == 0 && f.iteration_count != 0) ||
+      f.nodal_internal_order.size() != source_ids_.size() ||
+      f.nodal_source_ids.size() != source_ids_.size() ||
+      f.solid_results.size() != point_inventory_.size()) {
+    return Failure(
+        "Incomplete or out-of-order Solid endpoint inventory/schedule.");
+  }
+  for (std::size_t n = 0; n < source_ids_.size(); ++n)
+    if (f.nodal_internal_order[n] != n ||
+        !SameId(f.nodal_source_ids[n], source_ids_[n]))
+      return Failure("Solid nodal source identity differs from its inventory.");
+  const std::array<const Vector*, 10> high{
+      &f.displacement,   &f.velocity,
+      &f.acceleration,   &f.algorithmic_acceleration,
+      &f.inertia_force,  &f.internal_force,
+      &f.external_force, &f.residual,
+      &f.reaction,       &f.residual_scale};
+  const std::array<const Vector*, 10> low{
+      &f.low.displacement,   &f.low.velocity,
+      &f.low.acceleration,   &f.low.algorithmic_acceleration,
+      &f.low.inertia_force,  &f.low.internal_force,
+      &f.low.external_force, &f.low.residual,
+      &f.low.reaction,       &f.low.residual_scale};
+  for (std::size_t i = 0; i < high.size(); ++i)
+    if (!ValidPair(*high[i], *low[i], full_dof_count_))
+      return Failure(
+          "Solid endpoint pairs are incomplete, unnormalized or "
+          "nonfinite.");
+  using namespace math_internal;
+  for (std::size_t i = 0; i < full_dof_count_; ++i) {
+    const bool constrained = std::binary_search(constrained_dofs_.begin(),
+                                                constrained_dofs_.end(), i);
+    const HighLow inertia(f.inertia_force[i], f.low.inertia_force[i]);
+    const HighLow internal(f.internal_force[i], f.low.internal_force[i]);
+    const HighLow external(f.external_force[i], f.low.external_force[i]);
+    const HighLow residual(f.residual[i], f.low.residual[i]);
+    const HighLow scale(f.residual_scale[i], f.low.residual_scale[i]);
+    const HighLow reaction(f.reaction[i], f.low.reaction[i]);
+    const auto expected_residual = Subtract(Add(inertia, internal), external);
+    const auto expected_scale =
+        Add(Add(Abs(inertia), Abs(internal)), Abs(external));
+    if (!IsFinite(expected_residual) || !IsFinite(expected_scale) ||
+        Compare(residual, expected_residual) != 0 ||
+        Compare(scale, expected_scale) != 0 ||
+        Compare(reaction, constrained ? residual : HighLow(0)) != 0)
+      return Failure(
+          "Physical reaction/residual/uncancelled force scale disagree.");
+    if (constrained && (f.displacement[i] != 0 || f.low.displacement[i] != 0 ||
+                        f.velocity[i] != 0 || f.low.velocity[i] != 0 ||
+                        f.acceleration[i] != 0 || f.low.acceleration[i] != 0 ||
+                        f.algorithmic_acceleration[i] != 0 ||
+                        f.low.algorithmic_acceleration[i] != 0))
+      return Failure(
+          "Prescribed dynamic translations and rates must be exact zero.");
+    if (!constrained) {
+      if (Compare(scale, HighLow(0)) < 0)
+        return Failure("Negative physical force scale.");
+      if (Compare(scale, HighLow(0)) == 0) {
+        if (Compare(residual, HighLow(0)) != 0)
+          return Failure("Physical zero-scale residual must be exactly zero.");
+      } else {
+        const auto ratio = Divide(Abs(residual), scale);
+        if (!IsFinite(ratio) || !IsNormalized(ratio) ||
+            Compare(ratio, HighLow(1e-10)) > 0)
+          return Failure("Physical free endpoint equilibrium is incomplete.");
+      }
+    }
+    if (index == 0 &&
+        (f.displacement[i] != 0 || f.velocity[i] != 0 ||
+         f.low.displacement[i] != 0 || f.low.velocity[i] != 0 ||
+         Compare(HighLow(f.acceleration[i], f.low.acceleration[i]),
+                 HighLow(f.algorithmic_acceleration[i],
+                         f.low.algorithmic_acceleration[i])) != 0))
+      return Failure(
+          "Initial Solid displacement/velocity must be zero and A=a.");
+  }
+  for (std::size_t i = 0; i < point_inventory_.size(); ++i) {
+    const auto& row = f.solid_results[i];
+    const auto& expected = point_inventory_[i];
+    if (row.element != expected.element ||
+        row.integration_point != expected.integration_point ||
+        row.natural_coordinates != expected.natural_coordinates ||
+        row.reference_coordinates != expected.reference_coordinates ||
+        !Finite(row.cauchy_stress) || !Finite(row.strain))
+      return Failure(
+          "Solid point recovery has invalid identity or finite values.");
+  }
+  if (!ValidPair(f.strain_energy, f.low.strain_energy) ||
+      !ValidPair(f.kinetic_energy, f.low.kinetic_energy) ||
+      !ValidPair(f.external_work, f.low.external_work) ||
+      !ValidPair(f.energy_balance_error, f.low.energy_balance_error))
+    return Failure("Nonfinite or unnormalized Solid energy/work pair.");
+  if ((index == 0 &&
+       (f.external_work != 0 || f.low.external_work != 0 ||
+        f.energy_balance_error != 0 || f.low.energy_balance_error != 0)) ||
+      Compare(HighLow(f.kinetic_energy, f.low.kinetic_energy), HighLow(0)) < 0)
+    return Failure(
+        "Initial work/error or kinetic energy violates its definition.");
+  if (!ValidPair(f.linear_momentum, f.low.linear_momentum) ||
+      !ValidPair(f.angular_momentum, f.low.angular_momentum) ||
+      !ValidPair(f.linear_momentum_rate, f.low.linear_momentum_rate) ||
+      !ValidPair(f.angular_momentum_rate, f.low.angular_momentum_rate) ||
+      !ValidPair(f.linearization_current_internal_moment,
+                 f.low.linearization_current_internal_moment) ||
+      !ValidPair(f.reference_internal_moment,
+                 f.low.reference_internal_moment) ||
+      !ValidPair(f.current_free_residual_moment,
+                 f.low.current_free_residual_moment))
+    return Failure("Nonfinite or unnormalized physical momentum/moment pair.");
+  if (!ValidPair(f.global_equilibrium, f.low.global_equilibrium) ||
+      f.iterations.empty())
+    return Failure("Missing finite equilibrium or iteration evidence.");
+  for (std::size_t i = 0; i < f.iterations.size(); ++i) {
+    const auto& e = f.iterations[i];
+    if (e.iteration_index != i || !Finite(e.correction_metrics) ||
+        !Finite(e.worst_force_terms) || !Finite(e.worst_weighted_terms))
+      return Failure("Incomplete or nonfinite Newton correction evidence.");
+    for (double x : {e.residual_numerator, e.residual_scale, e.residual_metric,
+                     e.weighted_residual_numerator, e.update_metric,
+                     e.old_residual_numerator, e.recombined_residual_numerator,
+                     e.weighted_equivalence_metric})
+      if (!std::isfinite(x) || x < 0)
+        return Failure("Invalid raw Newton residual/update evidence.");
+    for (double x : e.correction_metrics)
+      if (x < 0) return Failure("Invalid correction scale.");
+  }
+  const auto& last = f.iterations.back();
+  if (f.iterations.size() != static_cast<std::size_t>(f.iteration_count) + 1 ||
+      !last.residual_pass || !last.correction_pass || !last.update_pass ||
+      !last.zero_scale_exact || last.residual_metric > 1e-10 ||
+      last.update_metric > 1e-12 || last.weighted_equivalence_metric > 1e-12 ||
+      std::any_of(last.correction_metrics.begin(),
+                  last.correction_metrics.end(),
+                  [](double x) { return x > 1e-10; }))
+    return Failure(
+        "Candidate does not carry complete accepted convergence evidence.");
+  accepted_frames_.push_back(f);
+  return Status::Ok();
+}
+
+Result<SolidDynamicAcceptedFrameState>
+SolidDynamicAnalysisState::BeginTrialFromCommitted() const {
+  if (accepted_frames_.empty())
+    return Result<SolidDynamicAcceptedFrameState>::Failure(
+        Failure("No accepted Solid endpoint exists."));
+  return Result<SolidDynamicAcceptedFrameState>::Success(
+      accepted_frames_.back());
+}
+const std::vector<SolidDynamicAcceptedFrameState>&
+SolidDynamicAnalysisState::AcceptedFrames() const noexcept {
+  return accepted_frames_;
+}
+const SolidDynamicAcceptedFrameState&
+SolidDynamicAnalysisState::CommittedState() const {
+  if (accepted_frames_.empty())
+    throw std::logic_error("No accepted Solid endpoint.");
+  return accepted_frames_.back();
+}
+bool SolidDynamicAnalysisState::IsInitialized() const noexcept {
+  return initialized_;
+}
+void SolidDynamicAnalysisState::RecordIteration(
+    SolidDynamicIterationRecord record) {
+  iteration_records_.push_back(std::move(record));
+}
+const std::vector<SolidDynamicIterationRecord>&
+SolidDynamicAnalysisState::IterationRecords() const noexcept {
+  return iteration_records_;
+}
+}  // namespace fesa

src/fesa/analysis/solid_hht_dynamic_procedure.cpp

@@ -0,0 +1,870 @@
+#include "fesa/analysis/solid_hht_dynamic_procedure.h"
+
+#include <algorithm>
+#include <array>
+#include <cmath>
+#include <initializer_list>
+#include <limits>
+#include <memory>
+#include <optional>
+#include <string>
+#include <utility>
+#include <vector>
+
+#include "fesa/analysis/hht_time_integration.h"
+#include "fesa/assembly/load_assembler.h"
+#include "fesa/assembly/parallel_for.h"
+#include "fesa/constraints/essential_constraint_policy.h"
+#include "fesa/elements/element_factory.h"
+#include "fesa/elements/hex8_element.h"
+#include "fesa/math/sparse_matrix.h"
+#include "fesa/solvers/linear/linear_solver.h"
+#include "fesa/solvers/linear/tangent_solver.h"
+#include "solid_hht_force_identity.h"
+#include "solid_hht_high_low.h"
+
+namespace fesa {
+namespace {
+using math_internal::Abs;
+using math_internal::Add;
+using math_internal::Compare;
+using math_internal::Divide;
+using math_internal::HighLow;
+using math_internal::IsFinite;
+using math_internal::IsNormalized;
+using math_internal::Multiply;
+using math_internal::Negate;
+using math_internal::Rounded;
+using math_internal::Subtract;
+
+HighLow Pair(const Vector& high, const Vector& low, std::size_t i) {
+  return {high[i], low[i]};
+}
+void Store(HighLow value, double* high, double* low) {
+  *high = value.high;
+  *low = value.low;
+}
+void Store(HighLow value, Vector* high, Vector* low, std::size_t i) {
+  Store(value, &(*high)[i], &(*low)[i]);
+}
+bool Valid(HighLow value) { return IsFinite(value) && IsNormalized(value); }
+HighLow Maximum(HighLow left, HighLow right) {
+  if (!Valid(left) || !Valid(right)) return math_internal::InvalidHighLow();
+  return Compare(left, right) >= 0 ? left : right;
+}
+void InitializeLowVectors(SolidDynamicAcceptedFrameState* f, std::size_t size) {
+  for (auto* v : {&f->low.displacement, &f->low.velocity, &f->low.acceleration,
+                  &f->low.algorithmic_acceleration, &f->low.inertia_force,
+                  &f->low.internal_force, &f->low.external_force,
+                  &f->low.residual, &f->low.reaction, &f->low.residual_scale})
+    *v = Vector(size);
+}
+struct PairedTangent {
+  std::vector<HighLow> values;
+};
+/// @brief Reports source-context procedure failure without committing a trial.
+Status Failure(const Domain& domain, const char* code,
+               const std::string& message) {
+  return Status::Failure(
+      FailureCategory::kSolver,
+      {{Severity::kError, code,
+        domain.Dynamic() == nullptr ? SourceLocation{}
+                                    : domain.Dynamic()->dynamic_location,
+        "DYNAMIC",
+        domain.Dynamic() == nullptr ? ""
+                                    : domain.Dynamic()->source_step_name_raw,
+        message}});
+}
+Status CheckArithmeticEnvironment(const Domain& domain) {
+  return math_internal::ArithmeticEnvironmentSupported()
+             ? Status::Ok()
+             : Failure(domain, "solid-dynamic-arithmetic-environment",
+                       "Paired HHT requires nearest rounding and gradual "
+                       "underflow, including after backend callbacks.");
+}
+/// @brief Owns immutable kernels and the stable full-space translation scatter.
+struct RuntimeSolid {
+  std::unique_ptr<Element> owner;
+  const Hex8Element* solid;
+  std::vector<std::size_t> scatter;
+};
+/// @brief Creates checked kernels in declaration order, with no mutable element
+/// state.
+Result<std::vector<RuntimeSolid>> BuildSolids(const Domain& domain,
+                                              const DofManager& dofs) {
+  std::vector<RuntimeSolid> solids;
+  ElementFactory factory;
+  for (std::size_t e = 0; e < domain.Elements().Size(); ++e) {
+    auto owner = factory.Create(domain.Elements()[e], domain);
+    if (!owner.HasValue())
+      return Result<std::vector<RuntimeSolid>>::Failure(owner.GetStatus());
+    const auto* solid = dynamic_cast<const Hex8Element*>(owner.Value().get());
+    if (solid == nullptr)
+      return Result<std::vector<RuntimeSolid>>::Failure(
+          Failure(domain, "solid-dynamic-element-family",
+                  "HHT requires exclusively Hex8 kernels."));
+    auto scatter = dofs.ElementScatter(solid->DofLayout());
+    if (!scatter.HasValue())
+      return Result<std::vector<RuntimeSolid>>::Failure(scatter.GetStatus());
+    solids.push_back(
+        {std::move(owner.Value()), solid, std::move(scatter.Value())});
+  }
+  return Result<std::vector<RuntimeSolid>>::Success(std::move(solids));
+}
+/// @brief Preserves all 576 structural entries in stable element/local order.
+void AppendMatrix(const Matrix& local, const RuntimeSolid& solid, std::size_t e,
+                  std::vector<CooContribution>* coo) {
+  for (std::size_t i = 0; i < 24; ++i)
+    for (std::size_t j = 0; j < 24; ++j)
+      coo->push_back(
+          {solid.scatter[i], solid.scatter[j], local(i, j), e, 24 * i + j});
+}
+/// @brief Assembles constant reference consistent mass once without lumping.
+Result<SparseMatrix> AssembleMass(const std::vector<RuntimeSolid>& solids,
+                                  const DofManager& dofs) {
+  std::vector<CooContribution> coo;
+  coo.reserve(solids.size() * 576);
+  for (std::size_t e = 0; e < solids.size(); ++e) {
+    auto mass = solids[e].solid->ComputeMass();
+    if (!mass.HasValue())
+      return Result<SparseMatrix>::Failure(mass.GetStatus());
+    AppendMatrix(mass.Value().values, solids[e], e, &coo);
+  }
+  return SparseMatrix::FromCoo(dofs.FullDofCount(), dofs.FullDofCount(),
+                               std::move(coo), dofs.GetSparsePattern());
+}
+/// @brief Evaluates independent local trials, then deterministically reduces
+/// recovery.
+Result<PairedTangent> EvaluateSolids(const Domain& domain,
+                                     const std::vector<RuntimeSolid>& solids,
+                                     const DofManager& dofs,
+                                     const ParallelFor& parallel,
+                                     SolidDynamicAcceptedFrameState* f) {
+  const auto environment = CheckArithmeticEnvironment(domain);
+  if (!environment.IsOk()) return Result<PairedTangent>::Failure(environment);
+  if (!detail::ValidSolidHhtPair(f->displacement, f->low.displacement,
+                                 dofs.FullDofCount()))
+    return Result<PairedTangent>::Failure(
+        Failure(domain, "solid-dynamic-invalid-paired-state",
+                "Local evaluation requires complete normalized pairs."));
+  std::vector<std::optional<Result<Hex8HighLowResponse>>> responses(
+      solids.size());
+  parallel.Execute(solids.size(), [&](std::size_t e) {
+    Vector local(24), low(24);
+    for (std::size_t i = 0; i < 24; ++i) {
+      local[i] = f->displacement[solids[e].scatter[i]];
+      low[i] = f->low.displacement[solids[e].scatter[i]];
+    }
+    responses[e] = domain.Dynamic()->nlgeom
+                       ? solids[e].solid->EvaluateNonlinearHighLow(local, low)
+                       : solids[e].solid->EvaluateLinearHighLow(local, low);
+  });
+  const auto reduction_environment = CheckArithmeticEnvironment(domain);
+  if (!reduction_environment.IsOk())
+    return Result<PairedTangent>::Failure(reduction_environment);
+  const auto& pattern = dofs.GetSparsePattern();
+  PairedTangent tangent;
+  tangent.values.resize(pattern.column_indices.size());
+  f->internal_force = Vector(dofs.FullDofCount());
+  f->low.internal_force = Vector(dofs.FullDofCount());
+  f->strain_energy = 0;
+  f->low.strain_energy = 0;
+  f->solid_results.clear();
+  for (std::size_t e = 0; e < solids.size(); ++e) {
+    if (!responses[e])
+      return Result<PairedTangent>::Failure(
+          Failure(domain, "solid-dynamic-local-execution",
+                  "Local worker did not evaluate its index."));
+    if (!responses[e]->HasValue())
+      return Result<PairedTangent>::Failure(responses[e]->GetStatus());
+    const auto& r = responses[e]->Value();
+    for (std::size_t i = 0; i < 24; ++i) {
+      const auto row = solids[e].scatter[i];
+      const auto force =
+          Add(Pair(f->internal_force, f->low.internal_force, row),
+              Pair(r.high.internal_force, r.internal_force_low, i));
+      if (!Valid(force))
+        return Result<PairedTangent>::Failure(
+            Failure(domain, "solid-dynamic-nonfinite-assembly",
+                    "Paired internal force reduction failed."));
+      Store(force, &f->internal_force, &f->low.internal_force, row);
+      for (std::size_t j = 0; j < 24; ++j) {
+        const auto column = solids[e].scatter[j];
+        const auto begin =
+            pattern.column_indices.begin() + pattern.row_offsets[row];
+        const auto end =
+            pattern.column_indices.begin() + pattern.row_offsets[row + 1];
+        const auto found = std::lower_bound(begin, end, column);
+        if (found == end || *found != column)
+          return Result<PairedTangent>::Failure(
+              Failure(domain, "solid-dynamic-sparse-pattern",
+                      "Local tangent is outside the fixed sparse pattern."));
+        const auto k =
+            static_cast<std::size_t>(found - pattern.column_indices.begin());
+        tangent.values[k] =
+            Add(tangent.values[k],
+                HighLow(r.high.tangent(i, j), r.tangent_low(i, j)));
+        if (!Valid(tangent.values[k]))
+          return Result<PairedTangent>::Failure(
+              Failure(domain, "solid-dynamic-nonfinite-assembly",
+                      "Paired tangent reduction failed."));
+      }
+    }
+    const auto energy = Add(HighLow(f->strain_energy, f->low.strain_energy),
+                            HighLow(r.high.strain_energy, r.strain_energy_low));
+    if (!Valid(energy))
+      return Result<PairedTangent>::Failure(
+          Failure(domain, "solid-dynamic-nonfinite-assembly",
+                  "Paired strain energy reduction failed."));
+    Store(energy, &f->strain_energy, &f->low.strain_energy);
+    f->solid_results.insert(f->solid_results.end(), r.high.points.begin(),
+                            r.high.points.end());
+  }
+  return Result<PairedTangent>::Success(std::move(tangent));
+}
+/// @brief Returns the finite positive reference nodal bounding-box diagonal.
+Result<double> ReferenceLength(const Domain& domain) {
+  auto lo = domain.Nodes().front().coordinates;
+  auto hi = lo;
+  for (const auto& node : domain.Nodes())
+    for (std::size_t c = 0; c < 3; ++c) {
+      lo[c] = (std::min)(lo[c], node.coordinates[c]);
+      hi[c] = (std::max)(hi[c], node.coordinates[c]);
+    }
+  const double length =
+      std::hypot(std::hypot(hi[0] - lo[0], hi[1] - lo[1]), hi[2] - lo[2]);
+  if (!std::isfinite(length) || length <= 0)
+    return Result<double>::Failure(
+        Failure(domain, "solid-dynamic-reference-length",
+                "Reference length is not finite positive."));
+  return Result<double>::Success(length);
+}
+/// @brief Forms M/(beta*h²)+(1+alpha)*Kt in the unchanged full sparse pattern.
+Result<SparseMatrix> EffectiveTangent(const SparseMatrix& m,
+                                      const PairedTangent& k,
+                                      const DofManager& dofs, double h,
+                                      const HhtParameters& p) {
+  std::vector<CooContribution> coo;
+  coo.reserve(m.Values().size());
+  const auto denominator =
+      Multiply(Multiply(HighLow(p.beta), HighLow(h)), HighLow(h));
+  const auto coefficient = Add(HighLow(1), HighLow(p.alpha));
+  for (std::size_t i = 0; i < m.Rows(); ++i)
+    for (std::size_t j = m.RowOffsets()[i]; j < m.RowOffsets()[i + 1]; ++j) {
+      const auto value = Add(Divide(HighLow(m.Values()[j]), denominator),
+                             Multiply(coefficient, k.values[j]));
+      if (!Valid(value))
+        return Result<SparseMatrix>::Failure(
+            detail::SolidHhtPairedKinematicsFailure());
+      coo.push_back({i, m.ColumnIndices()[j], Rounded(value), 0, j});
+    }
+  return SparseMatrix::FromCoo(m.Rows(), m.Columns(), std::move(coo),
+                               dofs.GetSparsePattern());
+}
+
+using Paired3 = std::array<HighLow, 3>;
+void CrossAdd(const Paired3& x, const Paired3& force, Paired3* sum) {
+  for (std::size_t c = 0; c < 3; ++c) {
+    const auto j = (c + 1) % 3;
+    const auto k = (c + 2) % 3;
+    (*sum)[c] = Add((*sum)[c], Subtract(Multiply(x[j], force[k]),
+                                        Multiply(x[k], force[j])));
+  }
+}
+bool Store3(const Paired3& values, std::array<double, 3>* high,
+            std::array<double, 3>* low) {
+  for (std::size_t c = 0; c < 3; ++c) {
+    if (!Valid(values[c])) return false;
+    Store(values[c], &(*high)[c], &(*low)[c]);
+  }
+  return true;
+}
+
+/// @brief Recovers full physical endpoint quantities before any publication or
+/// commit. The constant Float64 mass coefficients are lifted exactly once.
+Status RecoverEndpoint(const Domain& domain, const DofManager& dofs,
+                       const SparseMatrix& mass,
+                       const SolidDynamicAcceptedFrameState* old,
+                       HighLow initial_energy,
+                       SolidDynamicAcceptedFrameState* f,
+                       SolidDynamicIterationEvidence* e) {
+  auto inertia = detail::MultiplySolidHhtMassHighLow(mass, f->acceleration,
+                                                     f->low.acceleration);
+  auto momentum =
+      detail::MultiplySolidHhtMassHighLow(mass, f->velocity, f->low.velocity);
+  if (!inertia.HasValue()) return inertia.GetStatus();
+  if (!momentum.HasValue()) return momentum.GetStatus();
+  f->inertia_force = std::move(inertia.Value().high);
+  f->low.inertia_force = std::move(inertia.Value().low);
+  const auto size = dofs.FullDofCount();
+  for (auto* v : {&f->residual, &f->low.residual, &f->reaction,
+                  &f->low.reaction, &f->residual_scale, &f->low.residual_scale})
+    *v = Vector(size);
+  HighLow numerator, scale_maximum, metric;
+  for (std::size_t i = 0; i < size; ++i) {
+    const auto ma = Pair(f->inertia_force, f->low.inertia_force, i);
+    const auto internal = Pair(f->internal_force, f->low.internal_force, i);
+    const auto external = Pair(f->external_force, f->low.external_force, i);
+    const auto residual = Subtract(Add(ma, internal), external);
+    const auto scale = Add(Add(Abs(ma), Abs(internal)), Abs(external));
+    if (!Valid(residual) || !Valid(scale))
+      return Failure(domain, "solid-dynamic-nonfinite-residual",
+                     "Physical paired force arithmetic failed.");
+    Store(residual, &f->residual, &f->low.residual, i);
+    Store(scale, &f->residual_scale, &f->low.residual_scale, i);
+  }
+  for (auto i : dofs.ConstrainedDofs()) {
+    f->reaction[i] = f->residual[i];
+    f->low.reaction[i] = f->low.residual[i];
+  }
+  for (auto i : dofs.FreeDofs()) {
+    const auto residual = Pair(f->residual, f->low.residual, i);
+    const auto scale = Pair(f->residual_scale, f->low.residual_scale, i);
+    numerator = Maximum(numerator, Abs(residual));
+    scale_maximum = Maximum(scale_maximum, scale);
+    if (Compare(scale, HighLow(0)) == 0) {
+      e->zero_scale_exact =
+          e->zero_scale_exact && Compare(residual, HighLow(0)) == 0;
+    } else {
+      const auto ratio = Divide(Abs(residual), scale);
+      if (!Valid(ratio))
+        return Failure(domain, "solid-dynamic-nonfinite-residual",
+                       "Physical paired force ratio is unsupported.");
+      if (Compare(ratio, metric) > 0) {
+        metric = ratio;
+        e->worst_full_dof = i;
+        e->worst_force_terms = {
+            Rounded(residual),
+            Rounded(Pair(f->inertia_force, f->low.inertia_force, i)),
+            Rounded(Pair(f->internal_force, f->low.internal_force, i)),
+            Rounded(Pair(f->external_force, f->low.external_force, i)),
+            Rounded(scale)};
+      }
+    }
+  }
+  if (!Valid(numerator) || !Valid(scale_maximum) || !Valid(metric))
+    return Failure(domain, "solid-dynamic-nonfinite-residual",
+                   "Physical paired evidence is nonfinite.");
+  e->residual_numerator = Rounded(numerator);
+  e->residual_scale = Rounded(scale_maximum);
+  e->residual_metric = Rounded(metric);
+  e->residual_pass =
+      e->zero_scale_exact && Compare(metric, HighLow(1e-10)) <= 0;
+
+  HighLow kinetic;
+  HighLow work = old == nullptr
+                     ? HighLow(0)
+                     : HighLow(old->external_work, old->low.external_work);
+  for (std::size_t i = 0; i < size; ++i) {
+    kinetic =
+        Add(kinetic,
+            Multiply(Pair(f->velocity, f->low.velocity, i),
+                     Pair(momentum.Value().high, momentum.Value().low, i)));
+    if (old != nullptr) {
+      const auto average =
+          Multiply(HighLow(.5),
+                   Add(Pair(old->external_force, old->low.external_force, i),
+                       Pair(f->external_force, f->low.external_force, i)));
+      const auto delta =
+          Subtract(Pair(f->displacement, f->low.displacement, i),
+                   Pair(old->displacement, old->low.displacement, i));
+      work = Add(work, Multiply(average, delta));
+    }
+  }
+  kinetic = Multiply(HighLow(.5), kinetic);
+  const auto error = Subtract(
+      Subtract(Add(kinetic, HighLow(f->strain_energy, f->low.strain_energy)),
+               initial_energy),
+      work);
+  if (!Valid(kinetic) || !Valid(work) || !Valid(error))
+    return Failure(domain, "solid-dynamic-nonfinite-energy",
+                   "Paired energy or endpoint work is unsupported.");
+  Store(kinetic, &f->kinetic_energy, &f->low.kinetic_energy);
+  Store(work, &f->external_work, &f->low.external_work);
+  Store(error, &f->energy_balance_error, &f->low.energy_balance_error);
+
+  Paired3 linear{}, angular{}, linear_rate{}, angular_rate{}, applied_moment{},
+      linear_equilibrium{}, current_internal_moment{}, reference_moment{},
+      free_residual_moment{};
+  for (std::size_t n = 0; n < domain.Nodes().size(); ++n) {
+    Paired3 x{}, reference{}, mv{}, ma{}, applied{}, internal{},
+        free_residual{}, u{};
+    for (std::size_t c = 0; c < 3; ++c) {
+      const auto i = 3 * n + c;
+      reference[c] = HighLow(domain.Nodes()[n].coordinates[c]);
+      u[c] = Pair(f->displacement, f->low.displacement, i);
+      x[c] = Add(reference[c], u[c]);
+      mv[c] = Pair(momentum.Value().high, momentum.Value().low, i);
+      ma[c] = Pair(f->inertia_force, f->low.inertia_force, i);
+      internal[c] = Pair(f->internal_force, f->low.internal_force, i);
+      applied[c] = Add(Pair(f->external_force, f->low.external_force, i),
+                       Pair(f->reaction, f->low.reaction, i));
+      free_residual[c] = Subtract(Pair(f->residual, f->low.residual, i),
+                                  Pair(f->reaction, f->low.reaction, i));
+      linear[c] = Add(linear[c], mv[c]);
+      linear_rate[c] = Add(linear_rate[c], ma[c]);
+      linear_equilibrium[c] =
+          Add(linear_equilibrium[c], Subtract(applied[c], ma[c]));
+    }
+    CrossAdd(x, mv, &angular);
+    CrossAdd(x, ma, &angular_rate);
+    CrossAdd(x, applied, &applied_moment);
+    CrossAdd(u, internal, &current_internal_moment);
+    CrossAdd(reference, internal, &reference_moment);
+    CrossAdd(x, free_residual, &free_residual_moment);
+  }
+  if (!Store3(linear, &f->linear_momentum, &f->low.linear_momentum) ||
+      !Store3(angular, &f->angular_momentum, &f->low.angular_momentum) ||
+      !Store3(linear_rate, &f->linear_momentum_rate,
+              &f->low.linear_momentum_rate) ||
+      !Store3(angular_rate, &f->angular_momentum_rate,
+              &f->low.angular_momentum_rate) ||
+      !Store3(current_internal_moment,
+              &f->linearization_current_internal_moment,
+              &f->low.linearization_current_internal_moment) ||
+      !Store3(reference_moment, &f->reference_internal_moment,
+              &f->low.reference_internal_moment) ||
+      !Store3(free_residual_moment, &f->current_free_residual_moment,
+              &f->low.current_free_residual_moment))
+    return Failure(
+        domain, "solid-dynamic-nonfinite-momentum",
+        "Paired momentum or current/reference moment is unsupported.");
+  for (std::size_t c = 0; c < 3; ++c) {
+    const auto moment = Subtract(applied_moment[c], angular_rate[c]);
+    if (!Valid(linear_equilibrium[c]) || !Valid(moment))
+      return Failure(domain, "solid-dynamic-nonfinite-equilibrium",
+                     "Paired global equilibrium is unsupported.");
+    Store(linear_equilibrium[c], &f->global_equilibrium[c],
+          &f->low.global_equilibrium[c]);
+    Store(moment, &f->global_equilibrium[3 + c],
+          &f->low.global_equilibrium[3 + c]);
+  }
+  return Status::Ok();
+}
+
+/// @brief Uses full paired same-unit uncancelled formula terms.
+HighLow UpdateMetric(HighLow left, std::initializer_list<HighLow> terms) {
+  HighLow sum;
+  auto scale = Abs(left);
+  for (const auto x : terms) {
+    sum = Add(sum, x);
+    scale = Add(scale, Abs(x));
+  }
+  const auto defect = Abs(Subtract(left, sum));
+  if (!Valid(defect) || !Valid(scale)) return math_internal::InvalidHighLow();
+  if (Compare(scale, HighLow(0)) == 0)
+    return Compare(defect, HighLow(0)) == 0 ? HighLow(0)
+                                            : math_internal::InvalidHighLow();
+  return Divide(defect, scale);
+}
+HighLow CheckUpdates(const SolidDynamicAcceptedFrameState& old,
+                     const SolidDynamicAcceptedFrameState& f,
+                     const HhtParameters& p, double h) {
+  HighLow metric;
+  const auto step = HighLow(h);
+  const auto hh = Multiply(step, step);
+  for (std::size_t i = 0; i < f.displacement.Size(); ++i) {
+    const auto A =
+        Pair(f.algorithmic_acceleration, f.low.algorithmic_acceleration, i);
+    const auto oldA =
+        Pair(old.algorithmic_acceleration, old.low.algorithmic_acceleration, i);
+    const auto oldv = Pair(old.velocity, old.low.velocity, i);
+    metric = Maximum(
+        metric,
+        UpdateMetric(
+            Pair(f.displacement, f.low.displacement, i),
+            {Pair(old.displacement, old.low.displacement, i),
+             Multiply(step, oldv),
+             Multiply(Multiply(hh, Subtract(HighLow(.5), HighLow(p.beta))),
+                      oldA),
+             Multiply(Multiply(hh, HighLow(p.beta)), A)}));
+    metric = Maximum(
+        metric,
+        UpdateMetric(
+            Pair(f.velocity, f.low.velocity, i),
+            {oldv,
+             Multiply(Multiply(step, Subtract(HighLow(1), HighLow(p.gamma))),
+                      oldA),
+             Multiply(Multiply(step, HighLow(p.gamma)), A)}));
+    metric = Maximum(
+        metric,
+        UpdateMetric(A, {Multiply(Add(HighLow(1), HighLow(p.alpha)),
+                                  Pair(f.acceleration, f.low.acceleration, i)),
+                         Negate(Multiply(HighLow(p.alpha),
+                                         Pair(old.acceleration,
+                                              old.low.acceleration, i)))}));
+  }
+  return metric;
+}
+std::array<HighLow, 4> FreshCorrectionMetrics(const Vector& du, double length,
+                                              const HhtParameters& p) {
+  HighLow q;
+  for (std::size_t i = 0; i < du.Size(); ++i)
+    q = Maximum(q, Divide(Abs(HighLow(du[i])), HighLow(length)));
+  return {
+      q, Divide(Multiply(HighLow(p.gamma), q), HighLow(p.beta)),
+      Divide(q, Multiply(Add(HighLow(1), HighLow(p.alpha)), HighLow(p.beta))),
+      Divide(q, HighLow(p.beta))};
+}
+std::array<HighLow, 4> ObservedCorrectionMetrics(
+    const SolidDynamicAcceptedFrameState& before,
+    const SolidDynamicAcceptedFrameState& after, double h, double length) {
+  std::array<HighLow, 4> metric{};
+  const auto hh = Multiply(HighLow(h), HighLow(h));
+  for (std::size_t i = 0; i < after.displacement.Size(); ++i) {
+    metric[0] = Maximum(
+        metric[0],
+        Divide(Abs(Subtract(
+                   Pair(after.displacement, after.low.displacement, i),
+                   Pair(before.displacement, before.low.displacement, i))),
+               HighLow(length)));
+    metric[1] = Maximum(
+        metric[1],
+        Divide(Multiply(HighLow(h),
+                        Abs(Subtract(
+                            Pair(after.velocity, after.low.velocity, i),
+                            Pair(before.velocity, before.low.velocity, i)))),
+               HighLow(length)));
+    metric[2] = Maximum(
+        metric[2],
+        Divide(Multiply(hh, Abs(Subtract(Pair(after.acceleration,
+                                              after.low.acceleration, i),
+                                         Pair(before.acceleration,
+                                              before.low.acceleration, i)))),
+               HighLow(length)));
+    metric[3] = Maximum(
+        metric[3],
+        Divide(Multiply(hh, Abs(Subtract(
+                                Pair(after.algorithmic_acceleration,
+                                     after.low.algorithmic_acceleration, i),
+                                Pair(before.algorithmic_acceleration,
+                                     before.low.algorithmic_acceleration, i)))),
+               HighLow(length)));
+  }
+  return metric;
+}
+bool CorrectionPass(const std::array<HighLow, 4>& metrics) {
+  return std::all_of(metrics.begin(), metrics.end(), [](HighLow x) {
+    return Valid(x) && Compare(x, HighLow(0)) >= 0 &&
+           Compare(x, HighLow(1e-10)) <= 0;
+  });
+}
+Status SetRates(const SolidDynamicAcceptedFrameState& old,
+                const detail::SolidHhtPairedPredictor& predictor, double h,
+                const HhtParameters& p,
+                SolidDynamicAcceptedFrameState* candidate) {
+  auto rates = detail::EvaluateSolidHhtHighLowRates(
+      candidate->displacement, candidate->low.displacement, predictor,
+      old.acceleration, old.low.acceleration, h, p);
+  if (!rates.HasValue()) return rates.GetStatus();
+  candidate->velocity = std::move(rates.Value().velocity);
+  candidate->low.velocity = std::move(rates.Value().velocity_low);
+  candidate->acceleration = std::move(rates.Value().acceleration);
+  candidate->low.acceleration = std::move(rates.Value().acceleration_low);
+  candidate->algorithmic_acceleration =
+      std::move(rates.Value().algorithmic_acceleration);
+  candidate->low.algorithmic_acceleration =
+      std::move(rates.Value().algorithmic_acceleration_low);
+  return Status::Ok();
+}
+}  // namespace
+
+SolidHhtDynamicProcedureExecutor::SolidHhtDynamicProcedureExecutor(
+    const ParallelFor& parallel_for, LinearSolver& mass_solver,
+    TangentSolver& tangent_solver)
+    : parallel_for_(parallel_for),
+      mass_solver_(mass_solver),
+      tangent_solver_(tangent_solver) {}
+
+Result<SolidDynamicExecutionResult> SolidHhtDynamicProcedureExecutor::Execute(
+    const Domain& domain) const {
+  auto model = AnalysisModel::Create(domain);
+  if (!model.HasValue())
+    return Result<SolidDynamicExecutionResult>::Failure(model.GetStatus());
+  auto dofs = DofManager::Create(model.Value());
+  if (!dofs.HasValue())
+    return Result<SolidDynamicExecutionResult>::Failure(dofs.GetStatus());
+  auto state = SolidDynamicAnalysisState::Create(domain, dofs.Value());
+  if (!state.HasValue())
+    return Result<SolidDynamicExecutionResult>::Failure(state.GetStatus());
+  auto status = ExecuteInto(domain, state.Value());
+  if (!status.IsOk())
+    return Result<SolidDynamicExecutionResult>::Failure(status);
+  return Result<SolidDynamicExecutionResult>::Success(
+      {std::move(dofs.Value()), std::move(state.Value())});
+}
+Status SolidHhtDynamicProcedureExecutor::ExecuteInto(
+    const Domain& domain, SolidDynamicAnalysisState& state) const {
+  return ExecuteIntoForVerification(domain, state, -0.05);
+}
+Status SolidHhtDynamicProcedureExecutor::ExecuteIntoForVerification(
+    const Domain& domain, SolidDynamicAnalysisState& state,
+    double alpha) const {
+  SolidDynamicIterationRecord record;
+  auto fail = [&](const Status& failure) {
+    record.stage = "failed";
+    record.outcome = "failed";
+    record.failure_code = failure.Diagnostics().empty()
+                              ? "solid-dynamic-failure"
+                              : failure.Diagnostics().front().code;
+    state.RecordIteration(record);
+    return failure;
+  };
+  if (!state.IsInitialized() || !state.AcceptedFrames().empty() ||
+      !domain.IsSolid() || domain.Dynamic() == nullptr)
+    return fail(Failure(
+        domain, "solid-dynamic-invalid-execution",
+        "Execution requires an initialized empty Solid dynamic state."));
+  auto environment = CheckArithmeticEnvironment(domain);
+  if (!environment.IsOk()) return fail(environment);
+  auto model = AnalysisModel::Create(domain);
+  if (!model.HasValue()) return fail(model.GetStatus());
+  auto dofs_result = DofManager::Create(model.Value());
+  if (!dofs_result.HasValue()) return fail(dofs_result.GetStatus());
+  const auto& dofs = dofs_result.Value();
+  auto solids = BuildSolids(domain, dofs);
+  if (!solids.HasValue()) return fail(solids.GetStatus());
+  auto mass = AssembleMass(solids.Value(), dofs);
+  if (!mass.HasValue()) return fail(mass.GetStatus());
+  EssentialConstraintPolicy partition;
+  auto mass_partition = partition.Partition(mass.Value(), dofs);
+  if (!mass_partition.HasValue()) return fail(mass_partition.GetStatus());
+  if (dofs.FreeDofCount() != 0) {
+    const auto status = mass_solver_.Factorize(mass_partition.Value().k_ff);
+    if (!status.IsOk()) return fail(status);
+    environment = CheckArithmeticEnvironment(domain);
+    if (!environment.IsOk()) return fail(environment);
+  }
+  auto parameters = MakeHhtParameters(alpha);
+  if (!parameters.HasValue()) return fail(parameters.GetStatus());
+  const auto p = parameters.Value();
+  auto length = ReferenceLength(domain);
+  if (!length.HasValue()) return fail(length.GetStatus());
+  const double h = domain.Dynamic()->initial_increment;
+  SolidDynamicAcceptedFrameState initial;
+  initial.is_initial = true;
+  for (std::size_t n = 0; n < domain.Nodes().size(); ++n) {
+    initial.nodal_internal_order.push_back(static_cast<EntityIndex>(n));
+    initial.nodal_source_ids.push_back(domain.Nodes()[n].source_id);
+  }
+  for (auto* v : {&initial.displacement, &initial.velocity,
+                  &initial.acceleration, &initial.algorithmic_acceleration})
+    *v = Vector(dofs.FullDofCount());
+  InitializeLowVectors(&initial, dofs.FullDofCount());
+  auto force = LoadAssembler::AssembleFullNodalLoad(model.Value(), dofs, 0);
+  if (!force.HasValue()) return fail(force.GetStatus());
+  initial.external_force = std::move(force.Value());
+  auto tangent =
+      EvaluateSolids(domain, solids.Value(), dofs, parallel_for_, &initial);
+  if (!tangent.HasValue()) return fail(tangent.GetStatus());
+  SolidDynamicIterationEvidence evidence;
+  bool initial_converged = false;
+  for (std::uint32_t refinement = 0; refinement <= 100; ++refinement) {
+    initial.algorithmic_acceleration = initial.acceleration;
+    initial.low.algorithmic_acceleration = initial.low.acceleration;
+    evidence = {};
+    evidence.correction_pass = evidence.update_pass = true;
+    auto status = RecoverEndpoint(domain, dofs, mass.Value(), nullptr,
+                                  HighLow(0), &initial, &evidence);
+    if (!status.IsOk()) return fail(status);
+    if (evidence.residual_pass) {
+      initial_converged = true;
+      break;
+    }
+    record.stage = "initial-mass";
+    record.outcome = "residual-refinement";
+    record.evidence = evidence;
+    record.evidence.iteration_index = refinement;
+    state.RecordIteration(record);
+    if (refinement == 100) break;
+    Vector rhs(dofs.FreeDofCount()), a(dofs.FreeDofCount());
+    for (std::size_t j = 0; j < dofs.FreeDofCount(); ++j) {
+      const auto i = dofs.FreeDofs()[j];
+      rhs[j] = Rounded(Negate(Pair(initial.residual, initial.low.residual, i)));
+    }
+    status = mass_solver_.Solve(rhs, a);
+    if (!status.IsOk()) return fail(status);
+    environment = CheckArithmeticEnvironment(domain);
+    if (!environment.IsOk()) return fail(environment);
+    if (a.Size() != dofs.FreeDofCount())
+      return fail(
+          Failure(domain, "solid-dynamic-invalid-backend-result",
+                  "Initial mass solve returned incompatible acceleration."));
+    for (std::size_t j = 0; j < a.Size(); ++j) {
+      if (!std::isfinite(a[j]))
+        return fail(
+            Failure(domain, "solid-dynamic-invalid-backend-result",
+                    "Initial mass solve returned nonfinite acceleration."));
+      const auto i = dofs.FreeDofs()[j];
+      const auto value =
+          Add(Pair(initial.acceleration, initial.low.acceleration, i),
+              HighLow(a[j] == 0 ? 0.0 : a[j]));
+      if (!Valid(value)) return fail(detail::SolidHhtPairedKinematicsFailure());
+      Store(value, &initial.acceleration, &initial.low.acceleration, i);
+    }
+  }
+  if (!initial_converged)
+    return fail(Failure(domain, "solid-dynamic-initial-mass-nonconvergence",
+                        "Physical initial equilibrium did not converge in100 "
+                        "mass refinements."));
+  initial.iterations.push_back(evidence);
+  auto status = state.CommitAcceptedFrame(initial);
+  if (!status.IsOk()) {
+    record.evidence = evidence;
+    return fail(status);
+  }
+  record.stage = "accepted";
+  record.outcome = "initial";
+  record.evidence = evidence;
+  state.RecordIteration(record);
+  const auto initial_energy =
+      Add(HighLow(initial.strain_energy, initial.low.strain_energy),
+          HighLow(initial.kinetic_energy, initial.low.kinetic_energy));
+  for (std::uint32_t increment = 1;
+       increment <= domain.Dynamic()->increment_count; ++increment) {
+    const auto& old = state.CommittedState();
+    record = {};
+    record.increment_index = increment;
+    record.time = static_cast<double>(increment) * h;
+    auto predictor = detail::PredictSolidHhtHighLow(old, h, p);
+    if (!predictor.HasValue()) return fail(predictor.GetStatus());
+    auto candidate = old;
+    candidate.frame_index = candidate.increment_index = increment;
+    candidate.iteration_count = 0;
+    candidate.time = record.time;
+    candidate.is_initial = false;
+    candidate.iterations.clear();
+    // Begin from certified accepted geometry, not an unchecked predictor.
+    force = LoadAssembler::AssembleFullNodalLoad(model.Value(), dofs,
+                                                 candidate.time);
+    if (!force.HasValue()) return fail(force.GetStatus());
+    candidate.external_force = std::move(force.Value());
+    candidate.low.external_force = Vector(dofs.FullDofCount());
+    std::array<HighLow, 4> correction{}, fresh_correction{};
+    bool accepted = false;
+    for (std::uint32_t iteration = 0; iteration <= 100; ++iteration) {
+      record.stage = "evaluation";
+      record.outcome = "paired-candidate";
+      record.evidence = {};
+      record.evidence.iteration_index = iteration;
+      status = SetRates(old, predictor.Value(), h, p, &candidate);
+      if (!status.IsOk()) return fail(status);
+      tangent = EvaluateSolids(domain, solids.Value(), dofs, parallel_for_,
+                               &candidate);
+      if (!tangent.HasValue()) return fail(tangent.GetStatus());
+      status = RecoverEndpoint(domain, dofs, mass.Value(), &old, initial_energy,
+                               &candidate, &record.evidence);
+      if (!status.IsOk()) return fail(status);
+      auto MA = detail::MultiplySolidHhtMassHighLow(
+          mass.Value(), candidate.algorithmic_acceleration,
+          candidate.low.algorithmic_acceleration);
+      if (!MA.HasValue()) return fail(MA.GetStatus());
+      Vector rhs(dofs.FreeDofCount());
+      HighLow identity_metric, original_norm, solved_norm, old_norm;
+      for (std::size_t j = 0; j < dofs.FreeDofCount(); ++j) {
+        const auto i = dofs.FreeDofs()[j];
+        detail::SolidHhtPairedForceIdentity identity;
+        if (!detail::EvaluateSolidHhtHighLowForceIdentity(
+                mass.Value(), i, Pair(MA.Value().high, MA.Value().low, i),
+                candidate, old, p.alpha, &identity))
+          return fail(
+              Failure(domain, "solid-dynamic-nonfinite-residual",
+                      "Paired weighted force arithmetic is unsupported."));
+        rhs[j] = Rounded(Negate(identity.solved));
+        identity_metric = Maximum(identity_metric, identity.metric);
+        original_norm = Maximum(original_norm, Abs(identity.original));
+        solved_norm = Maximum(solved_norm, Abs(identity.solved));
+        old_norm =
+            Maximum(old_norm, Abs(Pair(old.residual, old.low.residual, i)));
+        if (i == record.evidence.worst_full_dof)
+          record.evidence.worst_weighted_terms = {
+              Rounded(identity.original), Rounded(identity.solved),
+              Rounded(Pair(old.residual, old.low.residual, i)),
+              Rounded(identity.scale)};
+      }
+      if (!Valid(identity_metric) || !Valid(original_norm) ||
+          !Valid(solved_norm) || !Valid(old_norm) ||
+          Compare(identity_metric, HighLow(1e-12)) > 0)
+        return fail(Failure(domain,
+                            "solid-dynamic-weighted-algebra-inconsistency",
+                            "Paired original and solved forces disagree beyond "
+                            "their uncancelled scale."));
+      record.evidence.weighted_equivalence_metric = Rounded(identity_metric);
+      record.evidence.weighted_residual_numerator = Rounded(original_norm);
+      record.evidence.recombined_residual_numerator = Rounded(solved_norm);
+      record.evidence.old_residual_numerator = Rounded(old_norm);
+      for (std::size_t c = 0; c < 4; ++c) {
+        if (!Valid(correction[c]) || !Valid(fresh_correction[c]))
+          return fail(detail::SolidHhtPairedKinematicsFailure());
+        record.evidence.correction_metrics[c] = Rounded(correction[c]);
+      }
+      record.evidence.correction_pass =
+          CorrectionPass(correction) && CorrectionPass(fresh_correction);
+      const auto update = CheckUpdates(old, candidate, p, h);
+      if (!Valid(update))
+        return fail(detail::SolidHhtPairedKinematicsFailure());
+      record.evidence.update_metric = Rounded(update);
+      record.evidence.update_pass = Compare(update, HighLow(1e-12)) <= 0;
+      candidate.iterations.push_back(record.evidence);
+      candidate.iteration_count = iteration;
+      if (record.evidence.residual_pass && record.evidence.correction_pass &&
+          record.evidence.update_pass) {
+        status = state.CommitAcceptedFrame(candidate);
+        if (!status.IsOk()) return fail(status);
+        record.stage = "accepted";
+        record.outcome = "increment";
+        state.RecordIteration(record);
+        accepted = true;
+        break;
+      }
+      state.RecordIteration(record);
+      if (iteration == 100)
+        return fail(Failure(
+            domain, "solid-dynamic-nonconvergence",
+            "Paired HHT endpoint did not converge in100 Newton solves."));
+      auto effective =
+          EffectiveTangent(mass.Value(), tangent.Value(), dofs, h, p);
+      if (!effective.HasValue()) return fail(effective.GetStatus());
+      auto reduced = partition.Partition(effective.Value(), dofs);
+      if (!reduced.HasValue()) return fail(reduced.GetStatus());
+      Vector du(dofs.FreeDofCount());
+      if (dofs.FreeDofCount() != 0) {
+        status = tangent_solver_.FactorizeFresh(reduced.Value().k_ff);
+        if (!status.IsOk()) return fail(status);
+        environment = CheckArithmeticEnvironment(domain);
+        if (!environment.IsOk()) return fail(environment);
+        status = tangent_solver_.SolveCorrection(rhs, du);
+        if (!status.IsOk()) return fail(status);
+        environment = CheckArithmeticEnvironment(domain);
+        if (!environment.IsOk()) return fail(environment);
+      }
+      if (du.Size() != dofs.FreeDofCount())
+        return fail(Failure(domain, "solid-dynamic-invalid-backend-result",
+                            "Tangent solve returned incompatible correction."));
+      for (std::size_t j = 0; j < du.Size(); ++j)
+        if (!std::isfinite(du[j]))
+          return fail(Failure(domain, "solid-dynamic-invalid-backend-result",
+                              "Tangent solve returned nonfinite correction."));
+      fresh_correction = FreshCorrectionMetrics(du, length.Value(), p);
+      const auto before = candidate;
+      for (std::size_t j = 0; j < du.Size(); ++j) {
+        const auto i = dofs.FreeDofs()[j];
+        const auto value =
+            Add(Pair(candidate.displacement, candidate.low.displacement, i),
+                HighLow(du[j] == 0 ? 0.0 : du[j]));
+        if (!Valid(value))
+          return fail(detail::SolidHhtPairedKinematicsFailure());
+        Store(value, &candidate.displacement, &candidate.low.displacement, i);
+      }
+      status = SetRates(old, predictor.Value(), h, p, &candidate);
+      if (!status.IsOk()) return fail(status);
+      correction =
+          ObservedCorrectionMetrics(before, candidate, h, length.Value());
+    }
+    if (!accepted)
+      return fail(Failure(domain, "solid-dynamic-nonconvergence",
+                          "No complete paired endpoint was accepted."));
+  }
+  return Status::Ok();
+}
+}  // namespace fesa

src/fesa/analysis/solid_hht_force_identity.h

@@ -0,0 +1,182 @@
+#ifndef FESA_ANALYSIS_SOLID_HHT_FORCE_IDENTITY_H_
+#define FESA_ANALYSIS_SOLID_HHT_FORCE_IDENTITY_H_
+
+#include <cmath>
+#include <initializer_list>
+#include <limits>
+
+#include "fesa/analysis/solid_dynamic_analysis_state.h"
+#include "fesa/math/sparse_matrix.h"
+#include "solid_hht_high_low.h"
+
+namespace fesa::detail {
+struct SolidHhtForceIdentity {
+  double original{0};
+  double solved{0};
+  double scale{0};
+  double defect{0};
+  double metric{0};
+};
+
+struct SolidHhtPairedForceIdentity {
+  math_internal::HighLow original;
+  math_internal::HighLow solved;
+  math_internal::HighLow scale;
+  math_internal::HighLow defect;
+  math_internal::HighLow metric;
+};
+
+/// @brief Keeps both physical parts and the actual old residual in the approved
+/// weighted identity. Its scale includes each uncancelled mass product.
+inline bool EvaluateSolidHhtHighLowForceIdentity(
+    const SparseMatrix& mass, std::size_t row,
+    math_internal::HighLow algorithmic_inertia,
+    const SolidDynamicAcceptedFrameState& next,
+    const SolidDynamicAcceptedFrameState& old, double alpha,
+    SolidHhtPairedForceIdentity* result) {
+  using namespace math_internal;
+  if (!ArithmeticEnvironmentSupported() || row >= mass.Rows() ||
+      !IsFinite(algorithmic_inertia) || !IsNormalized(algorithmic_inertia) ||
+      !IsFinite(HighLow(alpha)))
+    return false;
+  for (const auto* frame : {&next, &old})
+    if (frame->acceleration.Size() != mass.Columns() ||
+        frame->low.acceleration.Size() != mass.Columns() ||
+        frame->algorithmic_acceleration.Size() != mass.Columns() ||
+        frame->low.algorithmic_acceleration.Size() != mass.Columns() ||
+        frame->internal_force.Size() != mass.Rows() ||
+        frame->low.internal_force.Size() != mass.Rows() ||
+        frame->external_force.Size() != mass.Rows() ||
+        frame->low.external_force.Size() != mass.Rows() ||
+        frame->residual.Size() != mass.Rows() ||
+        frame->low.residual.Size() != mass.Rows())
+      return false;
+  const HighLow a(alpha);
+  const auto c = Add(HighLow(1), a);
+  const HighLow f(next.internal_force[row], next.low.internal_force[row]);
+  const HighLow F(next.external_force[row], next.low.external_force[row]);
+  const HighLow old_f(old.internal_force[row], old.low.internal_force[row]);
+  const HighLow old_F(old.external_force[row], old.low.external_force[row]);
+  const HighLow r(next.residual[row], next.low.residual[row]);
+  const HighLow old_r(old.residual[row], old.low.residual[row]);
+  result->original =
+      Subtract(Add(algorithmic_inertia, Multiply(c, Subtract(f, F))),
+               Multiply(a, Subtract(old_f, old_F)));
+  result->solved = Subtract(Multiply(c, r), Multiply(a, old_r));
+  HighLow mass_scale;
+  HighLow relation_action;
+  for (std::size_t k = mass.RowOffsets()[row]; k < mass.RowOffsets()[row + 1];
+       ++k) {
+    const auto j = mass.ColumnIndices()[k];
+    const HighLow coefficient(mass.Values()[k]);
+    const HighLow A(next.algorithmic_acceleration[j],
+                    next.low.algorithmic_acceleration[j]);
+    const HighLow physical_a(next.acceleration[j], next.low.acceleration[j]);
+    const HighLow old_a(old.acceleration[j], old.low.acceleration[j]);
+    const auto relation =
+        Add(Subtract(A, Multiply(c, physical_a)), Multiply(a, old_a));
+    mass_scale =
+        Add(mass_scale,
+            Add(Add(Abs(Multiply(coefficient, A)),
+                    Multiply(c, Abs(Multiply(coefficient, physical_a)))),
+                Multiply(Abs(a), Abs(Multiply(coefficient, old_a)))));
+    relation_action =
+        Add(relation_action, Abs(Multiply(coefficient, relation)));
+    if (!IsFinite(mass_scale) || !IsFinite(relation_action)) return false;
+  }
+  result->scale = Add(
+      Add(Add(mass_scale, relation_action), Multiply(c, Add(Abs(f), Abs(F)))),
+      Add(Add(Multiply(Abs(a), Add(Abs(old_f), Abs(old_F))),
+              Multiply(c, Abs(r))),
+          Abs(Multiply(a, old_r))));
+  result->defect = Abs(Subtract(result->original, result->solved));
+  if (!IsFinite(result->original) || !IsFinite(result->solved) ||
+      !IsFinite(result->scale) || !IsFinite(result->defect))
+    return false;
+  if (Compare(result->scale, HighLow(0)) == 0) {
+    if (Compare(result->defect, HighLow(0)) != 0) return false;
+    result->metric = HighLow(0);
+  } else {
+    result->metric = Divide(result->defect, result->scale);
+  }
+  return IsFinite(result->metric) && IsNormalized(result->metric);
+}
+
+/// @brief Evaluates one original/equivalent HHT row, retaining old equilibrium
+/// error. Checks each scalar before it can enter an evidence maximum.
+inline bool EvaluateSolidHhtForceIdentity(
+    const SparseMatrix& mass, std::size_t row, double algorithmic_inertia,
+    const SolidDynamicAcceptedFrameState& next,
+    const SolidDynamicAcceptedFrameState& old, double alpha,
+    SolidHhtForceIdentity* result) {
+  for (double x :
+       {alpha, algorithmic_inertia, next.internal_force[row],
+        next.external_force[row], old.internal_force[row],
+        old.external_force[row], next.residual[row], old.residual[row]})
+    if (!std::isfinite(x)) return false;
+  result->original =
+      algorithmic_inertia +
+      (1 + alpha) * (next.internal_force[row] - next.external_force[row]) -
+      alpha * (old.internal_force[row] - old.external_force[row]);
+  result->solved = (1 + alpha) * next.residual[row] - alpha * old.residual[row];
+  double mass_action_scale = 0;
+  double relation_defect_action = 0;
+  for (std::size_t entry = mass.RowOffsets()[row];
+       entry < mass.RowOffsets()[row + 1]; ++entry) {
+    const auto column = mass.ColumnIndices()[entry];
+    const double coefficient = mass.Values()[entry];
+    const double a = next.acceleration[column];
+    const double A = next.algorithmic_acceleration[column];
+    const double a_old = old.acceleration[column];
+    const double relation_defect = A - (1 + alpha) * a + alpha * a_old;
+    const double action_scale = std::abs(coefficient * A) +
+                                (1 + alpha) * std::abs(coefficient * a) +
+                                std::abs(alpha) * std::abs(coefficient * a_old);
+    const double defect_action = std::abs(coefficient * relation_defect);
+    for (double x : {a, A, a_old, relation_defect, action_scale, defect_action})
+      if (!std::isfinite(x)) return false;
+    mass_action_scale += action_scale;
+    relation_defect_action += defect_action;
+  }
+  result->scale = mass_action_scale + relation_defect_action +
+                  (1 + alpha) * (std::abs(next.internal_force[row]) +
+                                 std::abs(next.external_force[row])) +
+                  std::abs(alpha) * (std::abs(old.internal_force[row]) +
+                                     std::abs(old.external_force[row])) +
+                  (1 + alpha) * std::abs(next.residual[row]) +
+                  std::abs(alpha * old.residual[row]);
+  result->defect = std::abs(result->original - result->solved);
+  for (double x : {result->original, result->solved, mass_action_scale,
+                   relation_defect_action, result->scale, result->defect})
+    if (!std::isfinite(x)) return false;
+  result->metric =
+      result->scale == 0
+          ? (result->defect == 0 ? 0 : std::numeric_limits<double>::infinity())
+          : result->defect / result->scale;
+  return std::isfinite(result->metric);
+}
+/// @brief Finite checked failure-shadow accumulation, separate from physics.
+inline bool AccumulateSolidHhtForceShadow(double term, double product_error,
+                                          double* sum, double* correction,
+                                          double* absolute_sum) {
+  if (!std::isfinite(term) || !std::isfinite(product_error) ||
+      !std::isfinite(*sum) || !std::isfinite(*correction) ||
+      !std::isfinite(*absolute_sum))
+    return false;
+  const double next = *sum + term;
+  const double absolute_next = *absolute_sum + std::abs(term);
+  const double error_next = *correction + product_error;
+  const double lost = std::abs(*sum) >= std::abs(term) ? (*sum - next) + term
+                                                       : (term - next) + *sum;
+  const double correction_next = error_next + lost;
+  if (!std::isfinite(next) || !std::isfinite(absolute_next) ||
+      !std::isfinite(error_next) || !std::isfinite(lost) ||
+      !std::isfinite(correction_next))
+    return false;
+  *sum = next;
+  *correction = correction_next;
+  *absolute_sum = absolute_next;
+  return true;
+}
+}  // namespace fesa::detail
+#endif  // FESA_ANALYSIS_SOLID_HHT_FORCE_IDENTITY_H_

src/fesa/analysis/solid_hht_high_low.h

@@ -0,0 +1,190 @@
+#ifndef FESA_ANALYSIS_SOLID_HHT_HIGH_LOW_H_
+#define FESA_ANALYSIS_SOLID_HHT_HIGH_LOW_H_
+
+#include "fesa/analysis/hht_time_integration.h"
+#include "fesa/analysis/solid_dynamic_analysis_state.h"
+#include "fesa/math/sparse_matrix.h"
+#include "math/high_low_arithmetic.h"
+
+namespace fesa::detail {
+/// @brief Keeps both physical parts of the fixed-increment predictor.
+struct SolidHhtPairedPredictor {
+  Vector displacement{0};
+  Vector displacement_low{0};
+  Vector velocity{0};
+  Vector velocity_low{0};
+};
+
+/// @brief Owns physical and auxiliary rates without collapsing either part.
+struct SolidHhtPairedRates {
+  Vector velocity{0};
+  Vector velocity_low{0};
+  Vector acceleration{0};
+  Vector acceleration_low{0};
+  Vector algorithmic_acceleration{0};
+  Vector algorithmic_acceleration_low{0};
+};
+
+struct SolidHhtPairedVector {
+  Vector high{0};
+  Vector low{0};
+};
+
+inline Status SolidHhtPairedKinematicsFailure() {
+  return Status::Failure(FailureCategory::kSolver,
+                         {{Severity::kError,
+                           "solid-dynamic-invalid-paired-kinematics",
+                           {},
+                           "DYNAMIC",
+                           {},
+                           "Incomplete, unnormalized or nonfinite paired HHT "
+                           "kinematics."}});
+}
+
+inline bool ValidSolidHhtPair(const Vector& high, const Vector& low,
+                              std::size_t size) {
+  if (high.Size() != size || low.Size() != size) return false;
+  for (std::size_t i = 0; i < size; ++i) {
+    const math_internal::HighLow value(high[i], low[i]);
+    if (!math_internal::IsFinite(value) || !math_internal::IsNormalized(value))
+      return false;
+  }
+  return true;
+}
+
+inline bool ValidSolidHhtParameters(double h, const HhtParameters& p) {
+  if (!math_internal::ArithmeticEnvironmentSupported()) return false;
+  const auto expected = MakeHhtParameters(p.alpha);
+  if (!expected.HasValue() || p.beta != expected.Value().beta ||
+      p.gamma != expected.Value().gamma)
+    return false;
+  const math_internal::HighLow step(h);
+  const auto square = math_internal::Multiply(step, step);
+  return math_internal::IsFinite(step) &&
+         math_internal::Compare(step, math_internal::HighLow(0)) > 0 &&
+         math_internal::IsFinite(square) &&
+         math_internal::Compare(square, math_internal::HighLow(0)) > 0;
+}
+
+/// @brief Uses the authoritative constant Float64 mass coefficients without
+/// collapsing either physical vector part or the ordered row accumulation.
+inline Result<SolidHhtPairedVector> MultiplySolidHhtMassHighLow(
+    const SparseMatrix& mass, const Vector& high, const Vector& low) {
+  using namespace math_internal;
+  if (!ArithmeticEnvironmentSupported() || mass.Rows() != mass.Columns() ||
+      !ValidSolidHhtPair(high, low, mass.Columns()))
+    return Result<SolidHhtPairedVector>::Failure(
+        SolidHhtPairedKinematicsFailure());
+  SolidHhtPairedVector result{Vector(mass.Rows()), Vector(mass.Rows())};
+  for (std::size_t row = 0; row < mass.Rows(); ++row) {
+    HighLow sum;
+    for (std::size_t k = mass.RowOffsets()[row]; k < mass.RowOffsets()[row + 1];
+         ++k) {
+      const auto col = mass.ColumnIndices()[k];
+      sum = Add(sum, Multiply(HighLow(mass.Values()[k]),
+                              HighLow(high[col], low[col])));
+      if (!IsFinite(sum) || !IsNormalized(sum))
+        return Result<SolidHhtPairedVector>::Failure(
+            SolidHhtPairedKinematicsFailure());
+    }
+    result.high[row] = sum.high;
+    result.low[row] = sum.low;
+  }
+  return Result<SolidHhtPairedVector>::Success(std::move(result));
+}
+
+/// @brief Applies the unchanged Newmark predictor to the full physical state.
+inline Result<SolidHhtPairedPredictor> PredictSolidHhtHighLow(
+    const SolidDynamicAcceptedFrameState& old, double h,
+    const HhtParameters& p) {
+  using namespace math_internal;
+  const auto size = old.displacement.Size();
+  if (!ValidSolidHhtPair(old.displacement, old.low.displacement, size) ||
+      !ValidSolidHhtPair(old.velocity, old.low.velocity, size) ||
+      !ValidSolidHhtPair(old.algorithmic_acceleration,
+                         old.low.algorithmic_acceleration, size) ||
+      !ValidSolidHhtParameters(h, p))
+    return Result<SolidHhtPairedPredictor>::Failure(
+        SolidHhtPairedKinematicsFailure());
+  SolidHhtPairedPredictor result{Vector(size), Vector(size), Vector(size),
+                                 Vector(size)};
+  const auto hh = Multiply(HighLow(h), HighLow(h));
+  const auto u_coefficient =
+      Multiply(hh, Subtract(HighLow(.5), HighLow(p.beta)));
+  const auto v_coefficient =
+      Multiply(HighLow(h), Subtract(HighLow(1), HighLow(p.gamma)));
+  if (!IsFinite(u_coefficient) || !IsFinite(v_coefficient))
+    return Result<SolidHhtPairedPredictor>::Failure(
+        SolidHhtPairedKinematicsFailure());
+  for (std::size_t i = 0; i < size; ++i) {
+    const HighLow u(old.displacement[i], old.low.displacement[i]);
+    const HighLow v(old.velocity[i], old.low.velocity[i]);
+    const HighLow A(old.algorithmic_acceleration[i],
+                    old.low.algorithmic_acceleration[i]);
+    const auto up =
+        Add(Add(u, Multiply(HighLow(h), v)), Multiply(u_coefficient, A));
+    const auto vp = Add(v, Multiply(v_coefficient, A));
+    if (!IsFinite(up) || !IsNormalized(up) || !IsFinite(vp) ||
+        !IsNormalized(vp))
+      return Result<SolidHhtPairedPredictor>::Failure(
+          SolidHhtPairedKinematicsFailure());
+    result.displacement[i] = up.high;
+    result.displacement_low[i] = up.low;
+    result.velocity[i] = vp.high;
+    result.velocity_low[i] = vp.low;
+  }
+  return Result<SolidHhtPairedPredictor>::Success(std::move(result));
+}
+
+/// @brief Recovers A, v and physical a from the paired displacement difference.
+inline Result<SolidHhtPairedRates> EvaluateSolidHhtHighLowRates(
+    const Vector& high, const Vector& low,
+    const SolidHhtPairedPredictor& predictor, const Vector& old_acceleration,
+    const Vector& old_acceleration_low, double h, const HhtParameters& p) {
+  using namespace math_internal;
+  const auto size = high.Size();
+  if (!ValidSolidHhtPair(high, low, size) ||
+      !ValidSolidHhtPair(predictor.displacement, predictor.displacement_low,
+                         size) ||
+      !ValidSolidHhtPair(predictor.velocity, predictor.velocity_low, size) ||
+      !ValidSolidHhtPair(old_acceleration, old_acceleration_low, size) ||
+      !ValidSolidHhtParameters(h, p))
+    return Result<SolidHhtPairedRates>::Failure(
+        SolidHhtPairedKinematicsFailure());
+  SolidHhtPairedRates result{Vector(size), Vector(size), Vector(size),
+                             Vector(size), Vector(size), Vector(size)};
+  const auto denominator =
+      Multiply(Multiply(HighLow(p.beta), HighLow(h)), HighLow(h));
+  const auto velocity_coefficient = Multiply(HighLow(p.gamma), HighLow(h));
+  const auto physical_coefficient = Add(HighLow(1), HighLow(p.alpha));
+  if (!IsFinite(denominator) || Compare(denominator, HighLow(0)) <= 0 ||
+      !IsFinite(velocity_coefficient) || !IsFinite(physical_coefficient))
+    return Result<SolidHhtPairedRates>::Failure(
+        SolidHhtPairedKinematicsFailure());
+  for (std::size_t i = 0; i < size; ++i) {
+    const auto A = Divide(Subtract(HighLow(high[i], low[i]),
+                                   HighLow(predictor.displacement[i],
+                                           predictor.displacement_low[i])),
+                          denominator);
+    const auto v =
+        Add(HighLow(predictor.velocity[i], predictor.velocity_low[i]),
+            Multiply(velocity_coefficient, A));
+    const auto a = Divide(
+        Add(A, Multiply(HighLow(p.alpha),
+                        HighLow(old_acceleration[i], old_acceleration_low[i]))),
+        physical_coefficient);
+    if (!IsFinite(A) || !IsNormalized(A) || !IsFinite(v) || !IsNormalized(v) ||
+        !IsFinite(a) || !IsNormalized(a))
+      return Result<SolidHhtPairedRates>::Failure(
+          SolidHhtPairedKinematicsFailure());
+    result.algorithmic_acceleration[i] = A.high;
+    result.algorithmic_acceleration_low[i] = A.low;
+    result.velocity[i] = v.high;
+    result.velocity_low[i] = v.low;
+    result.acceleration[i] = a.high;
+    result.acceleration_low[i] = a.low;
+  }
+  return Result<SolidHhtPairedRates>::Success(std::move(result));
+}
+}  // namespace fesa::detail
+#endif  // FESA_ANALYSIS_SOLID_HHT_HIGH_LOW_H_

src/fesa/analysis/solid_hht_representability.h

@@ -0,0 +1,217 @@
+#ifndef FESA_ANALYSIS_SOLID_HHT_REPRESENTABILITY_H_
+#define FESA_ANALYSIS_SOLID_HHT_REPRESENTABILITY_H_
+#include <algorithm>
+#include <array>
+#include <cmath>
+#include <limits>
+#include <vector>
+
+#include "fesa/analysis/solid_dynamic_analysis_state.h"
+#include "fesa/math/sparse_matrix.h"
+namespace fesa::detail {
+struct SolidHhtRepresentableNeighbor {
+  bool found{false};
+  std::size_t full_dof{0};
+  double displacement{0};
+  double delta{0};
+  double predicted_metric{0};
+};
+/// @brief Predicts a strictly improving one-ULP neighbor using separate
+/// physical inertia/internal-force actions and every original free score.
+/// This selects a trial only; callers must recompute actual endpoint gates.
+inline bool SelectSolidHhtRepresentableNeighbor(
+    const SparseMatrix& mass, const SparseMatrix& tangent,
+    const std::vector<std::size_t>& free_dofs,
+    const SolidDynamicAcceptedFrameState& frame, double h, double alpha,
+    double beta, SolidHhtRepresentableNeighbor* selected,
+    std::vector<SolidHhtRepresentableNeighbor>* ranked = nullptr) {
+  *selected = {};
+  if (ranked != nullptr) ranked->clear();
+  const auto size = mass.Rows();
+  if (tangent.Rows() != size || mass.Columns() != size ||
+      tangent.Columns() != size || mass.RowOffsets() != tangent.RowOffsets() ||
+      mass.ColumnIndices() != tangent.ColumnIndices() ||
+      frame.displacement.Size() != size || frame.inertia_force.Size() != size ||
+      frame.internal_force.Size() != size ||
+      frame.external_force.Size() != size || frame.residual.Size() != size ||
+      frame.residual_scale.Size() != size)
+    return false;
+  const double inertia_factor = 1 / ((1 + alpha) * beta * h * h);
+  if (!std::isfinite(inertia_factor) || inertia_factor <= 0) return false;
+  const auto score = [](double ma, double f, double F, double* metric) {
+    const double scale = std::abs(ma) + std::abs(f) + std::abs(F);
+    const double r = ma + f - F;
+    if (!std::isfinite(scale) || !std::isfinite(r)) return false;
+    *metric = scale == 0
+                  ? (r == 0 ? 0 : std::numeric_limits<double>::infinity())
+                  : std::abs(r) / scale;
+    return std::isfinite(*metric);
+  };
+  std::vector<bool> is_free(size, false);
+  std::vector<double> scores(size, 0);
+  double best = 0;
+  for (auto i : free_dofs) {
+    if (i >= size || !std::isfinite(frame.displacement[i]) ||
+        !score(frame.inertia_force[i], frame.internal_force[i],
+               frame.external_force[i], &scores[i]))
+      return false;
+    is_free[i] = true;
+    best = (std::max)(best, scores[i]);
+  }
+  struct Action {
+    std::size_t row;
+    double inertia;
+    double internal;
+  };
+  std::vector<std::vector<Action>> columns(size);
+  for (auto i : free_dofs)
+    for (std::size_t k = mass.RowOffsets()[i]; k < mass.RowOffsets()[i + 1];
+         ++k) {
+      const auto j = mass.ColumnIndices()[k];
+      if (!is_free[j]) continue;
+      const double inertia = mass.Values()[k] * inertia_factor;
+      const double internal = tangent.Values()[k];
+      if (!std::isfinite(inertia) || !std::isfinite(internal)) return false;
+      columns[j].push_back({i, inertia, internal});
+    }
+  const double original_metric = best;
+  auto predicted = scores;
+  for (auto j : free_dofs)
+    for (const double direction : {-std::numeric_limits<double>::infinity(),
+                                   std::numeric_limits<double>::infinity()}) {
+      const double proposed = std::nextafter(frame.displacement[j], direction);
+      const double delta = proposed - frame.displacement[j];
+      if (!std::isfinite(proposed) || !std::isfinite(delta) || delta == 0)
+        continue;
+      predicted = scores;
+      for (const auto& action : columns[j]) {
+        const auto i = action.row;
+        const double ma = frame.inertia_force[i] + action.inertia * delta;
+        const double f = frame.internal_force[i] + action.internal * delta;
+        if (!score(ma, f, frame.external_force[i], &predicted[i])) return false;
+      }
+      double metric = 0;
+      for (auto i : free_dofs) metric = (std::max)(metric, predicted[i]);
+      if (ranked != nullptr && metric < original_metric)
+        ranked->push_back({true, j, proposed, delta, metric});
+      // Stable full-DOF/direction ordering resolves ties without a floor.
+      if (metric < best) {
+        best = metric;
+        *selected = {true, j, proposed, delta, metric};
+      }
+    }
+  if (ranked != nullptr)
+    std::stable_sort(ranked->begin(), ranked->end(),
+                     [](const auto& a, const auto& b) {
+                       return a.predicted_metric < b.predicted_metric;
+                     });
+  return true;
+}
+/// @brief Keeps complete eligible failed Newton entries for a fixed increment.
+class SolidHhtNewtonHistory {
+ public:
+  bool Capture(const Vector& u, const std::vector<bool>& holding_mask,
+               const std::array<double, 4>& incoming_metrics,
+               std::uint32_t iteration, bool physical_pass) {
+    if (!Valid(u, holding_mask, incoming_metrics)) return false;
+    if (!physical_pass &&
+        std::all_of(incoming_metrics.begin(), incoming_metrics.end(),
+                    [](double metric) { return metric <= 1e-10; }))
+      entries_.push_back({u, holding_mask, incoming_metrics, iteration});
+    return true;
+  }
+  bool Match(const Vector& proposed_u,
+             const std::vector<bool>& next_holding_mask,
+             const std::array<double, 4>& fresh_metrics, bool physical_pass,
+             SolidDynamicNewtonRepeatEvidence* evidence) const {
+    *evidence = {};
+    evidence->history_entry_count = entries_.size();
+    if (!Valid(proposed_u, next_holding_mask, fresh_metrics)) return false;
+    // A physically converged prospective endpoint must reach the normal gate.
+    if (physical_pass ||
+        !std::all_of(fresh_metrics.begin(), fresh_metrics.end(),
+                     [](double metric) { return metric <= 1e-10; }))
+      return true;
+    for (const auto& entry : entries_) {
+      evidence->changed_translations = 0;
+      evidence->changed_holding_mask = 0;
+      evidence->changed_correction_metrics = 0;
+      for (std::size_t i = 0; i < proposed_u.Size(); ++i)
+        evidence->changed_translations += proposed_u[i] != entry.u[i];
+      for (std::size_t i = 0; i < next_holding_mask.size(); ++i)
+        evidence->changed_holding_mask +=
+            next_holding_mask[i] != entry.holding_mask[i];
+      for (std::size_t i = 0; i < 4; ++i)
+        evidence->changed_correction_metrics +=
+            fresh_metrics[i] != entry.incoming_metrics[i];
+      // Numeric correction magnitudes remain evidence. Their finite/PASS
+      // controls, rather than their exact magnitudes, affect the transition.
+      if (evidence->changed_translations == 0 &&
+          evidence->changed_holding_mask == 0) {
+        evidence->matched_entry_iteration = entry.iteration;
+        evidence->repeat = true;
+        break;
+      }
+    }
+    return true;
+  }
+  std::size_t Size() const { return entries_.size(); }
+
+ private:
+  struct Entry {
+    Vector u;
+    std::vector<bool> holding_mask;
+    std::array<double, 4> incoming_metrics;
+    std::uint32_t iteration;
+  };
+  bool Valid(const Vector& u, const std::vector<bool>& mask,
+             const std::array<double, 4>& metrics) const {
+    if (mask.size() > u.Size() ||
+        (!entries_.empty() &&
+         (u.Size() != entries_.front().u.Size() ||
+          mask.size() != entries_.front().holding_mask.size())))
+      return false;
+    for (std::size_t i = 0; i < u.Size(); ++i)
+      if (!std::isfinite(u[i])) return false;
+    for (double metric : metrics)
+      if (!std::isfinite(metric) || metric < 0) return false;
+    return true;
+  }
+  std::vector<Entry> entries_;
+};
+/// @brief Checks the fresh complete Newton trial before representable fallback.
+/// Historical scores are deliberately absent: a lower half of a two-cycle
+/// still needs actual proposed-step evaluation. False keeps common Newton
+/// apply.
+template <class Newton, class Neighbor>
+Result<bool> TrySolidHhtActualStep(
+    bool residual_pass, const std::array<double, 4>& fresh_corrections,
+    Newton newton, Neighbor neighbor) {
+  if (residual_pass || !std::all_of(fresh_corrections.begin(),
+                                    fresh_corrections.end(), [](double metric) {
+                                      return std::isfinite(metric) &&
+                                             metric <= 1e-10;
+                                    }))
+    return Result<bool>::Success(false);
+  auto keep_newton = newton();
+  if (!keep_newton.HasValue())
+    return Result<bool>::Failure(keep_newton.GetStatus());
+  if (keep_newton.Value()) return Result<bool>::Success(false);
+  return neighbor();
+}
+/// @brief Tries ranked predictions until actual endpoint evaluation improves.
+/// Evaluator failure is terminal; a rejected prediction does not discard later
+/// candidates, and success means only that a local trial improved.
+template <class Evaluate>
+Result<bool> TrySolidHhtRankedNeighbors(
+    const std::vector<SolidHhtRepresentableNeighbor>& ranked,
+    Evaluate evaluate) {
+  for (const auto& neighbor : ranked) {
+    auto actual = evaluate(neighbor);
+    if (!actual.HasValue()) return Result<bool>::Failure(actual.GetStatus());
+    if (actual.Value()) return Result<bool>::Success(true);
+  }
+  return Result<bool>::Success(false);
+}
+}  // namespace fesa::detail
+#endif  // FESA_ANALYSIS_SOLID_HHT_REPRESENTABILITY_H_

tests/CMakeLists.txt

@@ -19,6 +19,8 @@ add_executable(
     unit/analysis/nonlinear_static_control_test.cpp
     unit/analysis/newton_iteration_log_test.cpp
     unit/analysis/solid_frequency_procedure_test.cpp
+    unit/analysis/solid_hht_dynamic_procedure_test.cpp
+    unit/analysis/solid_hht_temporal_verification_test.cpp
     unit/analysis/solid_static_procedure_test.cpp
     unit/analysis/solid_static_state_test.cpp
     unit/analysis/thermal_time_grid_test.cpp

tests/unit/analysis/solid_hht_temporal_verification_test.cpp

@@ -0,0 +1,653 @@
+#include <gtest/gtest.h>
+
+#include <algorithm>
+#include <cmath>
+#include <iostream>
+#include <limits>
+#include <vector>
+
+#include "fesa/analysis/analysis_model.h"
+#include "fesa/analysis/solid_hht_dynamic_procedure.h"
+#include "fesa/assembly/parallel_for.h"
+#include "fesa/elements/hex8_element.h"
+#include "fesa/solvers/linear/mkl_pardiso_solver.h"
+#include "fesa/solvers/linear/mkl_pardiso_tangent_solver.h"
+#include "solid_hht_test_fixture.h"
+
+namespace {
+using M3 = std::array<std::array<double, 3>, 3>;
+struct AffineMotion {
+  M3 b{}, velocity{}, acceleration{}, piola{};
+};
+AffineMotion Motion(double time, bool nonlinear) {
+  const double q = 1 - std::cos(2 * time);
+  const double dq = 2 * std::sin(2 * time);
+  const double ddq = 4 * std::cos(2 * time);
+  AffineMotion m;
+  constexpr double lambda = 4.8;
+  constexpr double mu = 4.8;
+  if (!nonlinear) {
+    const M3 d{{{.015, .008, -.004}, {.006, -.01, .003}, {.005, .004, .012}}};
+    double trace = 0;
+    for (unsigned i = 0; i < 3; ++i) trace += q * d[i][i];
+    for (unsigned i = 0; i < 3; ++i)
+      for (unsigned j = 0; j < 3; ++j) {
+        m.b[i][j] = q * d[i][j];
+        m.velocity[i][j] = dq * d[i][j];
+        m.acceleration[i][j] = ddq * d[i][j];
+        m.piola[i][j] =
+            (i == j ? lambda * trace : 0) + mu * q * (d[i][j] + d[j][i]);
+      }
+  } else {
+    const double s = .004 * q, ds = .004 * dq, dds = .004 * ddq;
+    const double theta = .5 * q, dt = .5 * dq, ddt = .5 * ddq;
+    const double c = std::cos(theta), sn = std::sin(theta);
+    const M3 r{{{c, -sn, 0}, {sn, c, 0}, {0, 0, 1}}};
+    const M3 jr{{{-sn, -c, 0}, {c, -sn, 0}, {0, 0, 0}}};
+    const M3 jjr{{{-c, sn, 0}, {-sn, -c, 0}, {0, 0, 0}}};
+    const double eg = .5 * ((1 + s) * (1 + s) - 1);
+    const double stress = (3 * lambda + 2 * mu) * eg;
+    for (unsigned i = 0; i < 3; ++i)
+      for (unsigned j = 0; j < 3; ++j) {
+        m.b[i][j] = (1 + s) * r[i][j] - (i == j ? 1 : 0);
+        m.velocity[i][j] = ds * r[i][j] + (1 + s) * dt * jr[i][j];
+        m.acceleration[i][j] = dds * r[i][j] + 2 * ds * dt * jr[i][j] +
+                               (1 + s) * (ddt * jr[i][j] + dt * dt * jjr[i][j]);
+        m.piola[i][j] = (1 + s) * r[i][j] * stress;
+      }
+  }
+  return m;
+}
+std::array<double, 3> Apply(const M3& matrix, const std::array<double, 3>& x) {
+  std::array<double, 3> result{};
+  for (unsigned i = 0; i < 3; ++i)
+    for (unsigned j = 0; j < 3; ++j) result[i] += matrix[i][j] * x[j];
+  return result;
+}
+fesa::ModelDefinition TwoBricks(double h, unsigned count, bool nonlinear) {
+  auto d = solid_hht_test::Brick(h, count, nonlinear);
+  const std::array<std::array<double, 3>, 4> added{
+      {{2, 0, 0}, {2, 1, 0}, {2, 0, 1}, {2, 1, 1}}};
+  for (unsigned j = 0; j < 4; ++j) {
+    unsigned n = 8 + j;
+    d.nodes.push_back({{"I", n + 1, std::to_string(n + 1)}, added[j], {}});
+    d.node_sets.push_back({"N" + std::to_string(n), {}, {n}, {}});
+  }
+  d.solid_elements.emplace_back(
+      fesa::SourceEntityId{"I", 9, "9"},
+      std::array<fesa::EntityIndex, 8>{1, 8, 9, 2, 5, 10, 11, 6}, 0, 0,
+      fesa::SourceLocation{});
+  return d;
+}
+// Closed polynomial brick integrals: integral Na*X and integral grad Na.
+// These do not call the production stiffness, mass, quadrature or response.
+std::vector<double> ExactForce(const fesa::ModelDefinition& d, double t,
+                               bool nonlinear) {
+  const auto m = Motion(t, nonlinear);
+  std::vector<double> f(3 * d.nodes.size());
+  for (unsigned e = 0; e < 2; ++e) {
+    const auto& nodes = d.solid_elements[e].node_indices;
+    for (unsigned a = 0; a < 8; ++a) {
+      const auto& x = d.nodes[nodes[a]].coordinates;
+      std::array<double, 3> integral_x{}, gradient{};
+      for (unsigned c = 0; c < 3; ++c) {
+        const double origin = c == 0 ? static_cast<double>(e) : 0;
+        const double local = x[c] - origin;
+        integral_x[c] = 2 * (origin / 8 + (1 + local) / 24);
+        gradient[c] = (2 * local - 1) / 4;
+      }
+      const auto inertia = Apply(m.acceleration, integral_x);
+      const auto internal = Apply(m.piola, gradient);
+      for (unsigned c = 0; c < 3; ++c)
+        f[3 * nodes[a] + c] += inertia[c] + internal[c];
+    }
+  }
+  return f;
+}
+std::vector<double> ExactMassProduct(const fesa::Domain& d,
+                                     const fesa::Vector& v) {
+  std::vector<double> result(v.Size());
+  for (unsigned e = 0; e < 2; ++e) {
+    const auto& nodes = d.SolidElements()[e].node_indices;
+    for (unsigned a = 0; a < 8; ++a)
+      for (unsigned b = 0; b < 8; ++b) {
+        double coefficient = 2.0 / 216;
+        for (unsigned c = 0; c < 3; ++c)
+          coefficient *= d.Nodes()[nodes[a]].coordinates[c] ==
+                                 d.Nodes()[nodes[b]].coordinates[c]
+                             ? 2
+                             : 1;
+        for (unsigned c = 0; c < 3; ++c)
+          result[3 * nodes[a] + c] += coefficient * v[3 * nodes[b] + c];
+      }
+  }
+  return result;
+}
+void AddCross(const std::array<double, 3>& x, const std::array<double, 3>& f,
+              std::array<double, 3>* result) {
+  (*result)[0] += x[1] * f[2] - x[2] * f[1];
+  (*result)[1] += x[2] * f[0] - x[0] * f[2];
+  (*result)[2] += x[0] * f[1] - x[1] * f[0];
+}
+std::array<double, 3> ManufacturedRun(double h, unsigned count,
+                                      bool nonlinear,
+                                      bool compare_parallel = false) {
+  auto d = TwoBricks(h, count, nonlinear);
+  for (unsigned i = 0; i < 3 * d.nodes.size(); ++i) {
+    fesa::TabularAmplitudeDefinition amplitude;
+    amplitude.name_raw = "A" + std::to_string(i);
+    amplitude.name_canonical = amplitude.name_raw;
+    for (unsigned n = 0; n <= count; ++n) {
+      const double t = static_cast<double>(n) * h;
+      amplitude.points.push_back({t, ExactForce(d, t, nonlinear)[i]});
+    }
+    d.amplitudes.push_back(amplitude);
+    d.dynamic_step->loads.push_back({"N" + std::to_string(i / 3),
+                                     static_cast<int>(i % 3 + 1),
+                                     1,
+                                     {},
+                                     i,
+                                     amplitude.name_raw,
+                                     {}});
+  }
+  auto domain = fesa::Domain::Create(std::move(d));
+  EXPECT_TRUE(domain.HasValue());
+  if (!domain.HasValue()) return {1, 1, 1};
+  fesa::SerialParallelFor parallel;
+  fesa::MklPardisoSolver mass;
+  fesa::MklPardisoTangentSolver tangent;
+  auto model = fesa::AnalysisModel::Create(domain.Value());
+  auto dofs = fesa::DofManager::Create(model.Value());
+  auto state =
+      fesa::SolidDynamicAnalysisState::Create(domain.Value(), dofs.Value());
+  auto status = fesa::SolidHhtDynamicProcedureExecutor(parallel, mass, tangent)
+                    .ExecuteInto(domain.Value(), state.Value());
+  EXPECT_TRUE(status.IsOk()) << solid_hht_test::Message(status);
+  if (!status.IsOk()) {
+    const auto& e = state.Value().IterationRecords().back().evidence;
+    std::cout << "Failed force metric " << e.residual_metric << " update "
+              << e.update_metric << '\n';
+    return {1, 1, 1};
+  }
+  if (count == 30 || compare_parallel) {
+    // Both element workers have distinct index-owned buffers; compare every
+    // accepted field, not merely the final displacement norm.
+    auto parallel_state =
+        fesa::SolidDynamicAnalysisState::Create(domain.Value(), dofs.Value());
+    fesa::TbbParallelFor tbb;
+    fesa::MklPardisoSolver parallel_mass;
+    fesa::MklPardisoTangentSolver parallel_tangent;
+    auto parallel_status =
+        fesa::SolidHhtDynamicProcedureExecutor(tbb, parallel_mass,
+                                               parallel_tangent)
+            .ExecuteInto(domain.Value(), parallel_state.Value());
+    EXPECT_TRUE(parallel_status.IsOk())
+        << solid_hht_test::Message(parallel_status);
+    if (parallel_status.IsOk()) {
+      const auto& serial_frames = state.Value().AcceptedFrames();
+      const auto& parallel_frames = parallel_state.Value().AcceptedFrames();
+      EXPECT_EQ(serial_frames.size(), parallel_frames.size());
+      bool has_nonzero_low = false;
+      for (std::size_t n = 0; n < serial_frames.size(); ++n) {
+        const auto& sf = serial_frames[n];
+        const auto& pf = parallel_frames[n];
+        EXPECT_DOUBLE_EQ(sf.kinetic_energy, pf.kinetic_energy);
+        EXPECT_DOUBLE_EQ(sf.strain_energy, pf.strain_energy);
+        EXPECT_DOUBLE_EQ(sf.external_work, pf.external_work);
+        EXPECT_EQ(sf.angular_momentum, pf.angular_momentum);
+        EXPECT_EQ(sf.low.strain_energy, pf.low.strain_energy);
+        EXPECT_EQ(sf.low.kinetic_energy, pf.low.kinetic_energy);
+        EXPECT_EQ(sf.low.external_work, pf.low.external_work);
+        EXPECT_EQ(sf.low.energy_balance_error, pf.low.energy_balance_error);
+        EXPECT_EQ(sf.low.linear_momentum, pf.low.linear_momentum);
+        EXPECT_EQ(sf.low.angular_momentum, pf.low.angular_momentum);
+        EXPECT_EQ(sf.low.linear_momentum_rate, pf.low.linear_momentum_rate);
+        EXPECT_EQ(sf.low.angular_momentum_rate, pf.low.angular_momentum_rate);
+        EXPECT_EQ(sf.low.global_equilibrium, pf.low.global_equilibrium);
+        EXPECT_EQ(sf.low.linearization_current_internal_moment,
+                  pf.low.linearization_current_internal_moment);
+        EXPECT_EQ(sf.low.reference_internal_moment,
+                  pf.low.reference_internal_moment);
+        EXPECT_EQ(sf.low.current_free_residual_moment,
+                  pf.low.current_free_residual_moment);
+        for (auto member : {&fesa::SolidDynamicFrameLowParts::displacement,
+                            &fesa::SolidDynamicFrameLowParts::velocity,
+                            &fesa::SolidDynamicFrameLowParts::acceleration,
+                            &fesa::SolidDynamicFrameLowParts::algorithmic_acceleration,
+                            &fesa::SolidDynamicFrameLowParts::inertia_force,
+                            &fesa::SolidDynamicFrameLowParts::internal_force,
+                            &fesa::SolidDynamicFrameLowParts::external_force,
+                            &fesa::SolidDynamicFrameLowParts::residual,
+                            &fesa::SolidDynamicFrameLowParts::reaction,
+                            &fesa::SolidDynamicFrameLowParts::residual_scale}) {
+          const auto& sl = sf.low.*member;
+          const auto& pl = pf.low.*member;
+          EXPECT_EQ(sl.Size(), pl.Size());
+          if (sl.Size() != pl.Size()) continue;
+          for (std::size_t i = 0; i < sl.Size(); ++i) {
+            EXPECT_EQ(sl[i], pl[i]);
+            has_nonzero_low = has_nonzero_low || sl[i] != 0;
+          }
+        }
+        for (std::size_t i = 0; i < sf.displacement.Size(); ++i) {
+          EXPECT_DOUBLE_EQ(sf.displacement[i], pf.displacement[i]);
+          EXPECT_DOUBLE_EQ(sf.velocity[i], pf.velocity[i]);
+          EXPECT_DOUBLE_EQ(sf.acceleration[i], pf.acceleration[i]);
+          EXPECT_DOUBLE_EQ(sf.algorithmic_acceleration[i],
+                           pf.algorithmic_acceleration[i]);
+          EXPECT_DOUBLE_EQ(sf.residual[i], pf.residual[i]);
+        }
+        for (std::size_t point = 0; point < sf.solid_results.size(); ++point) {
+          EXPECT_EQ(sf.solid_results[point].cauchy_stress,
+                    pf.solid_results[point].cauchy_stress);
+          EXPECT_EQ(sf.solid_results[point].strain,
+                    pf.solid_results[point].strain);
+        }
+      }
+      EXPECT_TRUE(has_nonzero_low);
+    }
+  }
+  std::array<double, 3> errors{}, scales{};
+  const auto exact = Motion(static_cast<double>(count) * h, nonlinear);
+  const auto& final = state.Value().CommittedState();
+  for (unsigned node = 0; node < domain.Value().Nodes().size(); ++node) {
+    const auto u = Apply(exact.b, domain.Value().Nodes()[node].coordinates);
+    const auto v =
+        Apply(exact.velocity, domain.Value().Nodes()[node].coordinates);
+    const auto a =
+        Apply(exact.acceleration, domain.Value().Nodes()[node].coordinates);
+    for (unsigned c = 0; c < 3; ++c) {
+      const std::array<double, 3> truth{u[c], v[c], a[c]};
+      const std::array<double, 3> actual{final.displacement[3 * node + c],
+                                         final.velocity[3 * node + c],
+                                         final.acceleration[3 * node + c]};
+      for (unsigned j = 0; j < 3; ++j) {
+        errors[j] += (actual[j] - truth[j]) * (actual[j] - truth[j]);
+        scales[j] += truth[j] * truth[j];
+      }
+    }
+  }
+  for (unsigned j = 0; j < 3; ++j) errors[j] = std::sqrt(errors[j] / scales[j]);
+  for (const auto& f : state.Value().AcceptedFrames()) {
+    const auto mv = ExactMassProduct(domain.Value(), f.velocity);
+    const auto ma = ExactMassProduct(domain.Value(), f.acceleration);
+    std::array<double, 3> p{}, angular{}, pdot{}, ldot{}, vv{};
+    double kinetic = 0, energy_scale = 0, momentum_scale = 0;
+    double rate_scale = 0, moment_scale = 0, angular_scale = 0, vv_scale = 0;
+    double free_residual_sum = 0;
+    for (unsigned n = 0; n < domain.Value().Nodes().size(); ++n) {
+      auto x = domain.Value().Nodes()[n].coordinates;
+      std::array<double, 3> nmv{}, nma{}, v{};
+      for (unsigned c = 0; c < 3; ++c) {
+        x[c] += f.displacement[3 * n + c];
+        v[c] = f.velocity[3 * n + c];
+        nmv[c] = mv[3 * n + c];
+        nma[c] = ma[3 * n + c];
+        p[c] += nmv[c];
+        pdot[c] += nma[c];
+        kinetic += .5 * v[c] * nmv[c];
+        energy_scale += .5 * std::abs(v[c] * nmv[c]);
+        momentum_scale += std::abs(nmv[c]);
+        rate_scale += std::abs(nma[c]) + std::abs(f.internal_force[3 * n + c]) +
+                      std::abs(f.external_force[3 * n + c]) +
+                      std::abs(f.reaction[3 * n + c]);
+        free_residual_sum +=
+            std::abs(f.residual[3 * n + c] - f.reaction[3 * n + c]);
+      }
+      double arm = 0, velocity_scale = 0, node_mv = 0, node_rate = 0;
+      for (unsigned c = 0; c < 3; ++c) {
+        arm +=
+            std::abs(x[c]) + std::abs(domain.Value().Nodes()[n].coordinates[c]);
+        velocity_scale += std::abs(v[c]);
+        node_mv += std::abs(nmv[c]);
+        node_rate += std::abs(nma[c]) + std::abs(f.internal_force[3 * n + c]) +
+                     std::abs(f.external_force[3 * n + c]) +
+                     std::abs(f.reaction[3 * n + c]);
+      }
+      angular_scale += arm * node_mv;
+      moment_scale += arm * node_rate;
+      vv_scale += velocity_scale * node_mv;
+      AddCross(x, nmv, &angular);
+      AddCross(x, nma, &ldot);
+      AddCross(v, nmv, &vv);
+    }
+    const double roundoff = 512 * std::numeric_limits<double>::epsilon();
+    EXPECT_NEAR(f.kinetic_energy, kinetic, roundoff * energy_scale);
+    for (unsigned c = 0; c < 3; ++c) {
+      EXPECT_NEAR(f.linear_momentum[c], p[c], roundoff * momentum_scale);
+      EXPECT_NEAR(f.angular_momentum[c], angular[c], roundoff * angular_scale);
+      EXPECT_NEAR(f.linear_momentum_rate[c], pdot[c], roundoff * rate_scale);
+      EXPECT_NEAR(f.angular_momentum_rate[c], ldot[c], roundoff * moment_scale);
+      EXPECT_NEAR(vv[c], 0, roundoff * vv_scale);
+      EXPECT_NEAR(f.global_equilibrium[c], 0,
+                  free_residual_sum + roundoff * rate_scale);
+      const double identity = f.reference_internal_moment[c] +
+                              f.linearization_current_internal_moment[c] -
+                              f.current_free_residual_moment[c];
+      EXPECT_NEAR(f.global_equilibrium[3 + c], identity,
+                  roundoff * moment_scale);
+      if (nonlinear)
+        EXPECT_NEAR(f.reference_internal_moment[c] +
+                        f.linearization_current_internal_moment[c],
+                    0, roundoff * moment_scale);
+      else
+        EXPECT_NEAR(f.reference_internal_moment[c], 0, roundoff * moment_scale);
+    }
+  }
+  return errors;
+}
+}  // namespace
+
+TEST(SolidHhtTemporal, PairedSerialAndTbbPreserveEveryLowComponent) {
+  for (bool nonlinear : {false, true}) {
+    SCOPED_TRACE(nonlinear);
+    ManufacturedRun(.02, 2, nonlinear, true);
+  }
+}
+
+TEST(SolidHhtTemporal, FreeConstantForceHasAnalyticalInitialAcceleration) {
+  auto d = solid_hht_test::Brick();
+  d.dynamic_step->boundaries = {{"ALL", 2, 3, 0, {}}};
+  d.dynamic_step->loads = {{"ALL", 1, 0.25, {}}};
+  auto domain = fesa::Domain::Create(std::move(d));
+  ASSERT_TRUE(domain.HasValue());
+  fesa::SerialParallelFor parallel;
+  fesa::MklPardisoSolver mass;
+  fesa::MklPardisoTangentSolver tangent;
+  auto result = fesa::SolidHhtDynamicProcedureExecutor(parallel, mass, tangent)
+                    .Execute(domain.Value());
+  ASSERT_TRUE(result.HasValue()) << solid_hht_test::Message(result.GetStatus());
+  const auto& initial = result.Value().state.AcceptedFrames().front();
+  // Brick mass row sum rho V/8 = 1/4; rigid translation has zero stiffness.
+  for (unsigned node = 0; node < 8; ++node) {
+    EXPECT_NEAR(initial.acceleration[3 * node], 1, 1e-13);
+    EXPECT_DOUBLE_EQ(initial.displacement[3 * node], 0);
+  }
+}
+
+TEST(SolidHhtTemporal,
+     AllFreeUniformForceRetainsExactRigidTranslationBehavior) {
+  auto d = solid_hht_test::Brick();
+  d.dynamic_step->loads = {{"ALL", 1, .25, {}}};
+  auto domain = fesa::Domain::Create(std::move(d));
+  ASSERT_TRUE(domain.HasValue());
+  fesa::SerialParallelFor parallel;
+  fesa::MklPardisoSolver mass;
+  fesa::MklPardisoTangentSolver tangent;
+  auto model = fesa::AnalysisModel::Create(domain.Value());
+  auto dofs = fesa::DofManager::Create(model.Value());
+  auto state =
+      fesa::SolidDynamicAnalysisState::Create(domain.Value(), dofs.Value());
+  auto status = fesa::SolidHhtDynamicProcedureExecutor(parallel, mass, tangent)
+                    .ExecuteInto(domain.Value(), state.Value());
+  if (!status.IsOk()) {
+    const auto& e = state.Value().IterationRecords().back().evidence;
+    std::cout << "Rigid-force failure worst DOF " << e.worst_full_dof
+              << " force ratio " << e.residual_metric
+              << " terms r,Ma,f,F,scale ";
+    for (auto term : e.worst_force_terms) std::cout << term << ' ';
+    std::cout << '\n';
+    for (const auto& rec : state.Value().IterationRecords()) {
+      if (rec.evidence.iteration_index < 10) {
+        std::cout << "Trace " << rec.increment_index << ' '
+                  << rec.evidence.iteration_index << ' ' << rec.outcome
+                  << " ratio=" << rec.evidence.residual_metric
+                  << " worst=" << rec.evidence.worst_full_dof
+                  << " correction=" << rec.evidence.correction_metrics[0]
+                  << '\n';
+      }
+    }
+  }
+  ASSERT_TRUE(status.IsOk()) << solid_hht_test::Message(status);
+  EXPECT_TRUE(std::any_of(
+      state.Value().IterationRecords().begin(),
+      state.Value().IterationRecords().end(),
+      [](const auto& record) { return record.outcome == "paired-candidate"; }));
+  const double roundoff = 512 * std::numeric_limits<double>::epsilon();
+  for (const auto& f : state.Value().AcceptedFrames())
+    for (unsigned n = 0; n < 8; ++n) {
+      EXPECT_NEAR(f.displacement[3 * n], .5 * f.time * f.time,
+                  roundoff * .5 * f.time * f.time);
+      EXPECT_NEAR(f.velocity[3 * n], f.time, roundoff * f.time);
+      EXPECT_NEAR(f.acceleration[3 * n], 1, roundoff);
+      for (unsigned c = 1; c < 3; ++c) {
+        EXPECT_NEAR(f.displacement[3 * n + c], 0,
+                    roundoff * .5 * f.time * f.time);
+        EXPECT_NEAR(f.velocity[3 * n + c], 0, roundoff * f.time);
+        EXPECT_NEAR(f.acceleration[3 * n + c], 0, roundoff);
+      }
+    }
+}
+
+TEST(SolidHhtTemporal,
+     PairedNewtonKeepsAllOriginalComponentsAcrossAxesAndScales) {
+  const double roundoff = 512 * std::numeric_limits<double>::epsilon();
+  for (unsigned axis = 0; axis < 3; ++axis)
+    for (double acceleration : {1e-4, 1.0, 1000.0}) {
+      SCOPED_TRACE(axis);
+      SCOPED_TRACE(acceleration);
+      auto d = solid_hht_test::Brick(.01, 8);
+      d.dynamic_step->loads = {
+          {"ALL", static_cast<int>(axis + 1), .25 * acceleration, {}}};
+      auto domain = fesa::Domain::Create(std::move(d));
+      ASSERT_TRUE(domain.HasValue());
+      fesa::SerialParallelFor parallel;
+      fesa::MklPardisoSolver mass;
+      fesa::MklPardisoTangentSolver tangent;
+      auto model = fesa::AnalysisModel::Create(domain.Value());
+      ASSERT_TRUE(model.HasValue());
+      auto dofs = fesa::DofManager::Create(model.Value());
+      ASSERT_TRUE(dofs.HasValue());
+      auto state =
+          fesa::SolidDynamicAnalysisState::Create(domain.Value(), dofs.Value());
+      ASSERT_TRUE(state.HasValue());
+      const auto status =
+          fesa::SolidHhtDynamicProcedureExecutor(parallel, mass, tangent)
+              .ExecuteInto(domain.Value(), state.Value());
+      if (!status.IsOk()) {
+        const auto& records = state.Value().IterationRecords();
+        for (const auto& record : records) {
+          if (record.increment_index != records.back().increment_index ||
+              (record.evidence.iteration_index > 8 &&
+               record.evidence.iteration_index < 95))
+            continue;
+          std::cout << "Rigid paired Newton trace "
+                    << record.evidence.iteration_index << ' ' << record.outcome
+                    << " metric=" << record.evidence.residual_metric
+                    << " worst=" << record.evidence.worst_full_dof << " terms=";
+          for (double term : record.evidence.worst_force_terms)
+            std::cout << term << ' ';
+          std::cout << " corrections=";
+          for (double metric : record.evidence.correction_metrics)
+            std::cout << metric << ' ';
+          std::cout << '\n';
+        }
+      }
+      ASSERT_TRUE(status.IsOk()) << solid_hht_test::Message(status);
+      for (const auto& frame : state.Value().AcceptedFrames()) {
+        EXPECT_LE(frame.iterations.back().residual_metric, 1e-10);
+        EXPECT_LE(frame.iterations.back().update_metric, 1e-12);
+        for (unsigned node = 0; node < 8; ++node)
+          for (unsigned c = 0; c < 3; ++c) {
+            const double a = c == axis ? acceleration : 0;
+            EXPECT_NEAR(frame.displacement[3 * node + c],
+                        .5 * a * frame.time * frame.time,
+                        roundoff * .5 * acceleration * frame.time * frame.time);
+            EXPECT_NEAR(frame.velocity[3 * node + c], a * frame.time,
+                        roundoff * acceleration * frame.time);
+            EXPECT_NEAR(frame.acceleration[3 * node + c], a,
+                        roundoff * acceleration);
+          }
+      }
+    }
+}
+
+TEST(SolidHhtTemporal,
+     ActualTwoBrickLinearAndRotatingSmallStrainTlAreSecondOrder) {
+  for (bool nonlinear : {false, true}) {
+    SCOPED_TRACE(nonlinear);
+    const auto coarse = ManufacturedRun(.04, 30, nonlinear);
+    const auto medium = ManufacturedRun(.02, 60, nonlinear);
+    const auto fine = ManufacturedRun(.01, 120, nonlinear);
+    for (unsigned j = 0; j < 3; ++j) {
+      const double p1 = std::log(coarse[j] / medium[j]) / std::log(2.0);
+      const double p2 = std::log(medium[j] / fine[j]) / std::log(2.0);
+      std::cout << "Hex8 " << (nonlinear ? "TL" : "linear") << " u/v/a channel "
+                << j << " errors " << coarse[j] << ' ' << medium[j] << ' '
+                << fine[j] << " orders " << p1 << ' ' << p2 << '\n';
+      EXPECT_GT(fine[j],
+                1e-8);  // Meaningful truncation interval, away from noise.
+      EXPECT_LT(coarse[j], .1);
+      EXPECT_GE(p1, 1.8);
+      EXPECT_GE(p2, 1.8);
+    }
+  }
+}
+
+TEST(SolidHhtTemporal, IndependentThreeDofBrickOscillatorHasSecondOrderUVA) {
+  // Node0's three translations are free. Exact polynomial integrals yield
+  // M=(rho/27)I, Kii=(lambda+4mu)/9, Kij=(lambda+mu)/12.
+  std::array<std::array<double, 3>, 3> errors{};
+  for (unsigned level = 0; level < 3; ++level) {
+    const double h = .02 / static_cast<double>(1U << level);
+    const unsigned count = 32U << level;
+    auto d = solid_hht_test::Brick(h, count);
+    for (unsigned n = 1; n < 8; ++n)
+      d.dynamic_step->boundaries.push_back(
+          {"N" + std::to_string(n), 1, 3, 0, {}});
+    d.dynamic_step->loads = {
+        {"N0", 1, .01, {}}, {"N0", 2, .02, {}}, {"N0", 3, .03, {}}};
+    auto domain = fesa::Domain::Create(std::move(d));
+    ASSERT_TRUE(domain.HasValue());
+    fesa::SerialParallelFor parallel;
+    fesa::MklPardisoSolver mass;
+    fesa::MklPardisoTangentSolver tangent;
+    auto result =
+        fesa::SolidHhtDynamicProcedureExecutor(parallel, mass, tangent)
+            .Execute(domain.Value());
+    ASSERT_TRUE(result.HasValue())
+        << solid_hht_test::Message(result.GetStatus());
+    const auto& f = result.Value().state.CommittedState();
+    const double m = 2.0 / 27;
+    const double diagonal = (4.8 + 4 * 4.8) / 9;
+    const double off = (4.8 + 4.8) / 12;
+    const std::array<double, 2> k{diagonal + 2 * off, diagonal - off};
+    for (unsigned c = 0; c < 3; ++c) {
+      std::array<double, 3> truth{};
+      const std::array<double, 2> force{.02, .01 * (c + 1) - .02};
+      for (unsigned mode = 0; mode < 2; ++mode) {
+        const double w = std::sqrt(k[mode] / m);
+        truth[0] += force[mode] / k[mode] * (1 - std::cos(w * f.time));
+        truth[1] += force[mode] / (m * w) * std::sin(w * f.time);
+        truth[2] += force[mode] / m * std::cos(w * f.time);
+      }
+      const std::array<double, 3> actual{f.displacement[c], f.velocity[c],
+                                         f.acceleration[c]};
+      for (unsigned j = 0; j < 3; ++j)
+        errors[level][j] += std::pow(actual[j] - truth[j], 2);
+    }
+  }
+  for (unsigned j = 0; j < 3; ++j) {
+    const double p1 =
+        .5 * std::log(errors[0][j] / errors[1][j]) / std::log(2.0);
+    const double p2 =
+        .5 * std::log(errors[1][j] / errors[2][j]) / std::log(2.0);
+    std::cout << "Independent MDOF channel " << j << " orders " << p1 << ' '
+              << p2 << '\n';
+    EXPECT_GE(p1, 1.8);
+    EXPECT_GE(p2, 1.8);
+    EXPECT_GT(errors[2][j], 1e-15);
+  }
+}
+
+TEST(SolidHhtTemporal, IndependentScalarBrickOscillatorHasSecondOrderUVA) {
+  std::array<std::array<double, 3>, 3> errors{};
+  const double m = 2.0 / 27;
+  const double k = (4.8 + 4 * 4.8) / 9;
+  const double w = std::sqrt(k / m);
+  for (unsigned level = 0; level < 3; ++level) {
+    const double h = .02 / static_cast<double>(1U << level);
+    auto d = solid_hht_test::Brick(h, 32U << level);
+    d.dynamic_step->boundaries.push_back({"N0", 2, 3, 0, {}});
+    for (unsigned n = 1; n < 8; ++n)
+      d.dynamic_step->boundaries.push_back(
+          {"N" + std::to_string(n), 1, 3, 0, {}});
+    d.dynamic_step->loads = {{"N0", 1, .01, {}}};
+    auto domain = fesa::Domain::Create(std::move(d));
+    ASSERT_TRUE(domain.HasValue());
+    fesa::SerialParallelFor parallel;
+    fesa::MklPardisoSolver mass;
+    fesa::MklPardisoTangentSolver tangent;
+    auto result =
+        fesa::SolidHhtDynamicProcedureExecutor(parallel, mass, tangent)
+            .Execute(domain.Value());
+    ASSERT_TRUE(result.HasValue())
+        << solid_hht_test::Message(result.GetStatus());
+    const auto& f = result.Value().state.CommittedState();
+    const std::array<double, 3> truth{.01 / k * (1 - std::cos(w * f.time)),
+                                      .01 / (m * w) * std::sin(w * f.time),
+                                      .01 / m * std::cos(w * f.time)};
+    const std::array<double, 3> actual{f.displacement[0], f.velocity[0],
+                                       f.acceleration[0]};
+    for (unsigned j = 0; j < 3; ++j)
+      errors[level][j] = std::abs(actual[j] - truth[j]);
+    EXPECT_NEAR(
+        f.kinetic_energy, .5 * m * f.velocity[0] * f.velocity[0],
+        512 * std::numeric_limits<double>::epsilon() * f.kinetic_energy);
+    EXPECT_NEAR(f.strain_energy, .5 * k * f.displacement[0] * f.displacement[0],
+                512 * std::numeric_limits<double>::epsilon() * f.strain_energy);
+  }
+  for (unsigned j = 0; j < 3; ++j) {
+    const double p1 = std::log(errors[0][j] / errors[1][j]) / std::log(2.0);
+    const double p2 = std::log(errors[1][j] / errors[2][j]) / std::log(2.0);
+    std::cout << "Independent SDOF channel " << j << " orders " << p1 << ' '
+              << p2 << '\n';
+    EXPECT_GE(p1, 1.8);
+    EXPECT_GE(p2, 1.8);
+    EXPECT_GT(errors[2][j], 1e-8);
+  }
+}
+
+TEST(SolidHhtTemporal,
+     BendingMeshRefinementQuantifiesFullIntegrationShearLocking) {
+  constexpr double thickness = .125;
+  constexpr double curvature = .01;
+  constexpr double lambda = 4.8;
+  constexpr double mu = 4.8;
+  const double continuum = .5 * (lambda + 2 * mu) * curvature * curvature *
+                           thickness * thickness * thickness / 12;
+  double previous_error = 0;
+  for (unsigned nx : {1U, 2U, 4U, 8U}) {
+    double numerical = 0;
+    for (unsigned e = 0; e < nx; ++e) {
+      auto d = solid_hht_test::Brick();
+      fesa::Hex8Element::Coordinates x{};
+      fesa::ElementDofLayout layout{
+          {"I", e + 1, std::to_string(e + 1)},
+          {0, 1, 2, 3, 4, 5, 6, 7},
+          {fesa::DofComponent::kUx, fesa::DofComponent::kUy,
+           fesa::DofComponent::kUz}};
+      fesa::Vector u(24);
+      for (unsigned a = 0; a < 8; ++a) {
+        x[a] = {static_cast<double>(e) / nx + d.nodes[a].coordinates[0] / nx,
+                thickness * (d.nodes[a].coordinates[1] - .5),
+                d.nodes[a].coordinates[2]};
+        u[3 * a] = -curvature * x[a][0] * x[a][1];
+        u[3 * a + 1] = .5 * curvature * x[a][0] * x[a][0];
+      }
+      auto element = fesa::Hex8Element::Create(e, layout, x, d.materials[0], 2);
+      ASSERT_TRUE(element.HasValue());
+      auto response = element.Value().EvaluateLinear(u);
+      ASSERT_TRUE(response.HasValue());
+      numerical += response.Value().strain_energy;
+    }
+    const double error = numerical / continuum - 1;
+    const double exact_parasitic =
+        mu / (lambda + 2 * mu) / std::pow(nx * thickness, 2);
+    EXPECT_NEAR(error, exact_parasitic, 2e-11);
+    if (previous_error != 0) EXPECT_NEAR(previous_error / error, 4, 1e-10);
+    std::cout << "Bending nx=" << nx
+              << " energy/continuum=" << numerical / continuum << '\n';
+    previous_error = error;
+  }
+}

tests/unit/analysis/solid_hht_test_fixture.h

@@ -0,0 +1,52 @@
+#ifndef FESA_TESTS_SOLID_HHT_TEST_FIXTURE_H_
+#define FESA_TESTS_SOLID_HHT_TEST_FIXTURE_H_
+
+#include <array>
+#include <string>
+
+#include "fesa/elements/hex8_definition.h"
+#include "fesa/model/domain.h"
+#include "fesa/properties/solid_section.h"
+
+namespace solid_hht_test {
+inline fesa::ModelDefinition Brick(double h = 0.01, unsigned count = 20,
+                                   bool nonlinear = false) {
+  fesa::ModelDefinition d;
+  d.source_path = "solid-hht-test.inp";
+  const std::array<std::array<double, 3>, 8> x{{{0, 0, 0},
+                                                {1, 0, 0},
+                                                {1, 1, 0},
+                                                {0, 1, 0},
+                                                {0, 0, 1},
+                                                {1, 0, 1},
+                                                {1, 1, 1},
+                                                {0, 1, 1}}};
+  for (unsigned i = 0; i < 8; ++i) {
+    d.nodes.push_back({{"I", i + 1, std::to_string(i + 1)}, x[i], {}});
+    d.node_sets.push_back({"N" + std::to_string(i), {}, {i}, {}});
+  }
+  d.node_sets.push_back({"ALL", {}, {0, 1, 2, 3, 4, 5, 6, 7}, {}});
+  d.materials = {{"M", 12.0, 0.25, {}}};
+  d.densities = {{0, 2.0, {}}};
+  d.solid_sections.emplace_back(fesa::SourceEntityId{"I", 1, "1"}, "E", 0,
+                                fesa::SourceLocation{});
+  d.solid_elements.emplace_back(
+      fesa::SourceEntityId{"I", 8, "8"},
+      std::array<fesa::EntityIndex, 8>{0, 1, 2, 3, 4, 5, 6, 7}, 0, 0,
+      fesa::SourceLocation{});
+  fesa::DynamicStepDefinition step;
+  step.source_step_name_raw = "Dynamic";
+  step.source_step_name_canonical = "DYNAMIC";
+  step.initial_increment = h;
+  step.time_period = static_cast<double>(count) * h;
+  step.increment_count = count;
+  step.maximum_increments = count;
+  step.nlgeom = nonlinear;
+  d.dynamic_step = step;
+  return d;
+}
+inline std::string Message(const fesa::Status& s) {
+  return s.Diagnostics().empty() ? "" : s.Diagnostics()[0].message;
+}
+}  // namespace solid_hht_test
+#endif